The National One Week Prevalence Audit of Universal Meticillin-Resistant Staphylococcus aureus (MRSA) Admission Screening 2012

Introduction The English Department of Health introduced universal MRSA screening of admissions to English hospitals in 2010. It commissioned a national audit to review implementation, impact on patient management, admission prevalence and extra yield of MRSA identified compared to “high-risk” specialty or “checklist-activated” screening (CLAS) of patients with MRSA risk factors. Methods National audit May 2011. Questionnaires to infection control teams in all English NHS acute trusts, requesting number patients admitted and screened, new or previously known MRSA; MRSA point prevalence; screening and isolation policies; individual risk factors and patient management for all new MRSA patients and random sample of negatives. Results 144/167 (86.2%) trusts responded. Individual patient data for 760 new MRSA patients and 951 negatives. 61% of emergency admissions (median 67.3%), 81% (median 59.4%) electives and 47% (median 41.4%) day-cases were screened. MRSA admission prevalence: 1% (median 0.9%) emergencies, 0.6% (median 0.4%) electives, 0.4% (median 0%) day-cases. Approximately 50% all MRSA identified was new. Inpatient MRSA point prevalence: 3.3% (median 2.9%). 104 (77%) trusts pre-emptively isolated patients with previous MRSA, 63 (35%) pre-emptively isolated admissions to “high-risk” specialties; 7 (5%) used PCR routinely. Mean time to MRSA positive result: 2.87 days (±1.33); 37% (219/596) newly identified MRSA patients discharged before result available; 55% remainder (205/376) isolated post-result. In an average trust, CLAS would reduce screening by 50%, identifying 81% of all MRSA. “High risk” specialty screening would reduce screening by 89%, identifying 9% of MRSA. Conclusions Implementation of universal screening was poor. Admission prevalence (new cases) was low. CLAS reduced screening effort for minor decreases in identification, but implementation may prove difficult. Cost effectiveness of this and other policies, awaits evaluation by transmission dynamic economic modelling, using data from this audit. Until then trusts should seek to improve implementation of current policy and use of isolation facilities.

File S1. The NOW Questionnaire admissions. This policy was introduced in April 2009 for electives and in December 2010 for emergencies. As models were populated by limited literature-based data, possibly unrepresentative of NHS hospitals overall, the impact assessment committed the DH to reviewing the implementation of this policy, its impact on patients and their management, its effectiveness and cost-effectiveness.

AIMS:
To report on: (1) implementation of mandatory screening in acute, teaching & specialist trusts another) were calculated. Strategies costing less than £30,000 (€35,500) per QALY gained were defined as cost effective (£30,000 being the upper limit of the usual National Health Service "willingness to pay" for health benefits).
(£9,964/QALY), and was also cost-effective in specialist trusts (£10,777/QALY). This reflects the fact that most infections (which carry the largest determinants of cost and health benefits) occur in this patient group. In teaching trusts, high-risk specialty screening was only slightly more costly than the £30,000 willingness to pay threshold, with an ICER of £31,077/QALY. The differences in costeffectiveness between acute and teaching trusts reflected the overall greater costs per admission in teaching trusts (causing any strategy to have a higher cost/QALY). In specialist trusts, screening all admissions with pre-emptive isolation of those known to be previously MRSA positive was costeffective (£14,324/QALY).
In acute and specialist trusts high-risk specialty screening remained cost-effective at both high and low MRSA admission prevalence levels and became a cost-effective strategy in teaching trusts at a high prevalence (£20,715/QALY). These results therefore indicate that the cost-effectiveness of high risk specialty screening would be robust if MRSA prevalence were to rise again, to twice the current prevalence level. At lower MRSA prevalence, screening of high-risk specialties combined with checklistactivated screening of all other admissions became cost-effective in acute trusts (£23,196/QALY) as did screening all admissions in specialist trusts (£14,224/QALY). No screening strategy was deemed optimal (cost-effective) at lower prevalence in teaching trusts. At current prevalence levels none of the decisions changed for any Trust type if the willingness to pay for a QALY gained was reduced to £20,000 However, for all scenarios and settings there was substantial uncertainty (due to inherent uncertainty in the effectiveness of isolation and decolonisation), with the probability of any one strategy being the most cost effective not exceeding 30% (except at higher prevalence).

Cost per QALY gained is shown for strategies considered both cost-effective † and non-cost-effective.
Any remaining strategies for each prevalence scenario were dominated † † . Strategy 2 (£14,224/QALY) - †An ICER of less than £30,000 per QALY is considered cost-effective. An ICER of more than £30,000 is not considered cost-effective. £30,000 is the upper limit of the usual NHS willingness to pay range. † †Dominated strategies are those that are more costly and provide less benefit than one other strategy or a combination of two other strategies. Since it can never be cost-effective to pay more for less benefit, ICERs were not calculated for these strategies.

CORE REPORT INTRODUCTION:
Following government and public concern at reported high levels of meticillin resistant Staphylococcus aureus over 2001-2004[HPA, 2004  Technology Assessment (Ritchie et al 2007), suggested that, although there was little to choose between long term prevalence levels achieved by different screening strategies, the actual costs varied substantially. A Department of Health (England) impact assessment (appendix 1) using modelling suggested that the most cost-effective strategy would be mandatory screening of all appropriate elective admissions (except for paediatric, maternity and same day case admissions) and all appropriate emergency admissions. This policy was introduced in April 2009 for elective admissions and in December 2010 for emergencies. However, as these modelling studies were populated by limited data from the literature, which were unrepresentative of NHS hospitals overall, the impact assessment committed the Department to reviewing the implementation, impact on patients and their management, and the effectiveness and cost-effectiveness of this policy.
The current study was, therefore, commissioned to perform a national prevalence audit of MRSA screening providing representative clinical and economic data which would be used to populate an individual-based transmission dynamic model evaluating different screening strategies. It was also able to incorporate findings from the Scottish Pathfinder study (Health Protection Scotland 2011 a, b, c, d), which published a prevalence model in 2011, evaluating different screening strategies, using data from three trusts that had been screening all admissions for a year. They explicitly recommended that future models should be individual-based to develop more complex, powerful tools for predictive modelling to inform decisions on costed management of healthcare associated infection. This we have done.
STUDY AIMS: There were six objectives:-Objective 1. To report how widely the policy of screening Emergency and Elective admissions was implemented across the NHS in the three different trust types (acute, teaching , specialist) Objective 2: To report the prevalence of MRSA carriage on admission in Emergency and Elective admissions in different types of trust and the proportion of carriage that was previously unknown.
Objective 3: To report on screening, isolation and decolonisation policies and practices, laboratory methods and costs Objective 4: To report on how patients were managed, how soon results were available, how many patients were isolated and / or decolonised pre-emptively and after the result was known, and how many were treated for infection. For the NHS as a whole, checklist activated screening would reduce the numbers of screens from 790 to 398 a week per trust, whilst identifying 9 out of an average of 11 MRSA positive admissions (new and previously known). In other words, halving the number of screens, whilst detecting just over 80% of MRSA. This did not vary significantly between trust types.
Screening of all admissions to high-risk specialties plus checklist activated screening of all other admissions, would yield similar results. Just over half (55%) of the admitted population would be screened (448 screens/week/trust). This varied little between the three trust types, except that the detection rate was 90% in acute trusts.
In contrast, if only high risk specialties were screened, there would be a 90% reduction in screening (to 87/week/trust), but only 9% of MRSA positives would be detected. Although acute trusts followed this pattern, detection rates were higher in teaching (16%) and specialist trusts (24%), as the proportions of high-risk specialty patients were higher. Reductions in screening were less marked (80% reduction in teaching and 70% reduction in specialist).

ACUTE TRUSTS (Appendix 7a):
At current MRSA admission prevalence (1.4%), only strategy 3 (screening admissions to "high-risk" specialties) was cost effective, at £9,964/QALY. Strategy 5 (screening admissions to high risk specialties plus checklist activated screening for low risk specialties) was slightly too costly and above the £30,000 willingness to pay threshold (£33,806/QALY) (table 1).
Results were very similar using the Scottish Pathfinder checklist to determine checklist activated screening (strategy 3 costing £9,731/QALY and strategy 5 (£33,206/QALY) (table 2). There was substantial uncertainty, however, with the probability that any one strategy was the most costeffective within a willingness to pay range of £20,000-30,000/QALY clustered together, and not exceeding 30%.
In higher prevalence settings (2.8%), strategy 3 became better value for money at £8,650/QALY (table   1). Although the probability that it was the most cost-effective option was approximately 50% at lower willingness to pay values (around £10,000/QALY), there was substantial uncertainty within the usual NHS range of £20,000-£30,000/QALY.
However, a move to strategy 5 from 3 also came below the threshold at £23,196/QALY. There was substantial uncertainty as above.
In a setting of reduced transmission in high risk specialties, strategy 3 was the most cost-effective option and, despite having a greater cost per QALY (at £26,551/QALY), it remained optimal in a setting of reduced death rates (table 2).
For acute trusts, compared to universal screening, restricting screening to high-risk specialty patients would reduce total annual costs by an average of £1,592,000 per acute trust, (£619 per admission), at the expense of 56.5 more transmission events (i.e. one extra colonisation per week per trust) and 1.7 more infections per year. At higher prevalence levels this equated to savings of £1,766,148 (£670 per admission), 63.5 more colonisations (1.2 extra per trust/week) and 2.9 more infections per year.

TEACHING TRUSTS (Appendix 7b):
At current MRSA admission prevalence (1.3%), strategy 3 (screening admissions to "high-risk" specialties only) was not cost-effective, but was only just above the £30,000 willingness to pay threshold at £31,077/QALY (table 1). However, at higher prevalence levels (2.6%), this strategy became cost-effective at £20,715/QALY (table 1), with a probability of cost-effectiveness exceeding 40%. At low prevalence levels (0.65%), no screening strategy was considered cost-effective, with strategy 1 (no screening) and isolation only of clinical cases being optimal (table 1).
For teaching trusts, restricting screening to high-risk specialty patients would reduce total annual costs per trust by £1,864,000 (£576 per admission), at the price of 47 more colonisations (less than one extra/week/ trust) and 1.2 more infections per year.

SPECIALIST TRUSTS (Appendix 7c):
At a current admission prevalence of 1.04%, strategies 3 and 6 (screening all admissions) plus preemptive isolation of those known to be previously MRSA positive) were cost-effective at £10,077/QALY and £14,324/QALY respectively (table 1). There was substantial uncertainty, with the probabilities that any one strategy was the most cost-effective clustering at less than 30%. At higher prevalence levels (2.1%), strategy 3 remained cost-effective at £9,745/QALY (table 1) and was the optimal option with a 40% probability of cost-effectiveness. In low prevalence settings (0.5%), both strategies 3 and 2 (screening all patients) were cost-effective at £10,566/QALY and £14,224/QALY respectively (table 1), although there is substantial uncertainty with almost all screening strategies having an approximately 25% probability of being cost-effective.
For specialist trusts, restricting screening to high-risk specialty patients would reduce total annual costs per trust by £438,000 (£600 per admission), at the price of 18.3 more colonisations (one extra/every three weeks/ trust) and 2.6 more infections per year.  - †An ICER of less than £30,000 per QALY is considered cost-effective. An ICER of more than £30,000 is not considered cost-effective. £30,000 is the upper limit of the usual NHS willingness to pay range. † †Dominated strategies are those that are more costly and provide less benefit than one other strategy or a combination of two other strategies. Since it can never be cost-effective to pay more for less benefit, ICERs were not calculated for these strategies.

Table 2. Cost-effectiveness of screening strategies in alternative scenarios (simulations performed
for an Acute Trust setting). Cost per QALY gained is shown for strategies considered both costeffective † and non-cost-effective. Any remaining strategies for each prevalence scenario were dominated † † .

Scenario
Cost-effective strategies Non-cost-effective strategies Lower transmission in high-risk specialties* Strategy 3(£12,382/QALY) Strategy 6 (£85,713/QALY) Reduced probability of death in high-risk specialties** Strategy 3 (£26,511/QALY) Strategy 6 (£296,859/QALY) Use of the Scottish Pathfinder checklist to identify those at risk of MRSA carriage*** Strategy 3 (£9,731/QALY) Strategy 5 (£33,206/QALY) Strategy 6 (£87,517/QALY) †An ICER of less than £30,000 per QALY is considered cost-effective, An ICER of more than £30,000 is not considered cost-effective. £30,000 is the upper limit of the usual NHS willingness to pay range. † †Dominated strategies are those that are more costly and provide less benefit than one other strategy or a combination of two other strategies. Since it can never be cost-effective to pay more for less benefit, ICERs were not calculated for these strategies.

Summary of findings:
There was an excellent response to the study with 86% of all trusts providing data needed to inform the modelling. It also yielded seven important findings.
Secondly, admission prevalence of MRSA was low at 1.5% (overall), 2.1% (Emergency), 0.9% (Electives) and 0.7% (day cases), with approximately half being newly identified MRSA. This meant that numbers needed to screen in order to identify one new positive were high, especially for Elective (180) and day case (186) admissions, but were lower (80) for Electives in specialist trusts.
Thirdly, over three quarters of trusts pre-emptively isolated those with previous MRSA and nearly half pre-emptively isolated high-risk specialty patients.
Fourthly, a third of MRSA positive and negative patients alike were discharged before the results of screening were available, as the mean turn round time was 2.87 days for positives and 1.75 days for negatives. Decolonisation was started in three-quarters of all patients screened and found to be newly MRSA positive. Just over half of new positives who were still in patients were isolated. Very few were using PCR test methodologies that could potentially produce faster results.
Fifthly, the point prevalence survey showed that, although the overall prevalence of MRSA was 3.3% (3.6% acute, 3% teaching and 2% specialist), a third of MRSA patients were not isolated. In specialist trusts nearly all MRSA positive patients were isolated.
Sixth, for the NHS as a whole, checklist activated screening would detect 80% of the MRSA positive patients detected by routine screening (an average of 2 less/week/trust) and would halve the number of screens required. Screening high-risk specialty admissions would detect only 10% of MRSA positive patients (although this proportion would be higher in teaching and specialist trusts) whilst reducing screening by 90%.
Finally, the cost effectiveness of different screening strategies, evaluated at a willingness to pay threshold of £30,000 per QALY gained, showed screening admissions to "high-risk" specialties performed best overall in different scenarios and settings. In the base case scenario none of the decisions changed for any Trust type if the willingness to pay for a QALY gained was reduced to £20,000. However, there was substantial uncertainty, with the probability that any one strategy was the most cost-effective generally not exceeding approximately 30%. This low probability of costeffectiveness associated with each of the screening strategies is due to the inherent uncertainty in the effectiveness of the accompanying intervention (isolation and decolonisation). If we had assumed that identifying MRSA positive patients led to placing them under an intervention that was 100% effective and prevented any further colonisations or infections, this uncertainty would be reduced. However, our estimates of the effectiveness of isolation and decolonisation represent the best available evidence, and we considered it to be important for decision makers to visualise this uncertainty within the decision making process.
At current admission MRSA prevalence levels, moving from the current strategy of routine screening to targeted screening of high-risk specialty patients would reduce total average annual costs by £1,592,000 per acute trust, £1,864,000 per teaching and £438,000 per specialist trust. This would be at the expense of one extra colonisation per week per trust and less than two extra infections per year in acute and teaching trusts. For specialist trusts there would be even fewer colonisations but slightly more infections (an extra 2.6 per trust per year).

Strengths
The first strength of the study was the high response rates from trusts. This enabled audit data representative of current NHS settings to populate the model. The availability of good quality economic data from the Pathfinder (Health Protection Scotland 2011b) and MECAMIP studies (Robotham et al 2011) also contributed to the model and ensured that it was relevant to, and representative of, current NHS settings, practice and prevalence levels.
The use of a powerful sophisticated individual based model, as suggested by the Scottish Pathfinder study as the next appropriate direction for modelling studies was a particular strength. The incorporation of dynamic transmission models, with robust estimates of transmission, into economic evaluation enabled consideration of population-level effects of screening strategies, which benefited not only the person screened but other patients. This avoided underestimating the effects of screening.
Use of a stochastic model, with 1000 simulations per parameter set, for each strategy, minimises the uncertainty due to chance effects, which were dominant in the small populations seen in NHS trusts.
Other strengths included simulation of the time to result delays, and the capture of more realistic patient movements between specialities and between hospital and community. Long-term effects were considered by adjusting the quality adjusted life expectancy of infected patients with long-term detriments to health. The additional length of stay and additional risk of mortality due to infection were dominant economic parameters, and the way in which these were modelled, specifically by determining each patient's daily probability of discharge and death, adjusted according to their infection status make this model more relevant to the real world.

Limitations
There were the assumptions that both MRSA transmission and probability of death in high-risk specialties were equivalent to that in Intensive Care Units (ICUs), and that in low risk specialties they were equivalent to general medical wards. However, this was adjusted within sensitivity analyses, assuming that transmission in high-risk settings was intermediate between ICUs and general medical wards and that probabilities of death for the whole patient population were equivalent to those in general medical wards. Results were robust to these parameter changes.
The second main limitation, was that transmission was modelled at a specialty level, with homogenous mixing assumed within specialties, which might have especially affected large teaching trusts. A greater level of 'granularity' i.e. including a ward-based structure, and inclusion of ICUs in particular, would enhance the model. Further modelling, incorporating data from the sentinel sites on inter-ward transfers and readmission rates of MRSA patients would lead to more realistic transmission dynamics, and thus more reliable evaluation of screening strategies. Given that screening of admissions to highrisk specialties was only just above the willingness to pay threshold for teaching trusts, and that in low prevalence teaching hospital settings, no strategy appeared to be cost effective, it will be important to re-run the models, considering transmission at a more detailed ward level. In addition, modelling of the effects of excluding day-case or elective screening and of the effects of pre-emptive isolation of all patients with a history of MRSA (which audit data showed that most trusts attempt) would also have enhanced the applicability of the model to the current NHS. Other potential limitations were assumptions that isolation had no adverse effects and that there was no increased mupirocin resistance due to its widespread use in decolonisation.

Comparison with other modelling studies:
The most important comparisons are with the Department of Health impact assessment (appendix 1) the Scottish HTA model (Ritchie et al 2007) and the subsequent Scottish Pathfinder study (Health Protection Scotland 2011b). Non-UK or Republic of Ireland modelling (Hubben et al 2012, Murthy et al 2010, Lee et al 2009 or clinical studies (Harbath et al 2008) are not directly compared to this study due to differences in setting, their concentration on a single or limited number of specialties, rather than considering a whole hospital, or their evaluation of universal PCR screening, which is little used in the UK, with a much smaller range of alternative screening policies than those considered in the NOW project.
The DH impact assessment (appendix 1) did not account for transmission, considering only patient level events limited to those patients colonised on admission, evaluated health benefits using deaths avoided (each death having a value of £250,000) and had a much higher estimation of the effectiveness of isolation and decolonisation at 90%, whereas our model estimated reduction in transmission for primary isolation at 64% (SD 14%), and at 24% (SD 12%) for contact precautions and decolonisation.
These parameters have previously been found to exert the greatest influence, and this may explain why the DH policy of screening all admissions rarely proved cost-effective in our models, which incorporated full uncertainty of these parameters in order to maximise robustness of results.
The Scottish HTA prevalence model (Ritchie et al 2007)was limited, as its authors acknowledged, by substantial uncertainty in parameter estimation, which they did not attempt to adjust for. Our model included full uncertainty distributions for intervention parameters and used national audit data whenever possible.
The subsequent Scottish Pathfinder study (Health Protection Scotland 2011 a b c d) provided an excellent review and valuable general model of MRSA screening and associated cost-effectiveness. The authors explicitly recommended that the next step for modelling should be an individual-based approach (see introduction above), as conducted here, with stochastic modelling calculating each patient's probability of colonisation or infection on a daily basis, which depended on how many such patients they were surrounded by, and what screening, isolation and decolonisation interventions these were receiving (which might also change on a daily basis). Other key differences included extensive parameterisation of transmission which could change daily, the modelling of uncertainty, and incorporation of real patient movement data and patient level differences in probabilities of discharge and mortality.

The cost effectiveness results
The relative cost-effectiveness of screening admissions to high-risk specialties probably derives from the fact that most infections occur in this population, who have a higher probability of progressing from colonisation to infection and it is infections that have the largest impact on length of stay and mortality, the largest cost and health benefit determinants. The reduction of these infections, combined with lower screening and isolation costs associated with a strategy of screening only admissions to high risk specialties, makes this a more cost-effective strategy, especially in higher prevalence settings. The differences in cost-effectiveness between acute and teaching trusts reflected the overall greater costs per admission in teaching trusts (causing any strategy to have a higher cost/QALY) and lower transmission and MRSA admission prevalence in the high risk specialty population (meaning the screening strategies had less ability to reduce transmission and infections, and therefore less ability to generate health benefits in teaching trusts). The differences in cost-effectiveness between specialist and acute hospitals reflected higher transmission rates in specialist hospitals meaning that strategies could better prevent transmission and infections and thus generate greater QALY gains, making all strategies better value for money (even the most costly ones, including screening all admissions). The small population size of specialist trusts in particular also meant that stochastic (i.e. chance) effects had a greater impact on the transmission dynamics.
It should also be noted that there existed substantial uncertainty, with the probability of any one strategy being a better option than any other of around only 30% (at willingness to pay values of £20-30,000/QALY).
Overall the results indicate that persisting with the current policy of routine screening of all admissions, does not appear to be cost effective. Reverting to the previous targeted screening strategy of screening only admissions to high-risk specialties may generate substantial savings (on average £250m per year) across the NHS for a very minimal rise in infections (approximately two per year per trust) and colonisations (approximately one per week per trust). The cost-effectiveness of this strategy was maintained even if prevalence increased to twice the current levels.

Generalisability:
Wherever possible model inputs came from audit data to reflect current status of trusts in England, with extensive sensitivity analyses undertaken demonstrating the robustness of findings to trust type, MRSA prevalence, and assumptions regarding transmission and mortality probability parameters. This enabled the model to have substantial generalisability. However, the certainty in choosing between strategies was low and changes in costs and effects were clustered.

BACKGROUND
Following government and public concern at reported high levels of meticillin resistant Staphylococcus aureus over 2001-4 (HPA 2004) many national infection control interventions were introduced. These included the cleanyourhands campaign (NPSA 2004, Stone et al, 2012 -2011(DH 2009 in view of its historically high levels and associated infection, mortality and cost (Coia et al 2006;HPA 2012). The basis of MRSA reduction is screening for asymptomatic carriers, isolation of those found to be MRSA positive (MRSA+ve), and suppression/decolonisation therapy .
Epidemiological and health economic modelling, largely based on data from the literature, suggests that increasing either the intensity of screening or the isolation capacity of a hospital is effective, provided neither is limited (Cooper et al 2003. There are no randomised controlled trials to provide guidance on the most effective and cost-effective screening strategies and clinical studies in the United Kingdom vary in the patient group screened (Rao et al 207;Hardy et al 2009, Jeyaratnam et al 2008, Creamer et al 2010, Smyth et al J 2008 and reported effectiveness of different screening strategies. National guidance, published in 2006 (Coia et al 2006), recommended targeted screening of patients in high risk specialties (Nephrology, Neurosurgery, Orthopaedics and Trauma, Haematology and Oncology, Vascular Surgery and Cardiothoracic Surgery) where infections were likely to be deep-seated and hard to treat and/or targeted screening of individual patients with known risk factors for MRSA carriage, and/or of patients . Hospitals had discretion to implement these guidelines according to local circumstance.
A Scottish Health Technology Assessment systematic review (Ritchie et al 2007) produced a prevalence model, populated by data from the literature, which showed that routine screening of all admissions was effective, especially when combined with pre-emptive isolation of high-risk specialty patients. Long term differences in prevalence levels differed little between screening strategies, although there were substantial variations in five-year costs. It appeared possible that teaching hospitals could save up to £2M and non-teaching hospitals more than £1M over five years by using other risk based screening strategies instead of routine admission screening. On the basis of its own impact assessment (Appendix 1) which modelled the cost effectiveness of different screening and decolonisation strategies in preventing MRSA bacteraemias, wound infections and deaths, the Department of Health decided to introduce mandatory screening of all appropriate elective admissions from April 2009 (by which time MRSA bacteraemia rates had fallen to 7.8/100,000 beddays) and of all appropriate emergency admissions from December 2010 (by which time rates had fallen further to 4.2/100,000 beddays). Certain day cases (ophthalmology, endoscopy, dental and minor dermatology), paediatrics (unless in a high risk speciality) and maternity/obstetric cases were to be excluded from routine screening because the model concluded screening these groups was not cost effective. The impact assessment committed the Department to review the effectiveness of this policy in the future, a commitment reflected in calls by both the National Audit Office (NAO 2009) and Public Accounts Committee ( Public accounts committee 2009) for a robust review of the implementation of the policy, its effectiveness and cost-effectiveness, and its impact on patients and their management.

The NOW Study
This research commissioned by the Department and whose aims, methods, results and implications for research and policy are described in this report, took the form of a national prevalence audit of MRSA screening providing representative clinical and economic data to populate an individual-based transmission dynamic model evaluating the effectiveness and cost effectiveness of different screening strategies. The study was able to incorporate key findings of the Scottish Pathfinder study (Health Protection Scotland 2011b, Stewart et al 2011 which reported during the study. Pathfinder produced a prevalence model of the effectiveness and cost effectiveness of routine admission screening of all admissions, populating their model by clinical and economic data taken from three NHS Boards (equivalent to three English trusts), where routine screening had been implemented for a year. In purely economic terms the most cost effective strategies were, in order: 1. checklist activated screening of all admissions (using a check-list of clinical risk factors for MRSA carriage to assess all admissions and screening those with at least one risk factor), 2. two swab (nasal and perineal) screening of all admissions to high-risk specialities combined with universal (routine) check-list activated screening of all other admissions, 3. universal nasal swab screening.
Their overall conclusion was that, taking public acceptability and the economic climate into account, the second strategy (two swab screening of all admissions to high-risk specialities combined with universal check-list activated screening of all other admissions) "appeared to offer the best clinical return for a similar level of financial investment to universal screening of all admissions.". They explicitly recommended that future models should move from prevalence models to individual-based ones, to facilitate development of more complex, powerful tools for predictive modelling to inform decisions on costed management of healthcare associated infection.

OBJECTIVES
The current study had six objectives:-Objective 1. To report how widely the policy of screening Emergency and Elective admissions was implemented across the NHS in the three different trust types (acute teaching and specialist) Objective 2: To report the prevalence of MRSA carriage on admission in Emergency and Elective admissions in different types of trust and the proportion of carriage that was previously unknown.
Objective 3: To report on screening, isolation and decolonisation policies and practices, laboratory methods and costs Objective 4: To report on how patients were managed, how soon were results available, and how many patients were isolated and / or decolonised pre-emptively and after the result was known, and how many were treated for infection.

Objective 5:
To determine the extra yield of MRSA positive patients achieved by routine admission screening for the NHS as a whole and for each of the three types of trust (acute, teaching and specialist) compared to (a) check list activated screening of all patients (b) screening "high-risk" specialties only (c) and screening all high-risk specialty patients with checklist activated screening of all low-risk patients.
Objective 6: To use these data and reliable cost data to populate an existing model of hospital MRSA transmission to provide predictions of the effectiveness and cost-effectiveness for each type of trust for six different screening strategies:

2)
screening all admissions (emergency and elective)

3)
screening admissions to "high-risk" specialties only

4)
checklist activated screening of all admissions

5)
strategy 3 plus checklist activated screening all other admissions 6) strategy 2 (screening all admissions) plus pre-emptive isolation of those known to be previously MRSA positive.

Methodology audit questionnaire.
A national one-week prevalence audit of MRSA screening with modelling was deemed to be the most practical way to evaluate routine admission screening in a timely manner. Questionnaires were sent to infection control teams in all English NHS acute trusts for completion in May 2011, allowing time for acute admission screening to bed down nationally.

Ethics and Research Governance
Since the study used anonymised, confidential patient data and effected no change in clinical management, the National Research Ethics Service considered that it did not require formal ethical approval.

Study Design
A national one-week prevalence audit of MRSA screening with epidemiological and health economic modelling data was carried out through a questionnaire sent to infection control teams in all English

Questionnaire design and piloting
The questionnaire was designed by the research team in December 2010 and piloted in 10 trusts in early 2011 to check for understanding (face validity) of individual items and to assess for the feasibility of data collection. Further changes were made following meetings in March/April 2011 to discuss participation in the audit with representatives from approximately 125 hospital infection control teams in nine regional meetings around the country. The steering group was closely involved in the conduct of the research and gave input into the questionnaire design. A paper copy of the questionnaire was sent to infection control teams in all 167 NHS acute trusts at the end of April 2011.
The questionnaire (see appendix 2) was divided into 5 sections as follows: Sections 1a: Trust level data on the number of emergency, elective, and day-case admissions for the week 11 th -17 th April 2011. Data was broken down into "high-risk" specialty admissions (Nephrology, Neurosurgery, Orthopaedics and Trauma, Haematology and Oncology, Vascular Surgery and Cardiothoracic Surgery) and "low-risk" specialty admissions (all other specialties). Trusts were defined as Acute, Teaching or Specialist according to standard HPA definitions used in mandatory reporting.
Section 1 b: The number of MRSA screens performed on emergency, elective and day case admissions that week with the numbers screening positive for each of these three groups, divided into new and previously known MRSA. These data were requested retrospectively, to facilitate questionnaire completion and rapid return as data for the audit week in May would not be available from trust informatics for two to three weeks, and it is a reasonable assumption this admission data would be similar to admission data for the audit week in May.
Section 2: Trust level data were also collected on local screening practice and other policies including: of these risk factors, the risk factor was considered present. If there was no documented or data base evidence of these risk factors then it was assumed the factor was not present. In this way it was hoped to mimic routine clinical practice Section 5: The same individual patient level data was sought for a randomly selected sample of 5-10 MRSA negative patients who were admitted or screened in that week . Respondents were asked to identify all relevant patients for that week, number them consecutively and identify a random sample using an online research randomiser tool (http://www.randomizer.org/form.htp). As well as the five risk factors for MRSA carriage mentioned above, trusts were also requested to record whether the patient was known previously to be MRSA positive (MRSA +ve).

Methodology: Modelling & Cost-Effectiveness.
In order to address objective 6, the above data were used to inform mathematical models of MRSA transmission in hospital populations. Three models were developed to represent different Trust types; Acute, Teaching and Specialist. Each of the models simulated patient movement within the hospital and between hospital and community populations, transmission of MRSA within the hospital, as well as alternative screening and control strategies.
The models were used to simulate MRSA transmission under each of these alternative strategies in order to compare both the effectiveness and cost-effectiveness of control.

The models (see appendix 3)
An existing dynamic model developed for the DH funded MECAMIP project [Robotham et al 2011], which simulated MRSA transmission within a single hospital ward, was further advanced. In order for the models to be appropriate for evaluation of hospital-level screening policies the following major extensions were performed: 1. The development of a whole hospital model.
2. The inclusion of realistic patient movements (ward transfers and readmissions).
3. Stratification of the admissions into elective or emergency admissions 4. The development of three distinct models representing the different Trust types.

Model structure
Models were stochastic, individual-based and discrete-time and simulated the transmission of MRSA in a whole hospital setting. The models were individual-based allowing the MRSA status of individual patients to be tracked over time.
The model structure is represented schematically in Figure 1, and model assumptions listed below.

Summary of model assumptions:
• Admissions may be colonised or susceptible according to prevalence (but not infected).
• Prevalence on admission is dependent on whether the patient is categorised as checklist positive (ie having at least one risk-factor for MRSA carriage on the six-item checklist) as well as whether they are admitted via an Elective or Emergency admission route • No specific assumptions about transmission routes are made.
-The instantaneous risk of a susceptible patient becoming colonised increased linearly with the ward-level MRSA prevalence.
• Colonised and infected patients are equally infectious.
• Transmission parameters (infectiousness of colonised/infected individuals, probability of progression and susceptibility to colonisation/infection) are specialty dependent.
• Direct infection from a susceptible state cannot occur in low risk specialty settings and patients must first become colonised.
• Once MRSA positive, patients remain so for the duration of their stay.
• All infected patients are suspected to be so, with a delay of 1 day before a clinical specimen is taken.
• Recovery may occur in the community.
• At any time patients may belong to either high-risk (HR) or low-risk (LR) specialties -Parameters may differ between specialties -No transmission can occur between specialties.
• Length of stay of colonised and susceptible patients is modelled using the same daily probabilities of discharge, only infection increases length of stay • Similarly, additional mortality is associated with infection only • Daily probability of discharge and death is dependent on whether the patient is in a high-risk or low-risk specialty as well as their infection status.

Uncertainty
For parameters determining effectiveness of the intervention method used, values were defined as probability distributions rather than point estimates. These distributions were chosen to represent the uncertainty in each of the parameters and are assumed independent. In each model simulation run, a parameter value was sampled from these distributions. Since different simulations draw different parameter values from these distributions, the model outcomes also vary between simulations. In this way, uncertainty could be propagated through the model.
In addition to the uncertainty in the parameter values, chance also enters the model due to the stochastic nature of the transmission process. While it is important to account for such stochastic effects when evaluating different strategies, uncertainty in model outcomes should reflect only parameter and structural uncertainties in the models. Therefore, for each sample of parameter values we performed a large number of runs and recorded the mean value of the outcomes of interest.
Specifically, we selected 50 parameter sets (each with a different value pulled from the probability distribution for intervention effectiveness) and ran the model 1000 times for each parameter set.
Therefore for each strategy, we performed a total of 50,000 model runs.
thousands of model runs were required to compare the strategy options; these were performed on a high performance cluster. Analysis and graphical representation of the large amount of generated output was performed in R 2.10.1 (www.r-project.org).

Cost-effectiveness analysis
Health economic data were incorporated into the model to explore the direction and size of changes in economic costs and health benefits due to interventions, through a cost-effectiveness analysis (Graves et al, 2004), allowing comparison of each screening policy. Incorporation of economic parameters into a transmission dynamic model (as opposed to a static model) allows population-level effects to be accounted for, such effects are important since preventing infection in one individual directly benefits that individual and indirectly benefits others by preventing transmission. Analyses failing to take account of indirect effects may underestimate benefits of interventions . Health benefits are described using quality adjusted life years (QALYs). The theory by which health benefits may be evaluated using a dynamic simulation process is outlined in Figure 2. Three types of costs incurred were considered in our analyses:  Infection related costs per day.
 Cost of a bed day  Intervention related costs.
Health Benefits: Health benefits are described using quality adjusted life years (QALYs). Three measurements were required for analysis of health outcomes:  bed days accrued  number of deaths  number of successful patient episodes (number of patients discharged alive) This estimate was further adjusted to account for long term effects of MRSA infections.

Cost-effectiveness outputs
We conducted a health-economic evaluation to predict outcomes of each strategy in terms of costs and health benefits, measured in Quality Adjusted Life Years (QALYs). The perspective for this analysis represents the healthcare decision maker at a regional or national level.
Competing interventions were compared against a baseline scenario in terms of their (ICER) incremental cost-effectiveness ratios (the ratio of the change in costs to the change in health outcome compared to the alternative). Strategies were considered cost-effective if they generate an ICER) that is less than the current NHS decision makers willingness to pay threshold of £30,000 per QALY.
Using the economic transmission model we compared policies in different scenarios and settings, firstly in terms of the clinical effectiveness of each policy (in terms of appropriateness of resource use, and number of transmission, infection and death events) followed by the costs of each policy. These effectiveness and costs results are then combined and depicted on cost-effectiveness planes and presented as mean ICERs, allowing direct comparison of alternative strategies.
We also present results in the form of cost-effectiveness acceptability curves (CEACs) and costeffectiveness acceptability frontiers (CEAFs) which show, respectively, the probability of each strategy having the highest net monetary benefit (NMB ) for different values of the willingness to pay per unit of health benefit gained, and the strategy returning the greatest expected NMB.

Model Parameterisation (see appendix 6)
Model parameters were estimated using data from the prevalence audit wherever possible, readmission data from the sentinel audit (see appendices 4a and 4b) with others estimated as in MECAMIP (Robotham et al 2011) or from previously published data.
Where uncertainty associated with model parameters was included, parameters were described using probability distributions. Where NOW audit data were used for parameter estimation, data were stratified according to Trust type.
1) Population parameters, the characteristics and sizes of populations modelled.
2) Movement parameters, ie patient transfers, discharge and readmission, the additional length of stay and mortality associated with MRSA infection.

Methodology sentinel questionnaire( See appendix 5a.)
Data were collected from a sample of trusts to allow more accurate parameterisation of the model for the present study and for future work. All English acute NHS trusts that had responded to the questionnaire were asked whether they would be willing to take part in the Sentinel study. In total 8 trusts returned data -3 teaching and 5 acute trusts. They represented a range of sizes and covered a wide geographic spread.
Data were collected in the autumn/winter of 2011 for the following parameters:

Objective 1
To report how widely the policy of screening Emergency and Elective admissions is implemented across the NHS in different hospitals and patient groups.
Response rates to the questionnaire were excellent, 144 (86.2%) of trusts returned a questionnaire, of which 143 (85.6%) returned a response to sections 1-3 of the questionnaire (Table 1). Where data are presented broken down by trust type (acute, teaching and specialist), data from two trusts, for which it was not possible to identify the trust type are included under the "All" category.  significantly more likely to be screened than Day-Cases (z-ratio=44.444, p=<0.0002.) Elective admissions were significantly more likely to be screened than Day cases (z-ratio=91.594, p=<0.0002).

To report the prevalence of MRSA carriage on admission in Emergency and Elective admissions in different types of hospital and the proportion of carriage that was previously unknown.
In those trusts for which data on numbers of admissions and admission screens were available, only 2.1% of emergency screens and less than 1% of elective and day-case screens were MRSA +ve (Table 3).
Just 1%, 0.6% and 0.5% of emergency, elective and day-case screens were from patients who were newly identified as MRSA +ve (Table 4). In order to identify one new MRSA +ve patient, therefore, 102, 180 and 186 screens would need to be taken respectively for emergency, elective and day-cases (Table   5). Differences in proportions of patients found to be MRSA +ve on admission were calculated comparing Emergency with Elective admissions and separately comparing Emergency with Day-Case admissions.
To report on screening, isolation and decolonisation policies and practices, laboratory methods and costs.
Questionnaire section 2: Trust practice around MRSA admission screening

Screening:
In section 2 of the questionnaire trusts were asked to report screening protocols. Following DH advice, more than two thirds of trusts did not screen dermatology-, ophthalmic-, and dental day-cases, low risk paediatrics or endoscopy patients (Table 6). Acute trusts were significantly more likely to exclude groups from admission screening than specialist trusts. Fifty per cent of specialist trusts excluded no patients from screening compared to 5% of acute trusts (z-ratio=-5.435, p=<0.0002). Acute trusts were no more likely to exclude paediatric (z-ratio=1.723, p=0.0849) or emergency maternity admissions (z-ratio=0.911, p=0.3623) compared to teaching trusts. Statistically significant differences for other categories not calculated. All 143 trusts reported that nasal swabs were taken as part of admission screening, with the majority also screening groin and/or perineum plus another site (Table 7). *"Other" includes wounds, indwelling devices, throat etc.

Laboratory Testing:
The most common laboratory technique for processing both elective and emergency admission swabs was chromogenic agar plating with more than 80% of trusts using this as their main technique (Table   8). Only a very small number of trusts (7: 4.9%) were routinely using PCR for the processing of emergency admission screens and just one trust used PCR as the routine method of screening elective admissions. Reported costs for laboratory processing per swab are also reported in Table 8. As might be expected, the most commonly used technique (chromogenic agar plating) was also one of the cheaper options.   Of the 3033 patients for whom data were available 1837 (60.5%) were isolated in side-rooms, 82 (2.7%) in a designated ward and 107 (3.5%) in a cohort. The remaining third were not isolated (Table 11).
Specialist trusts were significantly more likely than acute (Fisher's p<0.0001) and teaching trusts (Fisher's p<0.0001) to isolate MRSA+ve patients.
In addition, 127 (88.8%) of responding trusts were able to supply data on the numbers of MRSA +ve inpatients receiving antibiotic treatment for any MRSA infection on the audit day. They reported that 286/2680 (10.7%) of MRSA positive patients were receiving antibiotic treatment (vancomycin, teicoplanin, doxycycline, linezolid, rifampicin or fusidic acid). Acute trusts reported a higher proportion of patients receiving antibiotic therapy than teaching trusts (z-ratio=-2.339, p=0.0193).

MRSA -ves:
A total of 141 trusts returned forms with details of MRSA -ve patients who were admitted or screened during the audit week. Each trust selected a random sample of 5-10 relevant patients. Data were received for 951 patients.
Differences between the two groups were not significantly different (z-ratio = 1.525, p=0.1273). Patients admitted to specialist trusts were generally younger than those admitted to acute and teaching trusts (50.1 years: MRSA-ves, 51.9 years: MRSA+ves)  The proportion of MRSA +ve patients and MRSA -ve patients admitted to "high risk" specialties was 17.6% and 24.2% respectively) (Table 15). This difference was found to be statistically significant (zratio = -3.291, p=<0.001).

Patient Management:
Sample turnaround time (the time between swabbing and the result becoming available) was calculated by subtracting the date that the result was available from the date that the swab was taken.
Mean turnaround time was 2.87 days (median 3 days) for MRSA +ve results. The corresponding mean turnaround time for MRSA -ve results was 1.75 days, (median 2 days) (Table 16). This difference was statistically significant. A substantial proportion (219/596:36.6%) of admitted patients who were subsequently found to be MRSA +ve were already discharged before results were available (see Table 17). Despite the faster turnaround time for MRSA -ve samples, a similar proportion of MRSA -ve patients (213:640: 33.3%) had been discharged before the result was available.

Checklist items MRSA positive patients:
Checklists for risk factors for MRSA colonisation were completed (or partially completed) for 760 patients. Five items were identified (whether the patient had previously been an inpatient in that trust or another trust was combined into one item). (table 19) Any patient who was checklist positive to at least one of the items was defined as "checklist positive".
All other patients (including those for whom one or more checklist items were not completed) were defined as "checklist negative". A total of 458/760 (60.3%) patients were found to be "checklist positive". A higher proportion of emergency screens were "checklist positive" (376/542: 69%) compared to those screened electively (72/201: 35.8%) (Table 20). Those admitted via an emergency route to HR specialties were most likely to be "checklist positive".
MRSA +ve patients were more likely to be checklist positive compared to the MRSA -ves for the following items: Previously resident in a nursing home (z-ratio=9.742, p=0.0002), presence of a wound (z-ratio = 7.184, p=0.0002), presence of an in-situ device (z-ratio = 2.594, p=0.0095). No statistically significant difference was found between the two groups for the following checklist items: transfer from another trust (z-ratio = -0.031, p=0.9753), previous admission to the trust (z-ratio = 0.436, p=0.6628), previous admission to another trust (z-ratio = -1.029, p=0.03035).

Checklist positivity using Pathfinder checklist
The checklist used in the Scottish Pathfinder study (Stewart et al 2011), unlike the one used here, did not include a checklist item for patients that had been previously admitted.
As would be expected proportions of patients that were found to be checklist positive were lower for both MRSA +ve and MRSA -ve admissions. (see tables 23 and 24). These data were used to inform a sub analyses looking at the cost-effectiveness of checklist activated screening in acute trusts to allow comparison with results derived from the Pathfinder study.   and 5 of the questionnaire (see Table 21 and The use of checklist activated screening, therefore, would almost halve the numbers of laboratory samples required in a week (from 790 to 397.69) and identify 81% (9.10/11.33) of MRSA +ve admissions.  Sensitivity and specificity of the screening tool were calculated for the average trust. Sensitivity, i.e. the proportion of MRSA +ves who were checklist positive, was 81%: 9.10/11.33 and specificity, i.e. the proportion of MRSA -ves who were checklist negative, was 50.1%: 390.11/778.70.

Objective 5b.
To determine the extra yield of MRSA positive patients achieved by routine admission screening compared to screening selected "high-risk" specialty admissions.
The average weekly number of patients per trust admitted to high risk specialties is reported in Table   27. Numbers were derived as for objective 5a above. Since infection control teams were not asked to report what proportion of admission screens were collected from patients in high risk specialties, it is assumed that proportions were the same as the proportion of admissions to these specialties reported in section 1a of the questionnaire (i.e. 7.3% and 14.75% for emergency and electives respectively).

To determine the extra yield of MRSA positive patients achieved by routine admission screening compared to screening all admissions to high risk specialties plus checklist activated screening of admissions to other specialties.
Combining the two strategies outlined above (laboratory screening for all admissions to high-risk" specialties and checklist activated screening for all admissions to "low risk specialties would require more laboratory screening for these trusts and would increase the yield slightly compared to screening only those who were checklist positive. Using this strategy, a total of 86.76 screens would be taken from "high-risk" specialty patients [row A] plus a further 361 from "low-risk" checklist positive patients [rows F+H+I]. In total 9.34 screens would be MRSA +ve (1.05 from "high-risk" specialties and 8.29 from "low-risk" specialties [F+H]). This strategy would require 447.73 laboratory screens per week and, compared to checklist activated screening of all admissions, would increase the number of MRSA +ves detected from 9.10/11.33 (81%)to 9.34/11.33 (82.4%) per week, approximately equivalent to one high risk specialty patient per month. Equivalent numbers for different trust types are presented in table 29.  Objective 6:

1) no screening
2) screening all admissions (emergency and elective)

3)
screening admissions to "high-risk" specialties only

4)
checklist activated screening of all admissions

5)
strategy 3 plus checklist activated screening all other admissions 6) strategy 2 (screening all admissions) plus pre-emptive isolation of those known to be previously MRSA positive.
For each Trust type, results are presented results in terms of the effectiveness of each strategy, followed by the costs of each strategy (split into cost components). Costs and effects are then combined and the results presented as cost-effectiveness planes and as mean incremental costeffectiveness ratios (ICERs). Probabilistic sensitivity analyses are then presented to include the impact of uncertainty (namely the effectiveness of interventions, which has previously been shown to be the dominating parameter [Robotham et al 2011]), as the plots of multiple model runs under multiple parameter values, showing the full extent of uncertainty in the model outputs.
The uncertainty is considered in the comparison of policies in the cost-effectiveness acceptability curves (CEACs), which show the proportion of simulations in which each strategy is cost-effective i.e.
how likely each strategy is to be suboptimal for each willingness to pay. Finally, results are presented as cost-effectiveness acceptability frontiers (CEAFs) which show the probability that the strategy with the highest net monetary benefit (NMB) is cost-effective for a given willingness to pay.
Results are also presented for several scenarios:  For each Trust type we compare strategies under baseline prevalence for each trust type, as well as a prevalence of twice (high prevalence) and half this value (low prevalence).
 For Acute Trusts only (assuming baseline prevalence values) strategies are evaluated where : -Transmission parameters in high risk settings were reduced to the midpoint between ICU and general medical ward estimates.
-Daily death probabilities in high risk specialties were reduced to be equivalent to those in low risk specialties.
 Finally, for Acute Trusts only, under baseline prevalence, strategies are evaluated using the Scottish Pathfinder definition of 'checklist positive', which is slightly different to that used in the audit study, to determine who is screened in "checklist activated screening".
A summary of the cost-effectiveness of the various strategies for different scenarios is given in tables C5 and C6 at the end of the results section. Figure A1 shows the degree to which each of the strategies applied the primary intervention (i.e. use of side rooms accompanied by patients decolonisation) appropriately (i.e isolating MRSA positive patients), inappropriately (i.e. isolating MRSA negative patients) and the number of unisolated days ( i.e. days spent by MRSA positive patients out of side room isolation.)

Baseline analyses Effectiveness
Strategies 2 (screening all admissions) and strategy 6 (screening all admissions with pre-emptive isolation of those known to be previously MRSA positive) do particularly well in terms of appropriate isolation use, but also give most inappropriate isolation days (simply due to a greater number of patients being isolated).
Strategy 6 reduces unisolated MRSA positive bed days only slightly compared to simply screening all patients. This is because those who are infected typically have long stays, so catching a small proportion of the MRSA positive population 'early' through pre-emptive isolation makes little overall difference to unisolated days. This finding would be missed if infection status dependent lengths of stay were not included within the model.
Strategy 3 (screening admissions to high risk specialties) appropriately (and inappropriately) isolates very few patients, with the majority remaining unisolated. This is due to the relatively small proportion of admissions to high risk specialties (16% of beds in Acute Trusts belong to high risk specialties).
Strategies 4 (checklist-activated screening) and 5 screening all high risk specialty patients plus 'checklist activated screening of admissions to low risk specialties) perform between the two extremes described above. However, strategy 4, whilst reducing the amount of appropriate isolation ~30% (compared to screening all patients) reduced inappropriate isolation usage by over 50%. This was because the prevalence in the checklist positive group was approximately 2.6% compared to 1.4% of the overall Acute admission population. Therefore, if isolation capacity is a limiting factor, screening only checklist positive patients may be an option to 'free up' 50% of the isolation capacity.

Figure A1. Primary isolation usage under each screening strategy,
showing appropriate isolation (isolation of patients who are MRSA positive), inappropriate isolation (isolation of MRSA negative patients) and unisolated bed days of MRSA positive patients. Figure A2 shows the extent to which the ability of each strategy to identify patients for control translates into reduction in transmission, infections and deaths. Strategies 2 and 6 which appropriately isolated the greatest number of patients reduced transmission to the greatest extent. However, overall reduction in transmission did not translate directly into reduction in MRSA infections; this was because some strategies concentrate on high risk specialty patients in whom the probability of acquiring infection was greater in the model than in low risk specialties. Thus, small reductions in transmission in high risk settings had a greater effect on the number of infections and therefore deaths, as the probability of death is directly related to infection status. Strategies with a greater impact on high risk specialties will do most to reduce infections and deaths. Thus strategy 3 (screening all admissions to high risk specialties) only marginally reduced transmission, but reduced infection and death more than strategy 4 (checklist activated screening) although the latter reduced overall hospital transmission more. Strategy 6 (screening all plus pre-emptive isolation of those known to be previously MRSA positive) reduced infections and deaths the most . Figure A3 shows total costs of each of the strategies, broken down into their component parts. Aside

Costs
from Strategy 1 ( no screening), Strategy 3 is the cheapest because it screens and isolates the fewest patients. Total costs per admission are dominated by bed day costs.

Figure A2. Patient outcomes under each screening strategy,
showing new acquisitions of MRSA by hospital patients, total MRSA infections in the hospital, and total deaths (all per 100 admissions).

Cost-effectiveness
Each strategy was evaluated on a cost-effectiveness plane where the effectiveness of each strategy was measured in terms of health-benefits (measured in QALYs) per admission. A reduction in the number of infections decreases length of stay (hence cost per admission) and numbers of deaths, resulting in improvements in cost per QALYs gained.
Model results ( Figure A4) confirmed that any investment in screening, compared to no screening, was likely to lead to increases in health benefits. Strategy 6 (screening all plus pre-emptive isolation of those known to be previously MRSA positive) gave the highest health benefits, but was associated with the greatest costs.

Figure A4.
Incremental cost-effectiveness plot comparing each of the screening strategies.
Numbers refer to strategies Error bars represent random error brought about by stochasticity and parameter uncertainty, and corresponded to +/-one standard error. Table A1 describes all strategies in terms of mean change in costs and mean change in health benefits compared to baseline (strategy 1), and then combines these in terms of a mean cost per QALY gained by changing strategy from the baseline 'do nothing' approach. Each strategy was also evaluated using the techniques of dominance and extended dominance, allowing some strategies to be eliminated from further evaluation. Dominated strategies were those both more costly and providing less benefit than at least one other strategy. An extendedly dominated strategy was one that is more costly and provides less benefit than a combination of another two strategies. Since it would never be costeffective to pay more for less benefit, these strategies (2 and 4) were excluded from any further evaluation. Incremental cost-effectiveness evaluations were then applied to the remaining strategies, which form the 'cost-effectiveness frontier'. Each option was evaluated along the cost-effectiveness frontier, starting at the baseline, strategy 1, to determine whether it was cost-effective (in terms of some maximum threshold of willingness to pay a unit of health benefits, such as a QALY) to move from one strategy to the next most costly strategy on the frontier. This process was iterated, calculating the change in costs and health benefits in moving to the next strategy on the frontier, stopping when no move to a new strategy was cost-effective within the chosen threshold, which for the purposes of this study was the upper threshold of the £20-30,000 per QALY that NHS decision makers tend to use as a willingness to pay threshold (Rawlins et al 2004) This provided mean incremental cost-effectiveness ratios (ICERs) (Table A2). By this analysis strategy 3 (screening all admissions to high risk specialties), at a mean incremental cost per QALY of £9,964/QALY, would be considered the optimal approach (under the specific model parameters and assumptions used). Although strategy 5, (checklist activated screening) was not cost-effective, it was only marginally above the£30, 000 threshold at £33,806/QALY.
Consideration of this option depends on decision-makers willingness to pay for health benefits.  pay. This shows substantial uncertainty as at willingness to pay values between £20,000 to £30,000 the probability that any one strategy is the most cost-effective option does not exceed 30%, with the probabilities that any one strategy was cost effective clustered together.

Figure A5. Simulation results comparing each strategy on a cost-effectiveness plane for an Acute
Trust setting. Each dot represents the mean of 1000 simulation runs for each parameter set. The results of 50 parameter sets for each strategy are plotted, where each parameter set is obtained by taking the mean value for all parameters apart from the effectiveness of the intervention, which is sampled from its probability distribution.

Figure A6. Cost-effectiveness acceptability curves.
Each line represents the proportion of simulations, for a particular strategy, that are cost-effective, as a function of willingness to pay for health benefits.

Scenario analyses Scenario a. High MRSA Prevalence:
In a high prevalence setting (with admission prevalence twice that found in the audit for Acute Trusts: 2.8% compared to 1.4%) the number of MRSA positive isolated and unisolated bed days are higher as there are more positive patients, not all of whom can be accommodated in isolation facilities. The pattern of differences between strategies is the same as at baseline prevalence ( Figure A8 in Appendix   7). The greater prevalence leads to slightly more transmission (more infectious imports to the hospital) and thus slight increases in absolute numbers of infections and deaths ( Figure A8). Differences in costs between higher and baseline prevalence settings are minimal ( Figure A10 in Appendix 7) however.
The cost-effectiveness plane shows similar overall health benefits per cost accrued as at baseline prevalence ( Figure A9) although health benefits were slightly greater as more transmission events and therefore slightly more infections and deaths were prevented. The error bars show overlap between strategies, which is reflected in the uncertainty demonstrated in the CEAC and CEAF plots ( Figures A10   and A11).
Evaluation of the cost-effectiveness frontier (Table A3) shows that despite the ICERs shifting slightly, the decision remains the same as in the baseline prevalence setting, with Strategy 3 (screening all admissions to high-risk specialties) the only cost-effective option (at a willingness to pay threshold of £30,000/QALY gained).

Figure A10. Cost-effectiveness acceptability curves.
Each line represents the proportion of simulations, for a particular strategy, that are cost-effective, as a function of willingness to pay for health benefits . Figure A11. Cost-effectiveness acceptability frontier. Lines depict the strategies with the highest expected net monetary benefit, dependent on the willingness to pay for health benefits, while dotted vertical lines the willingness to pay values at which the decision changes.

Scenario b. Low MRSA Prevalence:
In a low prevalence setting (with admission prevalence half (0.7%) that found in the audit for Acute Trusts) there are less MRSA positive isolated and unisolated bed days as there are fewer positive patients. The pattern of differences between strategies is the same as at baseline prevalence (Figure A14 in Appendix 7). The lower prevalence leads to less transmission, infections and deaths (due to less infectious imports into hospital), whilst comparison between policies remains the same as at baseline prevalence (Figure 15 Appendix 7) The costs however are very slightly higher (Figure A16 in Appendix 7), as for the same level of screening effort the screening costs are not offset to the same extent by cost savings through reductions in infections.
The cost-effectiveness plane ( Figure A12) shows the ordering of strategies remains the same under a lower prevalence.
Evaluation of the cost effectiveness frontier shows the ICERs change only slightly at lower prevalence, strategy 3 (screening of patients admitted to high risk specialties) remaining cost-effective ( Figure A9, Table A6) at a willingness to pay threshold of £30,000/QALY. However, Strategy 5 (checklist activated screening of all admissions plus screening all admissions to high risk specialties) also becomes a cost-effective option if moving from strategy 3 to 5.  A14) show very similar levels of uncertainty in the decision between strategies in a lower prevalence setting as in a baseline prevalence setting with clustering around the 30% probability of cost effectiveness for several stategies at the usual NHS willingnes to pay threshold of £20-30,000. However this rose to nearly 50% for strategy 3 below a £10,000 threshold, where it was the optimal strategy. Adding checklist activated screening of other admissions to this (Strategy 5) was optimal at the upper end of the usual willingness to pay threshold (figure A14).

Figure A13. Cost-effectiveness acceptability curves.
Each line represents the proportion of simulations, for a particular strategy, that are cost-effective, as a function of willingness to pay for health benefits.

Figure A14. Cost-effectiveness acceptability frontier.
Lines depict the strategies with the highest expected net monetary benefit, dependent on the willingness to pay for health benefits, while dotted vertical lines the willingness to pay values at which the decision changes.

Scenario c. Low transmission in high-risk specialties:
If transmission in high risk specialties was reduced to be midway between ICU and general medical ward transmission probabilities this had very little effect on the ability of each strategy to effectively isolate MRSA positive patients but there was half the number of deaths and therefore infections. Total costs therefore were greater than under baseline transmission parameters because costs of screening were not offset to the same degree by savings through reductions in The cost effectiveness plane (Figure A23 in Appendix 7) shows no strategy provides the same gains in health benefits at this lower transmission rate but despite a higher ICER, Strategy 3 (screening all patients admitted to high risk specialties) is still cost-effective at £12,382/QALY (Table A7).
There is much less uncertainty around this decision ( Figure A15) with Strategy 3 having a probability of almost 60% of being cost effective, compared to other strategies. This is reflected in the CEAF ( Figure A16) where strategy 3 is the optimal option from willingness to pay values of just over £10,000/QALY to nearly £90,000/QALY.  Figure A15. Cost-effectiveness acceptability curves. Each line represents the proportion of simulations, for a particular strategy, that are cost-effective, as a function of willingness to pay for health benefits.

Figure A16. Cost-effectiveness acceptability frontier.
Lines depict the strategies with the highest expected net monetary benefit, dependent on the willingness to pay for health benefits, while dotted vertical lines the willingness to pay values at which the decision changes .

Scenario d. Probability of death is homogenous across the hospital:
In a setting where the daily probability of death in high risk specialties is reduced to that in low risk specialties, there is little change in reduction in transmission or infection events but there are differences in long term QALY accrual, as death events prevent patients being discharged and going on to accrue QALYs . Results and policies are therefore compared only in terms of costs per QALY gained.
In this scenario each screening strategy has reduced ability to prevent deaths and thus gain health benefits (QALYs). The cost-effectiveness plane shows a 10-fold reduction in health benefits gained under each strategy ( Figure A17) compared to settings with the baseline probability of death. Evaluation of the cost-effectiveness frontier (Table A8) shows ICER values (cost/QALY) are much higher. Even so, Strategy 3 remains cost-effective at £26,551/QALY. The CEAC (Figure A18) shows the probability of this being cost-effective within the NHS willingness to pay threshold is about 60%. The CEAF ( Figure A19), which accounts for both the magnitude of the potential benefit and its probability of cost-effectiveness, shows the decision changes dependent on willingness to pay. The "no screening" strategy is optimal, up to approximately £25,000/QALY, while strategy 3 becomes optimal above this.

Figure A18. Cost-effectiveness acceptability curves.
Each line represents the proportion of simulations, for a particular strategy, that are cost-effective, as a function of willingness to pay for health benefits .

Figure A19. Cost-effectiveness acceptability frontier.
Lines depict the strategies with the highest expected net monetary benefit, dependent on the willingness to pay for health benefits, while dotted vertical lines the willingness to pay values at which the decision changes .

Scenario e. Checklist activated screening using Scottish Pathfinder checklist:
The Scottish Pathfinder study (Stewart et al 2011) had found that both checklist activated strategies (Strategies 4 and 5) were cost effective but used a slightly different checklist of risk factors for MRSA carriage from that used in the NOW study. Strategies 4 and 5 were therefore re-evaluated for their cost effectiveness in acute trusts at current admission prevalence levels (1.4%) using the Pathfinder checklist. This increased the prevalence of MRSA in the checklist positive group to 5.6% (from 2.6% using the NOW checklist), and decreasing the percentage of admissions screened (from 44% to 10.5% for elective patients, and from 57% of 25% of emergency admissions).
The incremental cost-effectiveness plot (Figure A29 in Appendix 7) shows, as expected, that the incremental costs for Strategies 4 (in particular) and 5 were markedly reduced because of the reduction in numbers screened. However, gains in health benefits were also reduced as more MRSA positive patients were "missed" using the Pathfinder definition because of the reduction in numbers screened. The overall incremental cost-effectiveness evaluations of the costeffectiveness frontier (Table A9) were therefore very similar to those found using the NOW checklist. Strategy 3, screening admissions to high risk specialties, remained the only cost-effective strategy at £9,731 /QALY gained.  Figures A6 and A7) as at willingness to pay values between £20,000 to £30,000 the probabilities that any one strategy was the most cost-effective option clustered together and did not exceed 30%.

Baseline analyses at current admission prevalences
Effectiveness: In teaching trusts, admission prevalence and transmission parameters are slightly lower than in acute trusts which results in fewer isolated and unisolated bed days ( Figure B1 in Appendix 7), with less infections and deaths ( Figure   B1), although the relative differences between screening strategies are the same.
Costs: Two effects lead to teaching trusts having greater costs per admission for each screening strategy ( Figure B3 in Appendix 7). These are their larger size (median of 1113 beds compared to 553) leading to greater numbers of bed days over the simulation period and the larger number of patients to be screened (acute trusts have 22% less admissions with 50% less overall bed day costs over the five-year simulation period) . Teaching hospitals also have 44% more high-risk specialty beds, which alters movement parameters within hospital and between hospital and community.

Cost-effectiveness:
The incremental cost-effectiveness plot ( Figure B2), shows that each strategy generates fewer health benefits (compared to baseline 'no screening' ) than are achieved in Acute settings due to the lower transmission, infections and deaths. Strategy 3 (screening patients admitted to high risk specialties) is the most cost-effective option on the cost-effectiveness frontier (Table B1) at just over £1,000 above the NHS willingness to pay threshold of £30,000/QALY, at £31,077 /QALY

Consideration of uncertainty:
The degree of uncertainty in this decision is large ( Figure B5 in Appendix 7), the CEAC ( Figure B3) demonstrating that no strategy has a greater than 30% chance of cost-effectiveness within the NHS willingness to pay range of £20,000-£30,000/QALY. The CEAF ( Figure B4) shows the optimal option is the isolation and

Figure B3. Cost-effectiveness acceptability curves.
Each line represents the proportion of simulations, for a particular strategy, that are cost-effective, as a function of willingness to pay for health benefits.

Scenario Analyses Scenario a. High MRSA prevalence:
When prevalence is twice that of current prevalence (2.6% compared to 1.3%) both appropriate and inappropriate isolation increase (Figure B8 in Appendix 7) as do transmissions, infections and deaths (Figure B9 in Appendix 7). Total costs reduce slightly (Figure B10 in Appendix 7) as screening strategies achieve cost savings through their impact on infections leading to cost savings. The incremental costeffectiveness plot shows each strategy provide greater gains in health benefits in this setting ( Figure   B5) making, screening patients admitted to high risk specialties (strategy 3), the only cost-effective strategy at £20,715/QALY (Table B3). Although both the CEAC ( Figure B6) and CEAF ( Figure B7) still exhibit considerable uncertainty, strategy 3 is the optimal strategy throughout the NHS willingness to pay range, with a 40% to nearly 50% probability of cost-effectiveness.

Figure B6. Cost-effectiveness acceptability curves.
Each line represents the proportion of simulations, for a particular strategy, that are cost-effective, as a function of willingness to pay for health benefits.

Scenario b. Low MRSA prevalence:
Where the prevalence is half that found from the audit for Teaching Trusts 0.65% compared to 1.3%) there is less use of isolation facilities (Figure B14 in Appendix 7), less transmission, infection and death (Figure B15 in Appendix 7), due to the lower infectious assault on the hospital. Costs are very slightly higher, as the same amount of screening takes place but has less impact in a setting with fewer infections ( Figure B16 in Appendix 7). The incremental cost-effectiveness plot ( Figure B8) shows that incremental gains in health benefits were much lower than in an average prevalence setting especially for strategy 2 (screening all admissions). The cost-effectiveness frontier showed that all strategies had ICERs well above the usual NHS willingness to pay threshold of £30,000/QALY and were therefore not cost-effective (Table B4). The CEAC ( Figure B9) and CEAF (Figure B10) show that even up to a willingness to pay of £40,000/QALY no screening strategies is cost-effective. The baseline Strategy 1 (no admission screening and isolating only clinical cases) is in fact the optimal strategy, with a probability of being cost effective of 60-80% at the conventional willingness to pay threshold.  Each line represents the proportion of simulations, for a particular strategy, that are cost-effective, as a function of willingness to pay for health benefits.

SPECIALIST TRUSTS
Baseline analyses at current admission prevalence (1.04%)

Effectiveness:
Key characteristics of specialist trusts which impact on the effectiveness and costs of screening strategies are their smaller size (72% smaller than acute trusts), which is not proportionally matched by the reduction in numbers of admissions (specialist trusts have 59% fewer admissions), and their higher proportions of high risk specialty beds (44% compared to 16%), which affects patient movement parameters and gives the population higher probabilities of transmission, infection and progression from colonisation to infection.
All strategies isolate greater numbers of patients per 100 bed days in a Specialist setting (Figure C1 in Appendix 7) , with greater appropriate and less inappropriate isolation because of the increased isolation of clinical cases owing to the higher numbers of infections arising from higher transmission in this setting. This results in higher numbers of unisolated MRSA positive cases, as those missed by individual strategies go on to generate more cases.
While the ordering of strategies remains the same as in Acute Trusts in terms of effectiveness at reducing transmission, there are more infections and a 22% increase in deaths per 100 admissions than in Acute trusts ( Figure C1) despite the slightly lower admission prevalence. Costs are less than in acute settings due principally to the lower bed costs in the smaller specialist trusts (figure C3 in Appendix 7) .

Cost-effectiveness
Comparing incremental costs ( Figure C2) shows that while costs remain broadly similar to an Acute setting, the ordering of strategies shifts slightly, with Strategy 3, screening admissions to high risk specialties, becoming a more costly option (compared to the baseline) due to the greater proportion of patients admitted to high risk specialties. Incremental health benefits gained for all screening strategies (compared to the baseline 'no screening' strategy, strategy 1) are at least twice as great in a Specialist setting ( Figure C2), leading to ICERs less costly than the cost-effectiveness threshold of £30,000.
Strategy 6, which also has the greatest gains in health benefits (screening all patients on admission plus pre-emptive isolation of those patients known to have been previously MRSA positive) is also cost effective at £14,324/QALY.

Consideration of uncertainty
There is substantial uncertainty in costs and QALYs as represented in Figure C3, with the CEAC ( Figure   C4) showing very tight clustering of screening strategies at probabilities of only 20% that any one strategy is the most cost-effective at willingness to pay values of £20-30,000/QALY. This is reflected in the CEAF ( Figure C5) where even though strategy 6 appears optimal for willingness to pay values over £15,000/QALY, the certainty in this decision does not exceed 25% for any willingness to pay.
For specialist trusts, moving from the current policy of screening all admissions to restricting screening to high-risk specialty patients would reduce total annual costs per trust by £438,000 (£600 per admission), at the price of 18.3 more colonisations (one extra/every three weeks/ trust) and 2.6 more infections per year. (see table C2)    Each dot represents the mean of 1000 simulation runs for each parameter set. The results of 50 parameter sets for each strategy are plotted, where each parameter set is obtained by taking the mean value for all parameters apart from the effectiveness of the intervention, which is sampled from its probability distribution.

Figure C4. Cost-effectiveness acceptability curves.
Each line represents the proportion of simulations, for a particular strategy, that are cost-effective, as a function of willingness to pay for health benefits . Figure C5. Cost-effectiveness acceptability frontier.
Lines depict the strategies with the highest expected net monetary benefit, dependent on the willingness to pay for health benefits, while dotted vertical lines the willingness to pay values at which the decision changes.

Scenario Analyses
Scenario a. High MRSA prevalence:

Effectiveness:
When prevalence is twice that of current prevalence, appropriate isolation increases and inappropriate isolation decreases, as there are more cases of MRSA ( Figure C8 in Appendix 7).
Unisolated MRSA positive bed days slightly increase as there are more missed cases and more onwards transmission. The relative ability of the strategies to effectively identify and isolate (and decolonise) MRSA positive patients remains the same. There are very slight increases in transmission and thus infection and deaths due to the higher admission prevalence. Total costs per admission are slightly lower ( Figure C10 in Appendix 7) as all strategies prevent more infections and produce cost-savings.

Cost-effectiveness:
The incremental cost-effectiveness plot ( Figure C6) shows the strategies are much more tightly clustered for both incremental costs and incremental health benefits, making the choice between strategies more difficult. Evaluation of the cost-effectiveness frontier (Table C3) shows only strategy 3, screening patients admitted to high risk specialties, is cost-effective at £9,745/QALY. The CEAC ( Figure   C7) demonstrates the uncertainty in this decision. The CEAF ( Figure C8) shows this is the optimal policy, yielding the greatest monetary net benefits for all values of willingness to pay above approximately £10,000/QALY.

Figure C7. Cost-effectiveness acceptability curves.
Each line represents the proportion of simulations, for a particular strategy, that are cost-effective, as a function of willingness to pay for health benefits . Scenario b. Low MRSA prevalence (0.52%) In a low prevalence setting, half that of current admission prevalence, appropriate isolation and unisolated MRSA positive bed days reduce, whilst inappropriate isolation increases marginally ( Figure   C14 in Appendix 7). Transmissions, infections and deaths are less ( Figure C15 in Appendix 7), whilst the relative ordering of strategies to reduce these remains the same. The lower prevalence leads to higher costs as the screening effort remains the same, but identifies fewer MRSA positive patients and has reduced ability to prevent infections and provide cost-savings ( Figure C16 in Appendix 7). The incremental cost-effectiveness plot values ( Figure C9) are similar to those in an average prevalence setting, although there is a slight re-ordering of the screening strategies. Evaluation of the costeffectiveness frontier (Table C4). shows that strategy 3, screening of high risk specialty admissions, remains cost-effective at £10,566/QALY but that Strategy 2, screening of all patients, also becomes cost-effective at £14,224/QALY. There is substantial uncertainty however, with the CEAC ( Figure C10) showing strategies 2, 3, 5 or 6 had a 25% probability of being the most cost-effective within the usual NHS willingness to pay range (£20,000-£30,000/QALY). The CEAF (figure C11), shows Strategy 2 is optimal for all willingness to pay values above £15,000/QALY and Strategy 3 below that, but with probabilities of cost-effectiveness no more than 30%.

Figure C10. Cost-effectiveness acceptability curves.
Each line represents the proportion of simulations, for a particular strategy, that are cost-effective, as a function of willingness to pay for health benefits.  - †An ICER of less than £30,000 per QALY is considered cost-effective. An ICER of more than £30,000 is not considered cost-effective. £30,000 is the upper limit of the usual NHS willingness to pay range. † †Dominated strategies are those that are more costly and provide less benefit than one other strategy or a combination of two other strategies. Since it can never be cost-effective to pay more for less benefit, ICERs were not calculated for these strategies. Strategy 6 (£87,517/QALY) †An ICER of less than £30,000 per QALY is considered cost-effective, An ICER of more than £30,000 is not considered cost-effective. £30,000 is the upper limit of the usual NHS willingness to pay range. † †Dominated strategies are those that are more costly and provide less benefit than one other strategy or a combination of two other strategies. Since it can never be cost-effective to pay more for less benefit, ICERs were not calculated for these strategies. * MRSA transmission rates in high-risk specialties were reduced (to be midway between values for high-risk and low-risk specialties in the baseline model (see appendix 4, table 6). ** The probability of death in high-risk specialties was reduced to the level of that in low-risk specialities (see appendix 4, table 4). ***Assumes that 26% of all admissions have a risk factor compared to the assumption used in the baseline model that 56% of all admissions have a risk factor (see appendix 4, table 1).

Summary of findings:
There was an excellent response to the study with 86% of all trusts providing data to inform the modelling. It also yielded seven important findings. Firstly, implementation of emergency screening was poor at 61% (median 67% IQR 48%-85%) and even worse for eligible day case admissions (47%) (median 41%, IQR 23%-79%) although better for electives (81%) (median IQR 58%-

100%)
Secondly, admission prevalence of MRSA was low at 1.5% (overall), 2.1% (emergency), 0.9% (electives) and 0.7% (day cases), with approximately half being newly identified MRSA. This meant that numbers needed to screen in order to identify one new positive were high, especially for elective (180) and day case (186) admissions, but were lower (80) for electives in specialist trusts.
Thirdly, over three quarters of trusts pre-emptively isolated those with previous MRSA with nearly half pre-emptively isolating high-risk specialty patients.
Fourthly, a third of MRSA positive and negative patients alike were discharged before the results of screening were available, as the mean turn round time was 2.87 days for positives and 1.75 days for negatives. . Decolonisation was started in ¾ of all patients screened and found to be newly MRSA positive Just over half of new positives who were still in patients were isolated. Chromogenic-agar plating was used for routine admission screening in most (82%) trusts and broth enrichment in a further 10%. Very few were using PCR test methodologies that could potentially produce faster results.
Fifthly, the point prevalence survey showed that although the overall prevalence of MRSA was 3.3% (3.6% acute, 3% teaching and 2% specialist), a third of MRSA patients were not isolated, except for specialist trusts.
Sixth, for the NHS as a whole, checklist activated screening would detect 80% of the MRSA positive patients detected by routine screening (an average of 2 less/week/trust) whilst halving the number of screens. Screening high-risk specialty admissions would detect only 10% of MRSA positive patients (although higher in teaching and specialist trusts) whilst reducing screening by 90%.
Finally, the cost effectiveness of different screening strategies, evaluated at a willingness to pay threshold of £30,000 per QALY gained, showed screening admissions to "high-risk" specialties performed best overall in different scenarios and settings. In the base case scenario none of the decisions changed for any Trust type if the willingness to pay for a QALY gained was reduced to £20,000. However, there was substantial uncertainty, with the probability that any one strategy was the most cost-effective generally not exceeding approximately 30%. This low probability of costeffectiveness associated with each of the screening strategies is due to the inherent uncertainty in the effectiveness of the accompanying intervention (isolation and decolonisation). If we had assumed that identifying MRSA positive patients led to placing them under an intervention that was 100% effective and prevented any further colonisations or infections, this uncertainty would be reduced. However, our estimates of the effectiveness of isolation and decolonisation represent the best available evidence, and we considered it to be important for decision makers to visualise this uncertainty within the decision making process.
At current admission MRSA prevalence levels, moving from current universal screening to targeted screening of high-risk specialty patients would reduce total average annual costs by £1,592,000 per acute trust, £1,864,000 per teaching and £438,000 per specialist trust, at the expense of one extra colonisation per week and less than two extra infections per year in acute and teaching trusts. For specialist trusts there would be even fewer colonisations but slightly more infections (an extra 2.6 per trust per year).

Strengths and limitations
The analyses presented have a number of strengths which are important to emphasise when comparing the findings with other model-based economic analyses. The first strength is the high response rates from trusts which enabled audit data representative of current NHS settings to populate the model. The availability of good quality economic data from the Pathfinder (Health Protection Scotland b) and MECAMIP (Robotham et al 2011) studies also contributed to the model and ensured that it was relevant to, and representative of, current NHS settings, practice and prevalence levels. The use of a powerful sophisticated individual based model, as suggested by the Pathfinder study as the next appropriate direction for modelling studies, is a particular strength. The incorporation of dynamic transmission models, with robust estimates of transmission, into economic evaluation enables consideration of population-level effects of screening strategies, which benefit not only the person screened but other patients. This avoided underestimating the effects of screening. Use of a stochastic model, with 1000 simulations per parameter set, for each strategy, minimises the uncertainty due to chance effects, which are dominant in the small populations seen in NHS trusts. Transmission parameters from the MECAMIP model were robust, being estimated using multi-state modelling techniques to analyse detailed individual-level patient data from hospital wards in NHS hospitals (see Appendix 6).
Other strengths include simulation of the time to result delays, and the capture of more realistic patient movements, than seen in other individual based models, between specialities and between hospital and community based on detailed hospital data. Long-term effects are considered by adjusting the quality adjusted life expectancy of a proportion of infected patients who suffered long-term detriments to health associated with an MRSA infection. However, there are two important potential limitations, which must be borne in mind when using the results to inform policy and research recommendations. Firstly, there were the assumptions that both transmission and probability of death in high-risk specialties were equivalent to that in Intensive Care Units (ICUs), and that in low risk specialties was equivalent to general medical ward patients. However this was adjusted within sensitivity analyses, assuming that transmission in high-risk settings was intermediate between ICUs and general medical wards and that probabilities of death for the whole patient population are equivalent to those in general medical wards. Results were robust to these parameter changes and this is important since differences in transmission parameters had a far greater impact on the strategy evaluation than differences in MRSA admission prevalence.
The second main limitation was that transmission was modelled at a specialty level, with homogenous mixing assumed between all patients within the high-risk specialties and all patients within the low-risk specialties, which might have especially affected large teaching trusts. A greater level of 'granularity' i.e. including a ward-based structure, and inclusion of ICUs in particular, would enhance the model, as patients in different wards and geographic locations will have different abilities for cross -transmission.
Further modelling, incorporating data from the sentinel sites on inter-ward transfers from ICUs to the rest of the hospital (ICUs being potential reservoirs of MRSA, acting as a "carousel" to spread the organism round the hospital), data on inter-ward transfers in general throughout a patient's hospital journey and data on readmission rates of MRSA positive and negative patients, would have led to still more realistic transmission dynamics, and thus more reliable evaluation of screening strategies.
Incorporating these unique data, whose collection had not been envisaged at the start of the NOW study, into the current transmission models to consider transmission at a more detailed ward level would have been helpful. This would be especially relevant in teaching trusts, since, in this setting high-risk specialty screening was only slightly more costly than the willingness to pay threshold, and since under conditions of low MRSA prevalence, no strategy appeared to be cost effective. In addition, modelling of the effects of excluding day case or elective screening and of the effects of pre-emptive isolation of all patients with a history of MRSA , which the audit data showed most trusts tried to do, would have also enhanced the applicability of the model to the current NHS. Other potential limitations were assumptions there was no increased mupirocin resistance due to its widespread use in decolonisation and that isolation had no adverse effects, although some evidence suggests isolated  Rao et al's study (2007) in an English hospital (84.5%), although we found a slightly lower proportion of admissions would be checklist positive (51% v 57.6% (Rao et al 2007)] v 57% (Smith et al 2011) and require screening. There were some differences between studies in that Rao's (2007)checklist was virtually the same as ours whereas the full Pathfinder checklist was much larger (12 items versus 6).
Pathfinder also examined use of a three item checklist (previous history of MRSA, presence of wounds or indwelling devices, and admissions not from home), which in fact covers five of the six NOW checklist risk factors, (excluding admission to hospital within the last year). They found that only 10% of their admissions would be checklist positive, but that the checklist would detect 50.4% of MRSA positive patients. When we applied the three-item checklist to our data we found that 26% of admissions were checklist positive and that 68% of MRSA+ve admissions would be identified. This difference between the studies is not easy to explain but may reflect differences in sampling and the types of hospitals studied. The NOW study was based on data collected from nearly all acute English Trusts, whereas the Pathfinder study is based on 6 hospitals, in three Boards, where the annual number of admissions (81,438 between them) is much lower than in the average teaching or acute trust in England (nearly 100,000 and 60,000 respectively).

Implementation of screening
The proportion of patients screened on admission was lower In the NOW study than in Pathfinder [Health Protection Scotland 2011a] where 85% of emergency admissions and 98% of elective admissions were screened. This may reflect the fact that data collected for NOW is representative of English rather than Scottish hospitals, but it could also reflect the fact that the Pathfinder hospitals were enrolled in a specific study, with a team of research nurses following admissions and reminding wards to screen if they had overlooked individual patients. The relatively low compliance that we report in England, with a national mandatory universal admission screening programme, which should be the easiest of all screening policies to routinise and implement, should give considerable pause for thought in respect of how well any policy of checklist activated screening would be implemented. The Pathfinder study was able to use research nurses to administer the checklist whereas in the NOW study, infection control nurses used the standard prevalence surveillance techniques of interrogation of hospital data bases, review of notes and examination or interview of patients, and entered more don't-know responses, which may be more representative of actual clinical practice. This indicates that implementation of a checklist by admitting nurses on wards is likely to be poor and should not be considered as a feasible option. this study due to differences in setting, their concentration on a single or limited number of specialties, rather than considering a whole hospital, or their evaluation of universal PCR screening, which is little used in the UK, with a much smaller range of alternative screening policies than those considered in the NOW project.
The Department of Health impact assessment did not account for transmission, considering only patient level events limited to those patients colonised on admission, evaluated health benefits using deaths avoided (each death having a value of £250,000) and had a much higher estimation of the effectiveness of isolation and decolonisation at 90%, whereas our model estimated reduction in transmission for primary isolation at 64% (SD 14%), and at 24% (SD 12%) for contact precautions and decolonisation (Worby et al in submission) These parameters have previously been found to exert the greatest influence and this may explain why the DH policy of screening all admissions rarely proved cost-effective in our models, which incorporated full uncertainty of these parameters in order to maximise robustness of results.
The Scottish HTA prevalence (Ritchie et al 2007) model was limited, as its authors acknowledged, by substantial uncertainty in parameter estimation, which they did not attempt to adjust for. A key assumption of the HTA model and therefore incorporated into the Pathfinder model, is that isolation reduces transmission to 0%. Our model included full uncertainty distributions for intervention parameters and used national audit data whenever possible. The subsequent Pathfinder study (Health Protection Scotland 2011 a, b,c ,d), published during planning stages of our study, provided an excellent review and valuable general model of MRSA screening and associated cost-effectiveness. The authors explicitly recommended that the next step for modelling should be an individual-based approach (see introduction above), as conducted in this study, so that patient movements, MRSA transmission and control brought about through screening, isolation and decolonisation were modelled at an individual-patient level. The stochastic modelling used in this study calculated each patient's probability of colonisation or infection on a daily basis, which depended on how many such patients they were surrounded by, and what screening, isolation and decolonisation interventions these were receiving (which might also change on a daily basis). Other key differences include extensive parameterisation of transmission which could change daily, the modelling of uncertainty, and the use of the MECAMIP model to estimate the effectiveness of decolonisation and isolation (Robotham et al 2011) using Markov state modelling techniques. For instance a key assumption of the Scottish HTA study (Ritchie 2007) and the subsequent Pathfinder report (Health Protection Scotland 2011b) was that isolation reduces transmission to zero, this contrasts with the assumption used here that isolation is 64% effective.
Final important differences are incorporation of real patient movement data and patient level differences in probabilities of discharge and mortality , according to whether they are in a high or low risk specialty, their length of hospital stay and whether they were infected or not . Readmissions were modelled at an individual patient level, with probabilities of readmission calculated from real hospital data and dependent on previous admission history. Length of time between readmissions was modelled from a full distribution based on real hospital data.
These comparisons between modelling studies probably underlie the differences in results. The HTA model (Ritchie et al 2007) suggests routine screening of all admissions was effective, especially when combined with pre-emptive isolation of high-risk specialty patients. However, long term differences in prevalence levels differed little between screening strategies, but there were substantial variations in five-year costs, with the possibility that that teaching hospitals could save up to £2M and non-teaching hospitals more than £1M over five years by using other risk based screening strategies (checklist activated or specialty based) instead of routine admission screening. The NOW study reports not only that screening all admissions to high risk specialties is in general more cost effective, the savings to be made are considerably greater than estimated by the Scottish HTA model (Ritchie et al 2007).
It is harder to compare Pathfinder and NOW cost effectiveness modelling for two main reasons. Firstly their definition of high-risk specialties was much broader so that 68% of their admissions came under this category (Health Protection Scotland 2011a). If they had used the NOW definition of high-risk specialties the proportion would have been 21% for elective and emergencies combined. This is still nearly twice the proportion in the NOW audit where 11.75% of admissions were to high risk specialties across the entire population of the three trust types (13% if day cases, which are not specifically mentioned in Pathfinder, are excluded) and suggests a difference in clinical practice between England and Scotland.
Secondly, although modelling of screening admission to high risk specialties was undertaken (Health Protection Scotland 2011b, Stewart et al 2011), no cost effectiveness plane, CEAC or CEAF results are presented. Pathfinder found that although routine screening of all admissions reduced infections and acquisitions more than any other strategy, although the difference was not statistically significant. The most cost-effective strategy was checklist activated screening of all admissions (using the three item checklist, with nasal and perineal swabbing of those with at least one risk factor), equivalent to NOW Strategy 4, followed by routine screening of all high-risk specialty patients combined with checklist activated screening of all low-risk admissions (equivalent to NOW Strategy 5), followed by routine screening by nasal swab of all admissions (equivalent to NOW Strategy 2, except that in England screening by nasal swab only is done in less than 10% of trusts). The ICERs quoted all came below the conventional willingness to pay threshold of £30, 000 although the study set no such threshold arguing that this is for policy makers to decide (Stewart et al 2011) . Pathfinder's overall conclusion was that, taking public acceptability and the economic climate into account, combining two swab screening of all admissions to high-risk specialties combined with universal check-list activated screening of all other admissions "appeared to offer the best clinical return for a similar level of financial investment to universal screening of all admissions."

The NOW cost effectiveness results
The relative cost-effectiveness of screening admissions to high-risk specialties probably derives from the difference in transmission parameters in high and low-risk specialties, and the fact that most infections occur in this population (high-risk specialties), who have a higher probability of progressing from colonisation to infection. It is infections that have the largest impact on length of stay and mortality, the largest cost and health benefit determinants. The reduction of these infections combined with lower screening and isolation costs associated with a strategy of screening only admissions to high risk specialties, makes this a more cost-effective strategy, especially in higher prevalence settings.
The differences in cost-effectiveness between acute and teaching trusts reflected the overall greater costs per admission in teaching trusts (causing any strategy to have a higher cost/QALY) and lower transmission and MRSA admission prevalence in the high risk specialty population (meaning the screening strategies had less ability to reduce transmission and infections, and therefore less ability to generate health benefits in teaching trusts). The differences in cost-effectiveness between specialist and acute hospitals reflected higher transmission rates in specialist hospitals meaning that strategies could better prevent transmission and infections and thus generate greater QALY gains, making all strategies better value for money (even the most costly ones, including screening all admissions). The small population size of specialist trusts in particular also meant that stochastic (i.e. chance) effects had a greater impact on the transmission dynamics.
It should also be noted that there existed substantial uncertainty, with the probability of any one strategy being a better option than any other of around only 30% (at willingness to pay values of £20-30,000/QALY).
Overall the results indicate that persisting with the current policy of routine screening of all admissions, does not appear to be cost effective. Reverting to the previous targeted screening strategy of screening only admissions to high-risk specialties may generate substantial savings (on average £250m per year) across the NHS for a very minimal rise in infections (approximately two per year per trust) and colonisations (approximately one per week per trust). The cost-effectiveness of this strategy was maintained even if prevalence increased to twice the current levels.

Generalisability:
Wherever possible model inputs came from audit data to reflect current status of trusts in England, with extensive sensitivity analyses undertaken demonstrating the robustness of findings to Trust type, MRSA prevalence, and assumptions regarding transmission and mortality probability parameters. Parameterisation of the model was also undertaken using audit data wherever possible, thus maximising the reflection of the current status of Trusts in England. This enabled the model to have substantial generalisability. Reassuringly, application of our model under these different analyses indicated that, for the most part, decisions were robust to parameter changes. However, the certainty in choosing between competing strategies was low and changes in costs and effects were clustered.
For the cost-effectiveness evaluations we had taken the perspective of a decision maker who manages resources at a regional or national level, and who seeks to improve the economic efficiency of healthcare services -therefore placing value on bed days freed and QALYs gained. However, from the point of view of clinical or nursing directors for example, the prompter discharge of patients, who have fewer MRSA infections, may facilitate more admissions into the bed base, increasing cost and workload. Factors such as bed management and staffing may be considered more important, and thus change the decision. Conversely, the opportunity to perform contracted work in beds no longer occupied by infected patients may be considered important and provide further support for the decision. The NOW study-the National One Week prevalence audit of MRSA screening The Department of Health (England) has commissioned an independent review of the implementation, clinical and cost-effectiveness, and impact on patient management of the national MRSA screening programme. This review is supported by the Department of Health, the British Infection Association, the Infection Prevention Society, the Hospital Infection Society, the NHS Confederation and the patients' group National Concern for Healthcare Infections.
As of December 2010, national policy is to screen all relevant elective and emergency hospital admissions for MRSA. This policy was based on the best available evidence at the time, but it is important that we understand how policy is being implemented and review it as new data becomes available. This will help the NHS achieve real patient benefits from screening that are appropriate to local circumstances.

This is a real opportunity for you to influence future policy.
Although participation is voluntary, the success of this review, whose primary output will be a report to the Department of Health, depends on your input and we hope that you will take part.

Acknowledgements
Thank you for taking the time to complete this audit. The questionnaire is divided into 6 sections each of which is likely to take approximately one hour to complete, depending on the size of your trust and your data collection systems.
If there are sections or individual questions that you cannot answer please leave them blank.
Section 2: Description of your local policy and practice.  Section 3: MRSA Point Prevalence Audit Q40. If wards share specialties, include only those wards for which more than 50% of patients belong to that specialty. Q41: Siderooms: Include all rooms that are potentially available for isolating patients with MRSA (occupied and unoccupied). Do not include rooms that are currently out of commission i.e. for maintenance work. Q45, 49: MRSA ward: A designated ward for isolation of MRSA patients only. May be purpose built or improvised, with or without controlled ventilation. Isolation ward: A designated ward for isolation of patients with infectious diseases including, but not exclusively used by, MRSA patients.

Instructions
In this section we ask you to provide data on Trust compliance with MRSA screening and proportions of patients found to be MRSA positive on admission/pre-admission screening. Individually named Trust data will not be passed on to the DH or SHAs.

Selecting MRSA negative patients
We ask you to randomly identify 5-10 patients who were found to be MRSA negative on admission or following a pre-admission screen and to complete a form for each of these patients.
It is important that the patients are chosen randomly to prevent bias. We explain the protocol for doing this below. We have given two options for doing this:  Option Arandomising by screen date.  Option Brandomising by admission date.
You can use whichever you find easiest.
Option A: Randomisation by Screen Date STEP 1: Identify all MRSA admission, and pre-admission screens that were logged on Wednesday 11 th May.
STEP 2: Number each screen consecutively on a master list.

USING THE ONLINE RANDOMISER
Please fill in all the required fields as follows: 1) How many sets of numbers do you want to generate? 1 2) How many numbers per set? 5 or 10 (depending on whether you are selecting 5 or 10 patients).
3) Number range (e.g., 1-50) From: 1 To: enter the total number of patients identified on your master list.

6) How do you wish to view your outputted numbers? Place markers off
Click on the Randomize Now! button. You will now see a list of 5 or 10 random numbers.

STEP 4:
Match these numbers to your master list to identify 5-10 patients. And complete a form for each of these patients.

STEP 5:
When results are available check which of the randomised patients have a result confirming that they are MRSA negative. If you have completed a form for a patient who is found to be MRSA positive or whose result is not available, please put a line through the form.

Option B: Randomisation by Admission Date
STEP 1: Identify all admissions (elective and emergency) for Wednesday 11 th May.
STEP 2: Number each screen consecutively on a master list.

USING THE ONLINE RANDOMISER
Please fill in all the required fields as follows: 1) How many sets of numbers do you want to generate? 1 2) How many numbers per set? 5 or 10 (depending on whether you are selecting 5 or 10 patients).
3) Number range (e.g., 1-50) From: 1 To: enter the total number of patients identified on your master list.

6) How do you wish to view your outputted numbers? Place markers off
Click on the Randomize Now! button. You will now see a list of 5 or 10 random numbers.

STEP 4:
Match these numbers to your master list to identify 5-10 patients. And complete a form for each of these patients.

Introduction
In order to address objective 6, the above data were used to inform mathematical models of MRSA transmission in hospital populations. Three models were developed to represent different Trust types; Acute, Teaching and Specialist. Each of the models simulated patient movement within the hospital and between hospital and community populations, transmission of MRSA within the hospital, as well as alternative screening and control strategies.
The screening strategies evaluated using each model were: 1. no screening,

routine screening all Elective and Emergency admissions,
3. screening only admissions to "high-risk" specialties, 4. patient level risk-based screening of all admissions (checklist activated screening {CAS]), 5. screening admissions to "high-risk" specialties plus checklist activated screening of admissions to "low-risk" specialties, 6. routine screening of all admissions with pre-emptive isolation of those known to be previously MRSA positive.
The models were used to simulate MRSA transmission under each of these alternative strategies in order to compare both the effectiveness and cost-effectiveness of control.

The models
An existing dynamic model developed for the DH funded MECAMIP project [Robotham et al 2011], which simulated MRSA transmission within a single hospital ward, was further advanced such that screening and control policies could be evaluated at the whole hospital level, including internal hospital structure and patient journeys through it.

Model Extensions
In order for the models to be appropriate for evaluation of hospital-level screening policies the following major extensions were performed: 5. The development of a whole hospital model comprising a specialty-level structure. This involved the inclusion of two specialty groups within one model and therefore required parameterisation at a specialty level, as well as a whole hospital level, allowing, for example, discharge and death rates to differ between specialties -as well as to reflect whole hospital patterns.
6. The inclusion of realistic patient movements. This involved simulation of patient movement in and out of hospital, as well as between specialties, and inclusion of the readmission process (see 'Patient Movements').
7. Stratification of the admission process such that admissions were classified as Elective or Emergency. This distinction was important due to the differences in proportion of high-risk patients and proportion of colonised patients admitted in these two ways. Additionally, this model development could allow more flexibility in policy evaluation at a later stage e.g.
questions such as 'is screening Elective or Emergency admissions a better use of limited resources?' can be addressed. If a specialty-level discharge was scheduled, the probability of the discharged patient being transferred to another specialty, rather than being discharged fully from hospital, was calculated. If the patient was to be transferred they were placed in a transfer queue until a bed became available.
Until the transfer occurred the patient remained in their original specialty and could be discharged at any point. If a transfer did not occur then the patient was discharged from the hospital into the community population. In this case, the probability of that patient being readmitted and the duration of time between discharge and readmission were calculated, and a readmission was scheduled at some point in the future if appropriate.
Under an assumption of 100% bed occupancy, each discharge or death event triggered an admission to the freed bed, with first preference given to a patient awaiting a transfer, second preference given to a patient scheduled for readmission at that time, or lastly, a new admission was randomly selected from the community population.

Interventions
Screening was performed using a chromogenic agar test (test characteristics described in  (see table 7 appendix 4). A queuing system was simulated such that as soon as space became available in isolation rooms, due to either: a discharge, death or believed recovery event, a known MRSA positive patient undergoing secondary control were moved into the isolation room.
The model structure is represented schematically in Figure D1, and model assumptions listed below.

Summary of model assumptions:
• Admissions may be colonised or susceptible according to prevalence (but not infected).
• Prevalence on admission is dependent on whether the patient is categorised as at risk of MRSA colonisation ( determined by risk-factor analysis i.e. the checklist), as well as whether they are admitted via an Elective or Emergency admission route • No specific assumptions about transmission routes are made.
-The instantaneous risk of a susceptible patient becoming colonised increased linearly with the ward-level MRSA prevalence.
• Colonised and infected patients are equally infectious.
• Transmission parameters (infectiousness of colonised/infected individuals, probability of progression and susceptibility to colonisation/infection) are specialty dependent.
• Direct infection from a susceptible state cannot occur in low risk specialty settings and patients must first become colonised.
• Once MRSA positive, patients remain so for the duration of their stay.
• All infected patients are suspected to be so, with a delay of 1 day before a clinical specimen is taken.
• Recovery may occur in the community.
• At any time patients may belong to either high-risk (HR) or low-risk (LR) specialties -Parameters may differ between specialties -No transmission can occur between specialties.
• Length of stay of colonised and susceptible patients is modelled using the same daily probabilities of discharge, only infection increases length of stay • Similarly, additional mortality is associated with infection only • Daily probability of discharge and death is dependent on whether the patient is in a high-risk or low-risk specialty as well as their infection status.

Uncertainty
For parameters determining effectiveness of the intervention method used, values were defined as probability distributions rather than point estimates. These distributions were chosen to represent the uncertainty in each of the parameters and are assumed independent. In each model simulation run, a parameter value was sampled from these distributions. Since different simulations draw different parameter values from these distributions, the model outcomes also vary between simulations. In this way, uncertainty could be propagated through the model.
In addition to the uncertainty in the parameter values, chance also enters the model due to the stochastic nature of the transmission process. While it is important to account for such stochastic effects when evaluating different strategies, uncertainty in model outcomes should reflect only parameter and structural uncertainties in the models. Therefore, for each sample of parameter values we performed a large number of runs and recorded the mean value of the outcomes of interest. Specifically, we selected 50 parameter sets (each with a different value pulled from the probability distribution for intervention effectiveness) and ran the model 1000 times for each parameter set. Therefore for each strategy, we performed a total of 50,000 model runs.
The model was programmed using the C++ language and open-source libraries. As hundreds of thousands of model runs were required to compare the strategy options; these were performed on a high performance cluster. Analysis and graphical representation of the large amount of generated output was performed in R 2.10.1 (www.r-project.org).

Cost-effectiveness analysis
Health economic data were incorporated into the model to explore the direction and size of changes in economic costs and health benefits due to interventions, through a cost-effectiveness analysis (Graves et al, 2004), allowing comparison of each screening policy. Incorporation of economic parameters into a transmission dynamic model (as opposed to a static model) allows populationlevel effects to be accounted for, such effects are important since preventing infection in one individual directly benefits that individual and indirectly benefits others by preventing transmission.
Analyses failing to take account of indirect effects may underestimate benefits of interventions . Health benefits are described using quality adjusted life years (QALYs). The theory by which health benefits may be evaluated using a dynamic simulation process is outlined in Figure D2.

Fewer infections
Reduced length of stay Reduced mortality More patients going on to accrue quality adjusted life years (QALYs) post-discharge Greater health benefits (measured in QALYs)

Costs
The transmission dynamic model is used to estimate the number of occurrences (over the simulation period) of each event that incurs costs. Important parameters are the monetary valuation for these events and of associated resources, expressed as unit costs.
Essentially, three types of costs incurred were considered in our analyses:  Infection related costs each day whilst an individual is infected  Cost of a bed day  Intervention related costs determined by the particular strategy under consideration e.g.

Health Benefits
For estimating the total health benefits per admission (in terms of quality adjusted life years) a life year accrued by a patient following hospital discharge is the dominating factor. Therefore, an intervention which reduces the number of deaths and increased the number of successful patient episodes would result in a higher life expectancy and greater health benefits. Health benefits gained while in the hospital represent a very minor adjustment to this, but for completeness were considered in our analysis.
Therefore three measurements were required for analysis of health outcomes:  bed days accrued  number of deaths  number of successful patient episodes (number of patients discharged alive) Quality adjusted life years accrued following hospital discharge were taken from the MECAMIP report [Robotham et al 2011], where age and sex matched survival rates were adjusted for quality and discounted at a rate of 5% to give an average quality adjusted life expectancy for a discharged general patient (10 years).
For this evaluation, this estimate was further adjusted to account for long term effects brought about through long term detriment to quality of life in a proportion of MRSA infections.
The proportion of patients who acquired an infection during their hospital stay was calculated.
Assuming that 2.4% of those patients with infections have long-term health loss amounting to a further reduction of 30% (estimates from the Pathfinder study (Smith et al 2011), we then calculated the proportion of discharged patients with no long term health effects, and the proportion discharged with long-term health effects and adjusted the average quality adjusted life expectance on discharge accordingly.

Cost-effectiveness outputs
We conduct a health-economic evaluation to predict outcomes of each strategy in terms of costs and health benefits, measured in Quality Adjusted Life Years (QALYs) (Drummond et al 2005).
Competing interventions are compared against a baseline scenario in terms of their incremental cost-effectiveness ratios (ICER) and their net monetary benefits (NMBs). ICERS are given by the ratio of the change in costs to the change in health outcomes (QALYs) compared to the alternative.
The ratio is interpreted in light of a decision-maker's maximum willingness to pay for a unit of health outcome, such as a QALY. Strategies are considered cost-effective if they generate an ICER) that is less than the current NHS decision makers willingness to pay threshold of £30,000 per QALY [Rawlins et al 2004].
The information used for the ICER can be re-arranged to estimate net monetary benefits (NMB), which are given by ΔE xλ-ΔC. Where Δ is the difference in health benefits between the alternative and baseline strategy, λ the willingness to pay for health benefits, and ΔC the difference in costs between the alternative and baseline strategy. Therefore the NMB represent the value of the health benefits gained minus the amount the amount paid to achieve them.
The perspective for this analysis represents the healthcare decision maker who manages resources at a regional or national level, rather than the perspective of a manager of a single medical ward or hospital. We aim to represent the preferences of high level policy makers who seek to improve the economic efficiency of health care services.
Using the economic transmission model we compare policies in different scenarios and settings, firstly in terms of the clinical effectiveness of each policy (in terms of appropriateness of resource use, and number of transmission, infection and death events) followed by the costs of each policy.
These effectiveness and costs results are then combined and depicted on cost-effectiveness planes and presented as mean ICERs, allowing direct comparison of alternative strategies. Together, these results are useful for understanding how and why costs and health benefits change with different 1. Population parameters obtained from audit data (full tables of audit results and parameter estimations are given in Appendix 4a). As the proportion of patient days spent in an ICU was the same for high and low risk specialties (2.6% of patient days), we grouped ICU with non-ICU ward stays when calculating patient movement parameters. For each group of specialties we calculated the probability of discharge from the ward on their first, second and up to 22nd (and all subsequent) days of stay. The probability of ward discharge on day 1 for example, is calculated from the number of patients discharged from the ward on day 1, divided by the sum of patients discharged from the ward on day 1 and all subsequent days. Specialty dependent discharges from wards were adjusted to account for differences in lengths of stay between infection states according to expert opinion (as described in the MECAMIP report [Robotham et al 2011]), where experts believed that on average, daily probability of discharge for infected patients would be reduced by a mean of 25%. We assumed this adjustment applied across both high risk and low risk specialties.
We then calculated the proportion of patients who on discharge from the ward were subsequently discharged from the hospital for each day of stay. For those patients who were discharged from the ward but not fully from the hospital, we calculated the daily probability that the patient would transfer from a high to low specialty (and vice versa).
Overall discharge patterns were validated against data from the whole hospital (to ensure that using the specialty level estimates were accurately reflecting hospital level patient movements).
Furthermore, specialty dependent daily probabilities of ward discharge for susceptible/colonised patients derived from the individual-level data from the Royal Free hospital, and the adjusted values for daily probability of ward discharge for infected patients, were confirmed to provide comparable estimates to those used for the MECAMIP report (Robotham et al 2011) for general medical wards (Figure 1). If ward discharge occurred, and given that a transfer event was chosen, the probability of which specialty patients were transferred to, given their specialty of discharge is presented in Table 3. less chance of death in general medical wards compared to ICUs) gave a daily death probability of 0.0073.

Figure 1. Daily probabilities of discharge by infection status for high and low risk specialties compared to those used in the MECAMIP model.
For high risk specialties it was assumed that the daily probability of death was equivalent to that in ICUs, derived from robust individual-level data (as described in the MECAMIP report [Robotham et al 2011]), as these were considered to be the best available data on daily probabilities of death. However, as the mortality rate in ICUs may be higher than that in our high risk specialties, we also conducted a sensitivity analysis under much more conservative assumptions, where the daily probability of death in high risk specialities was equivalent to daily probability of death in low risk specialties.
Adjustment to the daily death probabilities according to infection status was carried out according to expert opinion (as in the MECAMIP report, where full details are provided [Robotham et al 2011]), where infected patients were considered on average 22% more likely to die on each day, than susceptible or colonised patients.

Readmission probabilities
We assumed no difference in probability of readmission according to whether the patient was discharged from a high risk or low risk specialty. This assumption can be justified using the more detailed data collected from Sentinel Trusts, where the mean probability of readmission (to any specialty within 30 days of discharge) was 0.29 (CI: 0.19,0.68) for high risk specialty patients and 0.16 (CI: 0.07, 0.31) for low risk (from sample sizes of 468 and 2474 readmitted patients respectively).
Therefore overall readmission probabilities for the whole patient population were calculated from individual-level data from the Royal Free Hospital Trust (as described above). Estimates of the probability of readmission, refined by number of previous hospital visits, are given in Table 5. Table 5. Readmission probabilities given number of admissions within the past 365 days.
Probability of readmission within 365 days, after the: 1st admission 0.21 2nd admission 0.6 3rd admission 0.5 The data were also interrogated to determine length of stay between discharge and subsequent admission, as shown in Figure 2.

Transmission parameters
Transmission parameters (table 6) were taken from the MECAMIP report (Robotham et al 2011), as these were estimated using multi-state modelling techniques which analysed detailed individual-level patient data (for intensive care and general medical settings). These estimates therefore represented the best available evidence for transmission in different settings, and provided sufficient level of detail needed for our individual-based models. We assumed transmission in high risk specialties to be equivalent to that in ICUs, and low risk specialties equivalent to that of hospital general wards. To take into account the possibility that transmission in high risk specialties may not be as high as that seen in intensive care, we conduct a sensitivity analysis where transmission in high risk specialties is assumed to be midway between the level of transmission seen in ICUs and that seen in general hospital wards.
Under the assumption that the risk of a susceptible patient becoming colonised increased linearly with the ward-level MRSA prevalence, and homogenous mixing, MECAMIP (Robotham et al 2011) transmission parameters were adjusted according to Trust dependent specialty sizes. For example, a doubling in the number of beds (i.e. patients) would reduce the probability of any given susceptible patient becoming colonized/infected by half (provided the number of infectious patients remained the same).  Effectiveness in terms of reduction in transmissibility (mean (SD)) 0.24 (0.12) MECAMIP report [Robotham et al 2011]

Health economic parameters
Costs used for the MECAMIP report (Robotham et al 2011) were updated in accordance with Pathfinder cost estimates [Stewart et al 2011]. Appendix 4a. Parameters derived from the NOW audit.        Total number of discharges from "low-risk"** specialties.

Acute trusts
Number of those discharges readmitted to any specialty within 30 days. Number of those discharges readmitted to the same specialty within 30 days. **= All other specialties not mentioned above Total number of discharges from "high risk"* specialties.
Number of those discharges readmitted to any specialty within 30 days. Number of those discharges readmitted to the same specialty within 30 days. * = Haematology/oncology, neurosurgery, nephrology, trauma/orthopaedics, cardiothoracic surgery, vascular surgery.
Total number of discharges from "low-risk"** specialties.
Number of those discharges readmitted to any specialty within 30 days. Number of those discharges readmitted to the same specialty within 30 days. **= All other specialties not mentioned above Instructions:  The infection control team will need to identify all patients flagged as MRSA +ve previous to 1 st April 2010. Include those patients for whom 3 negative screens had not been obtained. This list should then be supplied to the informatics team to run the readmission query.
 Include emergency and elective admissions/discharges  Include only those discharge episodes that occur between 1 st April 2010 and 31 st March 2011.
 Readmission episodes between the 1 st and 30 th of April that are also connected to a discharge occurring in March 2011 should be included in the dataset.

Infection prevention and control teams)
For

SECTION A
Proportions of readmissions were higher for patients originally admitted to HR specialties and for MRSA-ve patients. The lowest proportion of readmissions was for MRSA-ve patients admitted to low risk specialties (21%). The median proportion of readmissions per trust was highest for MRSA-ve patients admitted to HR specialties (49%: IQR 10-74%). This may be explained by the fact that one of the trusts was a tertiary referral centre and had a large number of patients with multiple admissions for renal dialysis. There was relatively little movement between wards of patients admitted to high-risk specialties. 87% remained on one ward during the whole admission compared to 74% of low-risk patients. patients of whom the majority (53%) were transferred to a low-risk specialty within the trust IV Vancomycin represented the largest proportion of this spend (36.7% of the total) closely followed by oral Linezolid.      This is due to the relatively small proportion of admissions going to high risk specialties (16% of beds in Acute Trusts belong to high risk specialties. Strategies 4 and 5 (checklist activated screening (CAS), and screening all high risk specialty patients plus 'CAS of patients in low risk specialties, respectively) perform between the two extremes described above. However, it is worth noting that strategy 4, screening of only 'checklist positive' patients (and isolating those identified as positive), while reducing the amount of appropriate isolation (compared to screening all patients) by ~30%, reduces inappropriate isolation usage by over 50%.
This is because the prevalence in the checklist positive group is approximately 2.6% compared to 1.4% of the overall Acute admission population. Therefore, if isolation capacity is a limiting factor, checklist activated screening may be an option to 'free up' 50% of the isolation capacity.    Figure A3. Cost per admission for each strategy, presented as total costs and broken down into component parts (note different scales of sub graphs).

Cost-effectiveness
To combine these costs and effects in strategy assessment, each strategy can be evaluated on a cost-effectiveness plane. Here, effectiveness of each strategy is measured in terms of health-benefits (measured in QALYs) per admission. A reduction in the number of infections has two effects: 1) it decreases length of stay which decreases cost per admission and 2) it reduces the number of deaths. Both effects will lead to improvements in cost per QALYs gained.
Model results ( Figure A4) confirmed that any investment in screening, compared to no screening, was likely to lead to increases in health benefits. While strategy 6 (screening of all patients plus pre-emptive isolation of those known to be previously MRSA positive) gave the highest health benefits, it was also associated with the greatest costs. Figure A4. Incremental cost-effectiveness plot comparing each of the screening strategies.
Numbers indicate strategy numbers as outlined above. Error bars represent random error brought about by stochasticity in the model and parameter uncertainty, and correspond to plus or minus one standard error. Table A1 describes all strategies in terms of mean change in costs and mean change in health benefits compared to the baseline (strategy 1), and then combines these in terms of a mean cost per QALY gained by changing strategy from the baseline 'do nothing' approach.
Each strategy is also evaluated using the techniques of dominance and extended dominance, allowing some strategies to be eliminated from further evaluation. Dominated strategies are those that are both more costly and provide less benefit than at least one other strategy. An extendedly dominated strategy is one that is more costly and provides less benefit than a combination of another two strategies. Since it would never be cost-effective to pay more for less benefit, these strategies can be excluded from any further evaluation. Incremental cost-effectiveness evaluations are then applied to the remaining strategies, which form the 'cost-effectiveness frontier'. We now evaluate each of the options along the cost-effectiveness frontier. We start at our existing strategy, initially our baseline strategy 1, and then ask whether it is cost-effective (in terms of some threshold of willingness to pay for health benefits) to move from the currently selected strategy to the next most costly strategy on the frontier. This process is iterated, calculating the change in costs and health benefits in moving to the next strategy on the frontier. We stop when no move from the current strategy to a new strategy would be cost-effective according to our threshold. This provides mean incremental cost-effectiveness ratios (ICERs).
Whether a move from one strategy to another is considered cost-effective is dependent on the decision-maker's maximum willingness to pay for a unit of health outcome, such as a QALY. NHS decision makers tend to use values between £20,000 and £30,000 per QALY for a willingness to pay threshold [Rawlins et al 2004.]. Here we use the upper threshold of £30,000 per QALY.
The evaluation of each move along the cost-effectiveness frontier is described in Table A2.
The mean ICERs for strategies 1 and 3 fall beneath the cost-effectiveness threshold. The mean incremental cost per QALY for strategy 3 (screening all admissions to high risk specialties), at £9,964/QALY, was the final strategy to be considered cost-effective and would therefore be the optimal approach (under the specific model parameters and assumptions used). Strategy 5 was only marginally more costly than the £30,000 willingness to pay threshold at £33,806/QALY, and therefore whether to consider this option would be dependent on the decision-makers willingness to pay for health benefits.  The importance of this uncertainty becomes clear when we consider the cost-effectiveness acceptability curves (CEACs), which show the proportion of simulations in which each strategy is cost-effective ( Figure A6). CEACs therefore show the probability of each intervention being cost-effective accounting for the variation in cost/QALY outcome for each model run at each value of willingness to pay. Figure A6 shows substantial uncertainty over which is the most cost-effective strategy as a result of the parameter uncertainty. Figure A6 shows that for willingness to pay values between £20,000 to £30,000 (the usual NHS threshold range) the probability that any one strategy is the most cost-effective option does not exceed 30%. Figure A6. Cost-effectiveness acceptability curves. Each line represents the proportion of simulations, for a particular strategy, that are cost-effective, as a function of willingness to pay for health benefits.
The decision is made slightly clearer by the CEAF (Figure A7), which shows only the strategy with the highest expected NMB over the full range of parameter uncertainty, and describes the probability that this strategy is cost-effective for a given willingness to pay. The CEAF therefore shows which strategy the model suggests we should choose for each willingness to pay. It can be seen that Figure A7 is split into 4 sections, meaning that the optimal strategy changes according to the willingness to pay for health benefits. At a willingness to pay of less than approximately £10,000/QALY, strategy 1 (the 'no screening' approach) is optimal.
However above this, and up to the NHS willingness to pay threshold of £30,000/QALY, strategy 3 (screening of only high risk specialty patients) is the optimal option. Figure A7. Cost-effectiveness acceptability frontier. Lines depict the strategies with the highest expected net monetary benefit, dependent on the willingness to pay for health benefits, while dotted vertical lines the willingness to pay values at which the decision changes.

Scenario analyses
The following results explore each of the strategies in an Acute Trust setting in a series of scenario analyses, namely: i) a high prevalence setting (where prevalence is twice that found in the audit for Acute Trusts: 2.8% compared to 1.4%); ii) a low prevalence setting (where the prevalence is half that found from the audit for Acute Trusts: 0.7% compared to 1.4%); iii) a setting where transmission in high risk specialties is reduced to be between ICU and general medical ward transmission probabilities (see appendix 4 table 6); and iv) a setting where the probability of death is homogenous across the hospital(see appendix 4 Table 4) .
High prevalence setting Figure A8 assesses the screening strategies ability to impose control measures on MRSA positive patients. The pattern seen is the same as that in the baseline prevalence setting ( Figure A1). However, overall the number of MRSA positive isolated bed days per 100 bed days is greater, as there are more MRSA positive patients to isolate, and more unisolated bed days, as more MRSA positive patients means a greater number will be missed.
Interestingly, in the higher prevalence setting there is around the same level of inappropriate isolation, this may be explained through the positive predictive value of the test -even though overall more patients will be isolated (and therefore more also inappropriately isolated) this will be offset as the chance of a false positive will be reduced.
The greater prevalence on admission leads to a slightly greater level of transmission (due to more infectious imports to the hospital) and, in turn, slight increases in the absolute numbers of infections and deaths ( Figure A9).
Differences in costs between the higher prevalence setting and baseline prevalence setting are minimal ( Figure A10), and when costs and effects are combined on the costeffectiveness plane it can be seen that overall health benefits per cost accrued are similar to those at a baseline prevalence setting ( Figure A11 compared to Figure A4). Overall, strategies provide slightly greater health benefits, as they are able to prevent more transmission events, leading to fewer infections and therefore deaths. While strategy 6 (screening all patients plus pre-emptive isolation of those known to have previously been MRSA positive) appears to do less well than strategy 2 (screening all patients) in this setting, it can be seen from the error bars that there is considerable uncertainty and overlap between the two policies. This uncertainty is dealt with in the CEAC and CEAF plots.    Evaluation of the cost-effectiveness frontier shows that despite the ICERs shifting slightly, the decision remains the same in the higher prevalence setting. With strategy 3 (screening of patients admitted to high risk specialties) the only costeffective option (at a willingness to pay threshold of £30,000/QALY gained) (Table   A3). Very similar levels of uncertianty in the decision between strategies can be seen in a higher prevalence setting ( Figure A12), compared to the baseline prevalence setting ( Figure A6). However at higher willingness to pay values (greater than approximately £30,000/QALY), strategy 2, screening all patients, has a greater probability of being cost-effective compared to the competing options. Figure A12. Cost-effectiveness acceptability curves. Each line represents the proportion of simulations, for a particular strategy, that are cost-effective, as a function of willingness to pay for health benefits.
The CEAF ( Figure A13) demonstrates that, as in the baseline prevalence setting, strategy 3 (screening all patients admitted to high risk specialties) is optimal within the usual NHS willingness to pay range of £20,000-£30,000/QALY. As the CEAC indicated ( Figure A12) Low prevalence setting Figure A14 shows that the overall pattern of the ability of strategies to identify MRSA positive patients and impose control remains the same in a low prevalence setting, compared to the baseline prevalence setting ( Figure A1). As might be expected, as there are fewer MRSA positive patients admitted to the hospital, overall there is less isolation and fewer unisolated patient days. However, the degree of inappropriate isolation increases slightly with a lower admission prevalence due to the lower positive predictive value of the test in this setting.
Again, while comparison between policies remains the same, in a lower prevalence setting there is less within hospital transmission, leading to fewer infections and deaths, due to a lessened infectious assault on the hospital ( Figure A15).
Interestingly, costs associated with each of the screening strategies are very slightly higher in a low prevalence setting ( Figure A16), compared to the baseline prevalence setting ( Figure A3), by approximately £200 per admission). This is due to the fact that in a lower prevalence setting for the same level of screening effort the yield will be less, meaning the screening costs will not be offset to the same extent by cost savings brought about though reduction in transmission (and therefore infections).   On the cost-effectiveness plane ( Figure A17) the ordering of strategies remains the same under a lower prevalence. ICERs, as described in the evaluation of the costeffectiveness frontier (Table A4) change only slightly with strategy 3 (screening of patients admitted to high risk specialties) remaining cost-effective. However, in a low prevalence setting, checklist activated screening in addition to screening all admissions to high risk specialties (strategy 5) becomes a cost-effective option.  However, again, as demonstrated in the CEAC ( Figure A18) there is considerable uncertainty in the decision, where within the NHS willingness to pay range no strategy has more than a 30% chance of being cost-effective. From the CEAF ( Figure   A19) it can be seen that the decision varies within the NHS range: screening patients admitted to high risk specialties (strategy 3) being optimal at the lower end of the willingness to pay range, and adding checklist activated screening of patients in low risk settings to this policy (strategy 5) at the higher end of the range.  Low transmission in high risk specialties A change in the degree of onward transmission (and probability of infection) within the hospital itself has no effect on the ability of each screening strategy to effectively isolate colonised patients, as screening and identification of colonisation occurs on admission to hospital ( Figure A20). The slightly greater numbers of appropriate isolation events seen in this setting of lower transmission/infection compared to that at baseline transmission ( Figure A1) is due to fewer numbers of infections and therefore fewer infected patients identified (and subsequently isolated) via clinical samples (as opposed to screens).
It can be seen from Figure A21 (compared to Figure A2 for the baseline transmission setting) that the reduction in transmission parameters in high risk settings has the most marked difference on the overall number of infections (which are approximately halved) and, in turn, deaths.
Total costs for the strategies ( Figure A22) are greater than under baseline transmission parameters ( Figure A3) as costs are not offset to the same degree by savings through reductions in infections.
From the cost-effectiveness plane ( Figure A23)it can be seen that compared to the baseline strategy of no screening , none of the screening strategies provide the same gains in health benefits as are achieved in a setting with increased transmission in high risk specialties ( Figure A4). However, despite a higher ICER, strategy 3 (screening all patients admitted to high risk specialties) is still cost-effective at £12,382/QALY (Table A5).  The CEAC ( Figure A24) shows strategy 3 (screening admissions to high risk specialties) to have the greatest probability of being cost-effective within the NHS willingness to pay range, compared to the competing strategies -with an almost 60% of being cost-effective. This is reflected in the CEAF ( Figure A25) where strategy 3 is the optimal option from willingness to pay values of just over £10,000/QALY to nearly £90,000/QALY. Figure A24. Cost-effectiveness acceptability curves. Each line represents the proportion of simulations, for a particular strategy, that are cost-effective, as a function of willingness to pay for health benefits.

Homogenous probability of death across all specialties
In a setting where the daily probability of death in high risk specialties is reduced to be equivalent to that in low risk specialties, there will be little, or no, difference to the effectiveness of interventions in terms of reduction in transmission or infection events. The differences arise due to differences in long term QALY accrual, as death events prevent patients being discharged successfully and going on to accrue QALYs post discharge. We therefore present results, and compare policies, in terms of costs per QALY gained only.
As, under the revised death parameters, prevention of infections has less impact on deaths in high risk specialties, each of the screening strategies has a reduced ability to prevent death events and therefore reduced ability to gain health benefits (QALYs). As can be seen from the incremental cost-effectiveness plot ( Figure A26) the health benefits gained under each of the strategies is reduced by approximately 10-fold (compared to baseline death parameters, Figure A4). This, in turn, is translated to the ICER values, where cost/QALY values are much greater (Table A6 compared to Table A2).
However, even with these conservative assumptions around mortality in high risk specialties, strategy 3 remains cost-effective at £26,551/QALY. Indeed, the CEAC ( Figure A27) depicts strategy 3 to have an approximately 60% probability of being cost-effective within the NHS willingness to pay threshold. The CEAF ( Figure A28), which accounts for the magnitude of the potential benefit as well as the probability of cost-effectiveness, shows the decision changes dependent on willingness to pay.
At the lower end of the NHS willingness to pay range the baseline 'no screening' approach is optimal, up to approximately £25,000/QALY, while strategy 3 becomes optimal above this value. Figure A26. Incremental cost-effectiveness plot comparing each of the screening strategies. Numbers indicate strategy numbers as outlined above. Error bars represent random error brought about by stochasticity in the model and parameter uncertainty, and correspond to plus or minus one standard error.

Comparison with Scottish Checklist
Simulation results for strategies 4 (checklist activated screening patients) and 5 (screening admissions to high risk specialities plus checklist activated screening of Through an incremental cost-effectiveness plot ( Figure A29) the differences in both costs and effects for each of the strategies can be compared in a rational way, through incremental comparison with the baseline 'no screening' strategy (strategy 1). As would be expected, the incremental costs (as compared to the baseline) for both strategies 4 and 5 are markedly reduced. This effect is likely to be largely due to the smaller number of patients being classified as checklist positive, and thus a reduction in numbers screened. This reduction in incremental costs is less pronounced for strategy 5 as in this strategy all high risk specialty admissions are still screened, irrespective of the definition of checklist positivity. Putting these changes to incremental costs and incremental effects (or health benefits) together, as both costs and effects for checklist-activated screening strategies (strategies 4 and 5), the overall incremental cost-effectiveness evaluations of the cost-effectiveness frontier (Table A7) are very similar to those found using the audit definition of checklist positivity. Strategy 3, screening admissions to high risk specialties, remains the only strategy to be cost-effective at £9,731 /QALY gained.  Figure A30 compared to Figure A6). The CEAF ( Figure A31) shows strategy 3, screening of admissions to high risk specialties, remains the optimal option for all values of willingness to pay in the £20,000-£30,000/QALY range, albeit with a level of certainty in this decision of less than 30% (due to uncertainty in the effectiveness of the intervention). Figure A30. Cost-effectiveness acceptability curves. Each line represents the proportion of simulations, for a particular strategy, that are cost-effective, as a function of willingness to pay for health benefits.

Effectiveness
In terms of the ability of the screening strategies to identify MRSA positive patients, the relative differences between strategies are approximately the same for Teaching Trusts ( Figure B1), as were seen in Acute Trusts ( Figure A1), with strategies 2 and 6 having the highest levels of isolation (both appropriate and inappropriate usage) and fewest unisolated MRSA positive bed days. However, due to a combination of a slightly reduced overall prevalence on admission in Teaching Trusts and lower transmission parameters leading to fewer clinical cases, absolute numbers of isolated bed days (per 100 bed days) are slightly lower in Teaching Trusts.

Due to the method by which transmission parameters have been estimated for each
Trust type (see Model Parameters section), there is less onward transmission in this setting, as shown in Figure B2, meaning that there are fewer unisolated MRSA positive bed days in Teaching Trusts compared to Acute Trusts ( Figure B1 compared to Figure A1). This is because each of the screening strategies is implemented on the admission population only, and therefore unable to influence the degree of onward transmission in the hospital caused by MRSA positive patients who are missed by screening. If these 'missed patients' go on to cause less onward transmission, as is the case here, it follows that there will be fewer unisolated MRSA positive days.
The reduced levels of onward transmission, infections events, and thus death events, are demonstrated in Figure B2 (as compared to A2). However, again, while absolute numbers may vary between Trust types, the ordering of the strategies remains the same.

Costs
Compared to an Acute setting, in a Teaching Trust setting each of the strategies are associated with greater costs per admission ( Figure B3 for Teaching, compared to Figure A3 for Acute). This is due to the larger size of the Teaching Trusts (a median of 1113 beds, compared to 553 for Acute) leading to greater numbers of bed days over the simulation period. Overall bed day costs for the 5 year simulation period in Acute Trusts are 50% less than those for Teaching Trusts. Therefore the numerator in the cost per admission equation is halved for Acute Trusts. Whereas, the differences in the number of admissions over the simulation period (the denominator) are not so pronounced, with Acute Trusts having only 22% fewer admissions compared to Teaching Trusts). Together, these two effects lead to Teaching Trusts having greater costs per admission.
The lower numbers of admissions to Teaching Trusts than might be expected given their size, are due to the distribution of high risk and low risk specialty beds -Teaching Trusts having approximately 44% more high risk specialty beds than Acute Trusts. Having more high risk specialty patients will result in altered movement parameters within the hospital and between the hospital and community (for movement parameters see Appendix 4).
It is worth noting that while the absolute costs of the screening itself (as a component of the total costs) will be greater in Teaching Trusts due to larger number of patients needing to be screened, the costs per admission are identical to those in Acute Trusts ( Figure B3, 'Combined screening costs' panel; compared to Figure A3).

Cost-effectiveness
From the incremental cost-effectiveness plot ( Figure B4), it can be seen that each of the strategies generates fewer health benefits (compared to the baseline 'no screening' strategy, strategy 1) than are achieved in Acute settings ( Figure A4). This is due to the lower transmission seen in a Teaching Trust setting ( Figure B2).
However, despite the fact that fewer QALYs are gained, strategy 3 (screening patients admitted to high risk specialties) -the most cost-effective option on the cost-effectiveness frontier in Acute Trusts (Table A2) -is only approximately £1,000 more costly than the NHS willingness to pay threshold of £30,000/QALY in this setting, at £31,077/QALY (Table B1).

Consideration of uncertainty
It must be noted that the degree of uncertainty in this decision is large (depicted graphically in Figure B5). Considering this uncertainty the CEAC ( Figure B6) demonstrates that within the NHS willingness to pay range of £20,000 -£30,000/QALY, no one screening strategy has a greater than 30% chance of costeffectiveness. Indeed, the CEAF ( Figure B7) shows that considering all uncertainty, the optimal option is the isolation and decolonisation of clinical cases only. The decision only changing at willingness to pay values slightly over £30,000/QALY.
However, under only slightly different assumptions for transmission parameters it is likely that the cost-effectiveness of strategies would shift, which (given the proximity of the decision to the threshold), would be likely to change the decision. Figure B5. Simulation results comparing each strategy on a cost-effectiveness plane for a Teaching Trust setting. Each dot represents the mean of 1000 simulation runs for each parameter set. The results of 50 parameter sets for each strategy are plotted, where each parameter set is obtained by taking the mean value for all parameters apart from the effectiveness of the intervention, which is sampled from its probability distribution. Figure B6. Cost-effectiveness acceptability curves. Each line represents the proportion of simulations, for a particular strategy, that are cost-effective, as a function of willingness to pay for health benefits. Figure B7. Cost-effectiveness acceptability frontier. Lines depict the strategies with the highest expected net monetary benefit, dependent on the willingness to pay for health benefits, while dotted vertical lines the willingness to pay values at which the decision changes.

Scenario analyses
The following results explore each of the strategies in a Teaching Trust setting in a series of scenario analyses, namely: i) a high prevalence setting (where prevalence is twice that found in the audit for Teaching Trusts: 2.6% compared to 1.3%); and ii) a low prevalence setting (where the prevalence is half that found from the audit for Teaching Trusts 0.65% compared to 1.3%).

High prevalence setting
As would be expected in a high prevalence setting, both appropriate and inappropriate isolation increase, compared to in a baseline prevalence (Figure B8 compared to Figure B1), due to a greater absolute burden of colonisation, meaning more MRSA positive patients will be both 'picked up' and missed through any screening approach. As in an Acute setting, despite more patients being screened and isolated the degree of inappropriate isolation remains the same as in a baseline prevalence setting, potentially due to the greater positive predictive value of the screening test in high prevalence settings.
The higher prevalence leads to greater numbers of transmission events, MRSA infections and therefore death events per 100 admissions ( Figure B9) compared to the baseline prevalence setting ( Figure B2). Total costs are reduced slightly ( Figure   B10 compared to Figure B3), due to screening strategies being more able to impact transmission and therefore reduce infections leading to cost savings. Indeed, each of the screening strategies (compared to the baseline strategy) are able to provide greater gains in health benefits in a high prevalence setting ( Figure B11), compared to gains seen in the baseline prevalence setting ( Figure B4). These greater gains in health benefits have the consequence of making the ICER for the optimal strategy, screening patients admitted to high risk specialties (strategy 3), cost-effective at £27,715/QALY.    Not cost effective Figure B12. Cost-effectiveness acceptability curves. Each line represents the proportion of simulations, for a particular strategy, that are cost-effective, as a function of willingness to pay for health benefits. Figure B13. Cost-effectiveness acceptability frontier. Lines depict the strategies with the highest expected net monetary benefit, dependent on the willingness to pay for health benefits, while dotted vertical lines the willingness to pay values at which the decision changes.
In a low prevalence of setting the amount of intervention usage is reduced ( Figure   B14), as are transmission, infection and death events ( Figure B15), due to the lower infectious assault on the hospital. This translates to very slightly higher total costs as, in a setting of reduced infections, the same level of effort will yield less of an impact ( Figure B16). From the incremental cost-effectiveness plot ( Figure B17), it can be seen that incremental gains in health benefits are lower (compared to the baseline strategy) in a lower prevalence setting, compared to an average prevalence setting ( Figure B4). The incremental effectiveness, in terms of QALY gain, of strategy 2 in particular, screening all patients on admission, is reduced in a low prevalence setting.
Overall, with a lower prevalence on admission none of the strategies are costeffective. It is much harder for the screening strategies to be cost-effective, with ICERs (Table B3) for all strategies on the cost-effectiveness frontier falling above the cost-effectiveness threshold of £30,000/QALY.
In this case, the CEAC ( Figure B18) shows that up to a willingness to pay of approximately £40,000/QALY none of the screening strategies are cost-effective, with strategy 1, isolating only clinical cases, being optimal (as seen from the CEAF, Figure B19).    Figure B18. Cost-effectiveness acceptability curves. Each line represents the proportion of simulations, for a particular strategy, that are cost-effective, as a function of willingness to pay for health benefits.

Effectiveness
All strategies isolate greater numbers of patients per 100 bed days in a Specialist setting, compared to in an Acute Trust setting ( Figure C1, compared to Figure A1).
Very similar numbers of patients are isolated in both Trust types, however the denominator of total patient days is much less over the simulation period in Specialist Trusts due to their smaller size.
The greater number of infections in this setting ( Figure C2) compared to an Acute setting (brought about due to the method of transmission parameter estimation -(see appendix 4), translates to increased isolation of clinical cases, leading to greater appropriate and reduced inappropriate isolation usage ( Figure C1).
Interestingly, compared to Acute Trusts, the relative difference between strategies 3 and 4 is altered in terms of their ability to correctly isolate patients. These differences are due to the difference in distribution of high risk specialty vs. low risk specialty beds: 44% of beds are in high risk specialties in Specialist Trusts, compared to 16% in Acute Trusts. Therefore in Specialist Trust settings, strategy 3, screening admissions to high risk specialties, will screen a greater proportion of the admission population -leading to greater appropriate and inappropriate isolation. In fact, this different distribution between specialties, coupled with the slightly lower prevalence on admission to Specialist Trusts, strategy 3 actually leads to more inappropriate isolation compared to strategy 4 in this setting ( Figure C1), whereas the opposite is seen in Acute Trusts ( Figure A1).
There are greater numbers of unisolated MRSA positive bed days per 100 bed days in Specialist Trusts compared to Acute Trusts, due to the greater levels of onward transmission in the Specialist setting ( Figure C2 compared to Figure A2), meaning that any cases missed by the admission screening strategies will go on to generate more cases within the hospital.
While the ordering of strategies remains the same as in Acute Trusts in terms of their effectiveness at reducing transmission, there is overall a greater level of transmission in a Specialist setting ( Figure C2). This is due to both the way in which

Costs
Overall costs per admission in Specialist Trusts (Figure C30 are less than in acute Trusts ( Figure A3). As was the case for Teaching Trusts compared to Acute Trusts, the reduction in bed day costs (the dominant cost) are proportional to the difference in Trust size i.e. bed day costs in Specialist settings are approximately 72% less than in an Acute setting, due to a 72% smaller Trust size. However, the difference in the number of admissions over the simulation period (the denominator in costs/admission) is not proportional to the difference in Trust size. Instead, due to more high risk specialty patients in Specialist Trusts, and therefore altered movement parameters, Specialist Trusts have only 59% fewer admissions. This discrepancy in the degree of change in the numerator vs. the denominator in the cost/admission calculation leads to overall costs/admission being less in Acute Trusts. Figure C4. Incremental cost-effectiveness plot comparing each of the screening strategies. Numbers indicate strategy numbers as outlined above. Error bars represent random error brought about by stochasticity in the model and parameter uncertainty, and correspond to plus or minus one standard error.  Figure C5. Simulation results comparing each strategy on a cost-effectiveness plane for a Specialist Trust setting. Each dot represents the mean of 1000 simulation runs for each parameter set. The results of 50 parameter sets for each strategy are plotted, where each parameter set is obtained by taking the mean value for all parameters apart from the effectiveness of the intervention, which is sampled from its probability distribution. Figure C6. Cost-effectiveness acceptability curves. Each line represents the proportion of simulations, for a particular strategy, that are cost-effective, as a function of willingness to pay for health benefits. Figure C7. Cost-effectiveness acceptability frontier. Lines depict the strategies with the highest expected net monetary benefit, dependent on the willingness to pay for health benefits, while dotted vertical lines the willingness to pay values at which the decision changes.

Scenario analyses
The following results explore each of the strategies in a Specialist Trust setting in a series of scenario analyses, namely: i) a high prevalence setting (where prevalence is twice that found in the audit for Specialist Trusts: 2.1% compared to 1.04%); and ii) a low prevalence setting (where the prevalence is half that found from the audit for Specialist Trusts: 0.52% compared to 1.04%).

High prevalence setting
In high prevalence setting Specialist Trusts, the relative ability of each of the strategies to effectively identify and therefore isolate (and decolonise) MRSA positive patients ( Figure C8) remains the same: the ordering of strategies can be seen to be the same as in an average prevalence setting ( Figure C1). However, the overall degree of appropriate isolation increases slightly, due to the higher infectious assault. Conversely, inappropriate isolation decreases slightly, due to the improved positive predictive value of the test under higher prevalence conditions (and due to the isolation of more clinical cases). Unisolated MRSA positive bed days per 100 bed days increase marginally due to more missed cases, and more onwards transmission and infections (see Figure C9 for transmission levels).
Only very slight increases ion transmission are seen in response to the increase in colonised admissions to the hospital, compared to baseline prevalence conditions ( Figure C9, compared to Figure C2). Translating also to slight increases in the number of infection and death events per 100 bed days. It is worth noting that, compared to the other strategies, strategy 2, screening all patients on admission, appears to fare slightly better in a higher prevalence setting in terms of reduction in infection events.
As might be expected, total costs per admission are slightly reduced in a higher prevalence setting ( Figure C10, compared to Figure C3) due to all strategies having the ability to prevent more infection events and thus produce cost-savings.  Putting these costs and effects together for a high prevalence setting, and comparing strategies to the baseline 'no screening; approach (strategy 1), the incremental costeffectiveness plot ( Figure C11) demonstrates that strategies are much more tightly clustered in terms of both incremental costs and incremental health benefits than in an average prevalence setting. Due to this tight clustering, the choice between strategies becomes more difficult. While evaluation of the cost-effectiveness frontier (Table C2) shows strategy 3, screening patients admitted to high risk specialties, is cost-effective at £9,745/QLAY (slightly better value than in a baseline prevalence setting) the CEAC ( Figure C12) demonstrates the uncertainty in this decision. Taking this uncertainty into account, the optimal policy, i.e. the policy that yields the greatest monetary net benefits, is strategy 3 for all values of willingness to pay above approximately £10,000/QALY. Figure C11. Incremental cost-effectiveness plot comparing each of the screening strategies. Numbers indicate strategy numbers as outlined above. Error bars represent random error brought about by stochasticity in the model and parameter uncertainty, and correspond to plus or minus one standard error.  Figure C12. Cost-effectiveness acceptability curves. Each line represents the proportion of simulations, for a particular strategy, that are cost-effective, as a function of willingness to pay for health benefits. Figure C13. Cost-effectiveness acceptability frontier. Lines depict the strategies with the highest expected net monetary benefit, dependent on the willingness to pay for health benefits, while dotted vertical lines the willingness to pay values at which the decision changes.
As for the other Trust types, in a low prevalence Specialist Trust, the level of appropriate isolation and MRSA positive bed days decreases compared to an average prevalence Specialist Trust, while inappropriate isolation increases marginally ( Figure   C14).
While the ordering of strategies, in terms of the level of transmission, remains the same as in the average prevalence setting, under low prevalence conditions absolute numbers of transmission events per 100 bed days are reduced, leading in tern to fewer infection and death events ( Figure C15).
Again, as seen for different prevalence scenario analyses for the other Trust types, a lower prevalence leads to higher costs due to the same level of intervention effort leading to the identification of fewer MRSA positive patients and therefore reduced ability to prevent infections and provide cost-savings ( Figure C16).
Overall, these small differences in effects and costs leads to similar incremental cost -effectiveness values in a low prevalence Specialist Trusts ( Figure C17) as seen in an average prevalence Specialist Trust ( Figure C4). However, a slight shift in the ordering of policies means that while strategy 3, screening of high risk specialty admissions, remains cost-effective at £10,566/QALY (albeit at a slightly higher cost/QALY than in an average prevalence setting), strategy 2, screening of all patients, becomes part of the cost-effectiveness frontier and cost-effective (Table   C3).
Indeed, strategy 2 is shown to be the optimal options for all willingness to pay values above approximately £15,000/QALY ( figure C19), but the uncertainty in this decision is high ( Figure C18) with the probability of any of strategies 2,3,5 or 6 having an approximately 25% chance of being the most cost-effective within the usual NHS willingness to pay range (£20,000-£30,000/QALY).      Figure C18. Cost-effectiveness acceptability curves. Each line represents the proportion of simulations, for a particular strategy, that are cost-effective, as a function of willingness to pay for health benefits.