Skip to main content
Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Improving mathematical modeling of interventions to prevent healthcare-associated infections by interrupting transmission or pathogens: How common modeling assumptions about colonized individuals impact intervention effectiveness estimates

  • Camden D. Gowler,

    Roles Formal analysis, Methodology, Software, Writing – original draft

    Affiliations Division of Healthcare Quality Promotion, Centers for Disease Control and Prevention, Atlanta, Georgia, United States of America, Department of Ecology and Evolutionary Biology, University of Michigan, Ann Arbor, Michigan, United States of America

  • Rachel B. Slayton,

    Roles Supervision, Writing – review & editing

    Affiliation Division of Healthcare Quality Promotion, Centers for Disease Control and Prevention, Atlanta, Georgia, United States of America

  • Sujan C. Reddy,

    Roles Supervision, Writing – review & editing

    Affiliation Division of Healthcare Quality Promotion, Centers for Disease Control and Prevention, Atlanta, Georgia, United States of America

  • Justin J. O’Hagan

    Roles Conceptualization, Formal analysis, Methodology, Software, Writing – review & editing

    justinohagan@gmail.com

    Affiliation Division of Healthcare Quality Promotion, Centers for Disease Control and Prevention, Atlanta, Georgia, United States of America

Abstract

Mathematical models are used to gauge the impact of interventions for healthcare-associated infections. As with any analytic method, such models require many assumptions. Two common assumptions are that asymptomatically colonized individuals are more likely to be hospitalized and that they spend longer in the hospital per admission because of their colonization status. These assumptions have no biological basis and could impact the estimated effects of interventions in unintended ways. Therefore, we developed a model of methicillin-resistant Staphylococcus aureus transmission to explicitly evaluate the impact of these assumptions. We found that assuming that asymptomatically colonized individuals were more likely to be admitted to the hospital or spend longer in the hospital than uncolonized individuals biased results compared to a more realistic model that did not make either assumption. Results were heavily biased when estimating the impact of an intervention that directly reduced transmission in a hospital. In contrast, results were moderately biased when estimating the impact of an intervention that decolonized hospital patients. Our findings can inform choices modelers face when constructing models of healthcare-associated infection interventions and thereby improve their validity.

Introduction

Healthcare-associated infections (HAIs) impose a substantial burden of disease and can increase healthcare costs considerably [15]. However, it is frequently challenging to estimate the effectiveness of prevention measures [6, 7]. Key barriers include the fact that there is generally a low incidence of infections in any one hospital, and therefore individual facilities have little opportunity to quantify the impact of an intervention in a short study. Multicenter studies have been conducted to help address this issue, but their cost can be prohibitive and the length of follow-up can be relatively short [8, 9]. Further complicating the problem, patients transfer between hospitals, nursing homes, other healthcare settings, and the broader community; thus, transmission and symptom onset in non-hospital settings can complicate efforts to evaluate the impacts of hospital-based interventions [7, 1012].

To address such challenges, investigators sometimes use simulations to help assess whether a previous intervention was effective or whether a new intervention is likely to be successful [13, 14]. Transmission models are relatively inexpensive and quick to create, but require additional assumptions compared to traditional epidemiologic approaches (e.g., randomized trials, cohort studies) [15, 16]. Concerningly, even seemingly innocuous assumptions that are introduced to simplify an analysis have the potential to alter results in ways that modelers did not expect. For instance, modelers often simulate asymptomatically colonized individuals as having higher rates of hospital admission [17, 18] or longer lengths of hospital stay [1935] compared to uncolonized individuals because they are asymptomatically colonized. While it is true that colonized individuals on average spend more time in hospitals per admission and are more readily admitted, there are likely several confounding factors that can explain the association [36]. For example, people asymptomatically colonized with an HAI causing pathogen tend to be older and sicker than non-carriers [37]. Consequently, a more plausible explanation is that people in poor health are more likely to be hospitalized and to have long lengths of stay, and thereby are more likely to become colonized with a HAI causing pathogen. Reverse causation can also explain some of the crude association between carriage status and length of stay [38]. As a result, transmission models could provide inaccurate estimates of intervention impact when asymptomatic carriage is simulated in these ways.

Therefore, our primary aim was to test whether the assumptions regarding higher rates of hospital admission or longer lengths of hospital stay for asymptomatic carriers can affect the results of simulations that evaluate the effects of interventions that work by preventing transmission of HAIs interventions.

Methods

General model description

We created a deterministic compartmental model with Uncolonized, Colonized (asymptomatic carriers), and Symptomatic disease states to simulate transmission of an HAI pathogen in a hospital and the surrounding community (Fig 1). For each state, we included separate compartments for each combination of location (hospital or community), carrier type (short or long durations of carriage), and risk group. In our examples, we used two age groups (persons aged 0 to 64 years and adults 65 or older) to define the risk groups since older age is generally recognized as a proxy for increased rates of hospitalization and longer lengths of stay, and age-specific hospitalization data are widely available (Table 1) [39]. We parameterized the model so that adults 65 or older spent more time in the hospital than persons aged 0 to 64 years because of their higher admission rates and longer lengths of stay (Table 2). Individuals could move between the hospital and community but, to simplify the model, they could not transition between age groups or carrier types. We parameterized the model using data for methicillin-resistant Staphylococcus aureus (MRSA, Table 2, [29, 40]). Short-term and long-term carrier types were based on the observed heterogeneity in carriage duration for S. aureus [41, 42].

thumbnail
Fig 1. Flow diagram of general model.

A flow diagram of the model with compartments for each combination of (i) Uncolonized (U), Colonized (C), or Symptomatic (S); (ii) persons aged 0 to 64 years or adults 65 or older; (iii) short-term or long-term carrier types; and (iv) subscript denoting location in a hospital (1) or community (2). Gray arrows show movement between the hospital and community, through admission and discharge, whereas black arrows show transmission, disease progression, recovery from disease, and loss of carriage. Solid and dashed borderlines denote compartments for each age strata, and gray shading denotes the long-term carrier compartments. Individuals could not move between age groups or carrier types. Full model equations are in S1 Appendix.

https://doi.org/10.1371/journal.pone.0264344.g001

thumbnail
Table 1. Notation and descriptions of total population sizes.

https://doi.org/10.1371/journal.pone.0264344.t001

Transmission could occur between patients in the hospital or between individuals in the community, and the infectiousness of those carrying the pathogen was allowed to differ between the hospital (β1) and community (β2). Colonized and symptomatic individuals in the same location were assumed to be equally infectious (i.e., had the same β1 or β2 values). However, contact rates could be assortative by age group, meaning individuals of one age group were more likely to contact members of the same age group than expected if contacts had occurred randomly. The degree of this assortativity could vary by location (δ1, δ2). Transmission from adults aged 65 or older in the community was reduced by a proportion (σ) to reflect the fewer overall number of social contacts of older adults compared to younger people [43]. Once colonized, individuals could become symptomatic at a rate α, and this rate could depend on age group and location. Asymptomatic carriers could clear colonization at rate γa for short-term carriers, or γb for long-term carriers. Symptomatic cases returned to the colonized state when they recovered from infection at rate γS.

Admission (a) and discharge (r) rates could depend on age group and whether the person was uncolonized, colonized, or symptomatic. Admissions and discharges were balanced to keep hospital and community sizes constant. The lengths of stay in the hospital and rates of admission for colonized, symptomatic, or adults aged 65 or older could be equal to or higher than the rate of admission of uncolonized, persons aged 0 to 64 years according to the following admission rate multipliers: adults aged 65 or older (ρ>65), colonized (ρC), or symptomatic (ρS). These multipliers interacted multiplicatively for individuals with more than one risk factor for admission (e.g., adults aged 65 or older who were colonized were admitted at a rate ρ>65C times that of uncolonized, persons aged 0 to 64 years).

We modeled two types of hospital interventions: one that decreased the infectiousness of carriers (hereafter termed “transmission-based intervention”) and another that increased the rate at which asymptomatic carriers cleared carriage (hereafter termed “decolonization”). These interventions can be thought of as simplifications of hand hygiene or use of barrier precautions (transmission-based interventions) and the use of antiseptic bathing and nasal decolonization (decolonization). To simulate the transmission-based intervention, the parameter θ proportionally reduced the infectiousness of asymptomatically colonized and symptomatic carriers in the hospital. The parameters γint,0–64 and γint,>65 represented the additional clearance rates due to hospital-based decolonization intervention. We assumed that the duration of symptoms would not be affected by the decolonization intervention.

Sub-models for comparison

We used this general model to test the effects of different assumptions regarding asymptomatic carrier admission rates and lengths of stay on estimates of intervention impact. To help motivate the problem, we can consider the following hypothetical scenario. Suppose there are limited data suggesting a particular HAI intervention has been effective during outbreaks, but it has not been studied in endemic settings. Consequently, researchers want to assess the possible impact of the intervention in hospitals where the pathogen is endemic in order to inform decisions about the likely generalizability of empirical results. Ideally, the simulation should meet some basic criteria in order for the example to be plausible (e.g., the simulated prevalence of the pathogen in the hospital and community should match data from well-designed, large-scale screening studies). However, there are many different ways to make the simulation fit the observed data, and certain approaches could potentially skew results. Of particular concern are common modeling assumptions that asymptomatic carriers have higher rates of hospital admission or longer lengths of stay precisely because they are asymptomatically colonized, which have no biological basis. We analyzed a set of models to help inform modeling decisions in such a scenario.

We first devised parameter values to create five sub-models for comparison (Table 2). (1) Age Group model: we stratified the population into two age strata: persons 0 to 64 years old and adults aged 65 years and older, where older individuals had higher admission rates and longer lengths of stay in the hospital. No other model stratified the population into age groups. In the next three models, asymptomatically colonized individuals had (2) higher admission rates (Colonized Admission model), (3) longer length of stays (Colonized LOS model), or (4) both higher admission rates and longer lengths of stays (Colonized Admission + LOS model) compared to uncolonized individuals. We used the Age Group model as our base model against which we compared other model variants. Our main focus was on comparing Models 1–4. We also included a fifth model because of its widespread use (5) Homogeneous Carriage model: we assumed there were no differences in the average duration of carriage across the population. The average duration of carriage for the Homogeneous Carriage model was calculated as a weighted average of the mean durations of colonization used in the other models.

For each model, we fitted the hospital and community transmissibility parameters using the least-squares method to obtain 3.4% MRSA prevalence in the hospital and 1.5% prevalence in the community in the absence of interventions. Then, we ran the model to equilibrium and introduced either the transmission-based intervention, decolonization, or both. For each model, we compared the reproduction numbers (R) and percent of symptomatic cases averted by the interventions. We calculated the number of symptomatic cases averted by subtracting the number of symptomatic cases in a specific model with an intervention from the number of symptomatic cases for the same model without an intervention.

Reproduction number calculation

A reproduction number (R) quantifies the spread of a pathogen in a population. We used the next-generation matrix method to calculate R [51], though we note that the reproduction numbers provided here do not have a standard R0 interpretation [52]. In addition to an overall R, we also calculated reproduction numbers that included only hospital transmission (Rhosp) by setting community transmission to zero (i.e., β2 = 0). We similarly calculated a community reproduction number (Rcomm) for each model. Separate reproduction numbers for hospital and community transmission are commonly reported in the literature [12, 5355] and help compare transmission dynamics in each setting.

Sensitivity analyses

We also examined the effects of assumptions about colonized length of stay (rC) and colonized admission rate (ρC) by manipulating the parameter ranges within a single model structure (Age group model). We divided each of the two multiplier parameters into ranges of values from 1 to 3, and then created combinations of each parameter value. For every combination of the two parameters, we used least squares to fit the hospital (β1) and community (β2) transmission rates to obtain 3.4% prevalence in the hospital [29] and 1.5% prevalence in the community [40] for each model in the absence of any intervention.

We then applied a transmission-based or decolonization intervention and calculated the number of symptomatic cases averted by each intervention. For each set of parameter values, we compared the number of cases averted to a base case when the multipliers were set to one (rC = ρC = 1). This base case either represented the Age group model or the Homogeneous model; in both models, asymptomatic carriers have the same admission rate and length of stay as uncolonized individuals of the same age. We calculated the percent change in cases averted as the percentage of cases averted with multipliers equal to 1 minus the percentage of cases averted with multipliers equal to “x”, scaled by the percentage of cases averted with multipliers equal to 1. We visualized the results with a contoured heatmap and assessed differences across assumptions and treatments.

We used R (version 3.5.0) to create and analyze the model [56].

Results

Impact of assumptions that asymptomatic carriage affects admission rates or lengths of stay

In the pre-intervention period, the equilibrium prevalence of MRSA was 3.4% in the hospital and 1.5% in the community across all five models (Table 3). The admission prevalence was approximately 1.5% for the Age Group, Homogeneous, and Colonized LOS models, whereas admission prevalence was 1.96% for both models that assumed that colonized individuals were more likely to be admitted to the hospital than uncolonized individuals (i.e., Colonized Admission, Colonized Admission + LOS). The rate of carriage acquisition in the hospital ranged from 2.33/1,000 uncolonized patient-days (Colonized Admission + LOS model) to 4.79/1,000 uncolonized patient-days (Age Group model). As expected, the Homogeneous model had the lowest R (1.021), while R was 1.046 for all other models (Table 3).

thumbnail
Table 3. Impact of model assumptions on intervention effects using baseline parametersa.

https://doi.org/10.1371/journal.pone.0264344.t003

The effect of the different model assumptions on the estimated impact of a hospital intervention depended on the type of intervention (Fig 2, Table 3). In the absence of any interventions, 8,223 cases occurred over 5 years in each model. For the transmission-based intervention, the percentage of symptomatic cases averted was highest for the Homogeneous Model, followed by the following models in order: Age Group, Colonized Admission, Colonized LOS, and Colonized Admission + LOS. (Fig 2A). The differences in intervention impact between the models occurred across a wide range of transmission-based intervention effectiveness values (Fig 2).

thumbnail
Fig 2. Impact of model assumptions on the percentage of symptomatic cases averted by hospital interventions.

Percent of symptomatic cases averted after 5 years of a hospital: (A) transmission-based intervention that reduced the infectiousness of carriers (both asymptomatically colonized and symptomatic cases) or (B) decolonization intervention that reduced the duration of carriage for asymptomatically colonized patients. The y-axis shows the percent of symptomatic cases averted over 5 years in the entire population under simulated interventions that began with the pathogen at equilibrium. We combined hospital and community cases (both asymptomatic and symptomatic), and calculated that 8,223 cases occurred over 5 years in each model in the absence of any intervention. In (A), the transmission-based intervention effectiveness was the proportional reduction in the infectiousness of all MRSA patients (i.e., proportional reduction in β1). In (B), the intervention’s decolonization rate was the average increase in the rate of loss of asymptomatic carriage in the hospital. Colors show the different models (Age Group = solid black; Colonized Admission = dashed light blue; Colonized LOS = dot-dash dark blue; Colonized Admission + LOS = dotted green; Homogeneous = solid yellow). See Table 2 for parameter values used for each model.

https://doi.org/10.1371/journal.pone.0264344.g002

For the decolonization intervention, the percent of symptomatic cases averted was again largest for the Homogeneous Carriage model, but was similar across the other models until the rate of decolonization became high (Fig 2B, Table 3). Results for scenarios where both interventions were applied concomitantly were in line with findings for each intervention applied separately (S1 Fig).

Analysis of model parameters’ influence on how carrier assumptions change model output

All else being equal, including assumptions where asymptomatically colonized individuals had increased length of stay or admission rates because of their colonization status heavily affected estimates of the impact of a transmission-based intervention but had modest effects for a decolonization intervention (Fig 3). Both multipliers had similarly strong effects for the transmission-based intervention. However, the length of stay multiplier (rC) had a stronger relative impact than the admission rate multiplier for the decolonization intervention. Higher values of length of stay multiplier or colonized admission rates generally led to greater differences in the number of cases averted. However, when using the homogeneous model as the baseline case, results were more consistent across treatment types (S2 Fig).

thumbnail
Fig 3. Sensitivity of number of symptomatic cases averted to uncertainty in parameters for increased length of stay or admission rate for asymptomatic individuals.

The percent change in number of symptomatic cases averted in the Age Group Model when increasing the colonized admission rate multiplier (ρC) and/or the colonized length of stay multiplier (rC) compared to the base case (i.e., rC = ρC = 1). Colors and contour lines denote the percentage change in cases averted compared to the base case. For each combination of values for the rC and ρC multipliers, the transmission parameters were refit to achieve an equilibrium prevalence of 3.4% colonized individuals in the hospital and 1.5% in the community. We ran models to equilibrium and then implemented a decolonization intervention (γint = 0.035) or a transmission-based intervention (θ = 30%).

https://doi.org/10.1371/journal.pone.0264344.g003

Discussion

We showed that assuming asymptomatic carriers have higher rates of hospital admission or longer lengths of stay can bias simulations of the impact of hospital-based HAI interventions designed to prevent spread of healthcare-associated pathogens, especially for transmission-based interventions (e.g., enhanced infection control efforts). These assumptions are typically included as if asymptomatic carriers had more healthcare exposure precisely because they were asymptomatically colonized [1719, 2135]. However, confounding and reverse causation are more likely explanations for observed differences in admission rates and lengths of stay [36].

In our main example (Fig 2), the carrier assumptions produced larger differences across models in the impact of a transmission-based intervention than for a decolonization intervention. Compared to the Age Group model, fewer cases were predicted to be averted by a transmission-based intervention for models where colonized individuals had longer lengths of stay or higher admission rates. The most likely explanation is that when we assumed higher admission rates or longer lengths of stay for colonized individuals, then hospital transmissibility did not need to be as high to reach the same equilibrium hospital prevalence compared to the Age Group model, and therefore the community became relatively more important for pathogen transmission. In other words, the transmission-based intervention, which proportionally reduced transmissibility in the hospital (β1), prevented fewer transmissions in the Colonized Admission, Colonized LOS, and Colonized Admission + LOS models than in the Age Group model because the latter had the highest value of β1 to begin with, the highest Rhosp value, and therefore the largest contribution of hospital transmission to the overall R value. In contrast, assuming homogeneous lengths of carriage resulted in much higher numbers of transmissions and cases averted than in the Age Group model because the Homogeneous Carriage model had a lower R and transmission in the Homogeneous Carriage model was even more heavily weighted towards transmission in the hospital. On the other hand, decolonization treatments had smaller differences across models, possibly because the number of colonized individuals in the hospital was always the same so only a certain number could be decolonized. For combinations of decolonization and transmission-based interventions, comparisons across models could give similar results depending on the value of each intervention parameter (S1 Fig).

A potentially valid reason for parameterizing higher admission rates or longer lengths of stay for individuals in model compartments containing asymptomatic carriers is when symptomatic and colonized individuals are pooled together. In this case, the combined compartments could be parameterized using the weighted average of parameter values for colonized and symptomatic individuals [57]. However, such averaged parameter values would need to change in response to an intervention that affected the ratio of asymptomatic versus symptomatic carriers, and it could be difficult to calculate what the new ratio and parameter values should be without representing asymptomatic and symptomatic carriers separately. As with analyses that use standard statistical models (e.g., case-control studies analyzed with logistic regression), we recommend that the choice of which variables to include in a transmission model should be guided by the research question as well as subject matter considerations [58].

Additionally, we assessed the influence of these assumptions over a range of potential values. In general, the stronger the assumption about asymptomatically colonized individuals and their propensity to stay in the hospital, the more one might miscalculate the impact of an intervention. In our simulations, even small changes in the multipliers led to substantial differences in the model results. However, the type of intervention is still important: we observed different results for an intervention to reduce transmission compared to an intervention that decolonized patients, likely because decolonization acts more through recovery and transmission-based interventions directly reduces transmissibility.

Our analyses have several limitations. For example, we omitted antibiotics, competition from other pathogen strains (e.g methicillin-sensitive Staphylococcus aureus) or species (e.g. Clostridioides difficile), and impact of patient transfers to or from additional healthcare facilities healthcare facilities, each of which could have affected the magnitude of our results. However, we expect that the same issues would remain if such factors were included because assuming that asymptomatic carriers are more likely to be admitted to hospital or have longer lengths of stay would still have no biological basis and would continue to artificially inflate pathogen prevalence in the hospital compared to a more appropriately structured model (e.g., finer age stratification, higher admission rates for recently discharged patients or nursing home residents). We also expect similar problems to occur if pathogen incidence data were used for fitting instead of prevalence data. Lastly, including higher rates of hospitalization and longer lengths of stay for individuals recently discharged from hospital compared to those who have not been hospitalized recently would have improved the accuracy of the model. However, such model complexity was not required to illustrate the problem, and therefore doing so would have made our example more difficult without clearly providing a commensurate benefit. In future work, it could be useful to include such additional complexity and assess how well different models can simultaneously fit contemporaneous estimates of community prevalence, admission prevalence, hospital prevalence, and/or hospital incidence, which could also help discriminate between simulations in addition to using subject matter knowledge to guide modeling decisions.

In our Age Group model, we stratified the population into groups of younger and older individuals, and each age group had different admission rates and lengths of stay. However, we do not expect that this would result in the same problems as including changes in admission rates and lengths of stay being a direct consequence of being an asymptomatic carrier. Age is a proxy for factors that more directly affect admission rates and lengths of stay (e.g., being diagnosed with a health condition, insurance status). Such factors are unlikely to change concurrently with the gain and loss of colonization, especially after having included additional states in the model (e.g., hospitalization, which is correlated with health status and the risk of acquiring a pathogen), and there are no interventions that affect date of birth. Therefore, we believe that the Age Group model is more realistic compared to the other models and it was useful as a simple base case against which to compare the other modeling assumptions. Building on this framework, additional risk groups could be considered for inclusion as necessary. For example, male gender, recent hospitalization, and residence in a nursing home are risk factors for S. aureus colonization and bacteremia due to an S. aureus infection [5962], and each could be considered for inclusion as additional risk groups.

A simple model cannot capture every possible factor, but our simplified model helped illustrate the central problem associated with attributing different admission rates or lengths of stay to asymptomatic carriers without a clear biological basis for doing so. Our results show that such assumptions can greatly affect estimates of the impact of interventions, and therefore should be justified. We have included our model code, which can be adapted to represent other HAI pathogens and risk groups by changing parameter values or adapting the model structure. While the pathogen parameters in our main example were based on MRSA, our conclusions are broadly applicable because assuming that asymptomatic carriers are more likely to be admitted to hospital or have longer lengths of stay generally have no biological basis.

Conclusions

It is important to assess how modeling assumptions influence estimates of intervention impact. We have shown that common assumptions regarding how colonized individuals are included in simulations can heavily bias estimates of the impact of transmission interruption and to a lesser extent decolonization interventions. Modelers ought to consider whether data in favor of such assumptions could also be biased, how including such assumptions can affect model results, and if a more complex model (e.g., that includes ≥1 risk factor) is necessary to address specific research aims. Decision-makers should also critically evaluate the biological assumptions underpinning parameter values used in models that inform policy.

Supporting information

S1 Fig. Heatmap of the percentage of cases averted for simultaneous decolonization and transmission-based interventions.

Lighter colors show higher percentages of cases averted, and the range of decolonization and transmission-based intervention effectiveness is the same as in Fig 2.

https://doi.org/10.1371/journal.pone.0264344.s002

(TIF)

S2 Fig. Sensitivity of number of symptomatic cases averted to uncertainty in parameters for increased length of stay or admission rate for asymptomatic individuals.

The percent change in number of symptomatic cases averted in the Homogeneous Model when run to equilibrium and then either a decolonization intervention (γint = 0.035) or a transmission-based intervention (θ = 30%) implemented.

https://doi.org/10.1371/journal.pone.0264344.s003

(TIF)

S1 Table. State variable table with description of notation.

https://doi.org/10.1371/journal.pone.0264344.s004

(DOCX)

Acknowledgments

We thank John Jernigan, Prabasaj Paul, Brandon Schneider, Brajendra Singh, and Hannah Wolford for helpful discussions of model design and results along the way. We also thank members of the Office of Prevention Research and Evaluation at the Centers for Disease Control and Prevention for their useful feedback.

Disclaimer: The findings and conclusions in this report are those of the authors and do not necessarily represent the official position of the Centers for Disease Control and Prevention.

References

  1. 1. The biggest antibiotic-resistant threats in the U.S. 2013. Available from: http://www.cdc.gov/drugresistance/threat-report-2013
  2. 2. Cosgrove SE. The Relationship between Antimicrobial Resistance and Patient Outcomes: Mortality, Length of Hospital Stay, and Health Care Costs. Clin Infect Dis. Oxford University Press; 2006 Jan 15;42:S82–9. pmid:16355321
  3. 3. Centers for Disease Control and Prevention. Active bacterial core surveillance report, emerging infections program network, methicillin resistant Staphylococcus aureus, 2014 [Internet]. Atlanta, GA. Available from: http://www.cdc.gov/abcs/reports-findings/survreports/mrsa14.html
  4. 4. Wiegand PN, Nathwani D, Wilcox MH, Stephens J, Shelbaya A, Haider S. Clinical and economic burden of Clostridium difficile infection in Europe: a systematic review of healthcare-facility-acquired infection. Journal of Hospital Infection. W.B. Saunders; 2012 May 1;81(1):1–14. pmid:22498638
  5. 5. Scott RD. The direct medical costs of healthcare-associated infections in US hospitals and the benefits of prevention. 2009.
  6. 6. Allegranzi B, Pittet D. Role of hand hygiene in healthcare-associated infection prevention. J Hosp Infect. 2009 Dec;73(4):305–15. pmid:19720430
  7. 7. Rubin MA, Samore MH, Harris AD. The Importance of Contact Precautions for Endemic Methicillin-Resistant Staphylococcus aureus and Vancomycin-Resistant Enterococci. JAMA. American Medical Association; 2018 Mar 6;319(9):863–4. pmid:29435582
  8. 8. Huang SS, Septimus E, Kleinman K, Moody J, Hickok J, Avery TR, et al. Targeted versus universal decolonization to prevent ICU infection. N Engl J Med. Massachusetts Medical Society; 2013 Jun 13;368(24):2255–65. pmid:23718152
  9. 9. Harris AD, Pineles L, Belton B, Johnson JK, Shardell M, Loeb M, et al. Universal glove and gown use and acquisition of antibiotic-resistant bacteria in the ICU: a randomized trial. JAMA. 2013 Oct 16;310(15):1571–80. pmid:24097234
  10. 10. Cooper BS, Medley GF, Stone SP, Kibbler CC, Cookson BD, Roberts JA, et al. Methicillin-resistant Staphylococcus aureus in hospitals and the community: stealth dynamics and control catastrophes. Proceedings of the National Academy of Sciences. National Academy of Sciences; 2004 Jul 6;101(27):10223–8. pmid:15220470
  11. 11. van Kleef E, Luangasanatip N, Bonten MJ, Cooper BS. Why sensitive bacteria are resistant to hospital infection control. Wellcome Open Res. 2017;2:16. pmid:29260003
  12. 12. Kouyos R, Klein E, Grenfell B. Hospital-community interactions foster coexistence between methicillin-resistant strains of Staphylococcus aureus. Levin BR, editor. PLoS Pathog. Public Library of Science; 2013 Feb;9(2):e1003134. pmid:23468619
  13. 13. van Kleef E, Robotham JV, Jit M, Deeny SR, Edmunds WJ. Modelling the transmission of healthcare associated infections: a systematic review. BMC Infectious Diseases 2010 10:1. 1st ed. BioMed Central; 2013 Dec 1;13(1):294. pmid:23809195
  14. 14. Kwok KO, Read JM, Tang A, Chen H, Riley S, Kam KM. A systematic review of transmission dynamic studies of methicillin-resistant Staphylococcus aureus in non-hospital residential facilities. BMC Infectious Diseases 2010 10:1. BioMed Central; 2018 Apr 18;18(1):188. pmid:29669512
  15. 15. Cremin Í, Watson O, Heffernan A, Imai N, Ahmed N, Bivegete S, et al. An infectious way to teach students about outbreaks. Epidemics. Elsevier; 2018 Jun 1;23:42–8. pmid:29289499
  16. 16. Bellan SE, Pulliam JRC, Scott JC, Dushoff J, Committee TMO. How to Make Epidemiological Training Infectious. Kerfeld CA, editor. PLOS Biology. Public Library of Science; 2012 Apr 3;10(4):e1001295. pmid:22509129
  17. 17. Lee BY, Yilmaz SL, Wong KF, Bartsch SM, Eubank S, Song Y, et al. Modeling the regional spread and control of vancomycin-resistant enterococci. American Journal of Infection Control. Mosby; 2013 Aug 1;41(8):668–73. pmid:23896284
  18. 18. Lee BY, McGlone SM, Wong KF, Yilmaz SL, Avery TR, Song Y, et al. Modeling the Spread of Methicillin-Resistant Staphylococcus aureus (MRSA) Outbreaks throughout the Hospitals in Orange County, California. Infection Control & Hospital Epidemiology. Cambridge University Press; 2011 Jun 1;32(6):562–72. pmid:21558768
  19. 19. Austin DJ, Bonten MJM, Weinstein RA, Slaughter S, Anderson RM. Vancomycin-resistant enterococci in intensive-care hospital settings: Transmission dynamics, persistence, and the impact of infection control programs. Proceedings of the National Academy of Sciences. National Academy of Sciences; 1999 Jun 8;96(12):6908–13.
  20. 20. D’Agata EMC, Webb G, Horn M. A Mathematical Model Quantifying the Impact of Antibiotic Exposure and Other Interventions on the Endemic Prevalence of Vancomycin-Resistant Enterococci. J Infect Dis. Oxford University Press; 2005 Dec 1;192(11):2004–11. pmid:16267774
  21. 21. D’Agata EMC, Webb GF, Horn MA, Moellering RC, Ruan S. Modeling the Invasion of Community-Acquired Methicillin-Resistant Staphylococcus aureus into Hospitals. Clin Infect Dis. Oxford University Press; 2009 Feb 1;48(3):274–84. pmid:19137654
  22. 22. D’Agata EMC, Webb GF, Pressley J. Rapid Emergence of Co-colonization with Community-acquired and Hospital-Acquired Methicillin-Resistant Staphylococcus aureus Strains in the Hospital Setting. Mathematical Modelling of Natural Phenomena. EDP Sciences; 2010 Apr 28;5(3):76–93. pmid:20740067
  23. 23. Doan TN, Kong DC, Marshall C, Kirkpatrick CM, McBryde ES. Modeling the impact of interventions against Acinetobacter baumannii transmission in intensive care units. Virulence. Taylor & Francis; 2015 Sep 25;7(2):141–52. pmid:26252184
  24. 24. Forrester M, Pettitt AN. Use of Stochastic Epidemic Modeling to Quantify Transmission Rates of Colonization With Methicillin-Resistant Staphylococcus Aureus in an Intensive Care Unit. Infection Control & Hospital Epidemiology. Cambridge University Press; 2005 Jul 1;26(7):598–606.
  25. 25. Grima DT, Webb GF, D’Agata EMC. Mathematical Model of the Impact of a Nonantibiotic Treatment for Clostridium difficile on the Endemic Prevalence of Vancomycin-Resistant Enterococci in a Hospital Setting. Computational and Mathematical Methods in Medicine. Hindawi; 2012 Jan 15;2012(13):1–8. pmid:22291852
  26. 26. Haber M, Levin BR, Kramarz P. Antibiotic control of antibiotic resistance in hospitals: a simulation study. BMC Infectious Diseases 2010 10:1. BioMed Central; 2010 Dec 1;10(1):254. pmid:20738872
  27. 27. Hogea C, Van Effelterre T, Cassidy A. A model-based analysis: what potential could there be for a S. aureus vaccine in a hospital setting on top of other preventative measures? BMC Infectious Diseases 2010 10:1. BioMed Central; 2014 Dec 1;14(1):291.
  28. 28. Lee BY, Bartsch SM, Wong KF, Singh A, Avery TR, Kim DS, et al. The Importance of Nursing Homes in the Spread of Methicillin-Resistant Staphylococcus aureus (MRSA) Among Hospitals. Medical care. NIH Public Access; 2013 Mar 1;51(3):205–15. pmid:23358388
  29. 29. Lee BY, Bartsch SM, Wong KF, McKinnell JA, Cui E, Cao C, et al. Beyond the Intensive Care Unit (ICU): Countywide Impact of Universal ICU Staphylococcus aureus Decolonization. Am J Epidemiol. Oxford University Press; 2016 Mar 1;183(5):480–9. pmid:26872710
  30. 30. Lesosky M, McGeer A, Simor A, Green K, Low DE, Raboud J. Effect of Patterns of Transferring Patients among Healthcare Institutions on Rates of Nosocomial Methicillin-Resistant Staphylococcus aureus Transmission: A Monte Carlo Simulation. Infection Control & Hospital Epidemiology. Cambridge University Press; 2011 Feb 1;32(2):136–47. pmid:21460468
  31. 31. Pelat C, Kardaś-Słoma L, Birgand G, Ruppé E, Schwarzinger M, Andremont A, et al. Hand Hygiene, Cohorting, or Antibiotic Restriction to Control Outbreaks of Multidrug-Resistant Enterobacteriaceae. Infection Control & Hospital Epidemiology. Cambridge University Press; 2016 Mar 1;37(3):272–80. pmid:26708383
  32. 32. Sadsad R, Sintchenko V, McDonnell GD, Gilbert GL. Effectiveness of Hospital-Wide Methicillin-Resistant Staphylococcus aureus (MRSA) Infection Control Policies Differs by Ward Specialty. Kluytmans J, editor. PLOS ONE. Public Library of Science; 2013 Dec 10;8(12):e83099. pmid:24340085
  33. 33. Starr JM, Campbell A, Renshaw E, Poxton IR, Gibson GJ. Spatio-temporal stochastic modelling of Clostridium difficile. Journal of Hospital Infection. W.B. Saunders; 2009 Jan 1;71(1):49–56. pmid:19013677
  34. 34. Wang X, Chen Y, Zhao W, Wang Y, Song Q, Liu H, et al. A data-driven mathematical model of multi-drug resistant Acinetobacter baumannii transmission in an intensive care unit. Scientific Reports 2015 5. Nature Publishing Group; 2015 Mar 25;5(1):9478.
  35. 35. Webb GF, Horn MA, D’Agata EMC, Moellering RC Jr, Ruan S. Competition of hospital-acquired and community-acquired methicillin-resistant Staphylococcus aureus strains in hospitals. Journal of Biological Dynamics. Taylor & Francis; 2009 Jul 8;4(1):115–29.
  36. 36. De Angelis G, Allignol A, Murthy A, Wolkewitz M, Beyersmann J, Safran E, et al. Multistate modelling to estimate the excess length of stay associated with meticillin-resistant Staphylococcus aureus colonisation and infection in surgical patients. Journal of Hospital Infection. W.B. Saunders; 2011 Jun 1;78(2):86–91. pmid:21481492
  37. 37. Blixt T, Gradel KO, Homann C, Seidelin JB, Schønning K, Lester A, et al. Asymptomatic Carriers Contribute to Nosocomial Clostridium difficile Infection: A Cohort Study of 4508 Patients. Gastroenterology. W.B. Saunders; 2017 Apr 1;152(5):1031–2. pmid:28063955
  38. 38. Nelson RE, Nelson SD, Khader K, Perencevich EL, Schweizer ML, Rubin MA, et al. The Magnitude of Time-Dependent Bias in the Estimation of Excess Length of Stay Attributable to Healthcare-Associated Infections. Infection Control & Hospital Epidemiology. Cambridge University Press; 2015 Sep 1;36(9):1089–94. pmid:26041436
  39. 39. Weiss AJ, Elixhauser A. Overview of Hospital Stays in the United States, 2012: Statistical Brief #180. Rockville (MD): Agency for Healthcare Research and Quality (US); 2006 Feb.
  40. 40. Gorwitz RJ. Understanding the success of methicillin-resistant Staphylococcus aureus strains causing epidemic disease in the community. J Infect Dis. 2008 Jan 15;197(2):179–82. pmid:18173362
  41. 41. Nouwen J, Boelens H, van Belkum A, Verbrugh H. Human factor in Staphylococcus aureus nasal carriage. Infect Immun. American Society for Microbiology; 2004 Nov;72(11):6685–8. pmid:15501803
  42. 42. van Belkum A, Verkaik NJ, de Vogel CP, Boelens H, Verveer J, Nouwen J, et al. Reclassification of Staphylococcus aureus Nasal Carriage Types. The Journal of Infectious Diseases. 2009 Jun 15;199(12):1820–6. pmid:19419332
  43. 43. Mossong J, Hens N, Jit M, Beutels P, Auranen K, Mikolajczyk R, et al. Social Contacts and Mixing Patterns Relevant to the Spread of Infectious Diseases. Riley S, editor. PLOS Medicine. Public Library of Science; 2008 Mar 25;5(3):e74. pmid:18366252
  44. 44. US Census Bureau. Quickfacts: United States. 2017.
  45. 45. Pedersen AB, Altizer S, Poss M, Cunningham AA, Nunn CL. Patterns of host specificity and transmission among parasites of wild primates. International Journal for Parasitology. Pergamon; 2005 May 1;35(6):647–57. pmid:15862578
  46. 46. Centers for Disease Control and Prevention. Hospitals, beds, and occupancy rates, by type of ownership and size of hospital: United States, selected years 1975–2009.
  47. 47. US Census Bureau. National Population Totals and Components of Change: 2010–2018.
  48. 48. Mossong J, Hens N, Jit M, Beutels P, Auranen K, Mikolajczyk R, et al. POLYMOD social contact data. 2017 Nov 7.
  49. 49. Jones M, Ying J, Huttner B, Evans M, Maw M, Nielson C, et al. Relationships Between the Importation, Transmission, and Nosocomial Infections of Methicillin-Resistant Staphylococcus aureus: An Observational Study of 112 Veterans Affairs Medical Centers. Clin Infect Dis. Oxford University Press; 2014 Jan 1;58(1):32–9. pmid:24092798
  50. 50. Marshall AL, Durani U, Bartley A, Hagen CE, Ashrani A, Rose C, et al. The impact of postpartum hemorrhage on hospital length of stay and inpatient mortality: a National Inpatient Sample–based analysis. American Journal of Obstetrics and Gynecology. Mosby; 2017 Sep 1;217(3):344.e1–344.e6. pmid:28502758
  51. 51. Diekmann O, Heesterbeek JAP, Roberts MG. The construction of next-generation matrices for compartmental epidemic models. Journal of The Royal Society Interface. The Royal Society; 2009 Nov 5;7(47):rsif20090386–885. pmid:19892718
  52. 52. Li J, Blakeley D, Smith RJ. The failure of R0. Comput Math Methods Med. 2011;2011. pmid:21860658
  53. 53. Chowell G, Blumberg S, Simonsen L, Miller MA, Viboud C. Synthesizing data and models for the spread of MERS-CoV, 2013: Key role of index cases and hospital transmission. Epidemics [Internet]. 2014;9:40–51. pmid:25480133
  54. 54. Legrand J, Grais RF, Boelle PY, Valleron AJ, Flahault A. Understanding the dynamics of Ebola epidemics. Epidemiol Infect. 2007;135(4):610–21. pmid:16999875
  55. 55. McLure A, Clements ACA, Kirk M, Glass K. Modelling diverse sources of Clostridium difficile in the community: Importance of animals, infants and asymptomatic carriers. Epidemiol Infect. 2019;147. pmid:31063089
  56. 56. R Core Team. R: A Language and Environment for Statistical Computing [Internet]. 3rd ed. Vienna, Austria. Available from: http://www.R-project.org/
  57. 57. Perencevich EN, Fisman DN, Lipsitch M, Harris AD, Morris JG, Smith DL. Projected Benefits of Active Surveillance for Vancomycin-Resistant Enterococci in Intensive Care Units. Clin Infect Dis. Oxford University Press; 2004 Apr 15;38(8):1108–15. pmid:15095215
  58. 58. Hernán MA, Hernández-Díaz S, Werler MM, Mitchell AA. Causal Knowledge as a Prerequisite for Confounding Evaluation: An Application to Birth Defects Epidemiology. Am J Epidemiol. Oxford University Press; 2002 Jan 15;155(2):176–84. pmid:11790682
  59. 59. Heijer den CD, van Bijnen EM, Paget WJ, Pringle M, Goossens H, Bruggeman CA, et al. Prevalence and resistance of commensal Staphylococcus aureus, including meticillin-resistant S aureus, in nine European countries: a cross-sectional study. The Lancet Infectious Diseases. Elsevier; 2013 May 1;13(5):409–15. pmid:23473661
  60. 60. McKinnell JA, Miller LG, Eells SJ, Cui E, Huang SS. A Systematic Literature Review and Meta-Analysis of Factors Associated with Methicillin-Resistant Staphylococcus aureus Colonization at Time of Hospital or Intensive Care Unit Admission. Infection Control & Hospital Epidemiology. Cambridge University Press; 2013 Oct 1;34(10):1077–86.
  61. 61. Wyllie DH, Walker AS, Peto TEA, Crook DW. Hospital exposure in a UK population, and its association with bacteraemia. Journal of Hospital Infection. W.B. Saunders; 2007 Dec 1;67(4):301–7.
  62. 62. Grigg C, Palms D, Stone ND, Gualandi N, Bamberg W, Dumyati G, et al. Burden of Invasive Methicillin-Resistant Staphylococcus aureus Infections in Nursing Home Residents. J Am Geriatr Soc. 2018 Aug;66(8):1581–6. pmid:30094828