Transition probabilities

Transition probabilities were obtained as a function of age and sex from incident cases between wave n and n+1 in ELSA. As with estimates of prevalence values, the transition probabilities obtained from pooling ELSA epochs were attributed to the mid-point of the data collection period. A new cohort of those reaching age 35 each year enters through the disease-free state (see model figure in Appendix in S1 File). The prevalence of cardiovascular disease and functional impairment is very low in this 35-year old cohort (<2% in total) therefore, the resulting error in misclassification is negligible. Movements between states occur every year in the model based on transition probabilities. The transition probabilities between states were calculated using logistic regression of 2-year incidence of CVD, cognitive impairment, functional impairment, and recovery from functional impairment, using the ELSA data with age, sex and current health state as coefficients (where dementia was classed as concurrent cognitive and functional impairment). A calendar effect was added where CVD incidence trends mirrored CVD mortality trends and cognitive impairment was set to decrease by 2.7% per year, based on trends in the ELSA data.

Probability of death was estimated using a three step model; for the first step, CVD and non-CVD mortality probabilities of CVD up 2025 in 5-year age bands were calculated using the Bayesian Age Period Cohort (BAPC) model, with ONS mortality and population estimates from 1982–2012 for England and Wales as inputs. In the second step, we calculated mortality rates from ELSA for the age groups 50–59, 60–69, 70–79, and 80–89 and fitted two logistic regression models, first including only sex, gender, and interactions, and secondly also including the model health state. The results of the first step gave probability of CVD and non-CVD death by sex, single year of age and calendar year, which were adjusted by the results of the second step to also give probability of CVD and non-CVD death by health state in the model. These methods were chosen to favour the population-level data from ONS but adjusting for the ELSA-specific data to estimate mortality risk by health state in the model.

The synthesised trends were projected from 2011 to 2029 (with outcomes measured for the remaining ten years of the NHS Plan from 2020–29) [13] based on trend data from ELSA waves 1–6 (2002/03 to 2012/13) and mortality trends from 1990–2016. There have been subsequent ELSA waves but these ones are the only ones that have been linked with resource use data. IMPACT BAM has eight health states: free of cardiovascular disease (CVD), cognitive impairment (CI) or functional impairment (FI); CVD; CVD and FI; CVD and CI; CVD and dementia; dementia; CI and FI; and two additional absorbing states of CVD death and non-CVD death. The main model outcomes were health and social care costs, value of informal care, and QALYs experienced. QALYs were valued at £60,000 based on UK Treasury Green Book [14] but with a sensitivity analysis using £30,000 which is often quoted as the threshold used by the National Institute for Health and Care Excellence (NICE).

Healthcare costs

Healthcare costs included all recorded costs for individuals, so were not specific to CVD and dementia, and include future healthcare costs, which is why the model is useful for answering whether preventing CVD saves costs in the longer term if it increases survival. The ELSA data was linked with Hospital Episode Statistics (HES) data (inpatient, outpatient, A&E) which was costed using NHS Healthcare Resource Groups (HRGs) for 2018/19 financial year. 80% of ELSA participants (14,789 of 18,529) gave consent for their records to be linked. Ordinary Least Squares (OLS) regression using data from the consenting sample was used to estimate the relationship between total hospital costs and health state, age and gender. These regression results were used to impute costs for those also in the non-consenting sample. ELSA respondents provide information on prescribed medications currently being taken. For each prescription, we use the British National Formulary (BNF) paragraph, section, and chapter number, and match this to the Net Ingredient Costs (NICs) contained in Prescription Cost Analysis compiled by the NHS Health and Social Care Information Centre. Total healthcare costs were then calibrated to estimates of total healthcare costs by age for the UK reported by the Office for Budget Responsibility [15] to account for missing costs like primary care, community, and other underreporting of costs in the linked ELSA-HES data. We do not assume any changes in costs over time due to new technologies, price or wage inflation, or other causes, so the modelling assumes that costs for each health state remain constant over time.

Social care and informal care costs

Age-related social care costs were estimated using reported social care contact hours from ELSA combined with Personal Social Service Research Unit (PSSRU) unit costs [16]. These were for five resources; cleaner, care/nursing home staff, other formal help, Local Authority-provided home care worker/ home help, and non-Local Authority home care worker/ home help. We added residential care costs to IMPACT-BAM from a logit regression of whether the ELSA member is currently living in institutional care with a representative sample of people living in institutions, but we ran the institutional care costs regressions by having the dependant variable as dummy of whether they were living in an institution at the time of the interview. We then assumed an average yearly cost of living in institutional care of £39,156 based on PSSRU reference costs (This is based on a 50:50 split between residential and nursing beds). The social care resource use does not include some costs such as costs of home improvements, and respite care. To account for the many ELSA respondents who report zero hours of social care use, we estimated a two-part model: (i) a probit for the presence of any social care use, and (ii) OLS regression for the amount of resource use for people who report non-zero hours of social care receipt. For daycare, data on receipt of daycare are available but not number of hours. We therefore used a probit model and applied an average annual cost from PSSRU reference costs (£7,280 in 2016/17 prices). The PSSRU estimates do not contain recent estimates for the costs of cleaners so we assumed an hourly cost of 1.5 times the national living wage.

We calculated informal care based on the number of hours of help that ELSA respondents reported they had received in the last week from up to 25 different people, ranging from spouses to neighbours. We assumed an average cost per hour of informal help of £7.76, based on data from ONS on gross value added of informal care less household inputs, and total number of hours of informal care. All health, social and informal care costs were inflated or deflated to 2019 prices using Treasury GDP Deflator (October 2018). The modelling was undertaken in real terms (i.e. in current prices) in line with the suggestion of the UK Treasury Green Book but with an additional sensitivity analysis where costs were discounted at 3.5% per annum. For probabilistic sensitivity analyses, costs were fitted to a beta distribution where the 95% uncertainty intervals represented +/- 20% of the median, which was applied in addition to epidemiological uncertainty around the proportion of the population in each health state in the model. The modelled costs for England were used as inputs for the England and Wales population in IMPACT BAM.

Calculating quality adjusted life years

Utility weights for QALYs were taken from the EQ-5D MEPS (Medical Expenditure Panel Survey) catalogue [17] and Health Survey for England [18]. Based on a linear regression of Health Survey for England 2012 (the most recent year that included all of these variables), the coefficient of EQ-5D index score for number of limitations in activities of daily living (ADLs) was -0.042 after controlling for health, age and gender. For each health state/age combination, we multiplied the population norm EQ-5D index score from Jannsen and Tzende [19] by an EQ-5D multiplier for cognitive impairment, dementia, or CVD (from the MEPS) and by the ADL decrements for the distribution of number of ADLs in that state.

QALYs were not discounted in the main scenario, but an additional scenario has QALYs discounted by 1.5% per annum and by 3.5% per annum, in line with NICE and UK Treasury Green Book. There was no probabilistic distribution added to the QALY weights because the uncertainty on QALY weights is very low, so any difference in QALYs in the results is driven only by epidemiological uncertainty.

Please see Technical Appendix for further details of our economic methods.

Scenarios modelled

We modelled undiscounted health and social care costs and QALYs for 2020–2029 under two scenarios:

1. Scenario 1. CVD Plateau–Assuming age-specific CVD incidence remains at 2011 levels, continuing recent trends.
2. Scenario 2. CVD Fall—Assuming age-specific CVD incidence continued to decline, following the long-term trends from 1991 to 2011.

The CVD plateau is the most likely of these two scenarios which were selected to give a comparison of the potential future trajectory for CVD trends. There are several countries such as Spain and France [20] where CVD has continued to decline beyond what has been achieved in England and Wales–although improvements have slowed across Europe—so it was felt that this comparison would be useful in understanding the costs of the slowdown in CVD improvements and the potential economic value of improvements that might be achieved in the NHS plan for England, if it was to produce a return to an improvement of the CVD trajectory. Fig 1 shows what the two scenarios mean in terms of CVD incidence, prevalence and mortality trends.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Modelled CVD incidence per 100,000 population aged 35–100, prevalence (% of people aged 35–100), and mortality per 100,000 population aged 35–100, from 2005 to 2030, comparing Scenario 2 (Continuing decline in CVD incidence) with Scenario 1 (plateaued CVD incidence).

https://doi.org/10.1371/journal.pone.0268766.g001

Forecasting future costs

We calculated total costs for the whole England and Wales population as well as the specific excess costs of dementia and CVD. We estimated specific excess costs of dementia and CVD by comparing costs of individuals with dementia or CVD with the costs of individuals who were identical in age, gender and other disabilities who did not have dementia or CVD. Dementia was defined as the presence of both cognitive and functional impairment and the excess costs of dementia were estimated by comparing the same people as if they only had functional impairment.

Summary of results

After previous, dramatic falls, CVD incidence and mortality have recently plateaued in the UK. This slowdown could substantially increase health and social care costs over the next ten years, and cumulatively cost approximately £54 billion by 2029. The additional £22 billion in health, social and informal care costs, would represent about a 1.6% increased demand on NHS and social care budgets, which are already strained. However, the biggest costs would be approximately 540,000 lost QALYs (reflecting worse quality-of-life from higher rates of CVD and disability, and more life-years being lost through increased mortality). The immediate impact would particularly hurt the NHS, with a more distal and delayed impact on informal care and social care.

Comparison with other studies

Our results generally endorse and expand on previous studies of CVD and dementia costs. The MODEM study estimated the total (not excess) costs of people with dementia in England to be approximately £24 billion in 2015, made up of £10 billion unpaid care, £10 billion social care and £4 billion in health care costs [21]. Healthcare costs per person per year were £3,025 for people with mild dementia, up to £4,800 for severe dementia, and £4,800 for all care home residents with dementia. These costs are similar to our excess healthcare costs of approximately £4,400 for dementia. Luengo Fernandez et al. [22] reported a similar figure of 17,000 Euros for the combined health and social care cost per dementia patient for the UK in 2007.

Total healthcare costs in our study for people aged 35 and over were around £80billion per year, similar to NHS data in 2017/18 showing costs of approximately £108 billion in England and £7billion for Wales for all ages [23].

Our total costs of social care for 35–100 year olds were approximately £14 billion per year. This was slightly lower than the £22 billion quoted for England in 2017/18 [24]. (However, almost half that spend was on people aged under 65 [25]).

Our total cost of informal care for adults aged 35 and over was approximately £28 billion per year for England and Wales, slightly lower than the corresponding ONS household accounts figure of some £55billion; that however included all adults aged 18 and over [26].

Strengths and weaknesses

Our study uses a single source of data, from a large representative survey of older people in England linked to administrative healthcare records, to estimate both incidence of CVD and dementia, and associated health and social care use. This enhances both precision and internal consistency. The use of administrative data to cost healthcare use also provides more accurate estimates than sample data alone. Conversely, most previous studies relied on a patchwork of data sources. The IMPACT-BAM model accounts for complex epidemiological interactions between CVD, dementia and disability, notably lag times and competing risks.

Our model estimates of dementia and CVD prevalence were previously validated by comparisons with real life data from HES and Cognitive Function and Ageing Studies (CFAS), showing a good level of agreement [12]. Furthermore, our model formally accounts for the uncertainty of input parameters by using rigorous probabilistic sensitivity analyses. Furthermore, being able to quantify excess costs meant that we could isolate the true cost impact of dementia.

This study also has limitations, notably that our cost estimates are likely to be conservative because we a) did not include the effect of healthcare or social care on changing the QALYs experienced; b) assumed no change in healthcare services efficiency over time (either reduced costs, or increased health per unit of spend), nor changes in costs due to new technologies or inflation; c) based some cost estimates on self-reported resource use rather than direct data collection [27]. Finally, while ELSA is reasonably representative of the non-institutionalised population, it may predict residential care prevalence and costs less well. We have included some one-way sensitivity analyses (value of QALY, discount rate), as well as the probabilistic sensitivity analysis; but more one-way sensitivity analyses may highlight the drivers of differences in outcomes in more detail. We plan to produce a future study to explore these drivers more, using decomposition analysis.

Implications for policy makers

The Covid-19 pandemic has clearly changed the overall mortality and health and social care costs trajectory for England and Wales. Given that CVD is a risk factor for many conditions and disabilities, including excess risk of covid-19 death [28] it is important for policy makers to prevent as much CVD as possible in the coming decades. If the recent plateau in CVD incidence and mortality is allowed to persist, that could substantially increase healthcare and social care costs, and the opportunity costs of informal care. Conversely, introducing CVD prevention policies of proven effectiveness could continue the historical decline in CVD and decrease healthcare and social care requirements. Such investments would actually be cost-saving [29].

There is ample room for further reductions in CVD incidence and mortality. Populations similar to ours have CVD rates 50% lower (Singapore) and even 80% lower (Qatar) as highlighted by the Global Burden of Disease studies [30]. The recent UK increases in obesity, diabetes and associated hypertension must therefore be considered key targets for a renewed CVD and dementia prevention strategy. Indeed, as 90% of CVD can be explained by dietary and behavioural risk factors [31], the focus on primordial and primary prevention is now more urgently needed than ever [13]. Improving diet by increasing fruit and veg intake, reducing salt and processed food intake, as well as tobacco control interventions, and reducing sedentary behaviour [3], are examples of cost effective interventions to reduce CVD. There are also clinical interventions that can reduce the case fatality ratio such as risk stratification, blood pressure and lipid control, and revascularisation [32]. Persistent policy stasis will likely worsen the already large inequalities in CVD and dementia [33]. It will also add further pressure to our strained healthcare system. Dementia is now the leading cause of death for women in England and Wales and since it shares risk factors with CVD, it may be that future dementia mortality will be even higher than predicted [34, 35].

Our study may help inform resource allocation decisions. Regardless of scenario, investments in social care may need to increase even more quickly than investments in healthcare. We hope our evidence regarding epidemiological shifts and future health and social care costs might also prove useful for policymakers planning greater integration of health and social care.

Future research

Future research could use our IMPACT-BAM model to look further at health inequalities, and produce more granular estimates at regional or at local authority levels, especially given the increasingly local input to resource allocation [36]. Future modelling studies could estimate the potential of policy and clinical interventions that may change future prevalence of CVD and dementia, for instance around diet, physical activity, or cognitive training, to provide policy makers with comparative options to more comprehensively inform their actions [37].

The excess costs of CVD and dementia by age used here may be of interest to health economists as potential model input parameters. For instance, an intervention to prevent CVD in people aged 65–79 might save approximately £2,300 per person-year in healthcare costs; or an intervention to prevent dementia in people aged 80–100 might save around £9,900 per person-year in QALY losses (Table 2).

It might also be useful to look further at productivity impacts. A high proportion of the ELSA sample are above typical working age so there may not be significant lost earnings through ill health. However, there are still household productivity impacts, and friends or relatives providing informal care who lose potential earnings. Modelling these dynamic relationships between social care and informal care in more detail, using time-use survey data might be useful.