Modelling the primary and hospital care costs associated with cardiovascular disease events

Aims and objectives

Our aim was to generate cost evidence for a patient-level microsimulation model that would be used in cost-effectiveness analyses of interventions to reduce CVD risk. This evidence had to be specific to patient characteristics and their experiences of CVD events. The objective was to estimate annual healthcare costs associated with patient characteristics and events, as used in the model, from the perspective of the UK NHS.

To achieve this, we developed separate cost regression models to predict the annual primary and hospital care costs based on the baseline characteristics of the participants and their time-updated experiences of key disease events. In particular, these cost models take into account the duration since adverse events, which can significantly impact healthcare costs.

Methods

Data

This study was conducted using data from the UKB, the cohort used in the calibration of the microsimulation model. All UKB participants with established linkage to primary care (38.5%) or hospital inpatient care (100%) records were included, with the exception of a small number of participants with end-stage renal disease for whom the linked hospital data did not include dialysis information. The analyses focused on the primary and hospital inpatient care costs. Primary care services were costed by identifying categories of consultations, monitoring tests and prescription medications and costing them based on the Unit Costs of Health and Social Care,37–39 the national reference costs,40 and NHS prescription cost analysis,41 respectively. Hospital inpatient care services were costed by identifying the hospital episodes, grouping the episodes into Healthcare Resource Groups42 and costing them using the NHS England reference costs.40,43 Costs were analysed in annual cycle from entry into the UKB and were generated by summing the costs incurred by each participant during each year of follow-up in the study. All costs were inflated to year 2020 using the NHS cost inflation index.37 The conventional participant risk factors included in the disease risk models were also included in the cost models. We assessed the impact of the following four CVD events, namely MI, stroke, CRV and VD, and three nonvascular events, namely diabetes, cancer and NVD, on annual healthcare costs. Follow-up data from participants’ entry into the UKB until 31 March 2016 contributed to these analyses.

Statistical methods

The following participant characteristics at entry into UKB were considered in the cost models: sex, ethnicity, smoking status, physical activity, diet quality, BMI, LDL-C, HDL-C, serum creatinine, systolic and diastolic BP, treated hypertension, and histories of diabetes, severe mental illness and CVD. The models also included annually updated participant characteristics such as current age and time since previous CVD events or incident diabetes or cancer. The study used generalised linear regression models to model annual primary care costs and two-part models to model annual hospital care costs, with the first part modelling the probability of incurring any costs and the second part modelling the costs conditional on incurring any. The study considered six different generalised linear models (GLMs) using three distributions (Gaussian, Poisson and gamma) and two link functions (identity and natural log), and the best-fitting models were chosen based on specification tests, predictive performance and parsimony. Finally, the study used cluster robust standard errors (SEs) to account for the lack of independence between annual periods for the same participant and performed stepwise bidirectional covariate selection at the 1% level of statistical significance. Analyses were performed using R version 4.1.3 (The R Foundation for Statistical Computing, Vienna, Austria).

Results

All 501,807 participants of the UKB contributed to the hospital cost analyses, while 192,983 (38.5%) contributed to the primary care cost analyses over 3,798,324 and 1,255,741 annual follow-up periods, respectively. Participants with previous CVD were older (mean age of 60 and 56 years, respectively), more likely to be men (57–59% and 44%, respectively), from more deprived socioeconomic backgrounds, and more likely to be smokers, be less physically active, have an unhealthy diet, have a higher BMI and have disease histories such as treated hypertension, diabetes, cancer and severe mental illness (Table 7).

During the 7.1-year mean follow-up period, a small percentage of participants experienced the key CVD events, with higher event risks among people with previous CVD (Table 8). Additionally, some participants without previous diabetes or cancer at recruitment were diagnosed with these conditions during the follow-up period. The duration of follow-up was similar between the primary and hospital care cost analyses, and between participants with and participants without previous CVD. On average, participants had several primary care consultations, diagnostic and monitoring tests, and prescription medications per year, with primary care costs totalling £409 per year and the hospital inpatient cost totalling £583 per year. People with previous CVD had higher rates and costs of primary and hospital care than those without (see Table 8).

Annual primary care consultation rates and annual hospital admission rates peaked in the year of events such as MI, stroke, CRV, diabetes or cancer, and then decreased over the subsequent years. The annual costs had a similar pattern across event types, except that annual primary care costs continued to increase after the year of stroke. Participants with previous CVD had higher annual primary care consultation rates and hospital admission rates, and higher annual primary care and hospital care costs (Figure 7).

FIGURE 7 Annual rates and costs of healthcare use by time since adverse event. a, Incident cases only. Mean annual costs are summarised for UKB participants experiencing the corresponding events. Annual periods are defined from date of participant’s entry into UKB. ‘Event’ years correspond to the annual periods with respective event; ‘pre-event’ years correspond to annual periods prior to annual periods with respective event. Thereafter data are presented for each subsequent annual period up to 5 years following the annual periods with events. Total numbers of corresponding first events in the study period are presented on the right-hand side. Reproduced with permission from Zhou et al.44 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original figure. (Figure not reproduced here; see the original article.)

After statistical model selection, primary care costs were modelled using one-part GLMs with Poisson distribution and identity link function, while hospital care costs were modelled using two-part models with logistic regression model (part 1) and GLM with Poisson distribution and identity link function (part 2). The annual primary and hospital care costs were £262 and £244, respectively, for the reference individual in models for people without previous CVD. The reference annual primary and hospital care costs were £302 and £376, respectively, in models for people with previous CVD with the same reference characteristics except having a history of MI at baseline. Participant characteristics associated with higher primary care cost included older age, female sex, south Asian ethnicity, higher socioeconomic deprivation, smoking, low level of physical activities, unhealthy BMI, and morbidity, including treated hypertension, severe mental illness or prior diabetes or CVD (Table 9). Similar characteristics, except south Asian ethnicity, were also associated with higher hospital care costs (Table 10).

Summary of findings

Our research offers valuable contemporary insights into the long-term primary and hospital care costs associated with CVD events, providing models to predict healthcare costs at the individual level. The cost models are intended to be used in cost-effectiveness assessments of therapies to reduce CVD risk, using CVD decision models.

The cost models can also be used to derive the marginal effects of CVD events on primary and hospital care cost. Both the cost models and the derived marginal effects can inform economic and policy assessments of the value of health interventions aimed at reducing CVD risk and burden, as well as cost analyses. Beyond the impact of conditions such as CVD, diabetes and cancer, our research identified key individual factors that play a significant role in primary and hospital care costs. These factors include smoking, obesity and low physical activity, and addressing them could help reduce the demand and costs associated with healthcare services.