Aims and objectives
Cardiovascular disease can have a significant impact on an individual’s HRQoL.45 To estimate this impact, we developed a regression model to predict an individual’s HRQoL based on their age, sex, socioeconomic status, BMI, smoking status, and history of CVD and other diseases. Our goal was to integrate this QoL model into the CVD microsimulation model to provide annual QoL prediction for individuals in the model.
Methods
Health Survey for England
The QoL model was developed using data from the HSE in 2006, 2011 and 2017. The HSE is an annual cross-sectional survey designed to monitor the national health. In addition to core questions, the HSE includes yearly questionnaires on different topics. In 2006, 2011 and 2017 the survey included both the EuroQoL-5 Dimensions (EQ-5D) questionnaire and the detailed CVD questionnaire. The EQ-5D questionnaire measures participants’ generic HRQoL and the CVD questionnaire queries their first and latest experiences of different CVD events.
Missing data
We aligned the specification of participants’ baseline characteristics and disease event histories with those in the CVD microsimulation model to ensure compatibility. The dependent variable, the ED-5D QoL utility index, was derived from the five scores of the five-dimension EQ-5D questionnaire, which measures mobility, self-care, usual activities, pain and anxiety. The EQ-5D-3L questionnaire was administered in HSE 2006 and 2011, with QoL utility values ranging from − 0.594 for the worst health state to 1 for full health, where 0 represents a health state equivalent to death and higher values indicate better QoL.46 In HSE 2017, the EQ-5D-5L questionnaire was used with a similar QoL utility range after mapping the 5L value set to the 3L values.47 The independent variables included age, sex, ethnicity, socioeconomic deprivation [indicated by the Index of Multiple Deprivation (IMD)], BMI, current smoking status, treated hypertension, mental illness, cancer, diabetes, MI in last 12 months, stroke in last 12 months, CRV in last 12 months, any angina, MI that happened more than 1 year ago, stroke that happened more than 1 year ago, and CRV that happened more than 1 year ago.
Apart from the EQ-5D (10% missing) and BMI (16.5% missing), only small proportions of missing values were observed for other participant characteristics. Based on the assumption of missing at random, multiple imputation was used to impute missing data while estimating the QoL model. The R package ‘mice’ was used, which imputes multivariate missing data based on fully conditional specification, where each incomplete variable is imputed by a separate model.48 All covariates included in the QoL model were included in the imputation models. The dependent variable, the EQ-5D utility index, was continuous and skewed, with lower and upper bounds. An ordered logit model has been shown more appropriate for imputing missing dimensions of the EQ-5D questionnaire data,49 and was thus used in the present study. For other missing variables, logistic regression was used for binary variables, multinomial logit for ethnicity, and ordered logit for BMI category. Twenty imputations and 100 iterations for each imputation were performed. Following imputation, the EQ-5D utility index was calculated.
Quality-of-life regression model
An ordinary linear regression is a popular choice for modelling EQ-5D utility, as it allows the estimation of direct QoL decrements associated with individual characteristics and disease events, and there is no convincing evidence that other estimators are clearly superior. A linear regression model was estimated for QoL utility following multiple imputation of missing data. Owing to the non-linear effect of age on EQ-5D utility, age splines with a knot at 70 years were used. In a sensitivity analysis, the QoL model coefficients remained similar following the exclusion of participants aged < 40 years. Further checks did not indicate important interactions between recent MI and CRV events or between CVD histories and recent events.
Results
There were 24,231 participants included in the QoL model estimation (Table 11). The average EQ-5D utility across study population was 0.83 (SD 0.24). Participants’ mean age was 54 years; 55% were women; 91% were of white ethnicity; 20% were current smokers; and 68% were overweight or obese. Among the study participants, 19% had treated hypertension, 7.6% had diabetes, 5.2% had long-term mental illness and 2.4% had a history of cancer. Additionally, 2.3% were diagnosed with angina; 1.3% with MI more than 1 year ago without other CVDs; 1.7% with stroke more than 1 year ago without other CVDs; 2% with more than 1 CVDs. 1.9% had received CRV more than 1 year ago. MI, stroke and CRV events that had occurred in the last 12 months were rare, accounting for 0.4%, 0.5% and 0.2% of participants, respectively.
| Characteristic | n (%)/mean (SD) (N = 24,231 participants) |
|---|---|
| EQ-5D utility | 0.83 (0.24) |
| Missing | 2436 (10%) |
| Age, years | 53.9 (15.2) |
| Male | 10,822 (44.7%) |
| Ethnicity | |
| White | 21,907 (90.7%) |
| Black | 584 (2.4%) |
| South Asian | 1102 (4.6%) |
| Others | 550 (2.3%) |
| Missing | 88 (0.4%) |
| Deprivation | |
| IMD 1 | 5214 (21.5%) |
| IMD 2 | 5390 (22.2%) |
| IMD 3 | 5033 (20.8%) |
| IMD 4 | 4544 (18.8%) |
| IMD 5 (most deprived) | 4050 (16.7%) |
| BMI category (kg/m2) | |
| < 18.5 | 166 (0.8%) |
| 18.5–25 | 6222 (30.8%) |
| 25–30 | 8046 (39.8%) |
| 30–35 | 3869 (19.1%) |
| 35–40 | 1351 (6.7%) |
| ≥ 40 | 573 (2.8%) |
| Missing | 4004 (16.5%) |
| Smoking status | |
| Current smoker | 4929 (20.4%) |
| Ex-smoker | 9283 (38.4%) |
| Missing | 65 (0.3%) |
| Treatment of hypertension | 4695 (19.4%) |
| Missing | 24 (0.1%) |
| Mental illness | 1254 (5.2%) |
| Missing | 8 (0.0%) |
| Cancer | 588 (2.4%) |
| Missing | 8 (0.0%) |
| Diabetes history | |
| Diabetes ≤ 10 years | 1204 (5.0%) |
| Diabetes > 10 years | 617 (2.6%) |
| Missing | 54 (0.2%) |
| CVD history | |
| Angina only (ever) | 564 (2.3%) |
| MI only ≥ 1 year | 318 (1.3%) |
| Stroke only ≥ 1 year | 421 (1.7%) |
| Two or more conditions | 483 (2.0%) |
| Missing | 21 (0.1%) |
| CRV ≥ 1 year ago | 448 (1.9%) |
| Missing | 15 (0.1%) |
| MI < 12 months ago | 85 (0.4%) |
| Missing | 4 (0.0%) |
| Stroke < 12 months ago | 114 (0.5%) |
| Missing | 4 (0.0%) |
| CRV < 12 months ago | 50 (0.2%) |
| Missing | 5 (0.0%) |
Table 12 presents the results of the QoL regression model. Previous experiences of CVD events were associated with lower QoL. MI was associated with 0.10 (95% CI 0.03 to 0.16) lower QoL in the year of the event and 0.07 (0.04 to 0.10) lower QoL in the following years. Stroke was associated with 0.09 (0.04 to 0.13) lower QoL in the year of the event and 0.13 (0.11 to 0.16) lower QoL in subsequent years. On the other hand, CRV was associated with higher QoL of 0.04 (0.02 to 0.07) in years following the procedure. Diabetes was associated with 0.04 (0.03 to 0.06) lower QoL in the first 10 years from diagnosis and 0.08 (0.06 to 0.10) lower QoL in the following years. Cancer affecting daily activities was associated with 0.13 (0.11 to 0.14) lower QoL. Mental illness was associated with 0.26 (0.24 to 0.27) lower QoL.
| Coefficient (95% CI) | |
|---|---|
| Intercept | 0.879 (0.869 to 0.889) |
| Age spline 1 (< 70 years)a | –0.028 (–0.031 to –0.025) |
| Age spline 2 (≥ 70 years)a | –0.057 (–0.067 to –0.048) |
| Male | 0.034 (0.028 to 0.040) |
| Ethnicity (ref: white) | |
| Black | –0.002 (–0.022 to 0.018) |
| South Asian | –0.026 (–0.041 to –0.011) |
| Others | –0.016 (–0.036 to 0.004) |
| Deprivation quintiles (ref: IMD 3) | |
| IMD 1 | 0.026 (0.017 to 0.035) |
| IMD 2 | 0.010 (0.001 to 0.019) |
| IMD 4 | –0.019 (–0.029 to –0.010) |
| IMD 5 | –0.055 (–0.065 to –0.045) |
| BMI categories (ref: 18.5–25 kg/m2) | |
| < 18.5 | –0.031 (–0.066 to 0.004) |
| 25–30 | –0.014 (–0.021 to –0.006) |
| 30–35 | –0.039 (–0.049 to –0.03) |
| 35–40 | –0.086 (–0.101 to –0.071) |
| ≥ 40 | –0.128 (–0.152 to –0.105) |
| Current smoker | –0.057 (–0.066 to –0.049) |
| Ex-smoker | –0.021 (–0.028 to –0.014) |
| Treatment of hypertension | –0.026 (–0.034 to –0.017) |
| Mental illness | –0.267 (–0.281 to –0.253) |
| Cancer | –0.128 (–0.147 to –0.109) |
| Diabetes ≤ 10 years | –0.043 (–0.059 to –0.028) |
| Diabetes > 10 years | –0.085 (–0.105 to –0.065) |
| CVD history | |
| Angina only (ever) | –0.118 (–0.14 to –0.096) |
| MI only ≥ 1 year | –0.071 (–0.100 to –0.042) |
| Stroke only ≥ 1 year | –0.129 (–0.155 to –0.104) |
| Two or more conditions | –0.187 (–0.212 to –0.162) |
| CRV ≥ 1 year ago | 0.041 (0.014 to 0.067) |
| MI < 12 months ago | –0.101 (–0.164 to –0.038) |
| Stroke < 12 months ago | –0.087 (–0.136 to –0.039) |
| CRV < 12 months ago | 0.025 (–0.051 to 0.102) |
Summary of findings
We developed a QoL model using data from HSE 2006, 2011 and 2017, after aligning the definitions of covariates and model specification with characteristics in the CVD microsimulation model. This allowed the QoL model to be integrated into the CVD microsimulation model and used to estimate QALYs during model simulation.
It is worth noting that although the estimate of the cancer-related utility decrement was 0.13, the incident cancer’s contribution in the CVD microsimulation model was revised to 0.03. The main reason is that the HSE only recorded cancer affecting daily life, whereas the incident cancer in the CVD microsimulation model included any cancer except non-melanoma skin cancer. For example, in the 2011 HSE questionnaire, the related question was ‘do you have any long-standing illness, disability or infirmity? By long-standing I mean anything that has troubled you over a period of time, or that is likely to affect you over a period of time?’. Further cross-tabulating the reported cancer with the follow-up question, ‘does this illness or disability/do any of these illnesses or disabilities limit your activities in any way’, more than 60% of participants reporting a cancer history reported limited activities due to the illness in all the 2006, 2011 and 2017 surveys of HSE. However, the percentages of limited activities associated with cancer reported in other studies ranged between 10% and 50%.50–52 This indicated that the estimate based on HSE data is likely to overstate the impact of any previous cancer on QoL, and a cancer-related utility decrement of 0.03, informed by the literature,53–59 was used, instead.