CV Disease Policy Model Using Socioeconomic Deprivation
CV Disease Policy Model Using Socioeconomic Deprivation
A total of 16 560 SHHEC participants were free of CVD at baseline, 8611 (52%) were women. The distribution of risk factors is shown in Table 1. The median survival time to first event was 20.8 years (IQR 14.5, 23.6 years). A total of 6175 people (37.3%) had a first event observed during the follow-up period.
We illustrate how the model moves from risk of events to remaining life expectancy using the following individual profile: 60-year-old man, no family history of CVD, non-diabetic, SIMD score of 60.8, SBP of 160 mm Hg, TC of 7 mmol/L, HDL of 1 mmol/L and 20 CPD. The predicted cumulative incidence of the four first events and survival probabilities after a first non-fatal event (as well as the survival model estimates that produce the predictions) are detailed in the online supplementary appendix http://heart.bmj.com/content/101/3/201/suppl/DC1.
Table 2 shows how life expectancy is estimated. The top panel of Table 2 shows remaining life expectancy estimated at the end of every cycle and for each of the four first events. For example, at the end of cycle 24 the estimated life expectancy for a man experiencing a non-fatal CHD event at that time is the 24 years alive and event free plus 3.3 years, which is the estimated life expectancy following the event. The probabilities of having a first event for each cycle are shown in the middle panel of Table 2. These are obtained from the cumulative incidence curves by subtracting the cumulative probability from the previous year from the current year. The bottom panel of Table 2 shows the weighted life expectancies which are obtained by multiplying together the values in the corresponding cells in the previous two panels. The sum of all the cells in the bottom panel of Table 2 is the remaining expected life expectancy, which for this illustration equals 14.07 years (so life expectancy is 60+14.07=74.07 years).
The discrimination of all statistical models was good with c-statistics in the range 0.65–0.80 (see online supplementary appendix http://heart.bmj.com/content/101/3/201/suppl/DC1). Discrimination was better for the first event models than the models following a first non-fatal event, better for the fatal CVD outcome compared with the other competing first events, and generally better for models for women than for men.
Figure 2 shows the comparison between observed cumulative incidence of first events from WOSCOPS with predictions made with our model. For the placebo arm, the predicted line falls well within the CI limits for non-fatal CBVD and fatal CVD events. However, the model underpredicts for non-fatal CHD and overpredicts for fatal non-CVD. The latter may be explained by the fact that WOSCOPS is a clinical trial where stringent exclusion criteria can result in lower mortality than observed in the general population. For the treatment arm, the agreement is good for all cardiovascular endpoints (given the exclusion of cancers from the trial), illustrating that the model has the potential to predict the impact of CVD interventions.
(Enlarge Image)
Figure 2.
A and B, Validation of model predictions using West of Scotland Coronary Prevention Study trial population. CBVD, cerebrovascular disease; CHD, coronary heart disease; CVD, cardiovascular disease.
Without applying a calibration factor, the original model overpredicted life expectancy. After applying a calibration factor (see online supplementary appendix http://heart.bmj.com/content/101/3/201/suppl/DC1 for details), all model estimates were within half a year of life table estimates with the exception of 80-year-old women.
The online supplementary table http://heart.bmj.com/content/101/3/201/suppl/DC1 shows predicted life expectancies for men and women across a wide range of individual profiles. The table shows how age, SBP, cholesterol, smoking (where 'smoker' is defined as somebody smoking 20 CPD) and SIMD influence life expectancy while holding diabetes and family history of heart disease constant at their average values. It can be seen that even after adjusting for traditional CVD risk factors, socioeconomic deprivation has a large impact on life expectancy. To illustrate the socioeconomic deprivation gradient, figure 3 shows the predicted life expectancies for 60-year olds across ranked fifths of SIMD scores (within each ranked group average values risk factor values were obtained from the Scottish Health Survey 2009). The difference in predicted life expectancy using the SII to compare the least and most deprived groups is 6.8 and 5.7 years for men and women, respectively.
(Enlarge Image)
Figure 3.
Predicted life expectancy for ranked fifths of socioeconomic deprivation (SIMD) scores.
Results
Description of Demographics and Event Outcomes
A total of 16 560 SHHEC participants were free of CVD at baseline, 8611 (52%) were women. The distribution of risk factors is shown in Table 1. The median survival time to first event was 20.8 years (IQR 14.5, 23.6 years). A total of 6175 people (37.3%) had a first event observed during the follow-up period.
Illustration of the Three Modelling Stages
We illustrate how the model moves from risk of events to remaining life expectancy using the following individual profile: 60-year-old man, no family history of CVD, non-diabetic, SIMD score of 60.8, SBP of 160 mm Hg, TC of 7 mmol/L, HDL of 1 mmol/L and 20 CPD. The predicted cumulative incidence of the four first events and survival probabilities after a first non-fatal event (as well as the survival model estimates that produce the predictions) are detailed in the online supplementary appendix http://heart.bmj.com/content/101/3/201/suppl/DC1.
Table 2 shows how life expectancy is estimated. The top panel of Table 2 shows remaining life expectancy estimated at the end of every cycle and for each of the four first events. For example, at the end of cycle 24 the estimated life expectancy for a man experiencing a non-fatal CHD event at that time is the 24 years alive and event free plus 3.3 years, which is the estimated life expectancy following the event. The probabilities of having a first event for each cycle are shown in the middle panel of Table 2. These are obtained from the cumulative incidence curves by subtracting the cumulative probability from the previous year from the current year. The bottom panel of Table 2 shows the weighted life expectancies which are obtained by multiplying together the values in the corresponding cells in the previous two panels. The sum of all the cells in the bottom panel of Table 2 is the remaining expected life expectancy, which for this illustration equals 14.07 years (so life expectancy is 60+14.07=74.07 years).
Discrimination, Validation and Calibration of Model
The discrimination of all statistical models was good with c-statistics in the range 0.65–0.80 (see online supplementary appendix http://heart.bmj.com/content/101/3/201/suppl/DC1). Discrimination was better for the first event models than the models following a first non-fatal event, better for the fatal CVD outcome compared with the other competing first events, and generally better for models for women than for men.
Figure 2 shows the comparison between observed cumulative incidence of first events from WOSCOPS with predictions made with our model. For the placebo arm, the predicted line falls well within the CI limits for non-fatal CBVD and fatal CVD events. However, the model underpredicts for non-fatal CHD and overpredicts for fatal non-CVD. The latter may be explained by the fact that WOSCOPS is a clinical trial where stringent exclusion criteria can result in lower mortality than observed in the general population. For the treatment arm, the agreement is good for all cardiovascular endpoints (given the exclusion of cancers from the trial), illustrating that the model has the potential to predict the impact of CVD interventions.
(Enlarge Image)
Figure 2.
A and B, Validation of model predictions using West of Scotland Coronary Prevention Study trial population. CBVD, cerebrovascular disease; CHD, coronary heart disease; CVD, cardiovascular disease.
Without applying a calibration factor, the original model overpredicted life expectancy. After applying a calibration factor (see online supplementary appendix http://heart.bmj.com/content/101/3/201/suppl/DC1 for details), all model estimates were within half a year of life table estimates with the exception of 80-year-old women.
Illustration of Life Expectancies
The online supplementary table http://heart.bmj.com/content/101/3/201/suppl/DC1 shows predicted life expectancies for men and women across a wide range of individual profiles. The table shows how age, SBP, cholesterol, smoking (where 'smoker' is defined as somebody smoking 20 CPD) and SIMD influence life expectancy while holding diabetes and family history of heart disease constant at their average values. It can be seen that even after adjusting for traditional CVD risk factors, socioeconomic deprivation has a large impact on life expectancy. To illustrate the socioeconomic deprivation gradient, figure 3 shows the predicted life expectancies for 60-year olds across ranked fifths of SIMD scores (within each ranked group average values risk factor values were obtained from the Scottish Health Survey 2009). The difference in predicted life expectancy using the SII to compare the least and most deprived groups is 6.8 and 5.7 years for men and women, respectively.
(Enlarge Image)
Figure 3.
Predicted life expectancy for ranked fifths of socioeconomic deprivation (SIMD) scores.