Predicting risk of cardiovascular disease and the cost-effectiveness of interventions in Thailand Stephen Lim On Behalf of the Setting Priorities using Information on Cost-Effectiveness (SPICE) Project
Top Ten Causes of Disability Adjusted Life Year (DALYs) by Sex, Thailand 1999
Thai Burden of risk factors, 1999
Prevention of CVD 2 different but complementary approaches to prevention: 1. Population-wide approach – aims to reduce levels of risk factor(s) across the whole population 2. High risk approach – targets prevention towards those who are at higher risk, e.g. high blood pressure, high cholesterol
Targeting high-risk How do we target those at high risk? Traditionally, by thresholds of individual risk factors, e.g. systolic blood pressure ≥ 140mmHg (hypertension) More recent approach uses absolute risk of CVD in, e.g. next 10 years E.g. using risk prediction equations from the Framingham study
Absolute risk Absolute risk of CVD takes into account 1. Multiple risk factors determine CVD risk age, sex, blood pressure, cholesterol, smoking, diabetes, etc
Absolute risk Absolute risk of CVD takes into account 1. Multiple risk factors determine CVD risk age, sex, blood pressure, cholesterol, smoking, diabetes, etc 2. Continuous measurements of risk factors e.g. relationship between blood pressure and CVD is not dichotomous (i.e. having hypertension or not having hypertension) but is continuous
Source: World Health Report 2002
Absolute risk Absolute risk of CVD takes into account 1. Multiple risk factors determine CVD risk age, sex, blood pressure, cholesterol, smoking, diabetes, etc 2. Continuous measurements of risk factors e.g. relationship between blood pressure and CVD is not dichotomous (i.e. having hypertension or not having hypertension) but is continuous An individual with moderately elevated levels of multiple risk factors may be at higher risk than an individual with high levels of a single risk factor
Risk prediction equations Determination of absolute risk is based on cohort studies examining the relationship between risk factors and CVD outcomes Uses survival analysis (Cox regression or Weibull models) to determine predictive risk equation Many of the risk equations in use are based on the Framingham study These have been validated and “adjusted” for use in other cohorts and settings, e.g. China, Australia, Europe, New Zealand
Risk prediction equations The Electricity Generating Authority of Thailand (EGAT) cohort study provides important information on the relationship between risk factors and CVD outcomes in a Thai population 3,499 employees of EGAT (2,702 males, 797 females) aged years Physical examinations (including blood) 1985, 1997, 2002 Information on a range of fatal and non-fatal CVD events 2 nd cohort of individuals followed from 1997
Risk prediction equations EGAT: Developed a range of risk prediction equations Coronary Heart Disease (CHD), Diabetes Equations used to develop a point score system for predicting absolute risk Validation of other risk prediction equations from the Framingham study and China cohorts Show, like other studies, that Framingham equations predict relative risk well, but overestimate absolute risk
EGAT-SPICE collaboration Use EGAT equations to determine predicted CHD risk for individuals in the National Health Examination Survey 3
Cox proportional hazards model from EGAT 2418 subjects, 74 CHD events Developed by Dr Sukit
Apply EGAT score to NHES 3 Using raw data from NHES No sample weights Not yet cleaned Apply to males aged only Excluding HDL as this is not measured Some inconsistencies between EGAT and NHES risk factors definitions NHES: Alcohol in last 12 months EGAT: Current alcohol use
Apply EGAT score to NHES 3 Cox-proportional harzards model was used to determine individual risk Risk estimate = 1 – S 0 (t) exp(∑βX- ∑βX) where,S 0 (t) is the average survival time at time t β’s are the Cox-regression coefficients X are the individual RF values X are the mean RF values
Apply EGAT score to NHES 3 For male aged 55, SBP 160, Chol 250mg/dl, diabetic, smoker, waist 102cm, no alcohol ∑βX = age* sbp* Tch* diabetes* smoke* alcohol* waist90* ∑βX = 55* * * * * * * = S 0 (10) from Kaplan-Meier estimate from EGAT is ∑βX is Risk estimate = 1 – exp( ) = % risk of CHD event over the next 10 years
Preliminary analysis – Please do not quote
Overall predicted 10-year CHD risk Predicted risk of CHD lower in NHES 3 (0.62% compared with 1.09% from EGAT) Preliminary analysis – Please do not quote
Comparison of mean RF values Preliminary analysis – Please do not quote
Ongoing work Repeated measures analysis Currently using only 1985 examination with 17 year follow-up Repeated measures allows us to use 1997, 2002 examination also Causal web estimation using hierarchical models
Causal web
Risk prediction equations Limitations Males aged not sufficient numbers to generate risk equation for women Time period is Risk of CVD in this period may be quite different from risk of CVD today Alternative approach is to calibrate Framingham risk prediction equations for use in Thailand
Calibration of absolute risk Population estimates of disease incidence, e.g. from Thai BOD Framingham Risk prediction + Absolute risk specific to the population Adjusted for local risk factor prevalence and underlying risk + Risk factor prevalence data e.g. from NHES3
Example Framingham 1-year CHD risk for this individual is Framingham 1-year CHD risk for NHES females aged 55 is Female, 55 years, total cholesterol 6.7, no diabetes, current smoker, SBP 140mmHg = 4.70 Risk for this individual relative to all Thai females aged 55: Population-level incidence of CHD for females aged 55 is Individual calibrated CHD risk is 4.7 * = In other words, this individual has a 2.4% chance of having a CHD event in the next year RR =
Risk prediction Approach adjusts for: Risk factor prevalence (NHES) Underlying risk of CVD (population-level incidence of CVD from Thai BOD) Underlying assumption is that relative risk of risk factors is the same across the two populations Supported by EGAT data for males
Issues for CVD prevention Many different strategies exist for reducing the risk of CVD How can we target high-risk individuals? Traditional approach using thresholds of individual risk factors, e.g. systolic blood pressure ≥ 140mmHg Absolute risk approach takes into account multiple risk factors e.g. age, sex, blood pressure, cholesterol, smoking, diabetes Should a cholesterol test be included to identify high- risk individuals?
Issues for CVD prevention Due to these difficulties, it is likely that the large amount of resources that are devoted to preventive strategies for CVD are not being used in an optimal manner. Cost-effectiveness analysis can tell us which interventions are optimal given currently available resources Which mix of strategies is most efficient in reducing the burden of CVD?
Modelling cost-effectiveness Rely on state transition (“Markov”) models Portions of a cohort move through different mutually exclusive states over time Movement between states is determine by transition probabilities Model the current Thai population in terms of CVD outcomes over time
Modelling cost-effectiveness Year 1 Year 2 Year 0 ALIVECHD DEAD
Modelling cost-effectiveness Transition probabilities for CVD can be determined in a similar way to calibration of CVD absolute risk equations Allows us to simulate individuals with different risk factor profiles / absolute risk through the state transition model
Modelling cost-effectiveness Repeat process under “no intervention” and “intervention” scenarios e.g. statins may reduce the transition between ALIVE and CHD by 30% Can then determine health years gained by the intervention cost of interventions potential cost savings due to reduced cases of CVD Cost-effectiveness
Natural history of CVD Model structure depends on natural history of the disease being modelled 2 major types of CVD events Acute coronary syndromes (ACS), including myocardial infarction and unstable angina pectoris Major sequelae are angina and heart failure Stroke including both hemorrhagic and ischemic sub- types Multiple risk factors for both ACS and stroke Age, sex, blood pressure, cholesterol, diabetes, etc
Natural history of CVD Prognosis of both ACS and Stroke are similar (very) high case-fatality in first 28-days risk of mortality in 28-days survivors remains elevated thereafter Need to differentiate between initial mortality (first 28-days or first year) and risk of mortality thereafter
T7 T5 T3 T1 T4 CVD model structure
Data sources EGAT Incidence:mortality ratios Some information on case-fatality rates Risk prediction equation Vital registration with cause of death corrections Mortality from CVD National Health Examination Survey Self-reported prevalence of CHD and stroke Risk factor prevalence ACS and Stroke registries In-hospital case fatality Major limitation is lack of information on out-of- hospital case-fatality
Results from analysis in Australia
Reasons for inefficiency of current practice in Australia Absolute risk vs Risk factor thresholds Not enough attention to lifestyle and public health interventions Community programs Dietary counselling Phytosterol supplementation Current resources directed at less efficient classes of BP lowering drugs e.g. ACE inhibitors
Summary There is potential to increase the efficiency of CVD prevention efforts with : The development of robust tools to predict absolute risk of CVD in clinical practice Estimates of the cost-effectiveness of different prevention strategies