The tyranny of the averages and the indiscriminate use of risk factors and population attributable fractions in Public Health: The case of coronary heart disease

INTRODUCTION: Modern medicine is overwhelmed by a plethora of both established and novel risk-factors for diseases. This knowledge is based on measures average association (i.e., differences in average risk between exposed and unexposed groups) like the relative risk (RR). We also use the RR to calculate the population attributable fraction (PAF). However, measures average association do not assess the discriminatory accuracy of the risk factors (i.e., its ability to discriminate the individuals who will develop the disease from those who will not). Focusing on coronary heart disease (CHD) and applying measures of discriminatory accuracy, we aim to revisit the role of risk factors and PAFs in public health and epidemiology. 

METHODS: We used information from 6103 individuals who participated in the Malmö Diet and Cancer study in Malmö, Sweden (1991-2009). We performed logistic regression models including (A) age and sex, (B) traditional risk factors (blood pressure, cholesterol, diabetes, smoking), and (C) biomarkers (CRP, NTBNP, Cystatin C, LpPLA₂ activity) and combinations of A, B, C. We calculated measures of discriminatory accuracy (e.g., AU-ROC, risk assessment plots) and PAF for different thresholds of a risk score for coronary heart disease

RESULTS: Compared with model A (AU-ROC=0.68), model A+B improved the discriminatory accuracy by 0.07 units and model A+B+C by 0.08 units. For a risk factor prevalence of 9% the PAF was 60% and the false positive fraction 30%. For a risk factor prevalence of 90% these values were 93% and 90% respectively

CONCLUSIONS: Neither traditional risk factors nor biomarkers substantially improved the discriminatory accuracy obtained by simple models considering only age and sex. The PAF measure provided misleading information for planning preventive strategies in the population. We need a fundamental change in the way we currently quantify and interpret risk factors in public health epidemiology.