Background The population attributable fraction (PAF) that estimates potentially community-level effect of risk factors can be useful in planning public health interventions. This study compared different methods for calculating adjusted PAFs for cardiovascular diseases (CVD) in a cohort study with 10 years of follow-up.

Methods Baseline data were employed from 6630 participants (3746 women) above 30 years old and 558 CVD events (238 women) detected during follow-up. Unadjusted approach using Levin's formula, Miettinen formula approach using adjusted OR and HR estimated from logistic and Cox regression and direct estimation of average PAF from logistic regression using Rückinger method, were applied.

Result Estimated PAFs, using HR comparing OR, in both Levin's and Miettinen's formula, with tiny decrease, gave very similar results. However, according to the average PAF method, frankly, we reach to lower fractions; highest modifiable cardiovascular risk factor PAFs, in sequence, was hypertension (16.2%), smoking (14.8%), diabetes (10.1%), hypercholesterolaemia (8.5%) for men, and hypertension (25.6%), diabetes (18%), hypercholesterolaemia (10.7%), for women. Also PAF of Age =60 years and premature family history of CVD, as most important non-modifiable CVD risk factors, were 18.4%, 11.2% and 4.4%, 6.9% for men and women respectively.

Conclusion If individual data (eg, cohort and case-control studies) are accessible then the direct estimation of average PAF provides more realistic results. Besides, it seems that the order of important risk factors is the same in men and women except smoking.