LAUSR

There has been much discussion in the epidemiological literature about the so-called ‘obesity paradox’, where being overweight or obese appears to be protective for mortality in studies including individuals with prior cardiovascular disease (CVD). These inferences are based on the J-shaped associations observed between body mass index (BMI) and mortality in such populations, but it has been argued that these are a result of biases (such as selection bias) rather than obesity conferring a true survival advantage. Indeed, several studies have shown that, with appropriate adjustment for confounding and reverse causality (i.e. restricting analyses to non-smokers with no prior disease and excluding the first 5 years of follow-up), overweight and obesity are associated with an increase in the risk of all-cause mortality compared with normal weight. In this issue of the journal, Iliodromiti and colleagues report similar findings for the association between BMI and incident CVD in the UK Biobank. After excluding participants with prior co-morbidities, the observed J-shaped associations had almost disappeared, with monotonic relationships being observed above the reference value of 22 kg/m. Interestingly, the associations between the other adiposity measures considered [waist circumference (WC), waist-hip-ratio, waist to height ratio, and body fat percentage] and incidence of CVD did not appear to be influenced by confounding or reverse causality. The associations were unchanged by the exclusion of smokers and individuals with prior co-morbidities, as well as by omitting the first 2 years of follow-up. This could be a result of BMI being a poor discriminator of body fat and lean mass. Chronic illness can lead to a reduction in lean mass and subsequent weight loss, which would have a larger impact on BMI than other adiposity measures. The associations between BMI (restricted to participants with a BMI >_22 kg/m), WC, and waist-hip-ratio and incidence of CVD reported by Iliodromiti and colleagues are weaker than those observed in previous analyses. The Emerging Risk Factors Collaboration (ERFC) estimate that 4.56 kg/m higher BMI is associated with a 23% increased risk of CVD, defined as a composite of coronary heart disease and ischaemic stroke [hazard ratio (HR) 1.23, 95% confidence interval (CI) 1.17–1.29]. They also estimate that 12.6 cm higher WC and 0.083 higher waist-hip-ratio are associated with increases in CVD risk of 27% and 25%, respectively (HR 1.27, 95% CI 1.20–1.33; and HR 1.25, 95% CI 1.19–1.31). There are several possible explanations for these discrepancies, including: (i) differences in covariate adjustment; (ii) lack of specificity in CVD outcome definition; and (iii) measurement error. The ERFC presented HRs with minimal adjustment for age, sex, and smoking status, so they may still have been subject to some residual confounding. Iliodromiti and colleagues additionally adjusted the associations between adiposity measures and CVD for socio-economic status and physical activity, both of which are potential confounders. Blood pressure and diabetes were also included in the models despite being likely to lie on the causal pathway between adiposity and CVD, so the HRs may have been overadjusted. However, the HRs were only marginally higher in sensitivity analyses adjusting for age, socio-economic status, smoking, alcohol, and physical activity, so adjustment for blood pressure and diabetes had little effect on the observed associations. A broad definition of any incident event with an ICD code in the range I00–I99 is used as the primary CVD outcome by Iliodromiti and colleagues. This includes a lot of different types of cardiovascular disease, each of which might have different causal mechanisms. Studies