Dear Editor,The role of body mass index (BMI) in relation to lung cancer risk has been the object of considerable interest over the years. For the most part, epidemiologists' primary concern has been to determine whether BMI has an effect on lung cancer that is independent of smoking, a key confounder. We, therefore, read with great interest the quantitative appraisal of the epidemiological evidence on obesity and incidence of lung cancer by Yang et al. 1 Their metaanalysis revealed a significant inverse relationship between BMI, used as a proxy measure of overweight and obesity, and lung cancer based on results from 20 cohort and 11 case-control studies. We would like to comment on some inter-related methodological and interpretational aspects of the analyses presented, namely, heterogeneity across studies, methodological quality of original studies and confounding by smoking.
HeterogeneityAn overall estimate of association of 0.79 [95% confidence interval (CI): 0.73-0.85] was reported for excess weight (BMI 25 kg/m 2 ) compared with normal weight (BMI ¼ 18.5-24.9 kg/m 2 ) based on 31 studies. However, the inconsistency coefficient I 2 statistic, an estimate of the proportion of total variation in study estimates due to heterogeneity, was 82.9%. In the presence of such heterogeneity, a pooled estimate, even from a random-effects model, may not be justifiable as it does not provide a fair representation of overall results. Visual inspection of the Forest plot (Fig. 2), where study-specific risk estimates and their CIs are plotted, also indicates a heterogeneous set of studies. In particular, heterogeneity could be attributed to six cohort 2-7 and five case-control 8-12 studies. It would have been informative to identify which aspects of the individual studies contributed to heterogeneity. This could have entailed (1) identifying distinctive characteristics of individual studies; (2) pinpointing sources of variability between studies and thereby (3) providing guidance for presenting a valid quantitative pooled estimate of association between obesity and lung cancer.There was still a strong indication of heterogeneity in most stratified analyses across several important features defined by methodological characteristics. Results varied substantially within studies defined by study design, gender, study population, body size assessment method and smoking status. Of note, in analyses restricted to histological subtypes, heterogeneity might have tapered off as a result of the small number of studies included.
Methodological Quality of Original StudiesThe synthesis of results from individual studies did not include a formal evaluation of study quality. Analyses stratified according to a score for methodological quality could have provided a way to explore the sources of heterogeneity between studies. If results of this meta-analysis were to be significantly affected by the quality of original studies, then its conclusion may be less meaningful. Moreover, although the lack of symmetry in the funnel plot (Fig. 3) could ...