Examining the Obesity-Mortality Association in Individuals with Chronic Conditions

Notwithstanding the established association between obesity and a plethora of diseases, several studies have reported a lower risk of all-cause or cause-specific mortality in overweight or obese individuals who already suffer from these conditions, as compared to their normal weight counterparts. These counterintuitive findings are termed as ‘obesity paradox’. Examples of this paradoxical survival benefits of obesity have been extensively reported in studies examining the association between obesity and mortality in individuals with type 2 diabetes mellitus (T2DM) and hypertension - two common consequences of obesity. A number of potential explanations have been proposed, most often around pathophysiological mechanisms but also around, though less often, several analytical biases. These include a special form of selection bias termed as collider bias; confounding due to illness-related weight loss (reverse causation); and/or insufficient adjustment for confounding due to smoking. Nevertheless, the reasons explaining this ‘paradox’ are still unclear and need to be unravelled. Noteworthy, in the light of these paradoxical findings, there are some suggestions that the weight management guidelines may need to be revised (that is, relaxed) for patients with established chronic diseases. However, it may be that the apparent protective effects of obesity for chronic conditions are merely due to artefacts or biases in the study designs or analyses. If so, then it is essential to first ascertain the extent of these biases and correct them before explanatory pathophysiological mechanisms are further explored and potentially deleterious weight management recommendations are advanced.The overall objective of this thesis was to further understand the relationship between obesity and mortality in individuals with chronic conditions (T2DM and hypertension). This thesis comprises of three main studies that use various statistical/epidemiological and data analysis methods. Statistical methods used in this thesis include multilevel meta-analysis and network meta-analysis methods to quantitatively summarise multiple effect estimates nested in the studies; sampling fractions technique to estimate the magnitude of selection bias and to obtain corrected unbiased estimates; and Cox proportional hazards regression to model the association between obesity and mortality.The first study is a systematic review and meta-analysis that reports on pooled estimates of studies exploring obesity-mortality association in participants with hypertension. Results of this meta-analysis demonstrated that obesity was associated with higher mortality risk in individuals with hypertension. Similar results were obtained in the subgroup analysis when individuals with and without hypertension from the same population were included in the analysis; however, when the analysis was restricted to individuals with hypertension no statistical difference for being overweight or obese was found as compared to normal weight.The second study, a systematic review and meta-analysis, quantitatively summarized the existing literature on the relationship between obesity and mortality in individuals with and without T2DM. The results of this meta-analysis revealed that obesity was associated with statistically significant higher mortality risk in the un-stratified analysis. Similar to the results of the first study, when we limited the analysis to participants with diabetes, obesity was not found as a significant predictor of mortality. Sampling fraction method demonstrated that selection bias was not of sufficient magnitude to reverse the direction of association.The third study examined a longitudinal population-based Australian cohort (the Australian Diabetes, Obesity and Lifestyle Study – AusDiab). The examined cohort included 10,575 Australian adults followed up for a median of 14.6 years. Cox proportional hazards regression approach was used to estimate mortality in the entire cohort as well as in subpopulations with and without diabetes. Results of this longitudinal study are in agreement with the meta-analyses included in this thesis and revealed no evidence of obesity paradox in the entire cohort or when the analyses were restricted to a subpopulation with the disease. In addition, we found no difference in the obesity-mortality association between individuals with or without diabetes.The findings of this research confirm the absence of obesity paradox in the subpopulations with hypertension and diabetes. It provided estimates of association when the analysis was restricted to individuals with the disease. Although the risk estimates do drop when analyses were limited to participants with diabetes or hypertension, we found no evidence that obesity behaved differently in those with and without the disease. The role of collider bias in the apparent protective effect of obesity is less clear, and better designed studies including clinical trials and weight trajectories exploration should be conducted to understand and inform, in an evidence-based manner the concepts surrounding the notion of adiposity as being healthy, to effectively mitigate the future burden of obesity.