Depression and Diabetes - Methodological Issues in Etiologic Research

Abstract

Cross-sectional studies showed that depression and diabetes often co-occur. However, the reason for this is unclear. Depression may be a risk factor for diabetes or diabetes may be a risk factor for depression. The first aim of this thesis was to investigate these two associations. The second aim was to discuss several methodological issues to improve the conduct and reporting of etiologic research. We conducted a meta-analysis of nine longitudinal studies that investigated whether depression was a risk factor for type 2 diabetes, and showed that depressed adults had a 37% increased risk of developing diabetes compared with non-depressed subjects. Using a large pharmacy database, we showed that the risk of diabetes was not increased in users of antidepressants. Within diabetes patients, we showed that antidepressant use did not influence the amount of insulin used. We found that the prevalence of depressive symptoms was increased among patients who reported to be diagnosed with type 2 diabetes but not among patients with impaired fasting glucose or undiagnosed type 2 diabetes. Finally, we found that the incidence of antidepressant and benzodiazepine use was only increased shortly before and after the initiation of diabetes treatment. We created missing values in a confounder to assess the bias in the exposure-outcome association when using the missing indicator method, complete case analysis, and multiple imputation. We concluded that the missing indicator method should never be used to handle missing confounder data, that complete case analysis can only be used in rare situations, while multiple imputation gives unbiased effect estimates in most situations. We provided the methods to assist epidemiologists to estimate interaction on an additive scale between two continuous determinants using logistic regression analysis. We assessed how interaction was reported in 225 published cohort and case-control studies and found that only one in ten studies reported sufficient data to interpret interaction effects on an additive and multiplicative scale. We found a lower prevalence of depression among diabetes patients compared with patients without diabetes and showed that these unexpected findings could at least in part be explained by differential selection of subjects who visited their general practitioner for a control visit. Examination of 150 published case-control studies showed that, based on the method of selecting cases and controls, 57 studies required a stable population to estimate a rate ratio, 48 studies estimated a rate ratio without needing any assumption, and 16 studies estimated a risk ratio requiring the rare disease assumption. Based on the results of this thesis and previous studies, we cannot draw a firm conclusion on whether depression is a risk factor for diabetes, because the studies were too heterogeneous. There seems to be no increased risk of depression after a period of at least two years after the diagnosis of diabetes. There may be an increased risk of depression shortly after the diagnosis of diabetes. More knowledge on the point in time that the risk of depression may be increased after a diagnosis of diabetes, might direct specific monitoring of depression in diabetes patients.