Are women really at a 2-fold increased risk to develop lifetime depression, compared to men?

BackgroundA gender gap in lifetime major depressive disorder (women to men ratio 2:1) has consistently been reported over roughly three decades. However, suggested artefacts, e.g., different biases between women and men regarding their reporting of depression symptoms, have never been taken into account quantitatively. This research gap may have led researchers to report overly confident estimates of a depression gender gap. Overconfident reports may have serious and manifold implications, e.g., the repeated reporting of a 2-fold increased risk to develop depression may be internalized by some women, maybe causing them to firmly believe that the female human nature is more vulnerable to depression than the male human nature. We sought to reassess the scientifically claimed 2-fold increased risk for depression in women, compared to men, while quantitatively taking the possibility of different reporting biases into account.MethodsWe employed the openly accessible National Survey on Drug Use and Health (NSDUH) data from the years 2012 to 2021. The NSDUH is an annual cross-sectional survey among the general US population. Our analyses included adult (age 18 years or older) survey participants, because only they were assessed for symptoms of a lifetime major depressive episode. We split the total sample into many subsets, according to the results of the required minimal sample size, given the assumed effect size (risk ratio and risk difference, respectively), statistical power, and the statistical significance level. We then estimated both effect sizes for each subsample, first not taking outcome misclassification bias into account, and then taking it into account, which is part of a quantitative bias analysis (QBA). Using meta-analyses, we visualized the possible change of scientific confidence in the empirically estimated depression risk of women, compared to men.ResultsThe total sample size of 399’751 (53.9% females) made it possible to estimate between 180 and 906 separate risk ratios and risk differences, respectively, e.g., 906 subsamples, each subsample consisting of 441 adults. Overall, the prevalence of a lifetime major depressive episode was 16.45%. One exemplary meta-analytic risk ratio (no. 10 out of 15) without applying QBA is 1.63 (99% Prediction Interval (PI) 1.60–1.66), as opposed to 1.69 (99% PI 1.31–2.17) when applying one out of 24 tested QBA scenarios (scenario no. 6). Applying QBA widens the prediction intervals, compared to ignoring systematic measurement error. Depending on which, if any, of the 24 outcome misclassification scenarios approximates real female and male reporting biases of depression symptoms, the prediction intervals may widen so much that the gender gap of depression can even disappear, based on the NSDUH data.DiscussionJudging by our presented visualizations of the meta-analytic risk ratios and risk differences, respectively, the confidence in the scientific claim of a depression gender gap declines, after quantitatively taking outcome misclassification bias into account, compared to ignoring this source of systematic measurement error. The decline in scientific confidence is due to the considerable widening of the meta-analytic prediction intervals, which take the heterogeneity of the bias adjusted effect estimates into account. We do not claim the absence of a gender gap of depression. Indeed, in the face of the many subtle discriminatory experiences which most women make every day in exceedingly neoliberal societies, no depression gender gap at all seemed implausible. We sought to relativize the scientific confidence with which the depression gender gap is claimed, by applying QBA. The main limitation of our study is our focus on only one specific type of misclassification bias, which may have widened the prediction intervals too much. If this was true, this would decline the scientific confidence too much, regarding the possible truth of a depression gender gap. The main strength of our study is that all study details, e.g., the raw data and analysis code, is openly available to everyone, in accordance with open data/open science recommendations.