Macro-level predictors and temporal persistence of global mental health burden: a machine-learning panel study, 1990–2023
Mental disorders contribute substantially to global disability and economic burden, yet relatively little is known about how macro-level national indicators relate to cross-country differences in mental health burden. We constructed a country–year panel for 183 countries from 1990 to 2023 by combining data from the Global Burden of Disease, the World Bank, and Our World in Data. Two composite indices were derived from age-standardized prevalence and DALY rates for depressive disorders, anxiety disorders, bipolar disorder, eating disorders, and schizophrenia. The analysis is exploratory and predictive rather than causal. With a time-based split, the models got trained on the 1990–2010 slice and then assessed using 2011–2023 observations. For the non-lagged macro variants, they were able to explain a notable part of the cross-national variation, but the lagged versions mostly ended up reflecting strong temporal carryover in mental health burden. Across the setups, GDP per capita, urbanization, life expectancy, alcohol consumption, unemployment, PM2.5 exposure, and refugee outflows consistently emerged as important predictive correlates. Regional models show heterogeneity. In East Asia and the Pacific, income, pollution, and alcohol use looked more prominent. In contrast, for Europe, the Middle East and North Africa, refugee flows and development indicators were the more prominent predictors. Robustness checks included alternative temporal splits, complete case analysis, bootstrap uncertainty intervals, calibration assessment, feature rank stability, and panel-style benchmarks. Overall the results point toward development, labour market conditions, environmental exposure, and displacement as important macro-level variables associated with aggregate mental health burden, even if the interpretation stays ecological and not causal. Clinical trial registration: Not applicable. This study was not a clinical trial.