Heat-Related Mental Health Hospitalizations in Italy:A Global Sensitivity Analysis Approach to Evaluate Generalized...

High temperatures are increasingly associated with adverse mental health outcomes, yet the influence of structural modeling assumptions on these estimates remains underexplored. This study examined the short-term association between high temperatures and mental health-related-hospitalizations in 21 major Italian cities from 2005 to 2023, using national hospital discharge data. Exposure-lag-response relationships were modeled through a Distributed Lag Nonlinear Model (DLNM) framework estimated within a Generalized Additive Model (GAM). Analyses focused on June-September and included hospitalizations with a primary diagnosis of mental disorders, considering the code 295-316 of the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM). A Global Sensitivity Analysis (GSA) assessed how structural decisions, such as the specification of temperature and lag splines, and the inclusion of September, affect risk estimates. A total of 210,310 hospitalizations were recorded. The cumulative exposure-response curve showed a marked nonlinear increase in risk, reaching relative risk values close to two at temperatures exceeding 40 ${}^{\circ }C$ . The GSA revealed that the number and placement of knots in the temperature dimension were the dominant contributors to output variability, with total-order sensitivity indices approaching one across most of the temperature range. Variations in the lag structure contributed minimally, while including or excluding September influenced model fit but only modestly affected risk estimates. Uncertainty quantified through GSA was substantially larger than that quantified by standard confidence intervals, indicating that structural assumptions meaningfully shape inference. Within the DLNM-GAM framework, high summer temperatures were consistently associated with increased psychiatric hospitalizations. Incorporating GSA clarified which modeling choices influence estimates, improving transparency and robustness in evaluating heat-related mental health impacts.