Socio-Material Vulnerability and Youth Mental Disorders in Lombardy: A Bayesian Mapping Study

Abstract

Introduction
Socio-economic inequalities are well-documented drivers of a mental health gradient, with children and adolescents from low socio-economic backgrounds being disproportionately vulnerable to mental disorders. Evidence suggests that childhood deprivation not only increases the prevalence of mental health conditions but also restricts access to routine healthcare, contributing to poorer health and economic outcomes in adulthood. While this association is well-established, small-area spatial analyses examining the geographic distribution of mental disorders in children and adolescents in relation to composite deprivation indices remain scarce, particularly in the Italian context.

Objective
To assess whether the geographical distribution of children and adolescents with at least one hospitalisation or Emergency Department (ED) visit for mental disorders in Lombardy is associated with the Social and Material Vulnerability Index (SMVI).

Methods
A population-based spatio-temporal ecological study was conducted using regional administrative healthcare data from the Lombardy (2016–2022). The study population included all residents aged 0–18 years as of December 31st of each year. A mental disorder case was defined as any individual with at least one ED visit or hospital discharge recording an ICD-9-CM code in the range 290–319 in any diagnostic field during the reference year. The SMVI, developed by the Italian National Institute of Statistics (Istat), was categorized into quintiles ranging from Q_1 (lowest vulnerability) to Q_5 (highest vulnerability). Observed and expected case counts were modelled using Bayesian hierarchical scaled Besag–York–Mollié (BYM2) models to estimate age- and sex-adjusted area-specific relative risks, interpretable as smoothed SMRs. Three model specifications were compared: a purely spatial model, a spatio-temporal model with a Random Walk of order 1 (RW1) temporal component, and a spatio-temporal model with a Knorr–Held Type I interaction. Area-specific relative risks (RR) and posterior probabilities (PP) were estimated. Analyses were performed in R using INLA framework.

Results
Findings were consistent across all three Bayesian BYM2 model specifications. Compared with the lowest SMVI quintile, only areas in the highest vulnerability quintile showed a credible association with increased risk of mental disorders diagnoses among youths, with posterior relative risks of 1.096 (95% CrI: 1.017-1.182), 1.097 (1.018-1.182), and 1.101 (1.019-1.190) in the spatial, spatio-temporal, and spatio-temporal interaction models, respectively. No statistically credible associations were found for intermediate quintiles. After adjusting for SMVI, residual spatial risk showed a marked west–east gradient across the region. Areas with high posterior probability of excess risk (PP>0.80) were concentrated in the Milan metropolitan area, the north-western pre-alpine belt, the Pavia area, and a few localized hotspots near Lake Garda, whereas central-eastern Lombardy and the alpine area of Sondrio showed consistently lower-than-expected risk (PP<0.20).

Conclusions
The study highlights that the SMVI captures a meaningful share of the phenomenon's variability, showing an increased risk of mental disorders in areas with the highest vulnerability. However, the persistence of strong spatial clusters in the residual risk maps suggests that the SMVI does not fully explain the heterogeneous territorial distribution of cases in Lombardy. This indicates the presence of additional local, environmental, or healthcare supply and access determinants that warrant further investigation.