Integrating ESG indicators and behavioral factors to estimate the mental health index in Italy Authors: Resta Emanuela, Logroscino Giancarlo, Peter Preethymol, Costantiello Alberto, Leogrande Angelo Published: 2026 ABSTRACT AbstractThis article examines the relationship between behavioral and Environmental, Social, and Governance (ESG) variables and the Mental Health Index (MHI) of Italian regions from 2016 to 2023. The use of descriptive statistics, cluster analysis, and forecasting models is employed for descriptive and exploratory purposes, as well as for forecasting regional disparities in mental health. Relative stability at the state level and persistent territorial variability are evident in the MHI derived from the ISTAT-BES indicators, with the North and Centre performing better than the South, except for notable improvements in disadvantaged regions such as Calabria and Campania. Behavioral factors, particularly sedentariness, smoking, and multimorbidity, are consistently associated with adverse trends for mental health, while adequate nourishment and exposure to urban green spaces are found to be protective. ESG-related factors, including occupational security and judicial efficacy, also contribute to these inequalities, attributing mental health status as part of the more general system effects of institutions and environments. Cluster analysis identifies distinct regional typologies based on aggregated behavioral and ESG patterns, while the predictive model highlights multimorbidity, smoking, and sedentary behavior as the most relevant factors negatively associated with the Mental Health Index (MHI), and urban green availability as a key factor positively associated with it. They point out that Italian mental health disparities are not only associated with behavioral determinants of individuals, but must be addressed from a multi-faceted ESG perspective. Policy reactions also mean that public health responses must attend not only to lifestyle risk factors but also structural determinants and incorporate active living promotion, reduction of smoking prevalence, investment in green infrastructures, and improvement of governance efficiency.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s13033-026-00716-x.Keywords: Mental health index; ESG (Environmental, Social, and Governance); Regional disparities; Machine learning analysis; Public policy and well-being Introduction The present research contributes to the existing literature by integrating conceptual, empirical, and methodological advances within a single analytical framework. First, from a conceptual perspective, the study introduces an ESG-based approach to the analysis of mental health inequalities, explicitly combining environmental quality, governance efficiency, and social conditions—dimensions that are typically examined separately in prior Italy-focused studies on health disparities [1, 2]. Second, from an empirical standpoint, the paper provides an updated regional panel analysis for the period 2016–2023 based on ISTAT-BES data, offering recent and consistent evidence on territorial heterogeneity across Italian macro-regions, including the more advanced North, the transitional Centre, and the structurally weaker South and Islands [3, 4]. Third, from a methodological perspective, the study adopts a combined clustering–prediction design, using k-means clustering to identify six homogeneous regional typologies and predictive modeling to assess the relative importance of the main determinants of the Mental Health Index. Within this framework, multimorbidity, smoking, and sedentariness emerge as the most relevant negative predictors, while urban green availability represents a key positive factor. Overall, the novelty of the paper lies not in each element taken in isolation, but in their joint integration—an approach that has not been previously applied in an integrated way to the analysis of mental health inequalities across Italian regions—thereby linking ESG conditions, regional disparities, and the main drivers of mental health outcomes, and supporting policy implications aligned with the SDGs, including health, sustainable cities, and social justice [5, 6]. Methodological integration of clustering and predictive modelling to analyze ESG and mental health interlinkages in Italian regions This study adopts a two-stage analytical framework to investigate regional patterns of mental health in Italy and their relationships with behavioral and Environmental, Social, and Governance (ESG) factors. In the first stage, unsupervised clustering methods are used to identify homogeneous typologies in mental health across Italy. Six clustering methods are used, namely Density-Based Clustering, Fuzzy C-Means Clustering, Hierarchical Clustering, Model-Based Clustering, Random Forest Clustering, and K-Means Clustering. The selection of clustering methods is informed by their performance across multiple evaluation criteria, including R2, AIC, BIC, Silhouette Score, Dunn Index, Entropy, and Calinski-Harabasz Index. The K-Means Clustering method is found to be the most suitable for this study, as concluded by Resta et al. [7]. The Elbow Method is used to determine the most suitable clusters for representing the Italian regions as homogeneous groups in terms of mental health typologies, considering both behavioral and ESG factors, as well as the distinction between northern and southern Italian regions, and it is concluded that six clusters are enough. Defining the ESG and behavioral foundations of mental health across Italian territories This section aims to demarcate the different variables used to measure the Mental Health Index (MHI) in relation to ESG behavioral, environmental, and social factors. The MHI measures the psychological well-being of individuals aged 14 and older. The Mental Health Index (MHI) is derived from the Mental Health (MH) component of the SF-36 (36-Item Short Form Health Survey), a widely used psychometric instrument in population-based studies. The index is based on the aggregation of responses to five specific items capturing psychological distress and well-being, including dimensions related to anxiety and depression. The resulting score ranges from 0 to 100, with higher values indicating better mental health conditions and higher levels of psychological well-being [8]. Major variables include sedentariness, alcohol consumption, smoking, nutrition, multimorbidity, availability of urban green spaces, job security, and the length of civil trials. These factors are associated with mental health in different ways. Behavioral factors like smoking and alcohol consumption have strong correlations with psychological well-being [9, 10]. In contrast, socioeconomic factors and access to green spaces are positively associated with psychological well-being [11]. Therefore, the variables provide a multidimensional ESG approach to analyzing territorial determinants of psychological well-being. See to Table 1. Table 1Description of variables and their relevance to the mental health index in the ESG frameworkAcronymVariableDescriptionMHIMental Health IndexComposite indicator of psychological well-being derived from five SF-36 mental health itemsSEDSedentarinessShare of individuals aged 14 + not engaging in regular physical activityALCAlcohol ConsumptionPercentage of individuals aged 14 + involved in risky drinking behaviorsSMKSmoking RatePrevalence of smoking among individuals aged 14 and aboveNUTRAdequate NutritionShare of individuals consuming at least four daily portions of fruit and vegetablesMORBMultimorbidityPercentage of elderly individuals with three or more chronic health conditionsUGAUrban Green AvailabilityUrban green space per capita, measuring environmental quality and accessibilityOSPOccupational Security PerceptionShare of workers perceiving high risk of job loss and limited reemploymentCTDCivil Trial DurationAverage duration of civil court proceedings indicating institutional efficiencySource: ISTAT-BES. Link: https://www.istat.it/statistiche-per-temi/focus/benessere-e-sostenibilita/la-misurazione-del-benessere-bes/gli-indicatori-del-bes/, accessed 10 October 2024 Description of variables and their relevance to the mental health index in the ESG framework Source: ISTAT-BES. Link: https://www.istat.it/statistiche-per-temi/focus/benessere-e-sostenibilita/la-misurazione-del-benessere-bes/gli-indicatori-del-bes/, accessed 10 October 2024 Regional dynamics and temporal evolution of mental health in Italy (2016–2023): a static analysis of spatial disparities and macro-area trends Below we present a static analysis relating to the ISTAT-BES findings regarding the trend of the mental health index in the Italian regions. The analysis is conducted to create rankings, capture the changes detected in the medium term, and also analyse the distribution of the mental health index at the macro-regional level to identify the existence of a gap between Centre-North Italy and southern Italy. The data presented in the table below offer a detailed view of the evolution of the Mental Health Index (MHI) in Italian regions between 2016 and 2023, providing important insights into both spatial disparities and temporal dynamics (Table 2). Table 2Temporal evolution and regional distribution of the mental health index in Italy (2016–2023)RegionIndex 2016Index 2023Δ (2023—2016)Mean IndexΔ %(Change)RegionIndex 2016Index 2023Δ (2023—2016)Mean IndexΔ % (Change)Sardegna69.871.5 + 1.770.65 + 2.44%Veneto70.271.0 + 0.870.60 + 1.14%Sicilia65.666.5 + 0.966.05 + 1.37%Trentino-Alto Adige73.472.5–0.972.95–1.23%Calabria65.867.8 + 2.066.80 + 3.04%Lombardia68.269.0 + 0.868.60 + 1.17%Basilicata67.968.0 + 0.167.95 + 0.15%Liguria69.769.0–0.769.35–1.00%Puglia66.567.0 + 0.566.75 + 0.75%Valle d’Aosta68.069.5 + 1.568.75 + 2.21%Campania65.266.8 + 1.666.00 + 2.45%Piemonte67.468.5 + 1.167.95 + 1.63%Molise68.268.3 + 0.168.25 + 0.15%Nord69.069.5 + 0.5069.25 + 0.72%Abruzzo68.968.5–0.468.70–0.58%Nord-ovest68.569.3 + 0.8068.90 + 1.17%Lazio67.768.8 + 1.168.25 + 1.62%Nord-est69.769.6–0.1069.65–0.14%Marche67.068.7 + 1.767.85 + 2.54%Centro68.869.1 + 0.3068.95 + 0.44%Umbria67.869.5 + 1.768.65 + 2.51%Mezzogiorno66.467.5 + 1.1066.95 + 1.66%Toscana68.069.8 + 1.868.90 + 2.65%Sud66.567.5 + 1.0067.00 + 1.50%Emilia-Romagna69.870.5 + 0.770.15 + 1.00%Isole67.568.6 + 1.1068.05 + 1.63%Friuli-Venezia Giulia70.470.0–0.470.20–0.57%The table shows a general improvement in MHI alongside persistent territorial disparities, with northern regions consistently outperforming southern ones despite partial convergence trends Temporal evolution and regional distribution of the mental health index in Italy (2016–2023) The table shows a general improvement in MHI alongside persistent territorial disparities, with northern regions consistently outperforming southern ones despite partial convergence trends The pattern of mental health in Italian regions has been complex, reflecting overall improvements in mental health accompanied by persisting regional disparities. While overall mental health has increased slightly, the traditional division between the north and the south, defined by economic strength and disadvantage, has persisted [12]. Regional disparities in mental health in Italy have been influenced by outliers, underscoring the importance of local governance and policy interventions, irrespective of macroeconomic factors [7]. Between 2016 and 2023, most regions showed improvements in the Mental Health Index (MHI), with increases ranging from + 0.1 to + 2.0, indicating a 0.5% to 2.5% increase in overall mental health [13]. While the overall increase has been significant in Calabria, other regions that have shown considerable improvements in mental health include Toscana, Marche, Umbria, Valle d’Aosta, and Campania [7, 14]. There have been slight declines in some traditionally higher-performing regions, including Trentino-Alto Adige, Liguria, Friuli-Venezia Giulia, and Abruzzo [15, 16]. While regional disparities in mental health in Italy have continued, the faster rates of improvement in southern and insular regions indicate a pattern of convergence, driven by policy interventions. Comparative evaluation of clustering algorithms: evidence for the superiority of k-Means To identify the best clustering technique for analyzing territorial data on mental health issues and ESG factors, six clustering algorithms were comparatively evaluated. These clustering algorithms include Density-Based Clustering, Fuzzy C-Means, Hierarchical Clustering, Model-Based Clustering, Random Forest Clustering, and K-Means. Each clustering algorithm was evaluated using a set of traditional performance indicators that cover both statistical and structural properties of clustering results. Statistical properties used in the clustering algorithm evaluation include the coefficient of determination (R2), Akaike information criterion (AIC), and Bayesian information criterion (BIC). In contrast, structural properties include the Silhouette score, Dunn score, Pearson’s Gamma, and entropy. Past studies have shown that comparative clustering algorithms can be used to analyze ESG indicators and territorial data [17–19]. Among the six clustering algorithms used in this study, the K-Means algorithm showed the best-balanced results across all clustering algorithm evaluation criteria. For example, the K-Means algorithm showed the highest normalized coefficient of determination (R.2) and high scores in the Silhouette and Calinski-Harabasz evaluations. In addition, the low AIC and BIC values indicate that the K-Means algorithm achieves the best balance between clustering quality and complexity [20, 21]. Therefore, the K-Means algorithm yielded the best clustering results when evaluating territorial data on mental health issues and ESG factors. Table 3 Table 3Normalized performance metrics (Min–Max Scaling) for Clustering AlgorithmsDensity basedFuzzy C-MeansHierarchicalModel basedRandom ForestK-MeansR20.0210.0000.0790.0890.0671.000AIC0.7740.3870.1280.0590.2450.000BIC1.0000.9570.3120.1980.4770.171Silhouette0.0000.2730.8180.4550.2420.545Max diameter1.0000.9100.4840.1470.4540.000Min separation0.5710.0001.0000.3070.4830.307Pearson’s γ0.0000.0311.0000.7010.2550.541Dunn index0.5690.0001.0000.6450.4890.758Entropy0.0000.1170.2950.6830.4511.000Calinski-Harabasz0.0000.8290.8370.8491.0000.918The results highlight the superior performance of the K-Means algorithm across most evaluation criteria, supporting its selection as the most robust method for identifying regional mental health typologies Normalized performance metrics (Min–Max Scaling) for Clustering Algorithms The results highlight the superior performance of the K-Means algorithm across most evaluation criteria, supporting its selection as the most robust method for identifying regional mental health typologies Two figures validate the k-Means clustering approach. The Elbow Plot identifies six clusters as optimal using AIC, BIC, and WSS. The t-SNE visualization shows clear group separation, Table 4 supporting the robustness of the six-cluster solution [22, 23]. See Fig. 1. Table 4Cluster profiles based on mental health index and ESG-related variablesCluster123456Size5515252129Explained proportion within-cluster heterogeneity0.0150.0420.1760.2830.2090.275Within sum of squares3.90211.28447.05775.83056.01373.738Silhouette score0.7050.4690.1780.3620.2350.258Center MHI2.445-0.7930.539-0.408-0.6000.223Center SED-1.6320.877-0.0581.431-0.336-0.830Center ALC1.2780.708-0.148-1.1630.0370.710Center SMK-0.9500.185-0.086-0.2531.285-0.535Center NUTR0.514-0.7820.157-1.3240.7820.540Center MORB-1.741-0.0830.4351.034-0.025-0.784Center UGA2.8932.868-0.363-0.345-0.224-0.346Center OSP-1.410-0.5060.6091.179-0.473-0.658Center CTD-1.1580.197-0.0481.419-0.263-0.843Clusters differ markedly, with Cluster 1 showing the most favorable conditions and Cluster 4 the highest concentration of risk factors Cluster profiles based on mental health index and ESG-related variables Clusters differ markedly, with Cluster 1 showing the most favorable conditions and Cluster 4 the highest concentration of risk factors Optimal cluster selection and spatial visualization of k-Means results. Six distinct and well-separated clusters reveal clear regional patterns, highlighting the existence of differentiated mental health and ESG profiles across Italian regions Clustering results reveal notable regional disparities in the Mental Health Index when combined with behavioral and environmental variables. Cluster 1 shows the healthiest profile, while Cluster 4 reflects the most critical conditions, highlighting varied risk and protective factors across regions. To improve the interpretability of the clustering results, the standardized cluster centers can be translated into concrete regional profiles. Cluster 1 represents regions characterized by high mental health levels, low sedentariness and multimorbidity, and relatively strong environmental and institutional conditions, reflecting a “high well-being and favorable ESG” profile typically associated with more developed areas. In contrast, Cluster 4 identifies regions with low mental health levels, high sedentariness and multimorbidity, and weaker governance and environmental conditions, corresponding to a “high-risk and structurally disadvantaged” profile. Intermediate clusters (e.g., Clusters 2, 3, 5, and 6) capture mixed configurations, where moderate mental health levels coexist with either behavioral risk factors or partial ESG strengths. Overall, the clustering results highlight that regional mental health disparities are associated with distinct combinations of behavioral and structural conditions, providing a basis for targeted and differentiated policy interventions.For instance, Cluster 2 reflects relatively good environmental conditions but moderate behavioral risks, while Cluster 3 captures average profiles with balanced ESG characteristics. Cluster 5 is characterized by relatively favorable nutritional patterns but weaker institutional conditions, whereas Cluster 6 reflects moderately positive mental health levels combined with lower exposure to risk factors. See Table 5. Table 5Cluster Centers for the mental health index and associated ESG-related variablesClusterMHISEDALCSMKNUTRMORBUGAOSPCTD11.278-1.1582.445-1.7410.514-1.410-1.632-0.9502.89320.7080.197-0.793-0.083-0.782-0.5060.8770.1852.8683-0.148-0.0480.5390.4350.1570.609-0.058-0.086-0.3634-1.1631.419-0.4081.034-1.3241.1791.431-0.253-0.34550.037-0.263-0.600-0.0250.782-0.473-0.3361.285-0.22460.710-0.8430.223-0.7840.540-0.658-0.830-0.535-0.346Substantial heterogeneity emerges across clusters, with Cluster 1 showing the most favorable profile and Cluster 4 the most critical, characterized by higher exposure to behavioral risk factors. Overall, the results reflect the coexistence of protective and risk factors across regions Cluster Centers for the mental health index and associated ESG-related variables Substantial heterogeneity emerges across clusters, with Cluster 1 showing the most favorable profile and Cluster 4 the most critical, characterized by higher exposure to behavioral risk factors. Overall, the results reflect the coexistence of protective and risk factors across regions The comparative evaluation clearly illustrates the superiority of K-Means Clustering to partition geographical units along mental health and ESG-related indicators. Among the six algorithms experimented with, K-Means consistently achieved first- or second-best ranking for the majority of the statistical, geometric, and structural indices, such as R2, Silhouette measure, Dunn index, and Calinski-Harabasz index. Due to its ability to explain maximum variance while creating compact and well-separated clusters, it is particularly efficient in this multidimensional study [21]. Even when evaluated by indices such as AIC and BIC, which are traditionally penalty-imposing, K-Means was competitive and thus further contributed to its robustness. The six-cluster configuration, having been validated by both criterion-based choice and t-SNE representation, aids the interpretative meaning of spatial mental health patterns. Overall, K-Means is found to emerge as the most balanced and robust algorithm for uncovering hidden structures in complex data, making it the optimal choice for the purposes of this study’s classification and building typology [24]. Predictive modeling of mental health: comparative insights from behavioral and ESG determinants The growing use of predictive modeling in mental health research highlights the need for robust methods capable of capturing the complex interactions among individual behaviors, clinical conditions, and broader ESG-related factors. Mental health outcomes increasingly reflect the interplay between biological vulnerability, lifestyle habits, and structural contexts [25]. Traditional approaches often fail to represent this complexity, encouraging the use of advanced machine learning and regression techniques such as regularized regression, neural networks, ensemble methods, and K-Nearest Neighbors [26]. Model performance is evaluated through indicators including MSE, RMSE, MAE, MAPE, and R2, which help identify the variables most strongly associated with mental health outcomes [27]. Model performance was assessed using a holdout validation framework. Specifically, 20% of the data was reserved as a test set, while the remaining observations were used for training and validation. Given the limited sample size, no cross-validation procedure was implemented. Performance metrics were computed on held-out data to ensure out-of-sample evaluation and reduce the risk of overfitting. Specifically, we analyzed the following Eq. (1), considering that the value of i is equal to the 20 Italian regions and the value of t refers to the period between 2016 and 2023. where i = 20 and t = [2016; 2023] (Table 6). Table 6Comparative Performance Metrics of Predictive Models for Mental Health Index (MHI)ModelMSEMSE(scaled)RMSEMAE/MADMAPER2Boosting Regression0.4330.4360.4710.6930.7000.423Decision Tree Regression0.2811.0000.2340.5700.5000.000KNN0.3620.0000.3470.6330.6001.000Linear Regression1.0000.8771.0001.0001.0000.021ANN0.149-0.1370.0000.000-Random Forest Regression0.4180.3280.4560.5320.4600.552Regularized Linear Regression0.0000.4110.0000.0700.0800.456Support Vector Machine0.9440.6790.9200.9020.8800.153Model performance varies across metrics. Scaled metrics (e.g., MSE(scaled)) are obtained using min–max normalization across models, with values ranging from 0 (best performance) to 1 (worst performance). This transformation facilitates comparability across models with different performance ranges but does not affect the relative ranking of models. Therefore, scaled metrics should be interpreted as relative performance indicators rather than absolute measures Comparative Performance Metrics of Predictive Models for Mental Health Index (MHI) Model performance varies across metrics. Scaled metrics (e.g., MSE(scaled)) are obtained using min–max normalization across models, with values ranging from 0 (best performance) to 1 (worst performance). This transformation facilitates comparability across models with different performance ranges but does not affect the relative ranking of models. Therefore, scaled metrics should be interpreted as relative performance indicators rather than absolute measures To make model selection more transparent and reproducible, an additional rank-based aggregation procedure was applied across the main predictive performance metrics reported in Table 6 (MSE, RMSE, MAE/MAD, MAPE, and R2). For error-based metrics, lower values received better ranks, whereas for R2 higher values received better ranks. The final score was computed as the average rank across all available metrics according to the following formula: where Rankij is the rank assigned to model i for metric j, and m is the number of evaluation metrics. Lower values of the rank score indicate better overall predictive performance. As reported in Table 7, Regularized Linear Regression achieved the best overall mean rank (1.8), followed by Random Forest Regression (3.6) and K-Nearest Neighbors (3.8). Artificial Neural Networks were excluded from the final ranking because the R2 statistic was not available, preventing a complete comparison across all evaluation criteria. The rank-based aggregation indicates that Regularized Linear Regression achieved the best overall performance across the selected evaluation metrics. In addition, the model offers substantial interpretability advantages through coefficient shrinkage and variable selection, making it particularly suitable for identifying the behavioral and ESG-related determinants of mental health. Therefore, Regularized Linear Regression was retained as the final predictive model. Table 7Rank-based aggregation of predictive performance metricsModelMSE RankRMSE RankMAE/MAD RankMAPE RankR2 RankMean RankRegularized Linear Regression112231.8Random Forest Regression553323.6KNN445513.8Decision Tree Regression334474.2Boosting Regression666645.6Support Vector Machine777756.6Linear Regression888867.6Lower values receive better ranks for MSE, RMSE, MAE/MAD, and MAPE, whereas higher values receive better ranks for R2. The final score corresponds to the average rank across all metrics. ANN was excluded because R2 was not available Rank-based aggregation of predictive performance metrics Lower values receive better ranks for MSE, RMSE, MAE/MAD, and MAPE, whereas higher values receive better ranks for R2. The final score corresponds to the average rank across all metrics. ANN was excluded because R2 was not available Based on this reproducible selection rule, Regularized Linear Regression was retained as the final predictive model. In addition to its aggregate predictive performance, the model provides direct interpretability through coefficient shrinkage and variable selection, which is particularly relevant for identifying the behavioral, environmental, and socio-health determinants of mental health in a policy-oriented framework. The selected model was then used to forecast the Mental Health Index (MHI) and to examine predictor importance through mean dropout loss, defined as the change in predictive error when each variable was excluded from the model. The results showed that multimorbidity was the most important predictor, reaffirming its strong association with mental health and quality of life [28, 29]. Likewise, smoking habits and physical inactivity were found to be significant negative predictors of mental health, consistent with previous studies linking these factors to depression and anxiety [30, 31]. Urban green availability was found to be a significant positive predictor of mental health [32]. See Table 8. Table 8Variable importance based on mean dropout loss in regularized linear regression for MHI predictionMean dropout lossMORB1.150SMK1.129SED1.092UGA1.078ALC1.056NUTR1.056OSP1.056CTD1.056The results show that multimorbidity, smoking, and sedentariness are the most influential predictors, confirming their strong association with mental health outcomes Variable importance based on mean dropout loss in regularized linear regression for MHI prediction The results show that multimorbidity, smoking, and sedentariness are the most influential predictors, confirming their strong association with mental health outcomes The regularized linear regression model includes predictors such as sedentariness, alcohol consumption, smoking rate, adequate nutrition, multimorbidity, urban green availability, occupational security perception, and civil trial duration. Through regularization, the coefficients of less informative variables are shrunk to zero, excluding alcohol consumption, nutrition, occupational security perception, and civil trial duration from the final specification [33]. The most influential predictors are multimorbidity, smoking, and sedentariness. Multimorbidity shows the strongest negative association with MHI, confirming its close relationship with mental health outcomes [25, 34]. See Table 9. Table 9Predicted Mental Health Index (MHI) values across cases using regularized linear regression coefficientsCasePredictedBaseSEDALCSMKNUTRMORBUGAOSPCTD168.00568.476-0.0620.000-0.0670.000-0.298-0.0440.0000.000268.35068.476-0.0250.0000.1200.000-0.177-0.0440.0000.000367.85968.476-0.1490.000-0.0350.000-0.4790.0450.0000.000468.09968.476-0.1810.000-0.0910.000-0.1530.0470.0000.000568.25968.476-0.2200.0000.3090.000-0.271-0.0350.0000.000Variations in predicted MHI reflect the negative contribution of behavioral risk factors and the positive role of urban green availability Predicted Mental Health Index (MHI) values across cases using regularized linear regression coefficients Variations in predicted MHI reflect the negative contribution of behavioral risk factors and the positive role of urban green availability The estimated coefficients of the regularized linear regression model are reported in Table 10. The intercept represents the baseline MHI value, while each coefficient indicates the estimated association between a predictor and the Mental Health Index, holding the other variables constant. Through regularization, less informative predictors are shrunk to zero. Sedentariness, smoking, and multimorbidity show negative coefficients, confirming their adverse association with mental health. In particular, smoking and multimorbidity display the largest negative coefficients, while urban green availability shows a positive coefficient, suggesting a protective association with mental health [35, 36]. See Table 10. Table 10Coefficients of regularized linear regression model for predicting the Mental Health Index (MHI)Coefficient (β)(Intercept)68.442SED-0.157ALC0.000SMK-0.279NUTR0.000MORB-0.278UGA0.126OSP0.000CTD0.000Key behavioral variables are negatively associated with MHI, while urban green availability shows a positive association; other predictors are shrunk to zero, indicating limited relevance Coefficients of regularized linear regression model for predicting the Mental Health Index (MHI) Key behavioral variables are negatively associated with MHI, while urban green availability shows a positive association; other predictors are shrunk to zero, indicating limited relevance The predictive performance of the selected regularized linear regression model is further illustrated in Fig. 2. The observed-versus-predicted plot shows that most observations are located close to the diagonal line, indicating satisfactory predictive alignment. The coefficient trace plot illustrates the effect of increasing regularization strength on the estimated coefficients. Smoking prevalence, multimorbidity, and sedentariness remain consistently negatively associated with MHI, whereas urban green availability shows a positive association with mental health [35, 37]. In contrast, alcohol consumption, nutrition, perceived job security, and civil trial duration are progressively shrunk toward zero, confirming their lower contribution within the selected model. See Fig. 2. Predictive Performance and Coefficient Paths of Regularized Linear Regression for MHI. Observed and predicted values align closely, while key predictors remain stable and others are shrunk toward zero under regularization Overall, the selected model confirms the relevance of multimorbidity, smoking, sedentariness, and urban green availability in explaining regional differences in MHI, while also supporting the use of regularized regression when interpretability and variable selection are required in predictive health applications [38–40]. Integrating ESG frameworks into mental health policy: a national and regional strategy for Italy The integration of environmental, social, and governance (ESG) factors into the mental health care policy in Italy requires a coordinated national-regional approach to address the existing territorial disparities. The mental health outcomes remain characterized by spatial disparities, with northern regions performing better than central and southern regions. This situation requires a governance strategy that takes a holistic approach to address the behavioral, environmental, and social determinants [41]. At the national level, the policy should aim to improve mental health infrastructure and address disparities in access to mental health care, with a focus on the disadvantaged regions of Campania, Calabria, and Sicily. ESG-aligned financial instruments, such as social and green bonds, can be used to fund the construction of sustainable health care infrastructure, telemedicine, and community mental health services [42]. Employee well-being can also be incorporated into ESG factors to address stress and burnout (Han & Lee, 2024). At the regional level, investments in green areas, pollution control, and sustainable urban planning can help improve the environment and, in turn, promote psychological well-being [43] (See Fig. 3). This figure summarizes the behavioral and ESG determinants associated with Italy’s Mental Health Index from 2016 to 2023, highlighting risk and protective factors identified through empirical analysis and visualized using NotebookLM. Discussion: governance, environment, and lifestyle drivers of mental well-being The results are consistent with the growing and strengthening consensus in the literature that mental health outcomes are associated not only with individual behaviors but also by structural, environmental, and institutional factors. Given the observational nature of the data, the results should be interpreted as associations rather than causal relationships. More recent studies have progressively emphasized the importance of social determinants of health and the role of governance in explaining regional disparities in psychological well-being. In the Italian scenario, Bruzzi et al. [12] and Landi et al. [44] report a strong positive correlation between health inequalities and regional disparities in institutional capacity, healthcare systems, and public resource allocation. In line with these interpretations, the current study shows that regions with more advanced institutions of governance and healthcare systems report more favorable and stable mental health outcomes. At the same time, more recent studies have progressively emphasized the importance of environmental factors in influencing psychological well-being. The presence of urban green spaces, overall urban quality, and environmental sustainability have been found to have a protective role in mental health, through mechanisms related to stress relief, social interaction, and support for healthy lifestyle behaviors. The current study confirms these findings by reporting a statistically significant positive correlation between the presence of urban green spaces and mental health outcomes. On the other hand, behavioral risk factors such as smoking, lack of physical activity, and multimorbidity are still significantly associated with unfavorable outcomes for mental health, as reported in previous studies. However, it is important to note that territorial health disparities are not fixed. The improvement in health outcomes in disadvantaged areas shows that health interventions and system changes are associated with changes in mental health outcomes, underscoring the need for integrated health intervention policies with governance and environmental approaches. In terms of policy formulation, the study’s findings suggest that mental health policies should not be limited to the healthcare system or to individual behavior change. Instead, policymakers should take a holistic approach that seeks to minimize behavioral risks and maximize environmental quality and system performance. Green infrastructure, equal access to healthcare, and preventive public health policies may have a pivotal role in reducing territorial disparities in mental health outcomes. In policy formulation, the ESG (Environmental, Social, and Governance) framework can serve as a useful starting point for public health policymakers seeking to integrate public health goals with sustainable development and equity objectives. This study has some limitations that should be acknowledged. First, the analysis is based on observational regional data, which does not allow for causal inference but only the identification of statistical associations. Second, the relatively small sample size (Italian regions over time) may limit the robustness and generalizability of the predictive models. Third, the use of aggregated regional indicators may mask within-region heterogeneity and individual-level variability. Finally, although the ESG framework provides a comprehensive perspective, it may not capture all relevant determinants of mental health. Future research could address these limitations by using micro-level data, longitudinal designs, and alternative modeling approaches. Conclusions Mental health outcomes in Italy are associated with complex interactions among behavioral, environmental, and governance factors rather than by individual behaviors. While the overall Mental Health Index in Italy has remained fairly stable from 2016 to 2023, there are still large regional gaps, with northern regions performing better than southern regions. The predictive model shows that multimorbidity, smoking, and a sedentary lifestyle are the most important negative predictors of mental health, while the presence of urban green spaces is a protective factor. Cluster analysis shows that the mental health of regions is more closely associated with ESG-related behavioral factors than with geographical factors. Taken together, these results emphasize the need for a holistic policy strategy that combines lifestyle changes with environmental upgrades and enhanced institutional capacity to address territorial gaps in mental health. Supplementary Information Supplementary Material 1 Author contributions Conceptualization, RE, LG, PP, CA, LA; methodology, RE, LG, PP, CA, LA; validation, RE, LG, PP, CA, LA.; formal analysis, RE, LG, PP, CA, LA.; investigation, RE, LG, PP, CA, LA; resources, RE, LG, PP, CA, LA; data curation, RE, LG, PP, CA, LA; writing—original draft preparation, RE, LG, PP, CA, LA; writing—review and editing, RE, LG, PP, CA, LA; visualization, RE, LG, PP, CA, LA; supervision, RE, LG, PP, CA, LA; project administration, RE, LG, PP, CA, LA; funding acquisition, RE, LG, PP, CA, LA. Funding No funds were received. Data availability The data used in the following article were acquired through the analysis of the ISTAT-BES database. The database is public, open and accessible at the following link:. Declarations Ethics approval Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement The data used in the following article were acquired through the analysis of the ISTAT-BES database. The database is public, open and accessible at the following link:.