Linking External Variables to the Prediction of Anxiety and Suicide Attempt Cases.

The increasing pressure on mental health systems requires innovative surveillance strategies. This study addresses the prediction of anxiety cases and suicide attempts by analyzing emergency call records, enriched with external variables (socioeconomic, educational, and meteorological) to identify comprehensive determinants of mental health crises at the population level. The methodology followed developed a dual predictive framework. For resource planning, time series algorithms (SARIMA, Prophet, ETS) were applied to predict the expected volume of cases. For Risk Profiling (Classification), multiple Supervised Machine Learning models (notably XGBoost and Random Forest) were used to classify specific incidents, distinguishing between Anxiety and Suicide Attempt. The relevance of the characteristics was rigorously analyzed using SHAP explainability techniques. In addition, a pilot study was introduced to explore the feasibility of Quantum Machine Learning (QML) algorithms. Time series models confirmed an upward trend and high volatility in 2023. In classification, nonlinear algorithms generally outperformed linear ones. SHAP analysis revealed that demographic factors (age, sex) and geographic location are more determinant predictors than economic variables, while solar radiation showed a significant inverse relationship with call volume. The results of the QML pilot, although limited, show a future avenue for expanding the study. The results obtained, using emergency service (PSAP) data combined with external determinants support evidence-based strategies for risk stratification and enable resource planning and targeted public health interventions, underscoring the need for locally adapted implementation and robust ethical safeguards in mental health crisis management.