Predicting suicidality in people living with HIV in Uganda: A machine learning approach

Abstract

Background Suicidality is a major risk factor for future suicide attempts and completed suicide. People living with HIV (PLWH) are at a higher risk of fatal suicide attempts compared to the general population due to the psychological distress associated with an HIV infection. Timely identification and referral can prevent suicide, but the stigma and discrimination associated with mental illness prevent affected persons from seeking psychiatric treatment. This study applied machine learning (ML) approaches to predict suicidality among PLWH in Uganda. Materials and Methods This retrospective case-control study used sociodemographic, psychological, and clinical data of 1126 study participants to predict prevalent and incident suicidality. In addition, suicidality polygenic risk scores (PRS) for a subset of 282 study participants were calculated and incorporated as an additional feature in the model. Model performance was determined using positive predictive value (PPV), sensitivity, specificity, Mathew’s correlation coefficient (MCC), and the area under the receiver operating characteristics curve (AUC). Results The best model for predicting suicidality among PLWH was logistic regression (LR) and it predicted prevalent suicidality with a PPV of 0.34, sensitivity of 0.68, specificity of 0.70, and MCC of 0.31. In predicting incident suicidality, model specificity increased to 0.83, but at the cost of reduced sensitivity (0.50), PPV (0.05), and MCC (0.12). Suicidality PRS were statistically significant (p=0.007) but only explained 4.2% of the phenotypic variance between cases and controls. Incorporating PRS as an additional feature resulted in a modest (26.5%) improvement in the model’s PPV. Conclusion We developed an explainable ML model for predicting suicidality using sociodemographic, psychosocial, and clinical data. This model can be refined and incorporated into Electronic Medical Records (EMR) to support routine suicidality screening. Suicidality PRS improves the PPV of the prediction model and with increasing availability, they will play an increasingly significant role in disease risk prediction.