Sexual dysfunction in peritoneal dialysis: associated factors and exploratory machine-learning classification

Sexual dysfunction (SD) is a common yet under-recognized complication in patients receiving peritoneal dialysis (PD). We investigated the prevalence of SD and its associated factors, and developed interpretable machine learning models to classify prevalent SD in PD patients.

We enrolled 209 PD patients (100 males, 109 females) in this retrospective cross-sectional study. Sexual function was assessed with the International Index of Erectile Function-5 (IIEF-5) in males and the Female Sexual Function Index (FSFI) in females. We collected demographic, clinical, biochemical, and psychosocial variables, and performed univariate analysis, Spearman correlation, and multivariate linear and logistic regression. Eight machine learning models were compared. SHapley Additive exPlanations (SHAP) were applied to the Random Forest model, selected for interpretability and balanced performance.

The overall prevalence of SD was 73.7% (154/209), with rates of 78% (78/100) in males and 69.7% (76/109) in females. In males, multivariable linear regression identified depression score (SDS; β = -0.24, P = 0.013), 24-h urine volume (β = + 0.48 per 100 mL/day, P = 0.021), marital status (β = + 8.77, P < 0.001), and α-blocker use (β = -4.14, P = 0.013) as associated variables of male erectile-function scores. In females, the associations were weaker and reflected socioeconomic and clinical factors, including education, hypertension, and α-blocker use. Among machine-learning models, the random forest + XGBoost ensemble achieved a test-set AUC of 0.630. In the SHAP analysis of the overall SD model, BMI had the highest mean absolute SHAP value, followed by 24-h urine volume and SDS.

Sexual dysfunction was common among respondents receiving peritoneal dialysis. In men, higher depression scores and lower 24-h urine volume were associated with male erectile function scores, whereas in women associations were weaker and primarily socioeconomic. The models and SHAP-derived feature rankings should therefore be considered exploratory and require prospective external validation before clinical use.