Machine learning prediction of SIRS after percutaneous nephrolithotomy in older adults: role of frailty, malnutrition, and sarcopenia

Abstract

Systemic inflammatory response syndrome (SIRS) after percutaneous nephrolithotomy (PCNL) is a critical complication in older adults. However, the impact of frailty, malnutrition, and sarcopenia (F-M-S) comorbidity on postoperative SIRS remains unclear. This study aimed to evaluate the predictive value of indicators of frailty, malnutrition, and sarcopenia and to develop machine learning models for perioperative risk stratification of SIRS after PCNL in older adults. We enrolled 426 elderly patients (≥ 60 years) undergoing elective PCNL (2022–2024). Preoperative frailty (laboratory-based frailty index, FI-Lab), nutritional risk (Geriatric Nutritional Risk Index, GNRI), muscle mass (skeletal muscle index, SMI), and muscle density (skeletal muscle radiodensity, SMD) were assessed using laboratory tests, anthropometry, and abdominal CT. Clinical features were selected using the least absolute shrinkage and selection operator (LASSO) regression. Five machine learning models were trained on three feature sets (clinical, F-M-S indicators, combined). Performance was evaluated by AUC, decision curve analysis, and SHAP (SHapley Additive exPlanations). A SHAP-derived scoring system stratified patients into risk groups. SIRS occurred in 37.6% (160/426); only 27.5% had no F-M-S abnormalities. All four abnormalities were associated with increased SIRS risk in univariate analysis: low muscle density (OR = 3.90, 95% CI: 2.44–6.22), low muscle mass (OR = 2.09, 95% CI: 1.39–3.14), frailty (OR = 1.86, 95% CI: 1.25–2.77), and nutritional risk (OR = 1.70, 95% CI: 1.11–2.60). A significant cumulative effect was observed (four abnormalities: OR = 7.00, 2.56–19.15). The Support Vector Machine (SVM) integrating F-M-S indicators with clinical features achieved optimal performance (AUC = 0.770; 0.681–0.855), with decision curve analysis indicating the highest net benefit at 20%- 75% thresholds. Notably, F-M-S indicators alone (AUC = 0.707, 0.615–0.802) performed comparably to clinical features alone (AUC = 0.721, 0.619–0.819). SHAP identified SMD as the second most important predictor after hydronephrosis grade. The SHAP-derived scoring system effectively stratified patients into low-, intermediate-, and high-risk groups (SIRS incidence: 15.7% vs. 33.1% vs. 63.8%, P < 0.001). F-M-S indicators were associated with postoperative SIRS and may help identify higher-risk older patients undergoing PCNL; prospective external validation is needed.