Background

Distal radius fractures (DRFs) are the most common upper-extremity fractures in older adults. Frailty may modulate postoperative risk, but the comparative prognostic value of widely used frailty indices in DRF surgery is unclear.

Objective

To compare the 5-item modified Frailty Index (mFI-5), Clinical Frailty Scale (CFS), and Charlson Comorbidity Index (CCI) for predicting complications and functional outcomes after DRF surgery in patients ≥65 years, and to assess whether machine learning (ML) enhances risk stratification.

Methods

We retrospectively analyzed 562 patients (mean age 75.1±4.9 years; 71.7% female) undergoing open reduction and internal fixation (67.1%), closed reduction percutaneous pinning (21.2%), or external fixation (11.7%), with ≥12 months of follow-up. Preoperative mFI-5, CFS, and CCI were collected. The primary endpoint was any postoperative complication (composite of surgical site infection, wound dehiscence, loss of reduction, tendon injury, nonunion/malunion, hardware failure, reoperation, 30-/90-day readmission, venous thromboembolism, and complex regional pain syndrome). Functional outcomes were DASH, PRWE, grip strength, and return to activities of daily living (ADLs). Logistic and Cox regression were used. Exploratory ML models (random forest, gradient boosting) employed 5-fold cross-validation, an 80/20 train–test split, and isotonic calibration.

Results

Over 17.9±4.5 months, 195 of 562 patients (34.8%) developed ≥1 complication. mFI-5 ≥2 was associated with higher complication rates (38.2% vs 31.9%) and worse 6-month disability (DASH 55.4 vs 44.5; PRWE 47.9 vs 36.4; all p<0.001). CFS ≥4 predicted lower 12-month grip strength recovery (67.8% vs 74.1%) and reduced ADL return (65.9% vs 78.6%). Discrimination for complications was modest (AUCs: mFI-5 0.552; CFS 0.534; CCI 0.507). ML substantially improved performance (gradient boosting AUC 0.878; random forest AUC 0.812) with superior calibration.

Conclusions

mFI-5, CFS, and CCI are associated with postoperative complications and functional recovery after DRF surgery but have limited individual discriminative power. ML-based multivariable models achieve markedly better risk stratification and may support more accurate preoperative counseling and personalized perioperative management.