Re: Methodological Considerations in Frailty Assessment for Distal Radius Fracture Surgery

Dear Editor,

We carefully reviewed the work of Özdemir et al ¹ on the predictive performance of frailty indices for complications and functional recovery after distal radius fracture (DRF) surgery in older adults. The authors report that the modified Frailty Index-5 (mFI-5), Clinical Frailty Scale (CFS), and Charlson Comorbidity Index (CCI) demonstrate limited discriminative ability for postoperative complications, with area under the curve (AUC) values ranging from 0.507 to 0.552, essentially no better than chance. Despite these findings, the authors conclude that these tools can “enhance preoperative risk stratification.” This disconnect between statistical evidence and clinical interpretation warrants careful consideration.

Several methodological concerns further temper the study’s conclusions. First, frailty is not a static condition but a dynamic syndrome that can improve or decline following surgery. A one-time preoperative assessment cannot capture these fluctuations, which may substantially influence recovery trajectories beyond baseline predictions. Recent research demonstrates that postoperative frailty assessment identifies patients at risk for poor recovery who were not identified by preoperative evaluation alone. ² This limitation restricts the findings to static risk assessment and precludes dynamic, personalized care planning based on evolving frailty status. Furthermore, the machine learning (ML) models remain exploratory, as they lack external validation, the gold standard for translating predictive models from research to clinical practice. ML prediction models frequently demonstrate high accuracy on internal datasets, but show diminished performance when applied to external populations. ³ In a systematic review of acute care surgery ML models, only 9.6% (5/52) performed external validation. ⁴ Similarly, a systematic review of ML models for predicting colorectal anastomotic leakage found that ML studies were smaller, less transparent, and more variable than logistic regression approaches in external validation. ⁵ Without external validation, the generalizability of the present ML findings remains uncertain, and clinical deployment would be premature. Finally, the composite primary outcome, which combines hardware complications, reoperation, CRPS, and other events, may mask individual complication drivers and fail to identify the risks most salient to patients. For instance, hardware complications (13.2%), reoperation (13.7%), and CRPS (10.0%) have different etiologies and patient impacts; a composite endpoint driven by the most common event may obscure the prediction of clinically meaningful individual complications. Furthermore, although stratified analyses by surgical technique were performed, these were underpowered to detect differences between subgroups. The retrospective design and single-center setting further limit generalizability.

Future prospective multicenter studies incorporating serial frailty assessments, externally validated ML tools meeting TRIPOD+AI criteria, and adequately powered subgroup analyses are needed to enable truly personalized perioperative management. ⁶ In conclusion, while Özdemir et al demonstrate associations between frailty indices and postoperative outcomes after DRF surgery in older adults, the discriminative ability is no better than chance, and the lack of external validation for ML models precludes routine clinical deployment. Future multicenter prospective studies should verify these findings and develop integrated risk assessment tools to improve individualized perioperative management.

ORCID iDs

Khansa Farooq https://orcid.org/0009-0005-5426-2887

Fiza Nazir https://orcid.org/0009-0009-4715-6401

Ethical Considerations

This letter does not involve original research with human or animal subjects.

Consent to Participate

This is a Letter to the Editor commenting on published literature; no patient data were collected.

Data Availability Statement

This letter does not involve original data.