What Was the Incremental Value of the AI Model in Elderly Inguinal Hernia Risk Assessment?

Dear Editor, We read with great interest the research letter by Asafa et al. 1 on artificial intelligence-based perioperative risk assessment in elderly patients undergoing inguinal hernia repair in a resource-limited setting. The study addresses an important problem, particularly because older surgical patients in low-resource environments are often assessed with tools that insufficiently capture frailty and functional vulnerability. The authors should be commended for highlighting the need to move beyond chronological age alone in perioperative decision-making. The manuscript also raises a specific methodological question that seems central to the interpretation of its findings: what was the true incremental contribution of the artificial intelligence model beyond the structured use of the same bedside clinical variables? The methods state that the AI model incorporated demographic, comorbidity, frailty, and operative variables, whereas the conventional assessment was based on age, comorbidities, operative urgency, and clinician judgment. However, the article does not clearly specify the algorithm used, the modeling workflow, the internal validation strategy, or the safeguards against overfitting. This is particularly relevant because the study included only 60 patients, of whom 14 developed postoperative complications. Under these conditions, the reported improvement in discrimination from an AUC of 0.66–0.81 is certainly interesting, but it remains difficult to determine whether it reflects genuine algorithmic gain or simply a comparison against a less formalized clinical reference. A related issue arises from the multivariable analysis. Table 3 identifies emergency surgery, multimorbidity, frailty indicators, and high AI risk score as independent predictors of adverse outcomes. Yet the AI score appears to have been derived from overlapping clinical variables, including frailty and comorbidity. This creates an important interpretive problem: can the AI score be treated as an independent predictor in the same model as variables that may already constitute its inputs, or does this introduce conceptual circularity and redundancy? This point is not merely statistical. It directly affects how readers should interpret the claimed superiority of the AI approach and whether the model is adding new prognostic information or mainly repackaging familiar predictors into a composite score. The question becomes even more relevant because frailty itself was operationalized pragmatically through locally observable markers, with the presence of at least one marker considered sufficient to classify a patient as frail. That approach may be reasonable in a low-resource setting, but it broadens the construct substantially and may combine chronic vulnerability with acute physiological compromise. In that context, one wonders whether the model is truly learning a frailty-based risk phenotype, or instead capturing a mixture of illness severity, emergency presentation, and functional impairment. For these reasons, we believe the most important issue raised by this otherwise valuable study is not whether AI can outperform routine clinical judgment in principle, but whether the present manuscript demonstrates a reproducible and methodologically distinct AI advantage over a transparent clinical model built from the same variables. Clarification from the authors on model specification, validation, and the interpretation of Table 3 would substantially strengthen the message of the paper and help readers understand what should be taken forward into future surgical risk research in similar settings. Carlos M. Ardila: conceptualization, investigation, funding acquisition, writing – original draft, methodology, validation, writing – review and editing, formal analysis, supervision. Iordana Mejia-Kambourova: conceptualization, funding acquisition, investigation, writing – original draft, methodology, validation, visualization, writing – review and editing, formal analysis, supervision. Daniel Gonzalez-Arroyave: conceptualization, investigation, funding acquisition, writing – original draft, methodology, validation, visualization, writing – review and editing, formal analysis, supervision. The authors have nothing to report. The authors have nothing to report. The authors declare no conflicts of interest. Data sharing is not applicable to this article as no new data were created or analyzed in this study.