Answers are grounded in this paper and cite their sources.
Structured Abstract Context and Objective Risk stratification in prostate adenocarcinoma is essential for identifying patients with aggressive disease who may require prioritized evaluation. We aimed to develop and validate a rule-based natural language processing (NLP) framework to identify high-risk prostate cancer from unstructured electronic medical record (EMR) data. Materials and Methods This retrospective diagnostic accuracy study included two patient-independent cohorts. The development cohort comprised 3000 patient-level EMR records collected between February and December 2025, manually reviewed by specialists to derive and refine keyword-based classification rules. Thirty-eight candidate keyword combinations were generated, and 11 predefined rules were retained based on a prespecified sensitivity criterion and expert review. The validation cohort included adult male patients with documented prostate adenocarcinoma from the Santa Casa de Porto Alegre hospital system (Brazil) during January 2026. The reference standard was established according to American Urological Association high-risk criteria through blinded independent chart review by two investigators, with adjudication by a third reviewer when necessary. Results Among 517 patients included in the validation cohort, 220 (42.55%) had high-risk disease. The predefined keyword-based rules demonstrated different sensitivity-specificity profiles. The rule with the most balanced diagnostic performance achieved an accuracy of 91.30%, a sensitivity of 90.45%, a specificity of 91.92%, a positive predictive value of 89.24%, and a negative predictive value of 92.86%. Rules with higher sensitivity achieved increased detection of high-risk patients, supporting different potential prioritization strategies. Limitations include the single-center design and lack of external validation. Conclusions A reproducible rule-based NLP framework enabled identification of high-risk prostate cancer from unstructured EMR data with high diagnostic accuracy. This approach may assist early patient identification within clinical workflows, particularly before specialist review or multidisciplinary evaluation, although prospective studies are required to assess its impact on care processes and outcomes.
No takes yet. Share an insight, caveat, or question.
Zawaski et al. (2026) studied this question.