We have read with interest the pilot study by Zayas-Bórquez et al.1 that develops a machine-learning model for predicting right-sided colon cancer recurrence. It provides a relevant initiative: individualized predictive models could refine postoperative risk stratification beyond TNM staging and guide differentiated surveillance regimens. However, certain methodological limitations should be discussed in an effort to strengthen subsequent iterations and their future clinical implementation.
First, the retrospective, single center, pilot design carries an appreciable risk for selection bias, nonrandom missingness, and statistical optimism. The number of events per candidate predictor variable critically determines model stability; when the number of recurrences observed is limited and numerous nonlinear algorithms (random forest, gradient boosting, neural networks) are explored with no a priori calculation of sample size, following the updated criteria of Riley et al.,2 there is probable overfitting, even with internal cross-validation.
Second, it would have been beneficial for Zayas-Bórquez et al. to have explicitly reported their adherence to the TRIPOD+AI 2024 statement,3 a framework that extends the original TRIPOD criteria to artificial intelligence models and incorporates specific dimensions of transparency, algorithmic fairness, and reproducibility. The detailed description of preprocessing, imputation strategy justification, class imbalance management, and code availability are essential for third-party model replication and validation.
Third, the clinical utility of a model is not limited to its discrimination. The area under the curve (AUC), although informative, should be complemented by smoothed calibration curves and calibration intercept/slope estimates, given that a model with good discrimination, but poor calibration, may lead to clinically inadequate decisions.4 By incorporating decision curve analysis, the net benefit across plausible thresholds can be evaluated, providing an actionable perspective for a multidisciplinary team.
Lastly, no predictive model should be implemented without external validation in geographically and temporally independent cohorts.5 The heterogeneity of Latin American populations, regarding healthcare service access, comorbidities, ancestry, and surgical practices, underscores the need for prospective Ibero-American consortia that evaluate performance and equity across subgroups.
We recognize the exploratory value of the pilot study by Zayas-Bórquez et al. as a promising starting point. We trust that our comments will contribute to the development of robust, transparent, and clinically useful predictive tools for gastrointestinal oncology in our region.
Author contributionsJ.S. Therán-León: conception of the manuscript, critical analysis of the article under discussion, literature search and selection of references, drafting of the manuscript, critical review of its intellectual content, and approval of the final version. D.A. Montagut-Bermón: critical analysis of the article under discussion, supplementary literature review, critical review of the intellectual content, and approval of the final version.
C.A. Calderón-Salazar: critical analysis of the article under discussion, contribution to the methodological discussion, critical review of the intellectual content, and approval of the final version.
All three authors meet the authorship criteria of the International Committee of Medical Journal Editors (ICMJE), approved the final version of the manuscript, and accept public responsibility for its content.
Ethical considerationsProtection of human subjects and animals. The authors declare that no experiments involving human subjects or animals were conducted for this research. This Letter to the Editor constitutes a critical methodological and conceptual analysis of the previously published article by Zayas-Bórquez et al. (Rev Gastroenterol Mex 2026;91:68–75), without intervention involving human participants, collection of primary data, or manipulation of biological material. Given the nature of this contribution, approval by a research ethics committee was not required. The authors have acted in accordance with the ethical principles set forth in the Declaration of Helsinki of the World Medical Association (Fortaleza revision, 2013), the guidelines of the Council for International Organizations of Medical Sciences (CIOMS), and Resolution 8430 of 1993 of the Colombian Ministry of Health, as well as with the applicable codes of professional ethics governing the practice of medicine in their country.
Confidentiality of data. The authors declare that this article contains no personal, clinical, or identifiable patient data. No medical records, diagnostic images, institutional records, or databases subject to regulations governing the processing of sensitive information (Colombia’s Law 1581 of 2012 and related provisions) were used.
Right to privacy and informed consent. The authors declare that this article contains no patient data and, consequently, obtaining informed consent for the publication of clinical information or identifiable graphic material was not necessary.
Academic integrity. The authors declare that the manuscript is original, has not been previously published—either in print, electronically, in part, or in another language—and is not simultaneously under consideration by another journal. They also declare that they have respected the copyrights of the cited works, which are appropriately referenced using numbered Vancouver style.
Declaration of Generative AI and AI-assisted technologies in the writing processThe authors did not use generative artificial intelligence tools in the preparation of this manuscript. All conceptual development, critical analysis, and final drafting are the sole responsibility of the undersigned authors.
Financial disclosureNo specific funding from public-sector agencies, commercial-sector entities, or nonprofit organizations was received in relation to this research.
The authors declare that they have no conflict of interest—personal, financial, institutional, or otherwise—that could have inappropriately influenced the content or conclusions of this manuscript.
