Comment on “From innovation to implementation: Addressing the AI adoption gap in cardiac surgery”

Dear Editor,

We read with great interest the systematic review by Ngwu et al. [1], on the artificial intelligence (AI) adoption gap in cardiac surgery.

Gap in the reference numbering

The bibliography numbers proceed from thirty-three directly to thirty-five, with no citation numbered thirty-four appearing anywhere in the reference list. Since in-text citations throughout the Results and Discussion sections rely on these numbers to direct readers to the correct supporting source, such a gap raises the possibility that a citation was dropped during the preparation of the manuscript or that a numbering shift occurred without full correction across the text. Readers cannot verify whether any in-text citation intended to point to a now-missing thirty-fourth reference, which affects traceability for the surrounding claims about atrial fibrillation prediction and related electrocardiographic screening.

Quality appraisal tools designed for systematic reviews applied to primary AI validation studies.

Methodological quality was assessed using domains adapted from AMSTAR-2 and ROBIS, both of which were developed specifically to appraise the conduct of systematic reviews and their risk of bias rather than to evaluate individual primary studies reporting a machine learning model's development and validation [2]. Neither instrument was designed with domains addressing dataset representativeness for model training, internal-versus-external validation splits, or reproducibility of an algorithm's reported performance, concerns that are central to judging an individual AI study rather than a review of AI studies [3]. Established appraisal frameworks built specifically for prediction models and AI studies exist and would more directly address these concerns [4]. Adapting review-level tools to primary AI studies without describing how each adapted domain maps onto AI-specific validation concepts leaves the basis for the reported confidence ratings only partially transparent.

Confidence ratings appear disconnected from the calibration reporting gap identified elsewhere in the review.

Several primary studies have been assigned a high confidence rating based on strong discrimination statistics and external or prospective validation. The discussion separately concludes that calibration metrics were inconsistently reported or absent in most of the included studies and that discrimination measures alone do not capture absolute risk estimation. Because the quality domains described for assigning confidence ratings do not specify calibration reporting as a required element, a study could plausibly receive a high rating while sharing the calibration gap that the review later identifies as a field-wide limitation. This creates a tension between the individual confidence ratings presented for specific studies and the review's own overall verdict that calibration assessment remains a widespread deficiency, without clarifying whether the highest-rated studies were exceptions to that pattern.

Against this background, this review offers a genuinely comprehensive and clearly organized synthesis of fast-moving literature across five distinct clinical domains, and its explicit distinction between technical maturity, implementation maturity, and durable clinical adoption provides a valuable conceptual framework rarely articulated this clearly in prior surgical AI reviews. Correcting the reference numbering gap, adopting or explicitly justifying the use of AI-specific appraisal criteria alongside AMSTAR-2 and ROBIS, and clarifying whether calibration reporting factored into the individual confidence ratings would strengthen the readers' ability to interpret the review's quality assessments.

CRediT authorship contribution statement

J. Josphin Mini: Conceptualization, Formal analysis, Data curation. E. Santhi Nisha: Writing – review & editing, Writing – original draft, Visualization. Ganeshkumar M.: Writing – review & editing, Writing – original draft, Visualization.

Consent to participate

Not applicable.

Consent for publication

All authors have read and approved the final version of the manuscript and consent to its publication.

Ethics approval

Not applicable, as this manuscript is a commentary on a previously published article.

Declaration of Generative AI and AI-assisted technologies in the writing process

No generative artificial intelligence tools were used in the preparation, writing, analysis, interpretation, or revision of this manuscript.

Funding

The authors received no specific funding for this work.

Declaration of competing interest

The authors declare that they have no competing interests.

Data availability

No datasets were generated or analysed during the current study.