PREreview del Artificial Intelligence in Global Health: Transforming the Diagnosis and Management of Infectious Diseases

Brief summary of the study - a sentence summarizing the study and general comments that apply across the full paper

This manuscript reviews the application of artificial intelligence (AI) in infectious disease management, highlighting its roles in diagnosis, surveillance, drug discovery, and outbreak prediction; however, while the topic is timely and relevant, the review tends to overstate clinical impact and lacks methodological transparency and critical appraisal across the paper.

Major comments - Comments on the validity or strength of the methodology, experiments and analyses, strength of the conclusions

• No systematic methodology – The review does not specify search databases, inclusion/exclusion criteria, date range, or study quality assessment. This makes the review unreproducible and potentially biased. The authors should follow PRISMA guidelines or clearly state the narrative review’s limitations.

• Overstated conclusions not supported by evidence – The abstract claims AI “has transformed the way of infectious disease management,” but most cited studies are retrospective or preclinical. Few AI tools are standard of care. The conclusion should be tempered to “demonstrates potential to transform.”

• Missing critical discussion of model generalizability – No mention of external validation, temporal drift, or performance decay across populations. This is the single biggest failure mode of clinical AI (e.g., Google Flu Trends, Epic sepsis model). The review presents AI performance as static, which is misleading.

• Inadequate handling of statistical claims – Reported accuracies (e.g., “above 90%” for malaria, “95%” for TB resistance) are presented without confidence intervals, sensitivity/specificity breakdowns, or details on training/test splits. This prevents assessment of precision and generalizability.

• Conflation of screening vs. diagnostic AI – The text treats AI for chest X-ray TB detection as equivalent to diagnostic confirmation, but these are clinically distinct use cases with different performance requirements. This distinction is absent.

• Missing discussion of AI failures and publication bias – Negative or null results are not mentioned. The field has a strong publication bias toward positive findings, which the review perpetuates.

Minor comments - Clarifications to statements in the text, interpretation of the results, presentation of the data/figures

• Funding disclosure statement is lacking in this preprint.

• The manuscript would benefit from improved sectional organization, especially separating: diagnostics, therapeutics, surveillance

• Several claims require specific citations rather than general statements. Figures/tables (if included) should: summarize model types and datasets, include performance metrics in a structured format

• Terminology such as “AI,” “machine learning,” and “deep learning” should be used more precisely.

Comments on reporting - information on the statistical analyses or availability of data.

• The manuscript lacks transparency in: study selection, data synthesis approach

• No standardized reporting framework (e.g., TRIPOD-AI/STARD-AI alignment) is followed.

• Performance metrics are inconsistently reported and lack statistical completeness.

• There is no mention of: data/code availability, reproducibility considerations

Suggestions for future studies

• Standardized reporting for AI in infectious diseases – Develop domain-specific extensions to TRIPOD+AI or STARD-AI guidelines for infectious disease applications, including pathogen-specific metrics (e.g., limit of detection, cross-reactivity).

• Greater emphasis on: external validation across populations, prospective clinical trials, real-world deployment studies

• Inclusion of negative results and model failures to reduce publication bias.Focus on implementation science, including: integration into healthcare systems, clinician-AI interaction, cost-effectiveness

• Exploration of pathogen-specific evaluation metrics, such as: limit of detection, cross-reactivity, resistance prediction accuracy

Conflicts of interest of reviewers

The reviewer declares non-financial conflicts of interest relevant to this preprint. The reviewer has no professional or personal relationships with the authors, institutions, or funding bodies associated with this work. No competing interests exist that could bias this review.

Lead: Diptarup Mallick - No conflict of interests.

Reviewer: Dr. Mwafaq Ramzi Haji - No conflict of interests.

Disclosure of AI assistance: Reviewer used an AI language model to assist with formatting, grammar checking, and structuring the review. All scientific judgments, critical assessments, and final recommendations were made solely by the reviewers, who assumes full responsibility for the content.

Competing interests

The authors declare that they have no competing interests.

Use of Artificial Intelligence (AI)

The authors declare that they used generative AI to come up with new ideas for their review.