Review Reports - Predictive Analysis of Drug-Resistant Tuberculosis: Integrating Molecular Markers, Clinical Governance, and Community-Engaged Education in Rural South Africa

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Major concern

The study addresses an important problem, but several methodological issues need attention before the final decision.

The single center study and sample size (207 patients, ~45 MDR‑TB) is modest for developing and validating prediction models, especially given the very high reported AUCs (0.96 and 0.99), which raises concern about overfitting and overly optimistic internal validation.

The predictor set is quite limited (age, sex, a handful of canonical mutations), so complex methods like Random Forest add little beyond logistic regression; this makes the “machine learning” angle less compelling unless the models are shown to outperform simple, interpretable approaches in a robust way.

The statistical analysis description should clarify exactly how 10‑fold cross‑validation was implemented (including whether all preprocessing and feature handling were nested within folds) and report calibration, not just discrimination.

The risk‑stratification and simulation components are soundly under‑specified; they need clearer assumptions, a clearer structure, and more explicit treatment of uncertainty, and should be presented as modeled projections rather than firm empirical findings.

Overall, the work is relevant, but it would benefit from more cautious claims, clearer statistical reporting, and, if possible, some form of external or temporal validation.

There are many good studies in the same direction, with much larger sample sizes.

Read these studies/databases with sufficient data and prediction power.

https://pmc.ncbi.nlm.nih.gov/articles/PMC6368788/

https://pmc.ncbi.nlm.nih.gov/articles/PMC8844416/

They developed 24 binary classifiers of MTB drug resistance status across eight anti-MTB drugs and three different ML algorithms: logistic regression, random forest, and 1D CNN using a training dataset of 10,575 MTB isolates collected from 16 countries across six continents, where an extended pan-genome reference was used for detecting genetic features.

https://pmc.ncbi.nlm.nih.gov/articles/PMC10483414/

They have developed DrPRG (Drug Resistance Prediction with Reference Graphs) using the bacterial reference graph method Pandora. First, we outline the construction of a Mycobacterium tuberculosis drug resistance reference graph. The graph is built from a global dataset of isolates with varying drug susceptibility profiles, thus capturing common and rare resistance- and susceptible-associated haplotypes. We benchmark DrPRG against the existing graph-based tool Mykrobe and the haplotype-based approach of TBProfiler using 44,709 and 138 publicly available Illumina and Nanopore samples with associated phenotypes.

https://pubmed.ncbi.nlm.nih.gov/36177394/

https://www.frontiersin.org/journals/genetics/articles/10.3389/fgene.2019.00922/full

https://pubmed.ncbi.nlm.nih.gov/41402024/

These papers can be used to (a) show that your findings on S315T/S450L are biologically consistent with larger cohorts, and (b) justify your call for cautious interpretation of very high AUCs from a 207‑patient dataset compared with models trained on thousands of isolates.

Author Response

Dear Reviewer 1

Please find the attached comments

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript entitled "Predictive Analysis of Drug-Resistant Tuberculosis: Integrating Molecular Markers, Clinical Governance, and Community-Engaged Education in Rural South Africa" reports the rate of some mutations responsible for DR in M. tuberculosis and machine learning approach to create a predictive model for timely diagnosis and the prevention of DR-TB in Eastern Cape Province, South Africa. This study has importance in terms of its contribution to diagnosis and prevention of DR-TB in such locations with high TB burden. However, there are some shortcomings related with the manuscript as listed below.

1) All mutations including the novel ones related to INH and RIF resistance as reported in "Catalogue of mutations in Mycobacterium tuberculosis complex and their association with drug resistance, 2nd ed" should be mentioned in the Introduction. Why only three mutations were used in the study? Besides, two rpoB variants were also mentioned in the Results (which were not mentioned in the Methods). This should be consistent.

2) In Results and Discussion, the authors say "rpoB S450L was the predominant rifampicin resistance mutation". However, only this mutation was analyzed. Therefore, it is not convenient to say "predominant".

3) What were the resistant profiles of the samples accoring to Xpert MTB/RIF and Line Probe Assay platforms? These rates should be compared with the mutation rates.

4) In predictor variables, it is stated that "in addition to genomic predictors, relevant clinical variables were incorporated into the analysis, including HIV status, body mass index (BMI), presence of diabetes mellitus, previous TB treatment history, treatment delay of ≥30 days from diagnosis to initiation, and hospital admission status, allowing for integrated assessment of molecular and host-related determinants of resistance and treatment outcomes". However, only age and sex variables were used in addition to mutations. This issue should be addressed. Related data should be presented in a Table. Why were the other variables not used in ML?

5) The term "any resistance" used in the manuscript is not suitable because there are too much varieties of DR in TB. Therefore, this terms should be specific to INH and/or RIF.

6) All gene and bacteria names should be italic. Also, after first use of Mycobacterium tuberculosis, it should be used as M. tuberculosis.

Author Response

Dear Reviewer 2

Please find the attached comments

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The manuscript represents a high-quality and innovative contribution to the field of tuberculosis control and precision public health.

With minor revisions focused on validation and implementation clarity, it is suitable for publication.

Specifically, inclusion of calibration metrics, confusion matrix parameters, clarification of class imbalance handling, and expanded description of reflex LPA simulation assumptions would further strengthen the manuscript and enhance its translational applicability.

Author Response

Dear Reviewer 3,

Please find the attached comments.

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

Dear Authors,

I was invited to review the article “Predictive Analysis of Drug-Resistant Tuberculosis: Integrating Molecular Markers, Clinical Governance, and Community-Engaged Education in Rural South Africa “ by Siphosihle Conham, Ncomeka Sineke, Ntandazo Dlatu, Lindiwe Modest Faye, Mojisola Clara Hosu, Teke Apalata.

I consider that the subject integration of different approaches and region described are key points of increased interest to the readers from different specialties. Also, this article is the prove on the way of medicine in present- integration of informatics, epidemiology and molecular biology, that must be understood by all the medical community.

I have some suggestions:

Line19 - please write Mycobacterium tuberculosis in italics

In the abstract, you haven,t clearly mentioned the involvement of the community health workers in this study, they appear only in the conclusion and result section.

Line 40- please detail the resource limited settings

Line 71- describe the population composition: number, sex, age, prevalence and incidence of tuberculosis, organization of the tuberculosis network

Chapters 2.2. and 2.3- which is the studied period?

Please present the source of the isolates by pathological product- sputum or extrapulmonary products

I particularly liked the figures stating the ML and molecular biology contribution

Author Response

Dear Reviewer 4

Please find the attached comments.

Author Response File: Author Response.pdf

Reviewer 5 Report

Comments and Suggestions for Authors

Major Revision Points

Sample Size and Generalizability
- The study is based on a relatively small sample (n = 207) from a single geographic region (Eastern Cape, South Africa). This limits the generalizability of the findings and may affect the stability of machine learning models, especially for less common mutations.
- Recommendation: Acknowledge this limitation more explicitly and discuss the need for multi-site external validation before clinical implementation.
Missing Clinical Covariates
- Key variables such as HIV status, prior TB treatment history, and diabetes were noted as incomplete or excluded from the final models. These are known to influence drug resistance and treatment outcomes.
- Recommendation: Either impute missing data using robust methods or clearly state how missingness may have biased the results. Discuss the potential impact on model performance and interpretation.
Model Validation and Overfitting
- Although 10-fold cross-validation was used, the sample size is modest for the number of predictors. Random forest models, in particular, are prone to overfitting without careful tuning and external validation.
- Recommendation: Report out-of-bag (OOB) error rates or perform bootstrapping for internal validation. Consider simplifying the model or using penalized regression to reduce overfitting risk.
Integration Framework Feasibility
- The proposed integration of predictive outputs into clinical governance (CG) and community-engaged education (CEE) is conceptually strong but lacks pilot data or implementation science evidence.
- Recommendation: Include qualitative feasibility data or stakeholder feedback to support the proposed operational framework. If unavailable, frame this as a hypothesis-generating model requiring further testing.
Clarity of Outcome Definitions
- The primary outcome is binary (successful vs. unsuccessful treatment), but the definition of "unsuccessful" includes loss to follow-up and non-conversion, which may have different determinants.
- Recommendation: Consider sensitivity analyses using alternative outcome definitions or multinomial modeling to capture heterogeneity in treatment failure pathways.

Minor Revision Points

Typographical and Formatting Issues
- There are inconsistent line breaks, numbering artifacts, and figure placement issues (e.g., repeated figure numbers, missing captions).
- Recommendation: Thoroughly proofread the manuscript and ensure all figures are correctly labeled and cited in the text.
Figure Quality and Interpretation
- Figures 3–6 are conceptually rich but lack clear legends or explanatory captions. Some appear duplicated or misplaced.
- Recommendation: Simplify network diagrams and provide detailed captions explaining nodes, edges, and flow of interventions.
Terminology Consistency
- Terms like "Hr-TB," "INH-resistant," and "isoniazid monoresistance" are used interchangeably, which may confuse readers.
- Recommendation: Standardize terminology throughout the manuscript (e.g., use "isoniazid-resistant TB" consistently).
Reference Updates
- Some references are from preprints or in-press articles (e.g., medRxiv, 2025). While acceptable for emerging research, ensure that peer-reviewed versions are cited where available.
- Recommendation: Update references to final published versions prior to resubmission.
Language and Readability
- Some sentences are overly long or dense, particularly in the Discussion and Methods sections.
- Recommendation: Simplify complex sentences for clarity and readability, especially for an international audience.

Author Response

Dear Reviewer 5

Please find the attached comments.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

My previous major concern was not addressed. Sample Size, Single center study, predictive variable.

Author Response

Dear Reviewer 1,

Attached are our responses to your comments. We sincerely appreciate your feedback, as it has significantly enhanced the rigour, flow, and overall quality of our manuscript.

Thank you once again for your insightful suggestions.

Best regards,
Dr Dlatu, representing the authors.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

Some of the concerns based on the previous review have not been addressed in the revised manuscript.

2) Why only three mutations were used in the study? Besides, two rpoB variants were also mentioned in the Results (which were not mentioned in the Methods). This should be consistent.

3) The resistant profiles of the samples (Table 1) should be compared with the mutation rates (Table 2).

4) The authors say "a descriptive table of the collected clinical variables was added" but it is not found in the manuscript.

5) The term "any resistance" is still used in the Fig.1.

6) All gene and bacteria names should be italic. Also, after first use of Mycobacterium tuberculosis, it should be used as M. tuberculosis. This issue is still valid.

7) The layout of the revised manuscript with bullet points is not suitable.

Author Response

Dear Reviewer 2,

Thank you for your insightful comments. Your feedback has significantly enhanced the rigour of our manuscript, and we appreciate your contribution.

Kind regards,

Dr Dlatu

Author Response File: Author Response.docx

Reviewer 4 Report

Comments and Suggestions for Authors

Dear Authors,

I read the second version of the article “Predictive Analysis of Drug-Resistant Tuberculosis: Integrating Molecular Markers, Clinical Governance, and Community-Engaged Education in Rural South Africa “ by Siphosihle Conham, Ncomeka Sineke, Ntandazo Dlatu, Lindiwe Modest Faye, Mojisola Clara Hosu, Teke Apalata.

I was impressed by the new presentation and I thank you for taking in consideration my suggestions.

In this form, it is clear and understandable having an increased scientific value.

Author Response

Thank you for the insightful comments you raised in our manuscript. We truly appreciate it.

Regards

Dr Dlatu

Author Response File: Author Response.docx

Reviewer 5 Report

Comments and Suggestions for Authors

Thank you for your thorough and thoughtful responses to the points I raised in my review. I appreciate the care you have taken in revising the manuscript.

I confirm that all my comments have been adequately addressed, and I have no further suggestions for improvement. I believe the manuscript is now ready for the next steps in the editorial process.

Author Response

Dear Reviewer 5,

We would like to take a moment to express our gratitude for the valuable insights you've shared regarding our manuscript. Your feedback is instrumental in enhancing the quality of our work, and we sincerely appreciate the time and effort you've dedicated to your review.

Thank you once again.

Best regards,

Dr Dlatu

Author Response File: Author Response.docx

Round 3

Reviewer 1 Report

Comments and Suggestions for Authors

The study sample size could not be justified.

Mutations considered in the study are very few.

The limitations of the study are very high.

Similar studied have been found with better results and findings

Author Response

Dear Reviewer 1,

Thank you for your insightful comments! We've made revisions based on your feedback and are excited to share the updated responses with you. Please find them attached.

Thank you.

Dr Dlatu

Author Response File: Author Response.docx