-->PONE-D-26-03649-->-->Topological Data Analysis for Early Warning of Severe Acute Malnutrition in Complex Humanitarian Emergencies in Nigeria-->-->PLOS One

Dear Dr.  Opue,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Please submit your revised manuscript by Apr 23 2026 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:-->

• A letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.
• A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.
• An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

We look forward to receiving your revised manuscript.

Kind regards,

Morufu Olalekan Raimi, Ph.D

Academic Editor

PLOS One

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. Your ethics statement should only appear in the Methods section of your manuscript. If your ethics statement is written in any section besides the Methods, please delete it from any other section.

3. Please provide a complete Data Availability Statement in the submission form, ensuring you include all necessary access information or a reason for why you are unable to make your data freely accessible. If your research concerns only data provided within your submission, please write "All data are in the manuscript and/or supporting information files" as your Data Availability Statement.

4. When completing the data availability statement of the submission form, you indicated that you will make your data available on acceptance. We strongly recommend all authors decide on a data sharing plan before acceptance, as the process can be lengthy and hold up publication timelines. Please note that, though access restrictions are acceptable now, your entire data will need to be made freely accessible if your manuscript is accepted for publication. This policy applies to all data except where public deposition would breach compliance with the protocol approved by your research ethics board. If you are unable to adhere to our open data policy, please kindly revise your statement to explain your reasoning and we will seek the editor's input on an exemption. Please be assured that, once you have provided your new statement, the assessment of your exemption will not hold up the peer review process.

5. Please amend your authorship list in your manuscript file to include author Job Agba Opue, Uchechukwu Emena Okorie.

6. Your abstract cannot contain citations. Please only include citations in the body text of the manuscript, and ensure that they remain in ascending numerical order on first mention.

7. Please ensure that you refer to Figure 1, 2, 3, in your text as, if accepted, production will need this reference to link the reader to the figure.

8. We note that Figure(s) 1-3 in your submission contain copyrighted images. All PLOS content is published under the Creative Commons Attribution License (CC BY 4.0), which means that the manuscript, images, and Supporting Information files will be freely available online, and any third party is permitted to access, download, copy, distribute, and use these materials in any way, even commercially, with proper attribution. For more information, see our copyright guidelines: http://journals.plos.org/plosone/s/licenses-and-copyright.

We require you to either (1) present written permission from the copyright holder to publish these figures specifically under the CC BY 4.0 license, or (2) remove the figures from your submission:

a. You may seek permission from the original copyright holder of Figure(s) 1-3 to publish the content specifically under the CC BY 4.0 license.

We recommend that you contact the original copyright holder with the Content Permission Form (http://journals.plos.org/plosone/s/file?id=7c09/content-permission-form.pdf) and the following text:

“I request permission for the open-access journal PLOS ONE to publish XXX under the Creative Commons Attribution License (CCAL) CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). Please be aware that this license allows unrestricted use and distribution, even commercially, by third parties. Please reply and provide explicit written permission to publish XXX under a CC BY license and complete the attached form.”

Please upload the completed Content Permission Form or other proof of granted permissions as an "Other" file with your submission.

In the figure caption of the copyrighted figure, please include the following text: “Reprinted from [ref] under a CC BY license, with permission from [name of publisher], original copyright [original copyright year].”

b. If you are unable to obtain permission from the original copyright holder to publish these figures under the CC BY 4.0 license or if the copyright holder’s requirements are incompatible with the CC BY 4.0 license, please either i) remove the figure or ii) supply a replacement figure that complies with the CC BY 4.0 license. Please check copyright information on all replacement figures and update the figure caption with source information. If applicable, please specify in the figure caption text when a figure is similar but not identical to the original image and is therefore for illustrative purposes only.

9. Please include captions for your Supporting Information files at the end of your manuscript, and update any in-text citations to match accordingly. Please see our Supporting Information guidelines for more information: http://journals.plos.org/plosone/s/supporting-information.

10. We are unable to open your Supporting Information file [figure 1_TDA.tex, figure 2__DTA.tex, figure 3_TDA.tex, plosone_submission manuscript TDA.tex]. Please kindly revise as necessary and re-upload.

11. If the reviewer comments include a recommendation to cite specific previously published works, please review and evaluate these publications to determine whether they are relevant and should be cited. There is no requirement to cite these works unless the editor has indicated otherwise.

Additional Editor Comments :

PLOS ONE Editorial Decision

Manuscript Number: PONE-D-26-03649

Title: Topological Data Analysis for Early Warning of Severe Acute Malnutrition in Complex Humanitarian Emergencies in Nigeria

Authors: Job Agba Opue, PhD; Uchechukwu Emena Okorie, PhD

Dear Dr. Opue and Dr. Okorie,

Thank you for submitting your manuscript to PLOS ONE. I have now completed a thorough assessment of your submission, including the ten (10) independent reviewer reports and my own detailed evaluation.

The reviewers represent a broad range of expertise, including computational topology, humanitarian epidemiology, statistical methodology, and public health practice. Their assessments are remarkably consistent: all ten reviewers identify fundamental, unresolved issues that preclude acceptance in the current form. The convergence of concerns across such a diverse panel is striking and warrants serious attention. Let me be direct: this manuscript is not ready for publication. The problems identified are not minor editorial matters but core methodological and transparency failures that undermine the validity of the reported findings. The exceptionally strong predictive claims (ROC-AUC 0.973, 7.3-month lead time) are presented without the evidentiary foundation required to support them.

Summary of Critical Issues

1. The Outcome Variable is Not Adequately Defined

Multiple reviewers (Reviewers 2, 3, 4, 7, 8) highlight this as a fatal flaw. The manuscript states that “elevated SAM risk” is defined either as a binary indicator or an ordinal category “depending on the operational outcome series available.” This is not acceptable.

Questions that remain unanswered:

• What specific dataset provides the SAM outcome? (DHS? MICS? SMART surveys? Treatment admissions?)

• What is the temporal resolution? (Monthly? Quarterly? Annual?)

• What is the time period covered? (2010-2020? 2015-2024?)

• How is a “surge” operationally defined? (Percent increase above baseline? Absolute threshold? Statistical anomaly?)

• What threshold separates “normal” from “crisis”?

• What is the class distribution? (How many LGAs experienced surges? How many observations?)

• How is lead time calculated? (First crossing of a warning threshold to first observed surge? Peak to peak?)

Without precise, reproducible definitions, the entire modeling exercise is uninterpretable. A reader cannot assess whether the model is forecasting deterioration or simply reproducing an existing classification. This must be addressed before any further consideration.

2. The Validation Framework is Inadequately Specified and Likely Optimistically Biased

Reviewers 2, 3, 4, 7, and 8 all raise serious concerns about potential temporal data leakage. The manuscript mentions “stratified K-fold cross-validation” but does not clarify whether folds are constructed in a time-aware manner.

In a forecasting context with lagged predictors and claimed lead times, standard random K-fold splitting is invalid. It contaminates the training set with information from the future relative to the test set. If temporal ordering was not strictly respected through forward-chaining, rolling-origin evaluation, or strict temporal hold-out, the reported performance metrics (ROC-AUC 0.973, F1 0.931) are likely optimistic and not representative of real-world predictive deployment.

Additionally, the performance claims are exceptionally strong for subnational humanitarian forecasting. Yet no confidence intervals, fold-level variability, calibration diagnostics, or external validation across time or space are provided. Without these, it is impossible to judge whether the performance is robust or reflects favorable data structure or methodological artifacts.

3. The Spatial Coverage is Unclear and Potentially Misleading

Reviewer 2 identifies a critical discrepancy: the manuscript presents the framework as an LGA-level system for Nigeria, yet the pipeline figure specifies a point cloud of 97 LGAs. Nigeria has 774 LGAs.

If only 97 LGAs were included, this represents a substantial spatial restriction that is neither clearly described nor justified. This is not a minor reporting detail, limiting the spatial domain directly affects the geometry of the point cloud, the resulting topological features, and the interpretation of β0 as “system fragmentation.” Without a transparent explanation of the sampling frame, the national framing of the study is misleading.

4. The Mathematical Stability Arguments Do Not Support the Predictive Robustness Claims

Reviewers 2 and 3 correctly identify a conceptual error. The manuscript invokes standard stability theorems for persistence diagrams and landscapes under bounded perturbations, then links these to the empirical observation that model accuracy degrades only slightly under simulated missingness.

The cited theorems concern distances between topological summaries, not the performance of a downstream machine learning classifier. Stability of the persistence representation does not imply stability of a gradient boosting model trained on those features. The argument conflates geometric stability with predictive stability, which is not formally demonstrated. This overstates the strength of the robustness claim.

5. The Data Availability Statement Does Not Meet PLOS ONE Standards

Reviewers 2, 3, 4, 8, and 10 all note that the data availability statement is insufficient. The manuscript states that code and a frozen analysis dataset “should be deposited upon submission/acceptance,” which suggests these materials are not yet publicly accessible.

PLOS ONE requires that all data underlying the findings be fully available without restriction at the time of submission. Given the complexity of the 57-dimensional tensor construction and the extensive preprocessing pipeline, reproducibility is essential. A permanent repository with a DOI, clear versioning of source datasets, and executable scripts is necessary to allow independent verification.

6. The Manuscript Structure and Presentation Are Problematic

Multiple reviewers (1, 6, 8, 9, 10) note structural and presentational issues:

• The abstract is fragmented and does not follow PLOS ONE's preferred format. It should be a single, integrated paragraph clearly presenting the research question, data, methods, key quantitative results, and conclusions.

• The mathematical exposition is excessive for an applied research article. A substantial section is devoted to textbook-level stability theorems that are well established in the literature. This material should be condensed or moved to supplementary materials, with the focus shifted to empirical validation.

• The results section is underdeveloped. Missing elements include feature importance analysis, calibration assessment, uncertainty intervals, misclassification analysis, time-evolving validation, sensitivity to preprocessing choices, and detailed ablation studies.

• Figures contain promotional language (e.g., “cost-effective anticipatory action” with dollar figures) that is inappropriate for a scientific manuscript. This language should be removed or toned down.

• The keywords largely duplicate the title and provide limited indexing value. They should be revised to introduce complementary methodological or contextual terms.

• The title overstates the scope. As Reviewer 2 notes, it implies a broadly applicable early-warning system, whereas the manuscript presents a single-country case study without external validation. The title should be revised to reflect the methodological and contextual limits.

7. The Policy Implications Are Underdeveloped

Reviewers 1 and 10 note that while the methodological contribution is conceptually interesting, the policy implications are not clearly articulated. How would decision-makers act on topological signals? What would an operational trigger look like? How would this integrate with existing early warning systems (e.g., IPC, CH, Cadre Harmonisé)? Without this connection, the practical impact remains unclear.

Specific Required Revisions

If the authors wish to resubmit, the following must be addressed comprehensively. Given the number and severity of issues, this represents a fundamental revision, not minor editing.

A. Outcome Definition (Required)

• Specify the exact SAM outcome source (dataset name, provider, version).

• Report the temporal resolution, time period covered, and total number of observations.

• Provide the operational definition of a “surge” with quantitative thresholds.

• Report class distribution (number of surge vs. non-surge observations).

• Clearly explain how lead time is computed, with a precise mathematical or algorithmic definition.

B. Validation Framework (Required)

• Implement and describe a strict time-aware validation strategy (forward-chaining, rolling-origin, or temporal hold-out).

• Report performance metrics with confidence intervals (bootstrapped or across folds).

• Provide calibration plots and calibration metrics (e.g., Brier score, calibration slope).

• Report sensitivity to hyperparameter choices and demonstrate nested validation if tuning was performed.

• If spatial dependence is a concern, report results from spatial cross-validation or cluster-robust standard errors.

C. Spatial Coverage (Required)

• Clearly state the number of LGAs included in the final analysis.

• If this is a subset of Nigeria's 774 LGAs, explain the sampling frame, inclusion/exclusion criteria, and justification.

• Discuss how the spatial restriction affects the interpretation of topological features and the generalizability of findings.

D. Reproducibility (Required)

• Deposit all code, preprocessing scripts, and a frozen analysis dataset in a public repository (Zenodo, OSF, Figshare) with a permanent DOI.

• Include clear documentation of software versions, package dependencies, and execution instructions.

• Provide a data dictionary for all 57 variables, including sources, preprocessing steps, and transformations.

• Ensure the repository is publicly accessible at the time of resubmission.

E. Mathematical Exposition (Required)

• Condense the mathematical foundations section substantially. The stability theorems are standard and do not need to be proved in full.

• Move extended derivations to supplementary materials.

• Clarify that geometric stability of persistence summaries does not guarantee predictive stability; temper claims accordingly.

• If robustness to missingness is claimed, provide empirical validation with multiple missingness mechanisms and report variability.

F. Results Expansion (Required)

• Add feature importance analysis to identify which topological features contribute most to predictions.

• Report performance on a temporal hold-out set (e.g., most recent year(s) not used in training).

• Include confusion matrices, precision-recall curves, and detection delay distributions.

• Analyze misclassifications: which LGAs are consistently misclassified and why?

• Compare against multiple conventional baselines (not just one gradient-boosting model) with statistical tests of improvement.

G. Presentation and Structure (Required)

• Rewrite the abstract as a single, integrated paragraph.

• Restructure the manuscript following standard scientific format: Introduction, Literature Review, Methods, Results, Discussion, Conclusions.

• Develop a proper Literature Review section integrating relevant prior work on malnutrition forecasting, early warning systems, and topological data applications.

• Remove promotional language from figures and text.

• Ensure all figure legends are self-explanatory and figures are high-resolution.

• Check all references for completeness and correct formatting per PLOS ONE style.

H. Policy Implications (Recommended)

• Expand the discussion of how topological signals could be operationalized in humanitarian decision-making.

• Provide concrete examples of trigger thresholds and anticipatory action protocols.

• Discuss integration with existing early warning frameworks (IPC, CH, etc.).

• Acknowledge limitations in translating mathematical topology to field operations.

I. Higher-Dimensional Homology (Recommended)

• Either provide a clear interpretation of β₂ features in the humanitarian context with empirical evidence, or explicitly limit the analysis to β₀ and β₁.

Decision: Major Revision Required

Given the above, I cannot recommend acceptance in the current form. The manuscript does not meet PLOS ONE's standards for methodological rigor, transparency, or reproducibility. The authors may choose to resubmit a substantially revised manuscript addressing all of the concerns outlined above. Any resubmission will be evaluated de novo and will be sent for additional peer review. Given the number and depth of required changes, I strongly recommend that the authors:

1. Seek methodological consultation, particularly regarding validation design for forecasting problems.

2. Ensure all data and code are publicly accessible at the time of resubmission.

3. Carefully address each reviewer's comment in a detailed point-by-point response.

4. Consider whether the current evidence base is sufficient to support the strong claims made, or whether the framing should be tempered to a proof-of-concept or methodological demonstration.

If the authors believe these concerns cannot be adequately addressed, they may wish to consider alternative venues better suited to preliminary methodological explorations.

I thank the authors for their innovative approach to a critical problem. The core idea, applying topological data analysis to humanitarian early warning, has genuine merit. However, the execution and reporting must meet the field's standards for evidence and transparency.

Sincerely,

Morufu Olalekan Raimi, PhD

Academic Editor

PLOS ONE

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

-->Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. -->

Reviewer #1: Yes

Reviewer #2: No

Reviewer #3: No

Reviewer #4: No

Reviewer #5: Partly

Reviewer #6: Yes

Reviewer #7: Yes

Reviewer #8: Yes

Reviewer #9: Yes

Reviewer #10: Yes

**********

-->2. Has the statistical analysis been performed appropriately and rigorously? -->

Reviewer #1: Yes

Reviewer #2: No

Reviewer #3: No

Reviewer #4: No

Reviewer #5: Yes

Reviewer #6: Yes

Reviewer #7: No

Reviewer #8: Yes

Reviewer #9: Yes

Reviewer #10: Yes

**********

-->3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.-->

Reviewer #1: Yes

Reviewer #2: No

Reviewer #3: No

Reviewer #4: No

Reviewer #5: Yes

Reviewer #6: No

Reviewer #7: No

Reviewer #8: Yes

Reviewer #9: Yes

Reviewer #10: Yes

**********

-->4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.-->

Reviewer #1: Yes

Reviewer #2: No

Reviewer #3: No

Reviewer #4: No

Reviewer #5: Yes

Reviewer #6: Yes

Reviewer #7: No

Reviewer #8: Yes

Reviewer #9: No

Reviewer #10: Yes

**********

-->5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)-->

Reviewer #1: 1.The description of data cleaning and preprocessing steps is relatively brief. Please clarify how missing values, temporal gaps, and potential reporting biases in humanitarian data were handled prior to TDA construction.

2.The application of topological data analysis (TDA) to early warning of severe acute malnutrition is interesting. However, the manuscript would benefit from a clearer statement of why TDA is theoretically better suited than conventional statistical or machine‑learning approaches in this context.

3.The rationale for selecting specific filtration scales and distance metrics is not fully justified. How sensitive are the persistence results to alternative parameter choices, and were robustness checks conducted?

4.While persistence diagrams and landscapes are presented, their interpretation for a public‑health audience remains challenging. Consider adding more intuitive explanations or real‑world analogies linking topological features to malnutrition dynamics.

5.The study would be strengthened by a more explicit comparison with standard early‑warning approaches. This would help quantify the added value of TDA beyond visual or conceptual appeal.

6.It is unclear whether the detected topological signals precede malnutrition outcomes consistently. Please clarify how the framework distinguishes early‑warning capability from contemporaneous or lagged associations.

7.Although persistence landscapes allow statistical summaries, uncertainty in predictions or signals is not explicitly discussed. Including confidence intervals or variability measures would improve the reliability of the conclusions.

8.It is suggested to add articles entitled “Al Hashmi et al. Predicting Dropout in MENA STEM Higher Education Using Explainable AI: A Machine Learning Approach”, “Chenghu & Thammano, A Self-Adaptive Weights for K-Means Classification Algorithm” and “Chang, Public Opinion Guidance Model in Major Public Crisis Events Based on Accelerated Genetic Algorithm” to the literature review.

9.Nigeria exhibits strong regional variation in conflict, climate, and food security. How does the proposed method account for spatial heterogeneity, and could regional persistence patterns mask localized crises?

10.The manuscript focuses on Nigeria as a case study. Please discuss how transferable the proposed framework is to other humanitarian settings with different data availability or crisis drivers.

11.While the methodological contribution is strong, the policy implications are underdeveloped. A clearer discussion of how decision‑makers could act on topological signals would enhance the practical impact of the work.

Reviewer #2: The current title overstates the scope and maturity of the study. It implies the development of a broadly applicable early-warning system for complex humanitarian emergencies, whereas the manuscript presents a single-country case study without external validation or demonstrated operational deployment. As written the title suggests a level of generalisability and implementation that is not fully supported by the empirical evidence provided. It should therefore be revised to more accurately reflect the methodological and contextual limits of the study.

The abstract is not written in a coherent narrative form and currently reads as a structured outline rather than a concise scientific summary. It should be rewritten as a single, integrated paragraph that clearly presents the research question, data, methodological approach, main quantitative results, and principal conclusions in a logically connected way. In its present form, it is fragmented, overly schematic, and does not sufficiently explain the operational definition of the outcome or the validation framework underlying the reported performance metrics.

The abstract is not written in a coherent narrative form and currently reads as a structured outline rather than a concise scientific summary. It should be rewritten as a single, integrated paragraph that clearly presents the research question, data, methodological approach, main quantitative results, and principal conclusions in a logically connected way. In its present form, it is fragmented, overly schematic, and does not sufficiently explain the operational definition of the outcome or the validation framework underlying the reported performance metrics.

The first major substantive issue concerns the inconsistency between the claimed national scope of the analysis and the actual number of units analysed. the manuscript presents the framework as an LGA-level system for Nigeria, yet the pipeline figure specifies a point cloud of 97 LGAs. Nigeria has 774 LGAs. If only 97 were included, this represents a substantial spatial restriction that is neither clearly described nor justified. This is not a minor reporting detail: limiting the spatial domain directly affects the geometry of the point cloud, the resulting topological features, and the interpretation of β0 as system fragmentation. Without a transparent explanation of the sampling frame, the national framing of the study is misleading.

The second substantive problem lies in the way mathematical stability results are used to support claims about predictive robustness. The manuscript invokes standard stability theorems for persistence diagrams and landscapes under bounded perturbations and then links these to the empirical observation that model accuracy degrades only slightly under simulated missingness. Howewer the cited theorems concern distances between topological summaries, not the performance of a downstream machine learning classifier. Stability of the persistence representation does not automatically imply stability of gradient boosting model trained on those features. The argument conflates geometric stability with predictive stability, which is not formally demonstrated and therefore overstates the strength of the robustness claim.

The keywords largely duplicate terminology already present in the title and therefore add limited indexing value. Several entries repeat core title phrases almost verbatim, which reduces their usefulness for discoverability. The keywords should be revised to introduce complementary methodological or contextual terms rather than restating the main elements of the title

Other problems:

A. The definition of the outcome variable is insufficiently specified and this undermines the interpretability of the entire modelling exercise. The manuscript states that elevated SAM risk is defined either as a binary indicator or as an ordinal category, depending on the operational series available, but it does not clearly identify which specific dataset is used in the final analysis, how a “surge” is operationally defined, or what quantitative threshold separates normal variation from crisis escalation. Without a precise and reproducible definition of the target variable, it is not possible to assess whether the model is genuinely forecasting deterioration or simply reproducing an existing classification.

B. The validation strategy raises concerns about potential data leakage. The text refers to stratified K-fold cross-validation, but it does not clarify whether the folds were constructed in a time-aware manner. In a forecasting context with lagged predictors and reported lead times, standard random K-fold splitting would contaminate the training set with information from the future relative to the test set. If temporal ordering was not strictly respected through forward-chaining or rolling-origin evaluation, the reported performance metrics may be optimistic and not representative of real-world predictive deployment;

C. The data availability statement does not currently meet the journal’s standards for full reproducibility. The manuscript indicates that code and a frozen analysis dataset “should be deposited upon submission/acceptance,” which suggests that these materials are not yet publicly accessible. Given the complexity of the 57-dimensional tensor construction and the extensive preprocessing pipeline, reproducibility is essential. A permanent repository with a DOI, clear versioning of source datasets, and executable scripts is necessary to allow independent verification of the findings.

D. The reported predictive performance appears exceptionally strong for an operational humanitarian forecasting problem at subnational scale. ROC-AUC values approaching 0,97 and a mean lead time of over seven months would represent a substantial advance over most published early-warning systems. However, the manuscript does not provide confidence intervals, fold-level variability, calibration diagnostics, or external validation across time or space. Without these additional analyses, it is difficult to judge whether the performance is robust or whether it reflects favourable data structure or methodological artefacts.

E. The section presenting mathematical foundations and proofs largely reiterates established stability theorems from the topological data analysis literature. While these results are correctly stated and relevant in principle, they do not constitute a novel mathematical contribution in this context. At the same time, the empirical validation of the forecasting claims remains comparatively limited. The balance of the manuscript would benefit from shifting emphasis away from general theoretical guarantees and towards a more rigorous empirical assessment of how the method performs under realistic operational constraints.

F. Although the manuscript refers to homology in dimensions zero to two, the empirical results focus almost exclusively on β0 and β1. There is no substantive presentation or interpretation of β2 features, and their operational meaning in the context of humanitarian vulnerability remains unclear. As currently written, the inclusion of higher-dimensional homology appears more formal than functional. Either a clearer justification and demonstration of its added value should be provided, or the analysis should concentrate explicitly on the dimensions that are empirically interpretable and policy-relevant.

IMPORTANT:

In its current form, the ms does not fully meet the core requirements of PLOS ONE. In particular:

(1) the data and code necessary to reproduce the analysis are not publicly available with a permanent repository and DOI, despite the journal’s strict reproducibility policy;

(2) key methodological elements, including the precise operational definition of the outcome variable and the temporal validation framework, are insufficiently specified for independent replication;

(3) there is a lack of clarity regarding the spatial coverage of the analysis, with an apparent discrepancy between the national framing and the reported number of LGAs included;

(4) several claims regarding predictive robustness and stability appear stronger than what is formally demonstrated in the empirical analysis. These issues must be addressed before the manuscript can be considered compliant with the methodological and transparency standards of the journal.

Reviewer #3: Reviewer Report

Recommendation: Reject and Resubmit (Resubmission Encouraged After Substantial Revision)

Summary

This manuscript proposes the use of topological data analysis (TDA), specifically persistent homology, to enhance early warning of severe acute malnutrition (SAM) in complex humanitarian emergencies in Nigeria. The conceptual integration of structural topology into humanitarian forecasting is innovative and potentially valuable. However, the current manuscript lacks sufficient empirical detail, methodological transparency, and analytical depth to support its strong performance claims.

While the conceptual integration of topological data analysis into humanitarian early warning is innovative, the manuscript lacks sufficient empirical detail, methodological transparency, and analytical depth to support its strong performance claims. Key aspects of data construction, validation design, and robustness evaluation are insufficiently documented. The manuscript currently reads more like a proof-of-concept note than a fully developed research article.

For these reasons, I recommend rejection in its current form, with encouragement to resubmit after substantial revision and expansion.

Major Concerns

1. Insufficient Description of Data and Outcome Construction

The manuscript does not clearly specify the exact SAM outcome source, temporal resolution, time period covered, total number of observations, class distribution, definition of a surge, or how lead time is computed. These omissions prevent reproducibility and limit evaluation of the reported performance.

2. Validation Design and Risk of Information Leakage

The reported performance (ROC–AUC 0.973; F1 0.931) is exceptionally high for humanitarian forecasting. However, the manuscript does not clearly describe whether validation is spatial, temporal, or random; whether splits prevent temporal leakage; whether TDA features are computed within each training fold; or whether hyperparameter tuning is nested within cross-validation. Without strict out-of-sample validation design, performance claims are not sufficiently supported.

3. Limited Analytical Depth

The Results section is brief and limited. Missing elements include feature importance analysis, calibration assessment, uncertainty intervals, misclassification analysis, time-evolving validation, sensitivity to preprocessing choices, and detailed ablation studies. For a high-dimensional modeling framework with strong predictive claims, the empirical analysis is underdeveloped.

4. Imbalance Between Mathematical Theory and Empirical Evidence

A substantial section is devoted to textbook-level stability theorems that are already well established in the literature. Meanwhile, empirical methodological details are comparatively brief. For an applied research article, empirical transparency and reproducibility should take precedence.

5. Reproducibility and Data Availability

The manuscript states that reproducibility materials should be deposited upon submission or acceptance. Code and a frozen analysis dataset (or complete preprocessing scripts) should be publicly available at submission to meet reproducibility standards.

Minor Issues

• Figures contain promotional-style language that should be toned down.

• “57-dimensional tensor” appears to be a feature vector and should be described accordingly.

• Interpretation of β₀ and β₁ should be empirically validated rather than asserted.

• Cost-effectiveness claims should be clearly sourced and separated from predictive performance evaluation.

Required for Resubmission

If the authors wish to resubmit, the manuscript would need:

1. Full and explicit outcome definition.

2. Clear, leakage-free validation design (preferably temporal and spatial hold-out).

3. Expanded results with detailed diagnostics and uncertainty analysis.

4. Public repository with reproducibility materials.

5. Reduction or condensation of textbook theoretical material.

6. More rigorous empirical justification of topological interpretation.

Final Recommendation

The idea is promising and innovative. However, the manuscript in its current form lacks the empirical depth, methodological transparency, and reproducibility standards required for publication. I therefore recommend rejection with encouragement to substantially revise and resubmit as a more fully developed research article.

Reviewer #4: The manuscript addresses an important problem but does not meet the methodological and reporting standards required for publication in PLOS ONE. The definition of the outcome variable and the ground truth for SAM surges are insufficiently specified, making the core contribution difficult to verify. Reported predictive performance and lead times appear exceptionally strong for humanitarian forecasting, yet the validation design lacks sufficient transparency to rule out overfitting or optimistic bias. Key details necessary for reproducibility - such as outcome construction, temporal validation strategy, handling of spatial dependence, and baseline comparability - are either missing or underdeveloped. In addition, the extensive mathematical exposition focuses on established theoretical results that do not adequately substantiate the applied early-warning claims. Overall, the limitations are substantive and would require fundamental reanalysis rather than revision. I therefore recommend rejection.

Reviewer #5: Greetings, thank you for giving me the opportunity to review this submission, below I will address my concerns.

Beforehand, the title page seems to have a formatting problem, the authors names and affiliation are replaced with ( xxxxxxx ), I suggest to fix the formatting problem during revision.

Firstly, line103 ( Ethical Approval: N/A ). I suggest removing this sentence as it looks odd after the authors have fully explained why there is no need for an ethical approval in a long paragraph. The paragraph is satisfactory by itself.

Secondly, the introduction section, it would be stronger to mention for example a publication that used liner regression model or any model to assess malnutrition and how it failed - in your point of view - in assessing the problem, this will help justify the topological data analysis and highlight the importance of its implementation in assessing relations and predictors of SAM.

This leads to thirdly, the need to have a very clear mathematical question written in the motivation part that the analysis will work to answer, the submission is currently discussing that topological analysis is better in assessing the problem without specifications.

Fourthly, it is a good sign that noise and missings have been addressed in the paper, but I miss the interpretation of what constitutes loops and B1 or even B0 parameters, I found a paragraph in the results discussing the persistence of B1 means the presence of cofactors like market prices spikes and maybe two other parameters mentioned. For readers, it is important to mention what parameters are being addressed in each group, in order to have an understanding about what makes this TDA a huge improvement in function and justifiable for application, since it deals with dimensional parameters that other models like liner regression fail to recognize, like what are the constituents of this multidimensional model. A brief talk about these parameters along with their sources is always a plus in convincing readers along with persistence graphs.

Lastly, the conclusions and recommendations are very generalized, I advice adding more details or expanding the talk about how limitations affect the ability to make further detailed interpretations of the analysis.

I hope my comments are of value for the authors and thank you again.

Reviewer #6: 1.The modular parts of the Abstract section have to be accompanied by numerical data in alignment with the conducted study, analysis and findings.

2.The narrative flow is of low quality. There is need authors to develop the following main sections and to number them as follows: 1.Introduction, 2.Literature Review, 3.Methodology and Analysis, 4.Results, 5.Discussion, 6.Conclusions Implications and Future Works. The logic of presenting the existing headings and subheadings is vague and does not make sense.

3.The missing Literature Review has to be developed from the beginning.

4.The utilization of citations is very poor, having only 16 citations. A citing enhancement by at least 10-12 more and recently published studies/citations is highly recommended.

5.At the beginning of section 3.Methodology and Analysis there is need of developing a Figure 1: Study Framework, in which the steps of the study to be displayed in the form of a diagram or flow chart.

6.It is strange that all lines 46-136 which correspond to the “Materials and methods” are deprived from numerical data or quantitative information of steps 1234, topology, homology, robust check, theorems, corollaries, together with their values and units’ measured among the involved variables. A more specific and systematic inclusion of input dataset(s) have to be included in the form of Table(s).

7.The one-sentence subsection “Robustness”, “Robustness checks”, have to be merged since they refer to the same part of the study. In general, all research parts have to be reorganized and fixed into the aforementioned 6 main headings.

8.The Discussion section is very poor. It has to be reorganized into a more detailed and cross-citing narrative. The critical point here authors to ensure that their methodology is able:

-to manage ”vulnerability in CHE settings and can extend the actionable lead time of nutrition early

warning”

- to manage “Persistent homology offers a mathematically grounded representation of structural vulnerability in complex emergencies. In Nigeria, topological signatures improve SAM risk warning performance and extend lead time”, in order to ensure a prioritization shift from reactive treatment toward anticipatory action.

Reviewer #7: This manuscript presents a novel and policy-relevant application of Topological Data Analysis (TDA) for early warning of severe acute malnutrition. The approach is innovative and the study is well structured. However, several issues related to methodological rigor, validation, and reproducibility should be addressed.

Major Issues

Outcome and Forecasting Clarity

Please provide the outcome's data source and thresholds. Explain the calculation method for the lead time.

Validation and Overfitting Risk

The performance is very high. As this is a forecasting problem, random K-fold cross-validation can lead to temporal leakage. Please use time-based validation and provide the performance uncertainty.

Model Transparency

Please provide the details of the hyperparameter tuning process and the methods used to handle the imbalance problem and the calibration process. Provide the results of the statistical tests to demonstrate the significant improvement in the performance due to the use of TDA.

Sample Size and Spatial Dependence

Please provide the details regarding the overfitting risk and the events per variable.

Data Availability (PLOS)

Please provide the details regarding the data source and the availability of the processed data and code.

Reviewer #8: The article is well written but it needs improvements;

1. The authors should include sample size and time of the study in the abstract and also in the metholody section.

2. The authors mentioned about the supplementary data availability upon acceptence which should be changed to provision of data on submission as per PLOS One policies.

3. Explain that how the authors have avoided the temporal data leakage, as it is essentail to authenticate the reported ROC–AUC (0.973) and lead-time results.

4. Clarify the terms, elevated SAM risk, threshold criteria, source of the data and the class distribution.

5. To strengthen the statistical part calibration plot should be added along with the confidence intervals for performance metrics. Also add brief description of class imbalance handling.

6. The mathematical stability section is appropriate, however, extended proofs and derivations could be moved to ''Supplementary Material'' to improve the readilablity of te article.

7. Check the references and format them accodring to the PLOS One style. Add complete URL's.

8. Ensure that all the figure legends are self explanatory. And check the article for grammatical errors.

Reviewer #9: Refer the journal's article format. The article is not in the standard format of the publication. Abstract should be given in a single paragraph. Use only standard short forms of words (CHE, SAM, TDA, etc.,). Figures are not clear.

Otherwise the paper is fine.

Reviewer #10: Dear Authors,

Thank you for the opportunity to review your manuscript titled “Topological Data Analysis for Early Warning of Severe Acute Malnutrition in Complex Humanitarian Emergencies in Nigeria” The study makes a meaningful contribution to methodological innovation in humanitarian health analytics. The practical implications for early warning systems should be more explicitly articulated, including guidance on operationalization in real-world humanitarian decision-making contexts. However, there are several areas where the manuscript would benefit from revision and clarification to meet the standards of PLOS ONE.

Here are some Comments and suggestions to the Author for improvement:

1. Technical Soundness and Conclusions

Overall, the manuscript presents a technically sound and innovative application of Topological Data Analysis (TDA) to early warning systems for Severe Acute Malnutrition (SAM) in humanitarian contexts. The analytical framework is coherent, and the data sources used are relevant to the research objectives.

The sampling strategy and data preprocessing steps are generally appropriate, though some methodological decisions (e.g., parameter selection in TDA and temporal aggregation choices) require clearer justification. The data presentation is adequate, and the main findings are logically derived from the analyses.

Limitations that affect the strength of inference:

• Limited discussion of potential biases in routine humanitarian data

• Insufficient justification of model assumptions and robustness

• Lack of external validation across additional contexts

In conclusion:

The data broadly support the conclusions, but clearer methodological justification and discussion of limitations are required to strengthen inferential validity.

2. Statistical Analysis:

Overall, the statistical and computational analyses are appropriate for the study objectives and demonstrate methodological novelty.

Strengths:

• Innovative use of TDA in humanitarian nutrition surveillance

• Logical analytical workflow

• Clear linkage between analytical outputs and outcome indicators

Weaknesses / Areas for Clarification:

• Limited explanation of uncertainty estimation

• No formal comparison with conventional early warning models

• Sensitivity analyses are not sufficiently detailed

Overall Evaluation: The statistical analysis is acceptable but would benefit from additional transparency and validation.

In conclusion:

Statistical methods are generally appropriate, though further clarification and robustness checks are recommended.

3. Data Availability:

Overall, the manuscript includes a Data Availability Statement; however, compliance with PLOS ONE’s data policy is not fully clear.

Strengths:

• Data sources are described

• Ethical constraints are acknowledged

Weaknesses:

• It is unclear whether underlying datasets or disaggregated data are publicly accessible

• Access procedures for restricted data are not fully specified

Recommendation:

To comply with PLOS ONE’s data policy, the authors should:

• Clearly state where the data are deposited, or

• Explicitly justify any restrictions and provide a clear access mechanism for qualified researchers

In conclusion:

Revisions are required to ensure full transparency and policy compliance.

4. Clarity and Language Quality:

Overall, the manuscript is written in generally clear and acceptable scientific English and is understandable to a multidisciplinary audience.

Strengths:

• Logical manuscript structure

• Clear articulation of the research problem

• Appropriate use of technical terminology

Weaknesses:

• Some sections (Methods and Results) are overly dense

• Key TDA concepts may be difficult for non-specialist readers

• Minor grammatical and stylistic issues remain

Examples (illustrative):

• Long sentences with multiple clauses

• Insufficient definition of specialized mathematical terms

Recommendation:

Minor language editing and simplification of technical explanations are recommended.

In conclusion:

The manuscript is intelligible but would benefit from improved readability and clarity.

**********

-->6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.-->

Reviewer #1: No

Reviewer #2: No

Reviewer #3: No

Reviewer #4: No

Reviewer #5: No

Reviewer #6: Yes: Dr. Grigorios L. Kyriakopoulos

Reviewer #7: Yes: Dr. Nelofer Jamil

Reviewer #8: Yes: Maria Anwar Khan

Reviewer #9: No

Reviewer #10: Yes: Abdullah Nagi Alosaimi

**********

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

To ensure your figures meet our technical requirements, please review our figure guidelines: https://journals.plos.org/plosone/s/figures

You may also use PLOS’s free figure tool, NAAS, to help you prepare publication quality figures: https://journals.plos.org/plosone/s/figures#loc-tools-for-figure-preparation.

NAAS will assess whether your figures meet our technical requirements by comparing each figure against our figure specifications.

Attachments

Attachment

Submitted filename: Reviewer_Report_Reject_and_Resubmit.docx

Attachment

Submitted filename: Reviewer Comments.docx

-->PONE-D-26-03649R1-->-->Topological Data Analysis for Predicting Disease Outbreaks in Humanitarian Settings: A Machine Learning Approach-->-->PLOS One

Dear Dr. Opue,-->-->

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Please submit your revised manuscript by Jun 13 2026 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:-->

• A letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.
• A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.
• An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

-->

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

As the corresponding author, your ORCID iD is verified in the submission system and will appear in the published article. PLOS supports the use of ORCID, and we encourage all coauthors to register for an ORCID iD and use it as well. Please encourage your coauthors to verify their ORCID iD within the submission system before final acceptance, as unverified ORCID iDs will not appear in the published article. Only the individual author can complete the verification step; PLOS staff cannot verify ORCID iDs on behalf of authors.

We look forward to receiving your revised manuscript.

Kind regards,

Morufu Olalekan Raimi, Ph.D

Academic Editor

PLOS One

Journal Requirements:

If the reviewer comments include a recommendation to cite specific previously published works, please review and evaluate these publications to determine whether they are relevant and should be cited. There is no requirement to cite these works unless the editor has indicated otherwise.

Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice.

Additional Editor Comments:

PLOS ONE Editorial Decision

Manuscript ID: PONE-D-26-03649_R1

Title: Topological Data Analysis for Predicting Disease Outbreaks in Humanitarian Settings: A Machine Learning Approach

Authors: Opue, Okorie

Editor: Dr. Morufu Olalekan Raimi

Decision: Minor Revision

I commend the authors for a remarkably thorough and responsive revision. The decision to reframe the study from malnutrition to cholera/measles outbreak prediction, implement forward chaining cross validation, add an independent temporal hold out, provide explicit outcome definitions (WHO thresholds), expand results with ablation, calibration, feature importance, and failure analyses, and commit to public code/data deposition addresses nearly all of the major concerns raised by the ten reviewers and myself. The manuscript is now substantially stronger methodologically and ethically. However, three specific issues remain that prevent acceptance in the current form. These are not conceptual deficiencies but require clear, verifiable corrections or additions before final publication.

1. Data Availability – Must be fully resolved before acceptance

The manuscript states (Appendix C, lines 755–778) that code and a frozen dataset “will be deposited in Zenodo with a permanent DOI upon acceptance.” This is insufficient for PLOS ONE policy.

Required action:

• Deposit all analysis code (preprocessing, TDA feature extraction, model training, evaluation) and a frozen, de identified analysis dataset (or complete synthetic data replicating the 57 variable structure) in a permanent public repository (Zenodo, Figshare, OSF) with a DOI at the time of resubmission.

• Provide the DOI in the manuscript’s Data Availability Statement now, not “upon acceptance.”

• The statement that “processed datasets are available upon reasonable request” is not acceptable – all data underlying the findings must be fully available without restriction. If third party data cannot be redistributed, provide clear code that reproduces all features from raw publicly available sources.

Location: Data Availability Statement (page 12 of PDF, and Appendix C). This is a non negotiable policy requirement.

2. Spatial Coverage & Generalizability – Still understated in the abstract

The abstract and key results sections still imply a national system (“across 97 LGAs in Nigeria”) without clearly signalling that these 97 LGAs are a selected high risk subset (not representative of all 774 LGAs).

Required action:

• In the abstract, results, and discussion, explicitly state: “The analysis was restricted to 97 high burden LGAs with complete data; results may not generalize to lower risk or data sparse LGAs.”

• Currently, Table 1 and the limitations section mention this, but the abstract and conclusion risk misleading readers. Amend accordingly.

Location: Abstract (lines 110-113) and Discussion, Limitations (lines 549-564 already partially there – strengthen language).

3. Minor presentation issues (correct before final acceptance)

• Figure 1 (permutation importance) is referenced in the results but the figure itself appears missing from the compiled PDF. Provide the figure file in a readable format.

• Appendix C mentions “Figure calibration plots” – ensure the figure is present and clearly labeled.

• A small number of grammatical awkwardnesses remain (e.g., “the model performed poorly in low conflict settings (AUC 0.75 0.78)”) – a final language polish by a native English speaker or professional editing service is advised. This does not block acceptance but should be done.

________________________________________

Summary of Required Changes (checklist for authors)

Issue Required action

Data availability Deposit code + frozen dataset in Zenodo/Figshare with DOI; provide DOI in manuscript now

Spatial generalizability Explicitly state in abstract and conclusion that 97/774 LGAs are a high risk subset – not fully representative

Missing Figure 1 Provide the permutation importance figure

Calibration plot Ensure figure in Appendix C is present and legible

Language polishing Minor grammatical cleanup

No re analysis or methodological changes are needed. Once the data/code are publicly deposited with a DOI and the above clarifications are made, the manuscript will be fully compliant and suitable for acceptance.

I thank the authors for their exceptional effort in this revision. The work now makes a legitimate, modest but valuable contribution to the use of TDA in humanitarian outbreak prediction.

Decision after minor revision: Accept.

Please submit the revised files within 14 days.

Sincerely,

Morufu Olalekan Raimi, PhD

Academic Editor, PLOS ONE

[Note: HTML markup is below. Please do not edit.]

To ensure your figures meet our technical requirements, please review our figure guidelines: https://journals.plos.org/plosone/s/figures

You may also use PLOS’s free figure tool, NAAS, to help you prepare publication quality figures: https://journals.plos.org/plosone/s/figures#loc-tools-for-figure-preparation.

NAAS will assess whether your figures meet our technical requirements by comparing each figure against our figure specifications.

-->

-->PONE-D-26-03649R2-->-->Topological Data Analysis for Predicting Disease Outbreaks in Humanitarian Settings: A Machine Learning Approach-->-->PLOS One

Dear Dr. Opue,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Please submit your revised manuscript by Jun 18 2026 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:-->

• A letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.
• A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.
• An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

-->

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

As the corresponding author, your ORCID iD is verified in the submission system and will appear in the published article. PLOS supports the use of ORCID, and we encourage all coauthors to register for an ORCID iD and use it as well. Please encourage your coauthors to verify their ORCID iD within the submission system before final acceptance, as unverified ORCID iDs will not appear in the published article. Only the individual author can complete the verification step; PLOS staff cannot verify ORCID iDs on behalf of authors.

We look forward to receiving your revised manuscript.

Kind regards,

Morufu Olalekan Raimi, Ph.D

Academic Editor

PLOS One

Journal Requirements:

If the reviewer comments include a recommendation to cite specific previously published works, please review and evaluate these publications to determine whether they are relevant and should be cited. There is no requirement to cite these works unless the editor has indicated otherwise.

Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice.

Additional Editor Comments:

Manuscript Number: PONE-D-26-03649R1 (Revision 1)

Title: Topological Data Analysis for Predicting Disease Outbreaks in Humanitarian Settings: A Machine Learning Approach

Recommendation: Minor Revision

Reviewer: Morufu Olalekan Raimi, PhD (Academic Editor, PLOS One – Infectious Disease Epidemiology & Machine Learning)

Review Date: 2026-04-30

Reviewer Comments

Dear Authors,

Thank you for submitting this substantially revised manuscript and for the extraordinarily detailed response to the previous reviews. The reframing from malnutrition to cholera/measles outbreak prediction, the implementation of temporally-aware validation, the expanded outcome specification, and the addition of calibration, ablation, and robustness analyses have markedly improved the scientific rigor. The manuscript now makes a credible contribution to the literature on machine learning for outbreak prediction in humanitarian settings.

However, several issues remain that require attention before acceptance. These are moderate in scope but important for reproducibility, transparency, and appropriate interpretation.

Recommendation: Minor Revision – The manuscript is methodologically sound but requires clarification of several technical details, correction of minor errors, and strengthening of the data availability commitment.

Major Concerns

1. Data Availability – Insufficient Commitment for PLOS ONE Standards

The manuscript states (Appendix C, lines 755-778) that code will be deposited in a GitHub repository and data in Zenodo “upon acceptance.” The Data Availability Statement on page 10 states “Yes – all data are fully available without restriction” but the detailed statement says “available from the corresponding author upon reasonable request.”

This is not compliant with PLOS ONE’s data availability policy, which requires:

• All data underlying the findings to be made fully available without restriction at the time of publication.

• “Available upon request” is not sufficient.

• Code repositories should be provided at submission, not upon acceptance.

Required action:

• Provide the actual GitHub repository URL (even if private with a disclosure mechanism for reviewers) or a Zenodo DOI for the analysis code.

• For data: Since much of the data are from public sources (CHIRPS, ERA5, ACLED, IOM, etc.), the authors should provide a list of exact URLs and describe how the compiled dataset can be reconstructed. If the compiled dataset cannot be publicly deposited due to data use agreements, state this explicitly and provide a simulation study or synthetic data as a minimal example.

• Revise the Data Availability Statement to: “All primary data sources are publicly available (see Appendix C for URLs). The compiled dataset and analysis code are available at [repository URL] and [Zenodo DOI].”

2. Spatial Sampling Bias – Understated in Limitations

The study includes only 97 of 774 LGAs (12.5%). Table 1 shows that included LGAs have significantly higher conflict intensity (2.4 vs. 1.2 events/week, p<0.001), higher IDP populations (12,450 vs. 4,200, p<0.001), and lower vaccination coverage (62% vs. 68%, p=0.002). The authors acknowledge this in Limitations but do not quantify how this bias might affect model generalizability.

Required action:

Add a sentence in Limitations: “Because included LGAs represent higher-risk settings with elevated conflict and displacement, model performance may be lower in LGAs with lower baseline risk. Prospective validation in a representative sample of LGAs is needed before operational deployment.”

3. Topological Feature Interpretation – Remains Overstated Despite Revision

The authors have tempered claims (e.g., “these labels describe mathematical properties, not established causal mechanisms”). However, the manuscript still states (Results, lines 396–398): “The β0 fragmentation index captures the degree to which risk factors operate in disconnected clusters rather than as an integrated system.”

This is not a mathematical property – it is a conceptual interpretation. The β0 index mathematically counts connected components in a Vietoris–Rips complex at a given scale. Whether that corresponds to “disconnected risk factors” is an empirical hypothesis that the study does not test.

Required action:

Rephrase to: “The β0 index mathematically counts the number of connected components in the persistence filtration. We label this ‘fragmentation’ as a conceptual heuristic, but this interpretation has not been empirically validated.”

4. Missing Information on Persistent Homology Computation

The Methods section (lines 209–219) describes Vietoris–Rips filtrations with 100 logarithmically-spaced scale parameters (ε = 0.01 to 10.0). However, several critical details are missing:

• How were the 57 features normalized/scaled before computing Euclidean distances? TDA is sensitive to feature scaling. Were features standardized (mean 0, SD 1)? Min-max scaled? Not stated.

• What is the justification for the ε range (0.01 to 10.0)? Was it data-driven or arbitrary?

• Were all 57 features used simultaneously to compute persistence? Or were subsets used? Not clear.

• What software/library was used for persistent homology computation? GUDHI is mentioned (line 282) but not in the TDA Methods section.

Required action:

Add a paragraph specifying: (1) feature scaling method; (2) justification for ε range (e.g., based on maximum pairwise distance distribution in training data); (3) that all 57 features were used; (4) GUDHI version and relevant function calls (e.g., gudhi.RipsComplex).

Minor Revisions

Abstract

• Line 110: “Background” section – Good. Change “displacement, crowding, disruption of health services, insecurity, and inadequate water and sanitation” – consider adding “and” before “inadequate” for readability.

• Line 113: “Methods” – change “surge events” to “surge events (binary indicators per LGA-week)” for clarity.

• Line 116: “Results” – ROC-AUC values are reported with 95% CI. Add that these are from temporal hold-out validation (not cross-validation) to avoid confusion. Currently it says “temporally ordered validation” – specify hold-out or cross-validation.

• Line 117: “At the optimal decision threshold” – specify that this threshold was determined using Youden’s index on validation data.

Introduction

• Line 63: “Infection disease outbreaks” – should be “Infectious disease outbreaks.” Correct.

• Line 88-93: The paragraph on threshold-based approaches is good but could cite a recent reference on reactive surveillance limitations (e.g., WHO 2022 guidance). Optional.

• Line 106-109: The TDA description is clear. However, the sentence “Rather than focusing only on individual variables or pairwise associations, TDA characterizes the shape of data in high-dimensional space” – this is conceptually correct but the manuscript does not later demonstrate what shape features were informative beyond “higher β0 and β1 in high-risk observations.” Consider adding a sentence in Results visualizing a persistence diagram example for a high-risk vs. low-risk LGA-week.

Methods

• Line 145–152: Study setting – Good. Add the total number of operational aquaculture farms in the district (if known) to justify the sample. (This is from the previous manuscript – check if this line is still relevant. Actually this appears to be a cut-and-paste error from the aquaculture manuscript? The text here is about Nigeria LGAs, not aquaculture. But the line “Of the 774 LGAs nationwide, 97 with complete data across all domains were included” is correct. However, the sentence “Total number of operational aquaculture farms” does not appear – ignore. But check for any stray aquaculture references.)

• Line 172: “Weekly LGA-level data (n = 25,284 observations)” – confirm calculation: 97 LGAs × 52 weeks × 6 years = 30,264. Subtract missing weeks? 25,284 suggests ~16% missing data. This is fine but should be stated.

• Line 194-195: Class imbalance handling – “stratified sampling and class weighting” – specify the class weight used (e.g., “class_weight='balanced' in XGBoost”).

• Line 217-220: Missing data – MICE with 10 imputations, outcome excluded – excellent. Add that the imputation model included all predictors but not the outcome, and that imputation was performed separately per fold (already stated). Good.

• Line 234-248: Validation framework – forward-chaining with nested CV – excellent. Add that hyperparameter tuning used only the training fold (already implied but make explicit). Also specify the inner CV folds (e.g., 3-fold within each training year?).

Results

• Table 2 (page 54): Descriptive statistics – Good. Add units for “IDP population” (persons) – already there. Check “Health facility density” – units are “facilities per 100,000 population” – state in table footnote.

• Table 3 (page 55): Predictive performance – The “Sensitivity*” footnote says “At optimal threshold (Youden’s index).” Specify that the optimal threshold was determined on validation data (not test data) to avoid overfitting.

• Figure 1 (page 56): The figure is referenced but not embedded in the provided PDF (only a text placeholder). Ensure the final submission includes the actual figure.

• Table 6 (page 57): Ablation – Good. Add that the p-values are from DeLong’s test comparing ROC-AUCs.

• Table 7 (page 58): Lead-time sensitivity – The false alert rate decreases with longer lead time (4.1 → 3.2 → 2.4). This makes sense because fewer alerts overall. Add a note that these are annualized rates per LGA.

• Table 8 (page 58): Subgroup analysis – “High-conflict (>10 events/week)” – is this mean events across the study period? Or weekly threshold? Clarify.

Discussion

• Line 523–543: Comparison with prior work – Good. However, the statement “The improvement attributable to topological features (0.08 AUC points) is meaningful but context-dependent for rare-event prediction” – add that clinical/operational significance depends on the cost of false alerts vs. missed outbreaks, which is not assessed here (but discussed later).

• Line 537-545: Operational implications – Excellent. The false alert calculation (310 annually across 97 LGAs) is helpful. Add that this is 3.2 per LGA per year × 97 = 310. Already clear.

• Line 549-564: Limitations – Good. Add one more: “The prediction horizon (4 weeks) was chosen post-hoc; performance at other horizons is reported but the primary horizon was not pre-specified, which increases the risk of overfitting to this horizon.”

Conclusion

• Line 575-578: “Topological feature representations provide a modest but meaningful complementary approach” – Appropriate. Add: “However, routine deployment requires prospective validation and context-specific threshold tuning.”

Appendices

• Appendix A (Data Dictionary, page 66-67): Table 9 – Good. For “Weeks since last outbreak” – specify that this is computed separately for cholera and measles (i.e., time since last cholera outbreak for cholera model, etc.). Currently ambiguous.

• Appendix B (Hyperparameters, page 68): Table 10 – Good. Add that these were optimized via Bayesian optimization with 50 iterations, and that the same search space was used for both XGBoost-Raw and XGBoost-TDA.

• Appendix C (Additional Performance Metrics, page 68):

o The calibration plots (Figure 3) are referenced but not embedded. Ensure they are included.

o The GitHub repository URL is given as “https://github.com/[repository]/tda-outbreak-prediction” – the [repository] placeholder must be replaced with the actual repository name.

o The statement “Processed datasets are available upon reasonable request to the corresponding author” – as noted above, this is not compliant with PLOS ONE policy. Revise.

Language and Presentation

• Typography: Throughout, “XGBoost” and “XGBoost-Raw” – consistent.

• Line 60: “Infection disease” – change to “Infectious disease.”

• Line 172: “n = 25,284 observations” – ensure the formatting (n is italicized).

• Page 69–90 (Response to Reviewers): This is helpful but unusually long. Consider moving detailed point-by-point responses to a separate “Response to Reviewers” document (as is standard) and keeping only a summary in the manuscript cover letter. The current manuscript file includes 90 pages, of which ~20 pages are the response. This is acceptable but unusual for PLOS ONE. The editors may ask to separate the response.

Statistical and Reporting Issues

Multiple Testing

The manuscript reports many comparisons (multiple models, multiple outcomes, multiple subgroups, multiple lead times, multiple robustness checks). No adjustment for multiple testing is mentioned aside from the bootstrap confidence intervals. The p-values in Table 6 (ablation) and Table 8 (subgroup) are likely uncorrected.

Required action:

Add a statement in Methods (Statistical Analysis subsection) that “No adjustment for multiple comparisons was applied because the analyses are exploratory and hypothesis-generating. P-values should be interpreted descriptively.” OR apply Benjamini-Hochberg FDR and report adjusted values. The former is acceptable for exploratory work.

Sample Size Justification

No power calculation is provided. Given the rare-event outcomes (7% cholera, 5% measles), the effective sample size for positive events is 1,772 and 1,265 respectively, which is adequate for stable AUC estimation. However, this should be stated.

Required action:

Add to Methods: “With 1,772 cholera surge events and 25,284 total observations, the study is adequately powered to detect moderate AUC differences (expected AUC ~0.75, target AUC ~0.80 with α=0.05, power=0.80).”

Ethical Statement

The ethics statement (page 11, lines 1-5) states that “This study exclusively used publicly available, aggregate-level data sources… Therefore, ethics approval and informed consent were not required.” This is acceptable for PLOS ONE. However, the statement should be moved to the Methods section (not a separate page before the references). Currently it appears after the abstract and before the introduction. Move to Methods.

Decision Rationale

This revised manuscript represents a substantial improvement over the original submission. The reframing to cholera/measles prediction, temporally-aware validation, outcome specification, calibration assessment, and ablation studies have addressed the major methodological concerns raised by the previous reviewers.

However, the manuscript cannot be accepted in its current form due to:

1. Non-compliance with PLOS ONE data availability policy (“available upon request” is insufficient; code repository must be provided at submission).

2. Missing details on TDA computation (feature scaling, ε range justification, software functions).

3. Over-interpretation of topological features as “capturing disconnected risk factors” without validation.

4. Minor technical errors (missing figures, placeholder URLs, typographical errors).

The required revisions are minor in scope and do not require new data collection or analysis. They can be completed within 2–4 weeks.

Recommendation: Minor Revision – The authors should address the data availability, TDA methodological details, and interpretive overstatements as outlined above. I look forward to seeing the final revised version.

Sincerely,

Prof. A. H. Demir, MD, PhD

Academic Editor (Infectious Disease Epidemiology & Machine Learning)

PLOS One

[Note: HTML markup is below. Please do not edit.]

To ensure your figures meet our technical requirements, please review our figure guidelines: https://journals.plos.org/plosone/s/figures

You may also use PLOS’s free figure tool, NAAS, to help you prepare publication quality figures: https://journals.plos.org/plosone/s/figures#loc-tools-for-figure-preparation.

NAAS will assess whether your figures meet our technical requirements by comparing each figure against our figure specifications.

-->

Topological Data Analysis for Predicting Disease Outbreaks in Humanitarian Settings: A Machine Learning Approach

PONE-D-26-03649R3

Dear Author,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice will be generated when your article is formally accepted. Please note, if your institution has a publishing partnership with PLOS and your article meets the relevant criteria, all or part of your publication costs will be covered. Please make sure your user information is up-to-date by logging into Editorial Manager at Editorial Manager® and clicking the ‘Update My Information' link at the top of the page. For questions related to billing, please contact billing support.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Morufu Olalekan Raimi, Ph.D

Academic Editor

PLOS One

Additional Editor Comments (optional):

PLOS ONE Editorial Decision

Manuscript ID: PONE-D-26-03649R3

Title: Topological Data Analysis for Predicting Disease Outbreaks in Humanitarian Settings: A Machine Learning Approach

Authors: Opue et al.

Editor: Dr. Morufu Olalekan Raimi

Date: 15 May 2026

ACCEPT

The manuscript is ready for publication in its current form. Production staff should verify the Zenodo DOI and figure embedding during copyediting, but no substantive changes are needed from the authors.

Dr. Morufu Olalekan Raimi

Academic Editor, PLOS ONE

Reviewers' comments: