Dynamic Biochemical Phenotypes in Hospitalized Patients with Pulmonary Tuberculosis

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This retrospective observational study characterizes the early dynamic changes in biochemical and hematological profiles of 187 hospitalized patients with pulmonary tuberculosis during the first 2–10 days of antituberculous treatment. By employing advanced statistical methods such as UMAP and hierarchical clustering, the authors identified three distinct clinical-biochemical phenotypes (Mild, Inflammatory, and Severe) and tracked patient migration between these states after treatment initiation. The results indicate that while inflammatory markers like CRP generally decreased, a significant proportion of patients migrated toward a more compromised phenotype characterized by worsening anemia and renal function. Overall, this work underscores the potential of using early dynamic biochemical phenotyping as a complementary tool for risk stratification and individualized clinical surveillance in hospitalized TB patients. The manuscript is recommended to be accepted by the journal SCI. Here are some concerns for the authors to improve the quality of the manuscript:

Minor concerns:

1. In line 287, "Regarding to hemoglobin" should be corrected as "Regarding hemoglobin".

2. In line 114, In the statistical analysis section, the text reads. There is an unnecessary space after the period (for “Ward. D2”). It should be corrected to "Ward.D2".

3. For Figure 2 in the PDF version of the manuscript, the picture reads too narrow.

Major concerns:

1. The study identifies that patients in "Cluster 3" (Severe) had a higher prevalence of HIV and active cancer (Lines 159–160). However, the authors explicitly state in the limitations (Lines 327–328) that the study was not adjusted for these clinical confounders. This is a significant limitation because the "dynamic phenotype" observed might simply be a reflection of the underlying comorbidity rather than a specific response to TB treatment.

2. Variability in the "After" Time Window The post-treatment sampling window is defined as between 2 and 10 days. In the context of acute pharmacokinetics and inflammatory response, Day 2 is vastly different from Day 10. Lumping these time points together creates significant noise in the data. For example, a creatinine spike at Day 2 might be dehydration, whereas at Day 10 it could be drug-induced nephrotoxicity. If the sample size allows, the authors should stratify the "After" group (e.g., Day 2–4 vs. Day 5–10) or include "days since treatment start" as a covariate in the model to ensure the observed phenotypes are consistent across this time range.

 

Author Response

Comments 1-3:

 

In line 287, "Regarding to hemoglobin" should be corrected as "Regarding hemoglobin".

 

In line 114, In the statistical analysis section, the text reads. There is an unnecessary space after the period (for “Ward. D2”). It should be corrected to "Ward.D2".

 

For Figure 2 in the PDF version of the manuscript, the picture reads too narrow.

 

Author´s response: We thank the reviewer for these helpful technical observations. In response, we have corrected the wording in line 287 from “Regarding to hemoglobin” to “Regarding hemoglobin,” fixed the formatting in the Statistical Analysis section by removing the unnecessary space in “Ward. D2” to read “Ward.D2,” and adjusted the layout of Figure 2 in the PDF version to improve readability by increasing its width.

 

Comment 4:

The study identifies that patients in "Cluster 3" (Severe) had a higher prevalence of HIV and active cancer (Lines 159–160). However, the authors explicitly state in the limitations (Lines 327–328) that the study was not adjusted for these clinical confounders. This is a significant limitation because the "dynamic phenotype" observed might simply be a reflection of the underlying comorbidity rather than a specific response to TB treatment.

 

Author´s response: We appreciate this reviewer´s comment for highlighting the potential confounding role of HIV infection and active malignancy in the characterization of the severe biochemical phenotype. We fully agree that these comorbidities are biologically relevant and can influence both baseline laboratory values and early biochemical evolution; for this reason, their distribution across clusters is explicitly reported and acknowledged in the limitations. However, the primary aim of this study was not to isolate a treatment-specific biochemical response independent of comorbidity, but rather to describe early, real-world biochemical phenotypes observed in hospitalized patients with pulmonary tuberculosis at treatment initiation. In this clinical context, conditions such as HIV and cancer are not external confounders but integral components of disease severity and host vulnerability frequently encountered in tertiary-care settings. The clustering approach was intentionally unsupervised and data-driven, allowing these vulnerability-associated profiles to emerge naturally from routine laboratory parameters without imposing causal assumptions.

While we recognize that the absence of multivariable adjustment limits causal inference regarding the independent contribution of treatment versus comorbidity, adjusting for these

factors would have altered the exploratory nature of the analysis and was further constrained by the small size of the most severe cluster. Accordingly, the identified phenotypes should be interpreted as descriptive clinical-biochemical patterns reflecting overall patient vulnerability rather than as isolated effects of antituberculous therapy, a point that is emphasized in the Discussion and limitations.

 

Comment 5:

Variability in the "After" Time Window The post-treatment sampling window is defined as between 2 and 10 days. In the context of acute pharmacokinetics and inflammatory response, Day 2 is vastly different from Day 10. Lumping these time points together creates significant noise in the data. For example, a creatinine spike at Day 2 might be dehydration, whereas at Day 10 it could be drug-induced nephrotoxicity. If the sample size allows, the authors should stratify the "After" group (e.g., Day 2–4 vs. Day 5–10) or include "days since treatment start" as a covariate in the model to ensure the observed phenotypes are consistent across this time range.

 

Author´s response: We appreciate the reviewer comment regarding the heterogeneity of the post-treatment sampling window and agree that biochemical measurements obtained at day 2 may differ biologically from those obtained at day 10. The 2–10 day interval was selected to reflect routine clinical practice in a retrospective hospital-based cohort, where laboratory reassessments are performed according to clinical indication and patient evolution rather than at standardized time points. Narrowing or stratifying this window would have resulted in a substantial reduction in sample size and compromised the stability of the clustering analysis, particularly given the modest number of patients with complete data at specific post-treatment days. Moreover, the objective of the study was not to model precise pharmacokinetic or inflammatory trajectories, but to capture early biochemical instability occurring shortly after treatment initiation, a clinically relevant period during which anemia, inflammatory modulation, and renal changes commonly coexist and inform monitoring decisions. While incorporating “days since treatment start” as a covariate or performing temporal stratification could provide finer resolution, such approaches are better suited to prospective designs with protocolized sampling. We therefore acknowledge this temporal heterogeneity as a limitation and highlight it as an important consideration for future studies, while maintaining that the chosen window appropriately captures early biochemical dynamics in real-world hospitalized TB patients.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The paper is relevant and addresses an important topic. Below are my comments:

1) To strengthen the methodology and ensure reproducibility, could the authors please provide the exact UMAP, FAMD and clustering parameters? Could the authors stratify or adjust analyses based on follow-up timing?

2) If possible please provide a flow diagram of patient inclusion.

3) If possible could authors carry out multivariable logistics or multinomial regression to identify independent predictors of cluster migration, and sensitivity analyses using different thresholds for biochemical deterioration?

4) I think that the statements about unfavorable course should be reframed to biochemical worsening unless the clinical outcomes are provided. Authors should avoid language that implies causality.

5) Could the authors include population specific factors such as nutrition, comorbities and others in the discussion

6) In the limitation sections, discuss generalizability because this is a single centre study in Colombia and this limits applicability.

Author Response

Reviewer 2

 

Comment 6:

To strengthen the methodology and ensure reproducibility, could the authors please provide the exact UMAP, FAMD and clustering parameters? Could the authors stratify or adjust analyses based on follow-up timing?

 

Author´s response: We thank the reviewer for this important suggestion aimed at improving methodological transparency and reproducibility. In response, we have expanded the Statistical Analysis section to explicitly report all parameters used in the dimensionality reduction and clustering procedures. We now specify that Factorial Analysis of Mixed Data (FAMD) was performed using the FactoMineR package, retaining three principal dimensions (ncp = 3), and that ascending hierarchical clustering was conducted on the Euclidean distance matrix of the FAMD individual coordinates using Ward’s minimum variance method (Ward.D2), with a three-cluster solution selected based on dendrogram inspection and interpretability. For visualization, Uniform Manifold Approximation and Projection (UMAP) was applied to standardized laboratory variables after robust numeric parsing, median imputation of missing values, and removal of zero-variance variables, using Euclidean distance (metric = "euclidean"), a minimum distance of

0.20 (min_dist = 0.20), n_neighbors = min(15, n − 1), and a fixed random seed (set.seed = 123), with identical preprocessing and parameters applied to both the baseline (“Before”) and

post-treatment (“After”) datasets. Paired Wilcoxon signed-rank tests were used to evaluate within-patient changes between time points, defining clinically relevant biochemical deterioration a priori as a decrease in hemoglobin ≥1 g/dL, an increase in C-reactive protein ≥2 mg/L, or an increase in serum creatinine ≥0.3 mg/dL, and patient redistribution between clusters was classified as stable, improved, or worsened. Between-cluster comparisons were performed using ANOVA, the Kruskal–Wallis test, or the chi-square test as appropriate, with p < 0.05 considered statistically significant. Finally, regarding follow-up timing, biochemical reassessment occurred within a predefined short window (2–10 days after treatment initiation), and therefore follow-up time was not stratified or adjusted for in the analyses; no additional analyses according to follow-up day were conducted.

 

Comment 7:

If possible please provide a flow diagram of patient inclusion.

 

Author´s response: We thank the reviewer suggestion to improve the clarity of cohort derivation. Accordingly, a flow diagram has been added to the manuscript to visually summarize the patient selection process. The diagram details the number of medical records screened, the exclusion criteria applied (including incomplete clinical data and ineligible diagnoses), and the final cohort of hospitalized patients included in the analyses at baseline (“Before”) and after treatment initiation (“After”).

 

Comment 8:

If possible could authors carry out multivariable logistics or multinomial regression to identify independent predictors of cluster migration, and sensitivity analyses using different thresholds for biochemical deterioration?

 

Author´s response: We appreciate the reviewer thoughtful suggestion to further explore predictors of cluster migration through multivariable logistic or multinomial regression analyses and to perform sensitivity analyses using alternative thresholds for biochemical deterioration. We agree that such approaches could provide additional insight into independent associations and robustness of the findings. However, these analyses were not performed in the current study due to limitations related to data structure and study design, including the relatively small size of some clusters, particularly the most severe phenotype, which would compromise model stability and risk overfitting. In addition, the predefined thresholds for biochemical deterioration were selected a priori based on clinical relevance and established criteria, and alternative cut-offs were not explored analytically. These points have been acknowledged in the Discussion and Limitations sections, and we consider multivariable modeling and sensitivity analyses important directions for future studies with larger, multicenter cohorts and more balanced cluster distributions.

 

Comment 9:

I think that the statements about unfavorable course should be reframed to biochemical worsening unless the clinical outcomes are provided. Authors should avoid language that implies causality.

 

Author´s response: We appreciate the reviewer important observation regarding the interpretation of outcomes. In response, we have revised the manuscript to consistently reframe statements referring to an “unfavorable course” or “unfavorable evolution” as biochemical worsening or less favorable biochemical profiles, in line with the scope of the data collected. We have also reviewed the Abstract, Results, and Discussion sections to remove language that could imply clinical outcomes or causal relationships, ensuring that interpretations are limited to observed biochemical changes during early treatment. These revisions clarify that the study describes biochemical dynamics rather than clinical prognosis or causality.

 

Comment 10:

Could the authors include population specific factors such as nutrition, comorbities and others in the discussion

 

Author´s response: We appreciate the reviewer suggestion to further consider population-specific factors such as nutritional status and comorbidities. These variables, including body mass index and major comorbid conditions, were incorporated into the multivariate framework through the FAMD analysis and therefore contributed to the global phenotypic structure and clustering. However, the primary aim of the study was to characterize early dynamic biochemical and hematological phenotypes rather than to provide a detailed interpretative analysis of population-level determinants. For this reason, we focused the Discussion on biochemical trajectories directly related to treatment initiation, while recognizing that population-specific factors are embedded in the multivariate model and may be explored in greater depth in future, specifically designed analyses.

 

Comment 11:

In the limitation sections, discuss generalizability because this is a single centre study in Colombia and this limits applicability.

 

Author´s response: We thank the reviewer comment regarding the generalizability of our findings. Although the manuscript does not include a separate Limitations section, we have addressed this point by explicitly adding a statement in the Discussion acknowledging that this is a single-center study conducted in a tertiary care hospital in Colombia, which may limit the generalizability of the findings to other geographic regions, healthcare systems, or tuberculosis populations. This clarification aims to appropriately frame the scope and applicability of the results.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript has been revised according to concerns. All concerns have been issued. The present version is recommended to be accepted by the journal SCI. Congratulations to the authors.