AI-BASED Tool to Estimate Sodium Intake in STAGE 3 to 5 CKD Patients—The UniverSel Study

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This is a clinically relevant, multicenter study addressing a practical need: estimating daily sodium intake in CKD (stages 3-5) using routinely available variables. The probabilistic Tree-Augmented Naive Bayes (TAN) approach is appropriate for handling conditional dependencies. However, some aspects need improvement to enhance the scientific clarity and impact of the manuscript:

1. It is suggested that the words in the keywords should be distinguished from the title content, so as to facilitate the retrieval of future researchers.
2. Authors verify completeness via creatininuria and interviews. Please report how many collections failed checks, how many were repeated, and add a sensitivity analysis excluding borderline samples to gauge reference-standard noise.
3. The “test-with-cases” procedure compares predictions to the training data structure. This is essentially resubstitution and over-optimistic. Please add proper internal validation (nested k-fold or bootstrapping), report 95% CIs, class-wise metrics (recall/precision, macro-F1, κ), and calibration (Brier score/plots).
4. Discretization of predictors may harm signals. Many continuous predictors (e.g., age, BP) are binned. Moreover, “all variables were formatted to be categorical” for model development. Please justify this choice, compare against models that retain continuity (or use supervised binning), and assess performance impact.
5. The authors cited relevant literature but could strengthen their discussion by more explicitly comparing their results to previous studies.
6. Conclusion could be better with more specific key findings of the results and needs to be rewritten to include the advantages and limitations of the current work.
7. More attention should be paid to word spelling, space characters, singular and plurals, punctuations, the upper- and lower-case letters, subscripts, special characters, etc.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This is an important and well-executed study; using machine learning to estimate sodium intake in stage 3–5 CKD is both clinically relevant and practically useful. To strengthen the paper, please consider the following:
1. line 196, please specify your predictor selection criteria.
2. Figure 1, the second and third boxes appear to contain inconsistent numbers—please reconcile the counts.
3. lines 348–350, expand the interpretation that the three strongest predictors—weight, height, sex—are objective and unlikely to be misreported, whereas dietary surveys often under/overestimate intake; the finding that weight is most strongly associated suggests caloric intake (if accurately captured) likely tracks sodium intake, consistent with portion size being retained in the model.
4. please include discussion of potential strategies to improve accuracy beyond 70%.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have responded and justified all comments. The paper can be accepted.

Author Response

Veuillez consulter la pièce jointe