Paper Info Reviews Author Feedback Meta-Review Back to top
List of Papers Browse by Subject Areas Author List

Abstract

Accurate phenotyping of hospital discharge diagnoses is clinically consequential because it directly informs secondary prevention, medication selection, follow up intensity, and referral pathways after hospitalization. In particular, underrecognition of acute high risk phenotypes can delay time sensitive therapy and risk mitigation, leading to avoidable adverse events and inconsistent postdischarge management. However, real world clinical records are heterogeneous and often incomplete, which can bias learning toward prevalent phenotypes and obscure clinically salient signals. We propose CW-B, a class weighted XGBoost framework that embeds clinically motivated misclassification costs into the learning objective, enabling an explicit and controllable balance between overall discrimination and suppression of high risk false negatives. CW-B further incorporates robust preprocessing with explicit missingness handling to improve reliability under incomplete documentation while preserving interpretable and auditable decision logic. Overall, CW-B provides a practical and deployment oriented approach for discharge phenotyping that supports more consistent downstream clinical decision making.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/1403_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

N/A

Link to the Dataset(s)

N/A

BibTex

@InProceedings{LiSij_CWB_MICCAI2026,
        author = { Li, Sijia AND Tan, Xiaoyu AND Zhan, Chen AND Ma, Yuanji AND Wang, Haoyu AND Qiu, Xihe},
        title = { { CW-B: Class Weighted Boosting Framework for Imbalance Resilient Multi Class Cardiac Phenotyping } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 16886},
        month = {September},
        page = {pending}
}

Reviews

Review #1

  • Please describe the contribution of the paper

    The paper addresses multi-class discharge phenotyping for cardiac conditions using structured clinical data under class imbalance. The authors propose CW-B, a class-weighted XGBoost framework with explicit missingness-aware preprocessing, aiming to improve recognition of clinically prioritized classes such as stableCAD, ACS, and CAS. The study benchmarks CW-B against several tree-based, ensemble, and neural baselines under a unified 5-fold stratified cross-validation protocol and reports both global metrics and prioritized class-specific metrics.

  • Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.

    The paper studies a clinically meaningful problem, namely discharge phenotyping for cardiac conditions where missed detection of certain classes may have downstream implications for management. The task definition and label mapping are clearly presented, and the emphasis on prioritized classes is clinically well motivated.

    Another strength is that the preprocessing pipeline is described in a leakage-aware and reproducible manner. The manuscript explicitly states that preprocessing statistics are estimated on the training split only, and it augments imputed features with missingness indicators, which is a reasonable design choice for structured clinical data with incomplete documentation.

  • Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.

    The main weakness is limited methodological novelty. CW-B appears to be an application of inverse-frequency class weighting on top of XGBoost with standard missingness-indicator augmentation, rather than a substantially new boosting framework. The manuscript presents the method as a novel framework, but the technical contribution seems closer to a well-engineered cost-aware training pipeline than to a new methodological advance. The motivation of handling class imbalance is clinically relevant. However, imbalance-aware learning itself is not new in medical AI, and prior work has already explored data-level correction strategies for imbalanced biomedical classification tasks, e. g. , GAN-based imbalance correction (DiGAN Breakthrough: Advancing diabetic data analysis with innovative GAN-based imbalance correction techniques). In this regard, the present manuscript is better positioned as a task-specific implementation of cost-sensitive tree boosting rather than a substantially new methodological advance. The empirical validation is also not deep enough to isolate the value of each component. There is no ablation showing the effect of class weighting alone, missingness indicators alone, or alternative weighting strategies. This makes it difficult to determine which part of the pipeline is responsible for the observed gains.

  • Please rate the clarity and organization of this paper

    Satisfactory

  • Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.

    The submission does not mention open access to source code or data, but provides a clear and detailed description of the algorithm to ensure reproducibility.

  • Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?

    N/A

  • Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html

    N/A

  • Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.

    (4) Weak Accept — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    I find the paper clinically relevant and generally well written, and the evaluation protocol is reasonably organized. However, the technical novelty is limited, and the empirical evidence does not yet fully support the strength of the paper’s methodological and deployment-oriented claims. The proposed method appears to be a careful integration of known techniques rather than a substantial methodological innovation.

  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.

    N/A

  • [Post rebuttal] Please justify your final decision from above.

    N/A



Review #2

  • Please describe the contribution of the paper

    Propose a framework for clinical phenotyping: (1) Integrate class‑weighted optimization into XGBoost to enable a controllable cost trade‑off under asymmetric risk, thereby improving classification reliability in imbalanced‑label settings. (2) Explicitly handle missing values and enforce consistent preprocessing constraints, allowing the model to adapt to incomplete documentation and high measurement variability commonly encountered in practice. (3) Perform comparisons within a unified baseline suite, with a focus on missed detections of clinically important phenotypes, and provide an interpretable evaluation perspective to deliver deployment‑oriented evidence.

  • Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.

    (1) Instead of reporting only overall metrics, the paper performs a detailed analysis of missed detections for clinically important phenotypes (stableCAD, ACS, CAS), providing actionable insights into which classes are most often confused. (2) The framework explicitly handles incomplete documentation and high measurement variability through robust preprocessing (median imputation plus missing indicators) and leverages XGBoost’s native missing‑value handling, making it suitable for noisy, real‑hospital data rather than idealized benchmarks. (3) The ground truth is strengthened by incorporating the Gensini score for disease burden and longitudinal follow‑up for CAD‑related events, improving the reliability of discharge phenotype labels beyond routine administrative codes.

  • Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.

    (1) The core technical components class‑weighted XGBoost and missing indicator with median imputation are well‑established practices, offering limited methodological novelty: Assigning instance weights inversely proportional to class frequencies is a native feature of XGBoost. This practice is extensively documented in the official XGBoost documentation and widely used in multi‑class imbalance scenarios. The missing indicator method (MIM) – imputing missing values with a simple statistic (e.g., median) while adding a binary indicator for missingness – is a classical missing data handling strategy. (2) Ablation study for the missingness handling strategy is completely missing, leaving its claimed benefit unsubstantiated: The paper states that explicit missingness modeling (median imputation + missing indicator concatenation) is a key component of the robust training pipeline. However, no ablation experiments are conducted to validate this claim. Specifically, the paper does not compare:

    • Median imputation alone (without missing indicators);
    • Missing indicators alone (with zero or constant imputation);
    • Alternative imputation methods (mean imputation, k‑NN, multiple imputation, or MissForest);
    • No explicit missingness handling (letting XGBoost natively handle missing values by learning the best split direction). (3) The clinical justification for the prioritized class set Cp = {0, 1, 3} (stableCAD, ACS, CAS) is insufficiently argued. The paper defines Cp = {0, 1, 3} as clinically prioritized phenotypes because they are “associated with established diagnostic and therapeutic pathways”. However, the rationale is weakly supported.
  • Please rate the clarity and organization of this paper

    Satisfactory

  • Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.

    The submission does not mention open access to source code or data, but provides a clear and detailed description of the algorithm to ensure reproducibility.

  • Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?

    N/A

  • Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html

    N/A

  • Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.

    (2) Reject — should be rejected, independent of rebuttal

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    Based on the paper’s content and the identified weaknesses, my overall recommendation is Weak Reject. The major factors leading to this decision are as follows. (1) The proposed CW‑B framework combines two well‑established techniques: class‑weighted XGBoost and missing indicator with median imputation. The paper does not introduce a new algorithm, a novel theoretical insight, or a non‑trivial adaptation of these techniques. (2) The experimental evaluation lacks rigorous validation: no ablation study isolates the contribution of the missingness handling component, the dataset is small (N=4354, 57 features) and imbalanced, no statistical significance testing is provided (performance differences are small and confidence intervals likely overlap), and there is no external validation beyond a single‑center, single‑disease cohort. These gaps severely undermine confidence in the claimed practical deployment. (3) The paper repeatedly emphasizes that CW‑B preserves interpretability and auditability via tree‑based decision logic. However, no interpretability analysis is provided (e.g. feature importance ranking or visualization of individual decision trees or decision paths). The claim remains theoretical and unsubstantiated by any experiment or example, which is insufficient for a clinically‑oriented paper.

  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.

    Reject

  • [Post rebuttal] Please justify your final decision from above.

    I appreciate the authors’ rebuttal. However, my main concerns remain. CW-B still appears to combine established components, including class-weighted XGBoost, missingness indicators, and standard leakage-aware preprocessing, rather than introducing a substantially new methodological contribution. The empirical advantage also appears mixed. Therefore, despite the clinical motivation, I maintain a Reject recommendation.



Review #3

  • Please describe the contribution of the paper

    This paper introduces CW-B (Class Weighted Boosting), a framework designed to address class imbalance in multi-class cardiac phenotyping using hospital discharge diagnoses. The main contributions are: 1.A class-weighted XGBoost framework that integrates clinically motivated misclassification costs into the learning objective to handle asymmetric risks (specifically reducing high-risk false negatives). 2.A robust preprocessing pipeline that explicitly handles missingness and data heterogeneity, ensuring reliability in real-world clinical record settings. 3.A systematic evaluation demonstrating the framework’s ability to maintain interpretable decision logic while improving the detection of clinically salient signals in skewed datasets.

  • Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.

    This work addresses a clinically important and practically relevant challenge in medical informatics, namely the underrecognition of acute high-risk phenotypes in the presence of class imbalance and incomplete data. The emphasis on minimizing false negatives for high-risk classes is particularly well aligned with clinical priorities, where missed detections can have significant consequences for patient safety and downstream care.

    A key strength of the paper lies in its focus on interpretability and trustworthiness. By building on XGBoost and emphasizing auditable decision logic, the proposed approach offers greater transparency compared to black-box deep learning models. This is especially valuable in clinical settings, where interpretability is essential for practitioner trust, deployment, and potential regulatory considerations.

    The paper also demonstrates a strong understanding of real-world data challenges. The explicit handling of missingness, including the use of missingness indicators and structured preprocessing, is well motivated and appropriate for electronic health record (EHR) data, which is often heterogeneous and incomplete. This contributes to the robustness and practical applicability of the proposed framework.

    Finally, the methodology is clearly presented. The inclusion of a detailed algorithm outlining the integration of class weighting and missingness handling into the boosting framework improves clarity and supports reproducibility. The step-by-step formulation makes it easier for readers to understand and potentially reimplement the approach.

  • Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.

    Incremental Methodological Novelty: The core technique of applying class weights to gradient boosting (XGBoost) is a standard feature in existing libraries (e.g., scale_pos_weight or sample weights). While the authors frame this as a “framework” (CW-B), the methodological innovation appears somewhat incremental. The paper needs to clearly distinguish its specific weighting strategy from standard cost-sensitive learning implementations available in off-the-shelf tools. Comparison to State-of-the-Art: The abstract mentions a “unified baseline suite,” but it is not validated by the results, it may have been a small increase of the results, not any major improvement, other complex models will outperform it. Scope of Evaluation: The focus is on “Cardiac Phenotyping,” but the abstract does not specify the granularity of the classes or the exact data source (e.g., specific cohort size or feature dimensions). The generalizability of the specific weighting scheme to other phenotyping tasks remains to be seen.

  • Please rate the clarity and organization of this paper

    Satisfactory

  • Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.

    The submission does not mention open access to source code or data, but provides a clear and detailed description of the algorithm to ensure reproducibility.

  • Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?

    N/A

  • Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html

    N/A

  • Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.

    (4) Weak Accept — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    Overall, the paper presents a clinically valuable and practically relevant framework for addressing class imbalance and missing data in cardiac phenotyping. Its strengths lie in its strong clinical motivation, focus on minimizing high-risk false negatives, and robustness to real-world data challenges. The emphasis on interpretability and auditable decision logic further enhances its suitability for clinical translation, aligning well with the priorities of the MICCAI community.

    However, the methodological contribution appears somewhat limited. The use of class weighting within XGBoost is well established, and the primary novelty of the work seems to lie more in the specific application context and pipeline design rather than in a fundamentally new algorithmic development. While the framework is thoughtfully constructed and systematically evaluated, it does not clearly demonstrate a significant methodological advancement over existing approaches.

    Based on these considerations, I assign a Weak Accept (4). The paper’s clinical relevance and practical contributions are meaningful, but the limited methodological novelty prevents a stronger recommendation. I would be open to increasing the score if the rebuttal provides clearer evidence of novel weighting strategies or demonstrates substantial improvements over strong baselines, including modern deep learning approaches.

  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.

    Accept

  • [Post rebuttal] Please justify your final decision from above.

    1.The rebuttal clarifies the clinical motivation and better positions CW-B as a clinically risk-aligned framework rather than a generic imbalance-handling approach.

    2.The additional ablation results strengthen the empirical support by demonstrating the contribution of both class weighting and missingness-aware modeling.

    3.The methodological novelty remains somewhat incremental, as the core framework is still based on established cost-sensitive XGBoost techniques.

    4.The paper’s primary strength lies in its practical clinical applicability, interpretability, and robustness to real-world EHR missingness, making it relevant for translational medical AI applications.



Author Feedback

We thank the meta-reviewer and all three reviewers for their careful evaluation. We appreciate their recognition of the clinical relevance of reliable cardiovascular discharge phenotyping under real-world class imbalance and missingness, especially for reducing missed detections of stableCAD, ACS, and CAS. CW-B is a clinical risk-aligned phenotyping pipeline that integrates clinical risk prioritization, class-balanced learning, missingness-aware modeling, and classwise error auditing within a unified five-class task. Under a fold-specific, leakage-free protocol, it performs missing-sentinel conversion, training-split-only preprocessing, missingness-indicator augmentation, and class-balanced instance weighting, and evaluates clinically important errors using prioritized recall, F1, miss rate, and confusion-pattern analysis. R1-W1, R2-W1, R3-W1: Novelty and method positioning The novelty of CW-B lies in its risk-aligned formulation for cardiac discharge phenotyping. In this task, different errors have asymmetric clinical consequences, so overall accuracy or generic imbalance handling alone is insufficient. CW-B targets high-risk missed detection by combining fold-specific class-balanced weighting, explicit missingness modeling, and prioritized evaluation of stableCAD, ACS, and CAS. Under the same 5-fold stratified cross-validation protocol, CW-B achieves the strongest or most competitive Accuracy, Macro-F1, Balanced Accuracy, and Prioritized F1 among strong tree-based, ensemble, and neural baselines. Its improvement over a higher-capacity XGBoost baseline further indicates that the gain is not simply due to model capacity. We will clarify this positioning in the revision. R1-W2, R2-W2: Empirical evidence, component contributions, and missingness handling Following the meta-review and the rebuttal policy, we provide preliminary additional ablation results to clarify the contribution of each component. These analyses were conducted under the same 5-fold stratified cross-validation protocol, random seed, and main model configuration. Removing class weighting reduces Macro-F1, Balanced Accuracy, and Prioritized F1 to 0.6968/0.6621/0.6631.Removing the missingness indicator gives 0.6804/0.6666/0.6444.Using only XGBoost native missing handling gives 0.7082/0.6909/0.6780.Alternative imputation strategies with indicators also underperform the main model; for example, KNN yields 0.7008/0.6791/0.6703.These preliminary results support the role of both core components: class-balanced instance weighting improves macro-level and prioritized-class performance, while explicit missingness modeling benefits clinically prioritized phenotypes. Among 4354 samples and 57 features, 55 features contain missing values, with 38571 missing entries out of 248178 total entries, corresponding to a 15.54% missingness rate, suggesting that documentation patterns carry clinically relevant information. In the revision, we will include the finalized ablation table and clarify that all preprocessing statistics are estimated only from the training split. R2-W3, R3-W2: Clinical rationale for $C_p={0,1,3}$ $C_p={0,1,3}$ is a pre-defined clinical-priority set based on clinical actionability and missed-detection risk. ACS is a time-sensitive acute ischemic phenotype; stableCAD informs secondary prevention, medication planning, and follow-up; and CAS in this cohort refers to clinically suspected CAD patients with less than 50% luminal narrowing who still require risk assessment and symptom management. $C_p$ provides a clinical-risk-oriented evaluation perspective without changing the training labels, model objective, or post-processing rules; the model always performs unified five-class prediction. In the revision, we will state earlier that $C_p$ is determined by clinical actionability and missed-detection risk, not by class frequency or experimental results. We thank the reviewers again and will revise the manuscript accordingly.



Meta-Review

Meta-review #1

  • Your recommendation

    Provisional Accept

  • Please justify your decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. In case of an invitation for rebuttal, clarify which points are important to address in the rebuttal.

    The referees are generally positive and all recognize the important clinical problem. Please in your rebuttal discuss the key contribution (novelty) and the empirical evidence irl the papers claims (Reviewer #1),

  • After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.

    Reject

  • Please justify your recommendation.

    The authors have written a good rebuttal and address the key points. However, concerns around novelty remain and reviewers judge this as incremental.



Meta-review #2

  • After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.

    Accept

  • Please justify your recommendation.

    This paper was initially invited for rebuttal following the primary review period, where reviewers raised several concerns regarding novelty and empirical evidence. After carefully evaluating the authors’ rebuttal, the reviewers’ revised feedback, and re-examining the paper, I conclude that the authors have effectively addressed the key concerns raised, and the work now meets MICCAI’s publication standards.



Meta-review #3

  • After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.

    Accept

  • Please justify your recommendation.

    The reviewers’ concerns have been successfully addressed in the rebuttal, making the work a solid contribution especially for what concerns methodological applications in MIC.



back to top