pith. sign in

arxiv: 2605.25050 · v1 · pith:FYXOV4RWnew · submitted 2026-05-24 · 📊 stat.AP · cs.LG· q-bio.QM· stat.ML

Multimodality Stacking with Blockwise missing values and application to the PIONeeR biomarkers study for prediction of resistance to immunotherapy

Mohamed Boussena , Florence Monville , Jacques Fieschi-Meric , Frederic Vely , Pierre Milpied , Julien Mazieres , Maurice Perol , Eric Vivier
show 3 more authors
Laurent Greillier Fabrice Barlesi Sebastien Benzekry
This is my paper

Pith reviewed 2026-06-29 23:44 UTC · model grok-4.3

classification 📊 stat.AP cs.LGq-bio.QMstat.ML
keywords multimodal datablockwise missingnesssurvival analysisstacking ensembleimmunotherapy predictionbiomarker integrationlung cancerprogression-free survival
0
0 comments X

The pith

A late-fusion stacking method improves survival prediction on multimodal cancer data with entire missing biomarker blocks.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents MSB, a framework that builds separate survival models for each data modality and then combines their outputs with a cross-validated meta-learner. This design is tested on the PIONeeR cohort of 443 advanced lung-cancer patients who received immunotherapy, using 378 biomarkers drawn from eight heterogeneous sources that exhibit blockwise missingness. The method raises the C-index relative to direct application of standard survival algorithms, with the largest gains seen for weaker base learners, and it shrinks the gap between training and test performance. A sympathetic reader would care because the approach lets analysts retain all patients rather than dropping incomplete cases, while still producing usable predictions for progression-free survival.

Core claim

MSB is a late-fusion framework for survival analysis that independently models modality-specific features before aggregating predictions via a cross-validated stacking meta-learner. Applied to the PIONeeR study, it produced higher C-index values than baseline algorithms, with relative gains of 15.9 percent for linear models, 5.4 percent for random survival forests, and 2.1 percent for gradient boosting methods; it also reduced the generalization gap between train and test folds. Permutation importance showed that missing-block indicators carried negligible weight, indicating that predictions were driven by the observed biomarker values themselves.

What carries the argument

The cross-validated stacking meta-learner that aggregates modality-specific survival predictions while tolerating blockwise missingness.

If this is right

  • Linear survival models receive larger relative gains from MSB than stronger tree-based or boosting baselines.
  • Routine laboratory markers, clinical features, and PD-L1 expression remain the dominant predictors once modalities are stacked.
  • The generalization gap between training and held-out performance narrows when stacking is used.
  • All patients can be retained for analysis without requiring complete data across every modality.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same late-fusion structure could be applied to other clinical tasks that combine heterogeneous data sources with blockwise missingness.
  • External validation on separate patient cohorts would be required to establish whether the observed gains hold outside the PIONeeR study.
  • The framework might reduce reliance on imputation techniques that can themselves introduce bias.
  • Inclusion of additional modalities would be straightforward provided each modality can be modeled independently first.

Load-bearing premise

The stacking meta-learner combines predictions on the basis of the actual biomarker values rather than on the pattern of which data blocks are missing for each patient.

What would settle it

On an independent cohort, if the permutation importance of missing-block indicators rises above negligible levels or if the reported C-index gains disappear, the claim that the meta-learner learns from values rather than missingness patterns would be refuted.

Figures

Figures reproduced from arXiv: 2605.25050 by Eric Vivier, Fabrice Barlesi, Florence Monville, Frederic Vely, Jacques Fieschi-Meric, Julien Mazieres, Laurent Greillier, Maurice Perol, Mohamed Boussena, Pierre Milpied, Sebastien Benzekry.

Figure 1
Figure 1. Figure 1: Multimodality Stacking with Blockwise missing values [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Multiple sources at baseline : To visualize these patterns, we sorted the data using a binary indicator based on a 50% missingness threshold per source. This arrangement highlights the blockwise missingness structure, where specific patient cohorts are associated with the presence or absence of entire data modalities. a: General overview of missingness in the PioNeeR dataset,b: Example of blockwise missing… view at source ↗
Figure 3
Figure 3. Figure 3: Kaplan-Meier curves stratified by source-level missingness. Patients [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Summary of the results from 1 for C-index obtained sorted by their [PITH_FULL_IMAGE:figures/full_fig_p017_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Top Permutation importance on iBSS for ComponentWise boosting as [PITH_FULL_IMAGE:figures/full_fig_p019_5.png] view at source ↗
read the original abstract

Integrating multimodal datasets in clinical oncology is frequently hindered by high dimensionality and blockwise missingness, where entire data sources are unavailable for specific patient subsets. Standard survival models often struggle with these gaps, leading to biased results or patient exclusion. We introduce Multimodality Stacking with Blockwise missing values (MSB), a late-fusion framework for survival analysis that independently models modality-specific features before aggregating predictions via a cross-validated stacking meta-learner. MSB was validated on the PIONeeR study (n=443 patients, 378 biomarkers across eight heterogeneous sources) to predict progression-free survival in advanced non-small cell lung cancer patients receiving immunotherapy. MSB yielded higher predictive performance (C-index) than baseline algorithms. Improvements varied by baseline strength: linear models showed a 15.9% increase (p<0.001 for the Wilcoxon signed-rank test), random survival forests gained 5.4% (p=0.002), and gradient boosting methods improved by 2.1% (p=0.030). Beyond discrimination, MSB reduced the generalization gap (train-test difference in 5 folds cross-validation repeated 3 times: 0.055 vs 0.380 for linear models). Permutation importance analysis identified routine laboratory markers, clinical features, and PD-L1 expression as primary predictive drivers. Missing block indicators showed negligible importance, suggesting the model learned from biomarker values rather than data availability patterns. MSB provides a statistically validated framework for multimodal survival prediction with blockwise missingness. By enabling systematic biomarker evaluation without requiring complete data, MSB offers a practical tool for predictive modeling in biomedical research, pending external validation. Implementation is available at https://github.com/MohamedBoussena/MSB under Inria license.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The manuscript introduces Multimodality Stacking with Blockwise missing values (MSB), a late-fusion survival modeling framework that fits modality-specific base learners and aggregates their predictions via a cross-validated stacking meta-learner. Applied to the PIONeeR cohort (n=443 patients, 378 biomarkers from eight sources) for progression-free survival prediction under immunotherapy, MSB is reported to improve C-index over linear, random survival forest, and gradient boosting baselines (gains of 15.9%, 5.4%, and 2.1% respectively, with Wilcoxon p-values), reduce the train-test generalization gap, and identify routine labs, clinical features, and PD-L1 as top predictors while assigning negligible importance to explicit missing-block indicators.

Significance. If the reported gains reflect genuine multimodal signal aggregation rather than artifacts of missingness patterns, MSB would provide a practical, late-fusion tool for survival modeling with blockwise missing multimodal data, a common issue in oncology cohorts. The open-source implementation and explicit comparison of generalization gaps are strengths that could aid adoption. The significance is limited by incomplete demonstration that performance improvements are independent of informative missingness.

major comments (2)
  1. [Abstract] Abstract: The assertion that the model learned from biomarker values rather than data availability patterns is supported only by the negligible permutation importance of explicit missing-block indicators. This does not address whether the meta-learner indirectly exploits missingness through systematic patterns in the base-model prediction vector (e.g., default outputs, imputations, or dropped models when a modality block is absent), which remains a plausible explanation for both the C-index gains and the reduced generalization gap in the PIONeeR cohort where missingness may be clinically informative.
  2. [Methods] Methods (cross-validation and base-learner description): The manuscript provides insufficient detail on the specific base learners chosen per modality, hyperparameter tuning, the precise nested cross-validation scheme used for the stacking meta-learner, and the exact mechanism by which blockwise missing values are handled inside each base model. These omissions prevent assessment of data leakage risk and reproducibility of the reported C-index differences and associated p-values.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed review of our manuscript on the MSB framework. We have carefully addressed each major comment below. Where the comments identify areas requiring greater clarity or qualification, we agree and have revised the manuscript accordingly.

read point-by-point responses

  1. Referee: [Abstract] The assertion that the model learned from biomarker values rather than data availability patterns is supported only by the negligible permutation importance of explicit missing-block indicators. This does not address whether the meta-learner indirectly exploits missingness through systematic patterns in the base-model prediction vector (e.g., default outputs, imputations, or dropped models when a modality block is absent), which remains a plausible explanation for both the C-index gains and the reduced generalization gap in the PIONeeR cohort where missingness may be clinically informative.

    Authors: We agree that the permutation importance analysis of explicit missing-block indicators provides only indirect support and does not fully rule out the possibility that the meta-learner could exploit systematic patterns in the base-learner prediction vectors that correlate with missingness. In the revised manuscript we will qualify the relevant sentence in the abstract and add a short paragraph in the Discussion section acknowledging this limitation and noting that dedicated ablation experiments on the prediction vectors would be needed to investigate it further. The core methodological contribution of MSB as a late-fusion approach remains unchanged. revision: yes

  2. Referee: [Methods] The manuscript provides insufficient detail on the specific base learners chosen per modality, hyperparameter tuning, the precise nested cross-validation scheme used for the stacking meta-learner, and the exact mechanism by which blockwise missing values are handled inside each base model. These omissions prevent assessment of data leakage risk and reproducibility of the reported C-index differences and associated p-values.

    Authors: We acknowledge that the original Methods section lacked the level of detail required for full reproducibility and leakage assessment. In the revised manuscript we will expand the Methods section to explicitly describe: the base learner type and configuration used for each of the eight modalities, the hyperparameter search procedure and ranges, the precise nested cross-validation design (outer folds for performance estimation and inner folds for meta-learner training), and the exact rule applied when a modality block is missing for a given patient (e.g., whether the corresponding base model is omitted from the prediction vector or a default is substituted). These additions will also allow readers to evaluate potential data leakage. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper introduces MSB as a late-fusion stacking method and reports empirical C-index gains plus permutation importance results on the external PIONeeR cohort via repeated cross-validation. No equations, derivations, or self-citations reduce the performance claims to quantities defined by the authors' own fitted parameters or by construction. The central validation rests on standard statistical procedures applied to held-out data rather than any of the enumerated circular patterns (self-definitional, fitted-input-called-prediction, load-bearing self-citation, etc.). The skeptic concern about indirect missingness encoding is a potential correctness or validation gap, not a circularity reduction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on standard survival-analysis assumptions and the effectiveness of cross-validated stacking; no new entities are postulated and no free parameters beyond ordinary model hyperparameters are introduced.

axioms (1)
  • domain assumption Censoring is non-informative given the observed covariates
    Implicit in all survival models (linear, RSF, gradient boosting) used as baselines and in the MSB pipeline.

pith-pipeline@v0.9.1-grok · 5918 in / 1356 out tokens · 36399 ms · 2026-06-29T23:44:37.689153+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

40 extracted references · 27 canonical work pages · 2 internal anchors

  1. [1]

    P52 - Survival models: Proper scoring rule and stochastic optimization with competing risks

    J. Alberge et al. “P52 - Survival models: Proper scoring rule and stochastic optimization with competing risks”. In:Journal of Epidemiology and Pop- ulation Health. EPICLIN 2025, 19` eme Conf´ erence francophone d’´Epid´ emiologie Clinique et 32` emes Journ´ ees des Statisticiens des Centres de Lutte Contre le Cancer, Bordeaux, France, 14-16 mai 2025 73 (...

    work page doi:10.1016/j.jeph.2025.203083.url:https:// 2025

  2. [2]

    Permutation importance: a corrected feature im- portance measure

    Andr´ e Altmann et al. “Permutation importance: a corrected feature im- portance measure”. In:Bioinformatics26.10 (Apr. 2010), pp. 1340–1347. issn: 1367-4803.doi:10.1093/bioinformatics/btq134. eprint:https: / / academic . oup . com / bioinformatics / article - pdf / 26 / 10 / 1340 / 48851160/bioinformatics_26_10_1340.pdf.url:https://doi.org/ 10.1093/bioin...

    work page doi:10.1093/bioinformatics/btq134 2010

  3. [3]

    Anastasiia Bakhmach et al.ROOFS: RObust biOmarker Feature Selection

  4. [4]

    arXiv:2601.05151 [stat.ML].url:https://arxiv.org/abs/ 2601.05151

    work page arXiv

  5. [5]

    ISSN: 3067-2007 Pages: 2026.01.09.26343779

    Fabrice Barlesi et al.An integrative multimodal machine learning signa- ture of primary resistance to immunotherapy in advanced non-small cell lung cancer: biomarker analysis from the PIONeeR study. ISSN: 3067-2007 Pages: 2026.01.09.26343779. Jan. 11, 2026.doi:10.64898/2026.01.09. 26343779.url:https://www.medrxiv.org/content/10.64898/2026. 01.09.26343779v...

    work page doi:10.64898/2026.01.09 2007

  6. [6]

    Five-Year Outcomes From the Randomized, Phase III Trials CheckMate 017 and 057: Nivolumab Versus Docetaxel in Previ- ously Treated Non-Small-Cell Lung Cancer

    Hossein Borghaei et al. “Five-Year Outcomes From the Randomized, Phase III Trials CheckMate 017 and 057: Nivolumab Versus Docetaxel in Previ- ously Treated Non-Small-Cell Lung Cancer”. In:Journal of Clinical On- cology: Official Journal of the American Society of Clinical Oncology39.7 (Mar. 1, 2021), pp. 723–733.issn: 1527-7755.doi:10 . 1200 / JCO . 20 . 01605

    2021

  7. [7]

    Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries

    Freddie Bray et al. “Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries”. In: CA: a cancer journal for clinicians74.3 (2024), pp. 229–263.issn: 1542- 4863.doi:10.3322/caac.21834

    work page doi:10.3322/caac.21834 2022

  8. [8]

    Boosting for high-dimensional linear models

    Peter B¨ uhlmann. “Boosting for high-dimensional linear models”. In:The Annals of Statistics34.2 (2006), pp. 559–583.doi:10.1214/009053606000000092. url:https://doi.org/10.1214/009053606000000092

    work page doi:10.1214/009053606000000092 2006

  9. [9]

    API design for machine learning software: experiences from the scikit-learn project

    Lars Buitinck et al. “API design for machine learning software: experiences from the scikit-learn project”. In:ECML PKDD Workshop. 2013, pp. 108– 122

    2013

  10. [10]

    van Buuren.Flexible Imputation of Missing Data

    S. van Buuren.Flexible Imputation of Missing Data. Second Edition.Boca Raton, FL.: CRC Press, 2018. S31

    2018

  11. [11]

    Integration of clinical, pathological, radiological, and transcriptomic data improves prediction for first-line immunotherapy outcome in metastatic non-small cell lung cancer

    Nicolas Captier et al. “Integration of clinical, pathological, radiological, and transcriptomic data improves prediction for first-line immunotherapy outcome in metastatic non-small cell lung cancer”. In:Nature Communica- tions16.1 (Jan. 12, 2025), p. 614.issn: 2041-1723.doi:10.1038/s41467- 025-55847-5.url:https://www.nature.com/articles/s41467-025- 55847-5

    work page doi:10.1038/s41467- 2025

  12. [12]

    New response evaluation criteria in solid tumours: Revised RECIST guideline (version 1.1)

    E. A. Eisenhauer et al. “New response evaluation criteria in solid tumours: Revised RECIST guideline (version 1.1)”. In:European Journal of Cancer 45.2 (Jan. 1, 2009). Publisher: Elsevier, pp. 228–247.issn: 0959-8049, 1879-0852.doi:10.1016/j.ejca.2008.10.026.url:https://www. ejcancer.com/article/S0959-8049(08)00873-3/abstract

    work page doi:10.1016/j.ejca.2008.10.026.url:https://www 2009

  13. [13]

    Pembrolizumab plus Chemotherapy in Metastatic Non–Small-Cell Lung Cancer

    Leena Gandhi et al. “Pembrolizumab plus Chemotherapy in Metastatic Non–Small-Cell Lung Cancer”. In:New England Journal of Medicine378.22 (May 31, 2018). Publisher: Massachusetts Medical Society, pp. 2078–2092. issn: 0028-4793.doi:10.1056/NEJMoa1801005.url:https://www.nejm. org/doi/full/10.1056/NEJMoa1801005

    work page doi:10.1056/nejmoa1801005.url:https://www.nejm 2018

  14. [14]

    Pembrolizumab Plus Pemetrexed and Plat- inum in Nonsquamous Non-Small-Cell Lung Cancer: 5-Year Outcomes From the Phase 3 KEYNOTE-189 Study

    Marina C. Garassino et al. “Pembrolizumab Plus Pemetrexed and Plat- inum in Nonsquamous Non-Small-Cell Lung Cancer: 5-Year Outcomes From the Phase 3 KEYNOTE-189 Study”. In:Journal of Clinical Oncol- ogy: Official Journal of the American Society of Clinical Oncology41.11 (Apr. 10, 2023), pp. 1992–1998.issn: 1527-7755.doi:10.1200/JCO.22. 01989

    work page doi:10.1200/jco.22 2023

  15. [15]

    Multiple Kernel Learning Algo- rithms

    Mehmet G¨ onen and Ethem Alpaydin. “Multiple Kernel Learning Algo- rithms”. In:Journal of Machine Learning Research12.64 (2011), pp. 2211– 2268.issn: 1533-7928.url:http://jmlr.org/papers/v12/gonen11a. html

    2011

  16. [16]

    Assessment and comparison of prognostic classifica- tion schemes for survival data

    Erika Graf et al. “Assessment and comparison of prognostic classifica- tion schemes for survival data”. In:Statistics in Medicine18.17 (1999), pp. 2529–2545.issn: 1097-0258.doi:10.1002/(SICI)1097-0258(19990915/ 30 ) 18 : 17 / 18<2529 :: AID - SIM274 > 3 . 0 . CO ; 2 - 5.url:https : / / onlinelibrary . wiley . com / doi / abs / 10 . 1002 / %28SICI % 291097...

    work page doi:10.1002/(sici)1097-0258(19990915/ 1999

  17. [17]

    Multivariable prognostic models: issues in developing models, evaluating assumptions and ade- quacy, and measuring and reducing errors

    F. E. Harrell, K. L. Lee, and D. B. Mark. “Multivariable prognostic models: issues in developing models, evaluating assumptions and ade- quacy, and measuring and reducing errors”. In:Statistics in Medicine15.4 (Feb. 28, 1996), pp. 361–387.issn: 0277-6715.doi:10.1002/(SICI)1097- 0258(19960229)15:4<361::AID-SIM168>3.0.CO;2-4

    work page doi:10.1002/(sici)1097- 1996

  18. [18]

    HEALNet: Multimodal Fusion for Heterogeneous Biomedical Data

    Konstantin Hemker, Nikola Simidjievski, and Mateja Jamnik. “HEALNet: Multimodal Fusion for Heterogeneous Biomedical Data”. In:Advances in Neural Information Processing Systems37 (Dec. 16, 2024), pp. 64479– 64498.url:https://proceedings.neurips.cc/paper_files/paper/ 2024/hash/765871e77d2ca65126d3d64d31aa6908-Abstract-Conference. html. S32

    2024

  19. [19]

    On the consistency of supervised learning with missing values

    Julie Josse et al. “On the consistency of supervised learning with missing values”. In:Statistical Papers65.9 (Dec. 1, 2024), pp. 5447–5479.issn: 1613-9798.doi:10.1007/s00362- 024- 01550- 4.url:https://doi. org/10.1007/s00362-024-01550-4

    work page doi:10.1007/s00362- 2024

  20. [20]

    Multimodal survival prediction in advanced pancreatic cancer using machine learning

    J. Keyl et al. “Multimodal survival prediction in advanced pancreatic cancer using machine learning”. In:ESMO Open7.5 (Aug. 18, 2022), p. 100555.issn: 2059-7029.doi:10.1016/j.esmoop.2022.100555.url: https://pmc.ncbi.nlm.nih.gov/articles/PMC9588888/

    work page doi:10.1016/j.esmoop.2022.100555.url: 2022

  21. [21]

    Review of multimodal machine learning approaches in healthcare

    Felix Krones et al. “Review of multimodal machine learning approaches in healthcare”. In:Information Fusion114 (Feb. 1, 2025), p. 102690.issn: 1566-2535.doi:10.1016/j.inffus.2024.102690.url:https://www. sciencedirect.com/science/article/pii/S1566253524004688

    work page doi:10.1016/j.inffus.2024.102690.url:https://www 2025

  22. [22]

    Super learner

    Mark J. van der Laan, Eric C. Polley, and Alan E. Hubbard. “Super learner”. In:Statistical Applications in Genetics and Molecular Biology6 (2007), Article25.issn: 1544-6115.doi:10.2202/1544-6115.1309

    work page doi:10.2202/1544-6115.1309 2007

  23. [23]

    Imputation of missing values in multi-view data

    Wouter van Loon et al. “Imputation of missing values in multi-view data”. In:Information Fusion111 (Nov. 1, 2024), p. 102524.issn: 1566-2535.doi: 10.1016/j.inffus.2024.102524.url:https://www.sciencedirect. com/science/article/pii/S1566253524003026

    work page doi:10.1016/j.inffus.2024.102524.url:https://www.sciencedirect 2024

  24. [24]

    Evaluation of Adaptive Mixtures of Competing Experts

    Steven Nowlan and Geoffrey E Hinton. “Evaluation of Adaptive Mixtures of Competing Experts”. In:Advances in Neural Information Processing Systems. Ed. by R.P. Lippmann, J. Moody, and D. Touretzky. Vol. 3. Morgan-Kaufmann, 1990.url:https : / / proceedings . neurips . cc / paper_files/paper/1990/file/432aca3a1e345e339f35a30c8f65edce- Paper.pdf

    1990

  25. [25]

    Benchmarking missing-values approaches for predictive models on health databases

    Alexandre Perez-Lebel et al. “Benchmarking missing-values approaches for predictive models on health databases”. In:GigaScience11 (Apr. 2022), giac013.issn: 2047-217X.doi:10.1093/gigascience/giac013. eprint:https://academic.oup.com/gigascience/article- pdf/doi/ 10.1093/gigascience/giac013/60706597/giac013.pdf.url:https: //doi.org/10.1093/gigascience/giac013

    work page doi:10.1093/gigascience/giac013 2022

  26. [26]

    scikit-survival: A Library for Time-to-Event Analysis Built on Top of scikit-learn

    Sebastian P¨ olsterl. “scikit-survival: A Library for Time-to-Event Analysis Built on Top of scikit-learn”. In:Journal of Machine Learning Research 21.212 (2020), pp. 1–6.issn: 1533-7928.url:http://jmlr.org/papers/ v21/20-729.html

    2020

  27. [27]

    Society for Immunotherapy of Cancer (SITC) con- sensus definitions for resistance to combinations of immune checkpoint in- hibitors with chemotherapy

    Naiyer Rizvi et al. “Society for Immunotherapy of Cancer (SITC) con- sensus definitions for resistance to combinations of immune checkpoint in- hibitors with chemotherapy”. In:Journal for ImmunoTherapy of Cancer 11.3 (Mar. 14, 2023).issn: 2051-1426.doi:10.1136/jitc-2022-005920. url:https://jitc.bmj.com/content/11/3/e005920. S33

    work page doi:10.1136/jitc-2022-005920 2023

  28. [28]

    High-Dimensional Survival Analysis: Meth- ods and Applications

    Stephen Salerno and Yi Li. “High-Dimensional Survival Analysis: Meth- ods and Applications”. In:Annual review of statistics and its application 10.1 (Mar. 2023), pp. 25–49.issn: 2326-8298.doi:10.1146/annurev- statistics- 032921- 022127.url:https://www.ncbi.nlm.nih.gov/ pmc/articles/PMC10038209/

    work page doi:10.1146/annurev- 2023

  29. [29]

    MissForest—non-parametric missing value imputation for mixed-type data

    Daniel J. Stekhoven and Peter B¨ uhlmann. “MissForest—non-parametric missing value imputation for mixed-type data”. In:Bioinformatics28.1 (Oct. 2011), pp. 112–118.issn: 1367-4803.doi:10.1093/bioinformatics/ btr597. eprint:https://academic.oup.com/bioinformatics/article- pdf/28/1/112/50568519/bioinformatics_28_1_112.pdf.url:https: //doi.org/10.1093/bioinfo...

    work page doi:10.1093/bioinformatics/ 2011

  30. [30]

    Multimodal data fusion for cancer biomarker dis- covery with deep learning

    Sandra Steyaert et al. “Multimodal data fusion for cancer biomarker dis- covery with deep learning”. In:Nature machine intelligence5.4 (Apr. 2023), pp. 351–362.issn: 2522-5839.doi:10.1038/s42256-023-00633-5. url:https://pmc.ncbi.nlm.nih.gov/articles/PMC10484010/

    work page doi:10.1038/s42256-023-00633-5 2023

  31. [31]

    Society for Immunotherapy of Cancer (SITC) checkpoint inhibitor resistance definitions: efforts to harmonize terminol- ogy and accelerate immuno-oncology drug development

    Hussein A. Tawbi et al. “Society for Immunotherapy of Cancer (SITC) checkpoint inhibitor resistance definitions: efforts to harmonize terminol- ogy and accelerate immuno-oncology drug development”. In:Journal for ImmunoTherapy of Cancer11.7 (July 24, 2023).issn: 2051-1426.doi: 10.1136/jitc- 2023- 007309.url:https://jitc.bmj.com/content/ 11/7/e007309

    work page doi:10.1136/jitc- 2023

  32. [32]

    Missing value estimation methods for DNA microarrays

    Olga Troyanskaya et al. “Missing value estimation methods for DNA microarrays”. In:Bioinformatics17.6 (June 2001), pp. 520–525.issn: 1367-4803.doi:10 . 1093 / bioinformatics / 17 . 6 . 520. eprint:https : / / academic . oup . com / bioinformatics / article - pdf / 17 / 6 / 520 / 48837104/bioinformatics_17_6_520.pdf.url:https://doi.org/10. 1093/bioinformat...

    2001

  33. [33]

    Good methods for coping with missing data in decision trees

    B.E.T.H. Twala, M.C. Jones, and D.J. Hand. “Good methods for coping with missing data in decision trees”. In:Pattern Recognition Letters29.7 (2008), pp. 950–956.issn: 0167-8655.doi:https://doi.org/10.1016/ j . patrec . 2008 . 01 . 010.url:https : / / www . sciencedirect . com / science/article/pii/S0167865508000305

    2008

  34. [34]

    Improved Mixture Cure Model Using Machine Learning Approaches

    Huina Wang, Tian Feng, and Baosheng Liang. “Improved Mixture Cure Model Using Machine Learning Approaches”. In:Mathematics13.4 (Feb. 8, 2025). Company: Multidisciplinary Digital Publishing Institute Distribu- tor: Multidisciplinary Digital Publishing Institute Institution: Multidisci- plinary Digital Publishing Institute Label: Multidisciplinary Digital P...

    2025

  35. [35]

    Stacked generalization

    David H. Wolpert. “Stacked generalization”. In:Neural Networks5.2 (Jan. 1, 1992), pp. 241–259.issn: 0893-6080.doi:10 . 1016 / S0893 - 6080(05 ) 80023- 1.url:https://www.sciencedirect.com/science/article/ pii/S0893608005800231. S34

    1992

  36. [36]

    Renjie Wu et al.Deep Multimodal Learning with Missing Modality: A Survey. Oct. 21, 2024.doi:10.48550/arXiv.2409.07825. arXiv:2409. 07825[cs].url:http://arxiv.org/abs/2409.07825

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2409.07825 2024

  37. [37]

    Bi-level Multi-Source Learning for Heterogeneous Block- wise Missing Data

    Shuo Xiang et al. “Bi-level Multi-Source Learning for Heterogeneous Block- wise Missing Data”. In:NeuroImage102 Pt 1 (Nov. 15, 2014), pp. 192– 206.issn: 1053-8119.doi:10.1016/j.neuroimage.2013.08.015.url: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3937297/

    work page doi:10.1016/j.neuroimage.2013.08.015.url: 2014

  38. [38]

    Chang Xu, Dacheng Tao, and Chao Xu.A Survey on Multi-view Learning

  39. [39]

    arXiv:1304.5634 [cs.LG].url:https://arxiv.org/abs/1304. 5634

    work page internal anchor Pith review Pith/arXiv arXiv

  40. [40]

    Deep Multimodal Data Fusion

    Fei Zhao, Chengcui Zhang, and Baocheng Geng. “Deep Multimodal Data Fusion”. In:ACM Comput. Surv.56.9 (2024), 216:1–216:36.issn: 0360- 0300.doi:10.1145/3649447.url:https://dl.acm.org/doi/10.1145/ 3649447. S35

    work page doi:10.1145/3649447.url:https://dl.acm.org/doi/10.1145/ 2024