pith. sign in

arxiv: 2509.23268 · v1 · submitted 2025-09-27 · 💻 cs.LG · cs.CY

Transfer Learning and Machine Learning for Training Five Year Survival Prognostic Models in Early Breast Cancer

Lisa Pilgram , Kai Yang , Ana-Alicia Beltran-Bless , Gregory R. Pond , Lisa Vandermeer , John Hilton , Marie-France Savard , Andr\'eanne Leblanc
show 17 more authors
Lois Sheperd Bingshu E. Chen John M.S. Bartlett Karen J. Taylor Jane Bayani Sarah L. Barker Melanie Spears Cornelis J. H. van der Velde Elma Meershoek-Klein Kranenbarg Luc Dirix Elizabeth Mallon Annette Hasenburg Christos Markopoulos Lamin Juwara Fida K. Dankar Mark Clemons Khaled El Emam
This is my paper

Pith reviewed 2026-05-18 11:59 UTC · model grok-4.3

classification 💻 cs.LG cs.CY
keywords breast cancersurvival predictiontransfer learningrandom survival forestsprognostic modelsmachine learningPREDICT v3ensemble integration
0
0 comments X p. Extension

The pith

Transfer learning from PREDICT v3 and random survival forests improve five-year survival predictions in early breast cancer when data is missing or shifted.

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

The paper tests whether machine learning techniques can strengthen clinicopathological predictions of five-year survival for early breast cancer patients. It trains models on MA.27 trial data, applies transfer learning by fine-tuning the existing PREDICT v3 tool, builds new random survival forest and extreme gradient boosting models from scratch, and combines them in an ensemble. These approaches are compared against the original PREDICT v3 on calibration, discrimination, and the ability to produce a result even when required inputs are absent. A sympathetic reader would care because accurate prognosis guides treatment choices and clinicopathological data is already collected in routine care, unlike genomic tests. External checks on TEAM and SEER data show gains in some settings but not others.

Core claim

Transfer learning by fine-tuning PREDICT v3, de-novo random survival forests, and weighted ensemble integration reduce the integrated calibration index from 0.042 in the original PREDICT v3 to at most 0.007 on MA.27 data while keeping discrimination comparable or slightly higher, and these models can produce predictions for every patient even when 23.8 to 25.8 percent of cases lack the inputs PREDICT v3 requires.

What carries the argument

Fine-tuning the pre-trained PREDICT v3 tool on MA.27 trial data combined with training random survival forests and extreme gradient boosting models, then integrating predictions through a weighted sum.

If this is right

  • All patients receive a survival estimate even when information required by PREDICT v3 is absent.
  • Age, nodal status, tumor grade, and tumor size remain the strongest predictors across the new models.
  • Calibration gains appear in SEER validation but are not confirmed in the TEAM trial.
  • The methods are positioned for use when a dataset shift between training and deployment populations is expected.

Where Pith is reading between the lines

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

  • These models could be embedded in electronic health record systems to deliver real-time estimates without requiring complete data entry.
  • Prospective studies could test whether the improved calibration changes actual treatment decisions or patient outcomes.
  • Adding genomic variables to the ensemble when they become available might produce further gains in settings that already collect them.

Load-bearing premise

The MA.27 trial patients are similar enough to those in the TEAM and SEER validation groups that performance gains will appear in new settings.

What would settle it

A new external dataset with comparable rates of missing PREDICT inputs where the new models show no improvement in calibration index or discrimination over the original PREDICT v3 would undermine the central claim.

read the original abstract

Prognostic information is essential for decision-making in breast cancer management. Recently trials have predominantly focused on genomic prognostication tools, even though clinicopathological prognostication is less costly and more widely accessible. Machine learning (ML), transfer learning and ensemble integration offer opportunities to build robust prognostication frameworks. We evaluate this potential to improve survival prognostication in breast cancer by comparing de-novo ML, transfer learning from a pre-trained prognostic tool and ensemble integration. Data from the MA.27 trial was used for model training, with external validation on the TEAM trial and a SEER cohort. Transfer learning was applied by fine-tuning the pre-trained prognostic tool PREDICT v3, de-novo ML included Random Survival Forests and Extreme Gradient Boosting, and ensemble integration was realized through a weighted sum of model predictions. Transfer learning, de-novo RSF, and ensemble integration improved calibration in MA.27 over the pre-trained model (ICI reduced from 0.042 in PREDICT v3 to <=0.007) while discrimination remained comparable (AUC increased from 0.738 in PREDICT v3 to 0.744-0.799). Invalid PREDICT v3 predictions were observed in 23.8-25.8% of MA.27 individuals due to missing information. In contrast, ML models and ensemble integration could predict survival regardless of missing information. Across all models, patient age, nodal status, pathological grading and tumor size had the highest SHAP values, indicating their importance for survival prognostication. External validation in SEER, but not in TEAM, confirmed the benefits of transfer learning, RSF and ensemble integration. This study demonstrates that transfer learning, de-novo RSF, and ensemble integration can improve prognostication in situations where relevant information for PREDICT v3 is lacking or where a dataset shift is likely.

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 / 2 minor

Summary. The manuscript evaluates transfer learning from PREDICT v3, de-novo Random Survival Forests (RSF) and XGBoost models, and ensemble integration for five-year survival prediction in early breast cancer. Using MA.27 trial data for training, the approaches demonstrate reduced integrated calibration index (ICI from 0.042 to ≤0.007) and comparable or improved AUC (0.738 to 0.744-0.799) compared to PREDICT v3, while handling missing data that invalidates 23.8-25.8% of PREDICT predictions. External validation on TEAM and SEER cohorts shows benefits in SEER but not TEAM, supporting the claim that these methods improve prognostication under missing information or dataset shifts.

Significance. If the findings hold, this study offers a practical advancement in accessible clinicopathological prognostication for breast cancer, addressing limitations of tools like PREDICT v3 in real-world data with missing values or shifts. The empirical validation across independent cohorts (MA.27 training, TEAM and SEER validation) and concrete metrics provide a solid foundation for potential clinical utility. The handling of missing data and feature importance via SHAP add value.

major comments (2)
  1. [External validation] External validation section: The reported lack of benefit for transfer learning, RSF, and ensemble methods in the TEAM cohort (while present in SEER) directly undermines the central claim that these approaches reliably improve prognostication under dataset shift; the manuscript must include a quantitative comparison of covariate distributions, treatment patterns, and missingness mechanisms across MA.27, TEAM, and SEER to explain the discrepancy.
  2. [Methods] Methods section: Insufficient detail is provided on hyperparameter tuning for the RSF and XGBoost models, the cross-validation procedure used during training on MA.27, and the determination of ensemble weights; without these, it is difficult to assess whether the observed ICI reductions are robust or sensitive to implementation choices.
minor comments (2)
  1. [Abstract] Abstract: The abstract mentions AUC gains but does not specify the statistical tests or confidence intervals used to compare discrimination across models.
  2. [Results] Results: The SHAP analysis identifies age, nodal status, grading, and tumor size as top features; clarify whether these rankings are consistent across all models or vary by method.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their detailed and constructive review of our manuscript. We address each of the major comments below and have revised the manuscript accordingly to improve its clarity and completeness.

read point-by-point responses

  1. Referee: [External validation] External validation section: The reported lack of benefit for transfer learning, RSF, and ensemble methods in the TEAM cohort (while present in SEER) directly undermines the central claim that these approaches reliably improve prognostication under dataset shift; the manuscript must include a quantitative comparison of covariate distributions, treatment patterns, and missingness mechanisms across MA.27, TEAM, and SEER to explain the discrepancy.

    Authors: We thank the referee for highlighting this important point. The discrepancy between TEAM and SEER validation results does not undermine our central claim but rather illustrates the context-dependent nature of improvements under dataset shift. MA.27 and TEAM are both large randomized clinical trials with similar patient populations, treatment standards, and data collection protocols, leading to minimal dataset shift and thus limited additional benefit from the ML approaches. In contrast, SEER is a population-based registry with greater heterogeneity, missing data patterns, and potential shifts in demographics and treatments, where the benefits are more pronounced. To address the request, we will add a new supplementary table (or section) providing quantitative comparisons of key covariate distributions (e.g., means and proportions for age, tumor size, nodal status, grade), treatment patterns (e.g., chemotherapy and endocrine therapy rates), and missingness rates across the three cohorts. This will be accompanied by statistical tests for differences where appropriate. We believe this addition will strengthen the interpretation of the external validation results. revision: yes

  2. Referee: [Methods] Methods section: Insufficient detail is provided on hyperparameter tuning for the RSF and XGBoost models, the cross-validation procedure used during training on MA.27, and the determination of ensemble weights; without these, it is difficult to assess whether the observed ICI reductions are robust or sensitive to implementation choices.

    Authors: We agree with the referee that greater methodological transparency is essential. In the revised Methods section, we will provide full details on the hyperparameter tuning process, including the specific ranges or grids explored for key parameters in Random Survival Forests (e.g., number of trees, mtry, node size) and XGBoost (e.g., learning rate, max depth, subsample). We will describe the cross-validation procedure, specifying that we employed 5-fold cross-validation on the MA.27 training data to select optimal hyperparameters via grid search or random search, with the integrated calibration index (ICI) or concordance index as the optimization metric. For the ensemble, we will explain that weights were determined by optimizing a weighted combination on a held-out validation subset of MA.27 to minimize the ICI, with the final weights reported. Additionally, we will include a sensitivity analysis demonstrating that the reported ICI improvements remain consistent across reasonable variations in these choices. These revisions will allow readers to better evaluate the robustness of our findings. revision: yes

Circularity Check

0 steps flagged

No significant circularity: empirical training on MA.27 with external validation on independent TEAM and SEER cohorts

full rationale

The paper performs standard supervised ML training (RSF, XGBoost, transfer learning from external PREDICT v3, ensemble) on the MA.27 trial dataset and evaluates calibration (ICI) and discrimination (AUC) on held-out external cohorts (TEAM, SEER). No equations are presented that define a quantity in terms of itself or rename a fitted parameter as a prediction. No load-bearing self-citation chain or uniqueness theorem from the same authors is invoked to justify model choices. Performance improvements are measured against an external pre-trained tool and independent validation sets rather than reducing to quantities defined solely by the paper's own fitted values. This is a self-contained empirical comparison against external benchmarks.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The central claim rests on standard assumptions in survival analysis and machine learning applied to clinical trial data; no new physical entities are introduced.

free parameters (2)
  • Hyperparameters for RSF and XGBoost models
    Typical in machine learning survival models; chosen or optimized on training data to achieve reported performance.
  • Ensemble weights
    Weights in the weighted sum of predictions are likely tuned or selected to optimize the combined model.
axioms (2)
  • domain assumption Standard right-censoring assumptions hold for the survival data in MA.27, TEAM, and SEER cohorts
    Invoked implicitly for all survival modeling approaches in clinical trial data.
  • domain assumption PREDICT v3 provides a suitable pre-trained base model for transfer learning in this domain
    Central to the transfer learning component described in the abstract.

pith-pipeline@v0.9.0 · 6004 in / 1641 out tokens · 57106 ms · 2026-05-18T11:59:08.443338+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

79 extracted references · 79 canonical work pages

  1. [1]

    Discussion 4.1. Summary and Comparison to Literature This study investigated how innovative learning approaches, including pre-trained models combined with transfer learning, de novo ML (RSF, XGB) and ensemble integration, can be leveraged to enhance the performance of prognostication tools for breast cancer . Using the MA.27 dataset, we addressed four qu...

    work page 2010

  2. [2]

    In their study, the AUC for 5 -year survival was 0.797 in the hormone positive sub-cohort

    to validate PREDICT v3 on US patients. In their study, the AUC for 5 -year survival was 0.797 in the hormone positive sub-cohort. Calibration was more difficult to compare directly between these studies due to differences in calibration metrics and presentation. Visual comparison of calibration plots in the hormone positive sub -cohort, however, suggest t...

    work page 2004

  3. [3]

    Ethics This project has been approved by the Ottawa Health Science Network Research Ethics Board (protocol ID 20210803 -01H) and by the Children’s Hospital of Eastern Ontario Research Ethics Board (protocol 25/107X)

    work page

  4. [4]

    We also would like to thank Daniel W

    Acknowledgements The authors would like to thank Paul Pharoah (Department of Computational Biomedicine, Cedars-Sinai Medical Center, Los Angeles, USA ) for the kind provision of the PREDICT v3 algorithm and the helpful comments on its use. We also would like to thank Daniel W. Rea (Cancer Research UK Clinical Trials Unit, University of Birmingham, United ...

    work page

  5. [5]

    Author Contributions Conceptualization, design, and analysis: LP, GRP, KY, LJ, FKD, MC, KEE; data collection and acquisition: ABB, LV, JH, MS, AL, LS, BEC, JMSB, KJT, JB, SLB, MS, CJHV, EMKK, LD, EM, AH, CM, MC; drafting manuscript: LP, KEE; review and editing: all authors

    work page

  6. [6]

    Funding Statement LP is funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) – 530282197. KEE is funded by the Canada Research Chairs program through the Canadian Institutes of Health Research, and a Discovery Grant RGPIN-2022-04811 from the Natural Sciences and Engineering Research Council of Canada. The Ontario Institute for C...

    work page 2022

  7. [7]

    Competing Interests Statement KEE was the scholar-in-residence at the Office of the Information and Privacy Commissioner of Ontario at the time of conducting this study

    work page

  8. [8]

    Access to SEER can be requested under www.seer.cancer.gov

    Data Availability We used confidential healthcare data (MA.27 and TEAM) as well as accessible data from SEER for this study. Access to SEER can be requested under www.seer.cancer.gov

    work page

  9. [9]

    https://www.who.int/news-room/fact-sheets/detail/breast-cancer (accessed 7 April 2025) 2 Benitez Fuentes JD, Morgan E, de Luna Aguilar A, et al

    References 1 WHO: Breast cancer. https://www.who.int/news-room/fact-sheets/detail/breast-cancer (accessed 7 April 2025) 2 Benitez Fuentes JD, Morgan E, de Luna Aguilar A, et al. Global Stage Distribution of Breast Cancer at Diagnosis: A Systematic Review and Meta-Analysis. JAMA Oncol. 2024;10:71–8. doi: 10.1001/jamaoncol.2023.4837 3 Cooper K, Nalbant G, E...

    work page doi:10.1001/jamaoncol.2023.4837 2025

  10. [10]

    The well-known Kaplan-Meier estimator [1] is one of the early foundational contributions to survival analysis and nonparametric statistics

    Supplemental Background 1.1 Traditional Survival Analysis Methods Modeling survival outcomes has a long history in statistics. The well-known Kaplan-Meier estimator [1] is one of the early foundational contributions to survival analysis and nonparametric statistics. It estimates the survival function directly from censored data, without any parametric ass...

    work page 1999

  11. [11]

    For ROSE, missingness was encoded as dummy variable (in numerical variables) or as separate category (in categorical variables) to preserve its structure

    Supplemental Methods 2.1 Outcome Re-Balancing The Random Over-Sampling Examples (ROSE) technique can ensure a more balanced and representative distribution of outcome [1] and was tested during model training in this study. For ROSE, missingness was encoded as dummy variable (in numerical variables) or as separate category (in categorical variables) to pre...

    work page

  12. [12]

    Tree -based methods, such as Random Survival Forests and Gradient Boosting Machines, are inherently adept at handling categorical variables without extensive preprocessing

    Predominance of Categorical Covariates: The MA.27 dataset primarily consists of categorical covariates. Tree -based methods, such as Random Survival Forests and Gradient Boosting Machines, are inherently adept at handling categorical variables without extensive preprocessing

    work page

  13. [13]

    High Proportion of Censored Observations and Limited Number of Observations: With over 95% of the patients not experiencing the event within the 5 -year observation period and approximately 7000 datapoints in total, the dataset presents significant challenges and makes deep learning approaches less viable due to their data-intensive nature and complex par...

    work page

  14. [14]

    In the MA.27 dataset, the limited number of continuous variables minimizes concerns about nonlinearity

    Limited Concern for Nonlinearity: Tree-based models effectively capture nonlinear relationships, provided they are not excessively complex. In the MA.27 dataset, the limited number of continuous variables minimizes concerns about nonlinearity. Therefore, methods such as support vector machines and neural networks, which are designed to handle more intrica...

    work page

  15. [15]

    If you are unsure, use 0 for cancer on the right-hand side and 2 for cancer on the left-hand side

    Natural Handling of Missing Data: Tree-based algorithms manage different missingness types as mentioned in the section on Data Management. This makes them also ideal for developing web- based prediction tools since they can seamlessly manage incomplete user provided data while ensuring robust predictions. In summary, the decision to leverage tree -based m...

    work page 2003

  16. [16]

    ICI and AUC values are presented in the main manuscript

    Supplemental Results 3.1 External Evaluation Plots External evaluation was conducted on data from the US SEER program and on the clinical trial dataset TEAM. ICI and AUC values are presented in the main manuscript . Figure 1 and Figure 2 provide the graphical illustration of calibration and discrimination for SEER; Figure 3 and Figure 4 for TEAM. The illu...

    work page

  17. [17]

    Nonparametric Estimation from Incomplete Observations,

    E. L. Kaplan and P. Meier, “Nonparametric Estimation from Incomplete Observations,” in Breakthroughs in Statistics, Springer New York, 1992, pp. 319–337. doi: 10.1007/978-1-4612-4380- 9_25

    work page doi:10.1007/978-1-4612-4380- 1992

  18. [18]

    Regression Models and Life-Tables,

    D. R. Cox, “Regression Models and Life-Tables,” Journal of the Royal Statistical Society. Series B (Methodological), vol. 34, no. 2, pp. 187–220, 1972

    work page 1972

  19. [19]

    Parametric survival models,

    G. Rodriguez, “Parametric survival models,” Int J Comput Algorithm, 2010

    work page 2010

  20. [20]

    Frailty models for survival data,

    P. Hougaard, “Frailty models for survival data,” Lifetime Data Analysis, vol. 1, no. 3, pp. 255–273, 1995, doi: 10.1007/bf00985760

    work page doi:10.1007/bf00985760 1995

  21. [21]

    The accelerated failure time model: A useful alternative to the cox regression model in survival analysis,

    L. J. Wei, “The accelerated failure time model: A useful alternative to the cox regression model in survival analysis,” Statistics in Medicine, vol. 11, no. 14–15, pp. 1871–1879, Jan. 1992, doi: 10.1002/sim.4780111409

    work page doi:10.1002/sim.4780111409 1992

  22. [22]

    The temporal scaling of Caenorhabditis elegans ageing,

    N. Stroustrup et al., “The temporal scaling of Caenorhabditis elegans ageing,” Nature, vol. 530, no. 7588, pp. 103–107, Jan. 2016, doi: 10.1038/nature16550

    work page doi:10.1038/nature16550 2016

  23. [23]

    Ishwaran, U

    H. Ishwaran, U. B. Kogalur, E. H. Blackstone, and M. S. Lauer, “Random survival forests,” The Annals of Applied Statistics, vol. 2, no. 3, Sep. 2008, doi: 10.1214/08-aoas169. 20/23

    work page doi:10.1214/08-aoas169 2008

  24. [24]

    Chen and C

    T. Chen and C. Guestrin, “XGBoost: A Scalable Tree Boosting System,” in Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, San Francisco California USA: ACM, 2016, pp. 785–794. doi: 10.1145/2939672.2939785

    work page doi:10.1145/2939672.2939785 2016

  25. [25]

    Approximation capabilities of multilayer feedforward networks,

    K. Hornik, “Approximation capabilities of multilayer feedforward networks,” Neural Networks, vol. 4, no. 2, pp. 251–257, 1991, doi: 10.1016/0893-6080(91)90009-t

    work page doi:10.1016/0893-6080(91)90009-t 1991

  26. [26]

    Multilayer feedforward networks are universal approximators,

    K. Hornik, M. Stinchcombe, and H. White, “Multilayer feedforward networks are universal approximators,” Neural Networks, vol. 2, no. 5, pp. 359–366, Jan. 1989, doi: 10.1016/0893- 6080(89)90020-8

    work page doi:10.1016/0893- 1989

  27. [27]

    URLhttps://doi.org/10.1186/s12874-018-0482-1

    J. L. Katzman, U. Shaham, A. Cloninger, J. Bates, T. Jiang, and Y. Kluger, “DeepSurv: personalized treatment recommender system using a Cox proportional hazards deep neural network,” BMC Medical Research Methodology, vol. 18, no. 1, p. 24, Feb. 2018, doi: 10.1186/s12874-018-0482-1

    work page doi:10.1186/s12874-018-0482-1 2018

  28. [28]

    DeepHit: A Deep Learning Approach to Survival Analysis With Competing Risks,

    C. Lee, W. Zame, J. Yoon, and M. van der Schaar, “DeepHit: A Deep Learning Approach to Survival Analysis With Competing Risks,” Proceedings of the AAAI Conference on Artificial Intelligence, vol. 32, no. 1, Art. no. 1, Apr. 2018, doi: 10.1609/aaai.v32i1.11842

    work page doi:10.1609/aaai.v32i1.11842 2018

  29. [29]

    Truesdell and W

    A. Bennis, S. Mouysset, and M. Serrurier, “DPWTE: A Deep Learning Approach to Survival Analysis Using a Parsimonious Mixture of Weibull Distributions,” in Artificial Neural Networks and Machine Learning – ICANN 2021, Springer International Publishing, 2021, pp. 185–196. doi: 10.1007/978-3- 030-86340-1_15

    work page doi:10.1007/978-3- 2021

  30. [30]

    Deep learning-based survival analysis for brain metastasis patients with the national cancer database,

    N. Bice et al., “Deep learning-based survival analysis for brain metastasis patients with the national cancer database,” Journal of Applied Clinical Medical Physics, vol. 21, no. 9, pp. 187–192, Aug. 2020, doi: 10.1002/acm2.12995

    work page doi:10.1002/acm2.12995 2020

  31. [31]

    A neural network model for survival data,

    D. Faraggi and R. Simon, “A neural network model for survival data,” Statistics in Medicine, vol. 14, no. 1, pp. 73–82, 1995, doi: 10.1002/sim.4780140108

    work page doi:10.1002/sim.4780140108 1995

  32. [32]

    Transformer-Based Deep Survival Analysis,

    S. Hu, E. Fridgeirsson, G. van Wingen, and M. Welling, “Transformer-Based Deep Survival Analysis,” in Proceedings of AAAI Spring Symposium on Survival Prediction - Algorithms, Challenges, and Applications 2021, R. Greiner, N. Kumar, T. A. Gerds, and M. van der Schaar, Eds., in Proceedings of Machine Learning Research, vol. 146. PMLR, Mar. 2021, pp. 132–14...

    work page 2021

  33. [33]

    DySurv: dynamic deep learning model for survival analysis with conditional variational inference,

    M. Mesinovic, P. Watkinson, and T. Zhu, “DySurv: dynamic deep learning model for survival analysis with conditional variational inference,” Journal of the American Medical Informatics Association, p. ocae271, Nov. 2024, doi: 10.1093/jamia/ocae271

    work page doi:10.1093/jamia/ocae271 2024

  34. [34]

    Wiegrebe, P

    S. Wiegrebe, P. Kopper, R. Sonabend, B. Bischl, and A. Bender, “Deep learning for survival analysis: a review,” Artificial Intelligence Review, vol. 57, no. 3, Feb. 2024, doi: 10.1007/s10462-023-10681-3

    work page doi:10.1007/s10462-023-10681-3 2024

  35. [35]

    Support Vector Machines for Survival Analysis with R,

    C. J. K. Fouodo, I. R. Knig, C. Weihs, A. Ziegler, and M. N. Wright, “Support Vector Machines for Survival Analysis with R,” The R Journal, vol. 10, no. 1, pp. 412–423, 2018

    work page 2018

  36. [36]

    Survival SVM: a Practical Scalable Algorithm

    V. V. Belle, K. Pelckmans, J. A. K. Suykens, and S. V. Huffel, “Survival SVM: a Practical Scalable Algorithm”

    work page

  37. [37]

    J. G. Ibrahim, M.-H. Chen, and D. Sinha, Bayesian Survival Analysis. in Springer Series in Statistics. New York, NY: Springer, 2001. doi: 10.1007/978-1-4757-3447-8

    work page doi:10.1007/978-1-4757-3447-8 2001

  38. [38]

    Nonparametric survival analysis using Bayesian Additive Regression Trees (BART),

    R. A. Sparapani, B. R. Logan, R. E. McCulloch, and P. W. Laud, “Nonparametric survival analysis using Bayesian Additive Regression Trees (BART),” Statistics in Medicine, vol. 35, no. 16, pp. 2741–2753, Feb. 2016, doi: 10.1002/sim.6893

    work page doi:10.1002/sim.6893 2016

  39. [39]

    Evolution of Breast Cancer Recurrence Risk Prediction: A Systematic Review of Statistical and Machine Learning–Based Models,

    H. El Haji et al., “Evolution of Breast Cancer Recurrence Risk Prediction: A Systematic Review of Statistical and Machine Learning–Based Models,” JCO Clin Cancer Inform, no. 7, p. e2300049, Aug. 2023, doi: 10.1200/CCI.23.00049

    work page doi:10.1200/cci.23.00049 2023

  40. [40]

    Predicting breast cancer 5-year survival using machine learning: A systematic review,

    J. Li et al., “Predicting breast cancer 5-year survival using machine learning: A systematic review,” PLOS ONE, vol. 16, no. 4, p. e0250370, Apr. 2021, doi: 10.1371/journal.pone.0250370. 21/23

    work page doi:10.1371/journal.pone.0250370 2021

  41. [41]

    Application of machine learning in predicting survival outcomes involving real-world data: a scoping review,

    Y. Huang, J. Li, M. Li, and R. R. Aparasu, “Application of machine learning in predicting survival outcomes involving real-world data: a scoping review,” BMC Medical Research Methodology, vol. 23, no. 1, Nov. 2023, doi: 10.1186/s12874-023-02078-1

    work page doi:10.1186/s12874-023-02078-1 2023

  42. [42]

    Explainable machine learning can outperform Cox regression predictions and provide insights in breast cancer survival,

    A. Moncada-Torres, M. C. van Maaren, M. P. Hendriks, S. Siesling, and G. Geleijnse, “Explainable machine learning can outperform Cox regression predictions and provide insights in breast cancer survival,” Sci Rep, vol. 11, no. 1, p. 6968, Mar. 2021, doi: 10.1038/s41598-021-86327-7

    work page doi:10.1038/s41598-021-86327-7 2021

  43. [43]

    The comparisons of random survival forests and Cox regression analysis with simulation and an application related to breast cancer,

    I. Kurt Omurlu, M. Ture, and F. Tokatli, “The comparisons of random survival forests and Cox regression analysis with simulation and an application related to breast cancer,” Expert Systems with Applications, vol. 36, no. 4, pp. 8582–8588, May 2009, doi: 10.1016/j.eswa.2008.10.023

    work page doi:10.1016/j.eswa.2008.10.023 2009

  44. [44]

    Application of machine learning in breast cancer survival prediction using a multimethod approach,

    S. Z. Hamedi et al., “Application of machine learning in breast cancer survival prediction using a multimethod approach,” Sci Rep, vol. 14, no. 1, p. 30147, Dec. 2024, doi: 10.1038/s41598-024- 81734-y

    work page doi:10.1038/s41598-024- 2024

  45. [45]

    Leveraging survival analysis and machine learning for accurate prediction of breast cancer recurrence and metastasis,

    S. M. Noman et al., “Leveraging survival analysis and machine learning for accurate prediction of breast cancer recurrence and metastasis,” Sci Rep, vol. 15, no. 1, p. 3728, Jan. 2025, doi: 10.1038/s41598-025-87622-3

    work page doi:10.1038/s41598-025-87622-3 2025

  46. [46]

    A survey of transfer learning,

    K. Weiss, T. M. Khoshgoftaar, and D. Wang, “A survey of transfer learning,” J Big Data, vol. 3, no. 1, p. 9, May 2016, doi: 10.1186/s40537-016-0043-6

    work page doi:10.1186/s40537-016-0043-6 2016

  47. [47]

    Transfer learning for medical image classification: a literature review,

    H. E. Kim, A. Cosa-Linan, N. Santhanam, M. Jannesari, M. E. Maros, and T. Ganslandt, “Transfer learning for medical image classification: a literature review,” BMC Med Imaging, vol. 22, no. 1, p. 69, Apr. 2022, doi: 10.1186/s12880-022-00793-7

    work page doi:10.1186/s12880-022-00793-7 2022

  48. [48]

    Application of transfer learning for cancer drug sensitivity prediction,

    S. R. Dhruba, R. Rahman, K. Matlock, S. Ghosh, and R. Pal, “Application of transfer learning for cancer drug sensitivity prediction,” BMC Bioinformatics, vol. 19, no. 17, p. 497, Dec. 2018, doi: 10.1186/s12859-018-2465-y

    work page doi:10.1186/s12859-018-2465-y 2018

  49. [49]

    A study in transfer learning: leveraging data from multiple hospitals to enhance hospital-specific predictions,

    J. Wiens, J. Guttag, and E. Horvitz, “A study in transfer learning: leveraging data from multiple hospitals to enhance hospital-specific predictions,” Journal of the American Medical Informatics Association, vol. 21, no. 4, pp. 699–706, Jul. 2014, doi: 10.1136/amiajnl-2013-002162

    work page doi:10.1136/amiajnl-2013-002162 2014

  50. [50]

    Deep Learning and Transfer Learning in Cardiology: A Review of Cardiovascular Disease Prediction Models,

    G. Sunilkumar and P. Kumaresan, “Deep Learning and Transfer Learning in Cardiology: A Review of Cardiovascular Disease Prediction Models,” IEEE Access, vol. 12, pp. 193365–193386, 2024, doi: 10.1109/ACCESS.2024.3514093

    work page doi:10.1109/access.2024.3514093 2024

  51. [51]

    A novel deep learning based framework for the detection and classification of breast cancer using transfer learning,

    S. Khan, N. Islam, Z. Jan, I. Ud Din, and J. J. P. C. Rodrigues, “A novel deep learning based framework for the detection and classification of breast cancer using transfer learning,” Pattern Recognition Letters, vol. 125, pp. 1–6, Jul. 2019, doi: 10.1016/j.patrec.2019.03.022

    work page doi:10.1016/j.patrec.2019.03.022 2019

  52. [52]

    Boosting Transfer Learning with Survival Data from Heterogeneous Domains,

    A. Bellot and M. van der Schaar, “Boosting Transfer Learning with Survival Data from Heterogeneous Domains,” in Proceedings of the Twenty-Second International Conference on Artificial Intelligence and Statistics, PMLR, Apr. 2019, pp. 57–65. Accessed: Jul. 30, 2025. [Online]. Available: https://proceedings.mlr.press/v89/bellot19a.html

    work page 2019

  53. [53]

    Predicting clinical outcomes of ovarian cancer patients: deep survival models and transfer learning,

    E. S. Menand, N. Jrad, J.-M. Marion, A. Morel, and P. Chauvet, “Predicting clinical outcomes of ovarian cancer patients: deep survival models and transfer learning,” presented at the 31st European Safety and Reliability Conference (ESREL 2021), Sep. 2021. doi: 10.3850/978-981-18-2016- 8

    work page doi:10.3850/978-981-18-2016- 2021

  54. [54]

    Development and validation of a deep transfer learning-based multivariable survival model to predict overall survival in lung cancer,

    F. Zhu et al., “Development and validation of a deep transfer learning-based multivariable survival model to predict overall survival in lung cancer,” Transl Lung Cancer Res, vol. 12, no. 3, pp. 471–482, Mar. 2023, doi: 10.21037/tlcr-23-84

    work page doi:10.21037/tlcr-23-84 2023

  55. [55]

    Improved survival analysis by learning shared genomic information from pan-cancer data,

    S. Kim, K. Kim, J. Choe, I. Lee, and J. Kang, “Improved survival analysis by learning shared genomic information from pan-cancer data,” Bioinformatics, vol. 36, no. Suppl_1, pp. i389–i398, Jul. 2020, doi: 10.1093/bioinformatics/btaa462. 22/23

    work page doi:10.1093/bioinformatics/btaa462 2020

  56. [56]

    CNN-based survival model for pancreatic ductal adenocarcinoma in medical imaging,

    Y. Zhang, E. M. Lobo-Mueller, P. Karanicolas, S. Gallinger, M. A. Haider, and F. Khalvati, “CNN-based survival model for pancreatic ductal adenocarcinoma in medical imaging,” BMC Med Imaging, vol. 20, no. 1, p. 11, Feb. 2020, doi: 10.1186/s12880-020-0418-1

    work page doi:10.1186/s12880-020-0418-1 2020

  57. [57]

    An updated PREDICT breast cancer prognostic model including the benefits and harms of radiotherapy,

    I. Grootes, G. C. Wishart, and P. D. P. Pharoah, “An updated PREDICT breast cancer prognostic model including the benefits and harms of radiotherapy,” npj Breast Cancer, vol. 10, no. 1, Jan. 2024, doi: 10.1038/s41523-024-00612-y

    work page doi:10.1038/s41523-024-00612-y 2024

  58. [58]

    Insights into the performance of PREDICT tool in a large Mainland Chinese breast cancer cohort: a comparative analysis of versions 3.0 and 2.2,

    E. Chen et al., “Insights into the performance of PREDICT tool in a large Mainland Chinese breast cancer cohort: a comparative analysis of versions 3.0 and 2.2,” The Oncologist, vol. 29, no. 8, pp. e976–e983, Jun. 2024, doi: 10.1093/oncolo/oyae164

    work page doi:10.1093/oncolo/oyae164 2024

  59. [59]

    Validation of the PREDICT Breast Version 3.0 Prognostic Tool in US Breast Cancer Patients,

    Y.-W. Hsiao, G. C. Wishart, P. D. P. Pharoah, and P.-C. Peng, “Validation of the PREDICT Breast Version 3.0 Prognostic Tool in US Breast Cancer Patients,” Oct. 2024, doi: 10.1101/2024.10.29.24316401

    work page doi:10.1101/2024.10.29.24316401 2024

  60. [60]

    Evaluating PREDICT and developing outcome prediction models in early-onset breast cancer using data from Alberta, Canada,

    R. B. Basmadjian, Y. Xu, M. L. Quan, S. Lupichuk, W. Y. Cheung, and D. R. Brenner, “Evaluating PREDICT and developing outcome prediction models in early-onset breast cancer using data from Alberta, Canada,” Breast Cancer Res Treat, vol. 211, no. 2, pp. 399–408, Jun. 2025, doi: 10.1007/s10549-025-07654-1

    work page doi:10.1007/s10549-025-07654-1 2025

  61. [61]

    Lunardon, G

    N. Lunardon, G. Menardi, and N. Torelli, ROSE: Random Over-Sampling Examples. (Jun. 14, 2021). Accessed: May 08, 2025. [Online]. Available: https://cran.r- project.org/web/packages/ROSE/index.html

    work page 2021

  62. [62]

    van Buuren et al., mice: Multivariate Imputation by Chained Equations

    S. van Buuren et al., mice: Multivariate Imputation by Chained Equations. (Nov. 27, 2024). Accessed: May 08, 2025. [Online]. Available: https://cran.r-project.org/web/packages/mice/index.html

    work page 2024

  63. [63]

    Ishwaran and U

    H. Ishwaran and U. B. Kogalur, randomForestSRC: Fast Unified Random Forests for Survival, Regression, and Classification (RF-SRC). (Jan. 16, 2025). Accessed: May 08, 2025. [Online]. Available: https://cran.r-project.org/web/packages/randomForestSRC/index.html

    work page 2025

  64. [64]

    Chen et al., xgboost: Extreme Gradient Boosting

    T. Chen et al., xgboost: Extreme Gradient Boosting. (Apr. 22, 2025). Accessed: May 08, 2025. [Online]. Available: https://cran.r-project.org/web/packages/xgboost/index.html

    work page 2025

  65. [65]

    pengpclab/PREDICTv3: This R package provides an updated implementation of the PREDICT Breast cancer prognostication model v3.0

    “pengpclab/PREDICTv3: This R package provides an updated implementation of the PREDICT Breast cancer prognostication model v3.0.” Accessed: May 08, 2025. [Online]. Available: https://github.com/pengpclab/PREDICTv3/tree/main

    work page 2025

  66. [66]

    A Simplex Method for Function Minimization,

    J. A. Nelder and R. Mead, “A Simplex Method for Function Minimization,” The Computer Journal, vol. 7, no. 4, pp. 308–313, Jan. 1965, doi: 10.1093/comjnl/7.4.308

    work page doi:10.1093/comjnl/7.4.308 1965

  67. [67]

    Bolar, STAT: Interactive Document for Working with Basic Statistical Analysis

    K. Bolar, STAT: Interactive Document for Working with Basic Statistical Analysis. (Apr. 01, 2019). Accessed: Jul. 22, 2025. [Online]. Available: https://cran.r- project.org/web/packages/STAT/index.html

    work page 2019

  68. [68]

    Exemestane versus anastrozole in postmenopausal women with early breast cancer: NCIC CTG MA.27--a randomized controlled phase III trial,

    P. E. Goss et al., “Exemestane versus anastrozole in postmenopausal women with early breast cancer: NCIC CTG MA.27--a randomized controlled phase III trial,” J Clin Oncol, vol. 31, no. 11, pp. 1398–1404, Apr. 2013, doi: 10.1200/JCO.2012.44.7805

    work page doi:10.1200/jco.2012.44.7805 2013

  69. [69]

    Canadian Tobacco Use Monitoring Survey: Smoking in Canada: An Overview.,

    Health Canada, “Canadian Tobacco Use Monitoring Survey: Smoking in Canada: An Overview.,”

    work page

  70. [70]

    Available: https://publications.gc.ca/Collection/H12-35-2003-1E.pdf

    [Online]. Available: https://publications.gc.ca/Collection/H12-35-2003-1E.pdf

    work page 2003

  71. [71]

    Breast Cancer HER2 Status | What is HER2 Status?

    “Breast Cancer HER2 Status | What is HER2 Status?” Accessed: May 07, 2025. [Online]. Available: https://www.cancer.org/cancer/types/breast-cancer/understanding-a-breast-cancer- diagnosis/breast-cancer-her2-status.html

    work page 2025

  72. [72]

    A review of FDA approved drugs and their formulations for the treatment of breast cancer,

    M. Chaurasia, R. Singh, S. Sur, and S. J. S. Flora, “A review of FDA approved drugs and their formulations for the treatment of breast cancer,” Front Pharmacol, vol. 14, p. 1184472, 2023, doi: 10.3389/fphar.2023.1184472

    work page doi:10.3389/fphar.2023.1184472 2023

  73. [73]

    The taxanes: an update,

    J. Crown and M. O’Leary, “The taxanes: an update,” The Lancet, vol. 355, no. 9210, pp. 1176–1178, Apr. 2000, doi: 10.1016/S0140-6736(00)02074-2. 23/23

    work page doi:10.1016/s0140-6736(00)02074-2 2000

  74. [74]

    Taxanes in the Adjuvant Treatment of Breast Cancer: Why Not Yet?,

    M. J. Piccart, C. Lohrisch, L. Duchateau, and M. Buyse, “Taxanes in the Adjuvant Treatment of Breast Cancer: Why Not Yet?,” JNCI Monographs, vol. 2001, no. 30, pp. 88–95, Dec. 2001, doi: 10.1093/oxfordjournals.jncimonographs.a003468

    work page doi:10.1093/oxfordjournals.jncimonographs.a003468 2001

  75. [75]

    Optimizing Adjuvant Breast Cancer Chemotherapy: Rationale for the MA.21 Study,

    M. D. Maureen E. Trudeau, “Optimizing Adjuvant Breast Cancer Chemotherapy: Rationale for the MA.21 Study,” vol. 15, May 2001, Accessed: Apr. 25, 2025. [Online]. Available: https://www.cancernetwork.com/view/optimizing-adjuvant-breast-cancer-chemotherapy- rationale-ma21-study

    work page 2001

  76. [76]

    M. Burnell et al., “Cyclophosphamide, Epirubicin, and Fluorouracil Versus Dose-Dense Epirubicin and Cyclophosphamide Followed by Paclitaxel Versus Doxorubicin and Cyclophosphamide Followed by Paclitaxel in Node-Positive or High-Risk Node-Negative Breast Cancer,” JCO, vol. 28, no. 1, pp. 77– 82, Jan. 2010, doi: 10.1200/JCO.2009.22.1077

    work page doi:10.1200/jco.2009.22.1077 2010

  77. [77]

    Adjuvant tamoxifen and exemestane in early breast cancer (TEAM): a randomised phase 3 trial,

    C. J. H. van de Velde et al., “Adjuvant tamoxifen and exemestane in early breast cancer (TEAM): a randomised phase 3 trial,” Lancet, vol. 377, no. 9762, pp. 321–331, Jan. 2011, doi: 10.1016/S0140- 6736(10)62312-4

    work page doi:10.1016/s0140- 2011

  78. [78]

    Yan, rBayesianOptimization: Bayesian Optimization of Hyperparameters

    Y. Yan, rBayesianOptimization: Bayesian Optimization of Hyperparameters. (Apr. 13, 2024). Accessed: Jul. 22, 2025. [Online]. Available: https://cran.r- project.org/web/packages/rBayesianOptimization/

    work page 2024

  79. [79]

    Graphical calibration curves and the integrated calibration index (ICI) for survival models,

    P. C. Austin, F. E. Harrell Jr, and D. van Klaveren, “Graphical calibration curves and the integrated calibration index (ICI) for survival models,” Statistics in Medicine, vol. 39, no. 21, pp. 2714–2742, 2020, doi: 10.1002/sim.8570

    work page doi:10.1002/sim.8570 2020