SurvSHAP(t): Time-dependent explanations of machine learning survival models

Mateusz, Krzyziński,; Spytek, Mikołaj; Baniecki, Hubert; Biecek, Przemysław

doi:10.1016/j.knosys.2022.110234

Cited by 40 publications

(29 citation statements)

References 41 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…59 In comparison to the existing explainable methods for survival prediction models using LIME, 18,19,35 the presented approach allows for obtaining time-specific explanations and does not rely on a restrictive proportional hazards explanation model. Compared to SurvSHAP(t), 20 we also extend an existing explainer for classification to capture time-dependent effects in a survival model. However, we take advantage of LIME's fast estimation and its additive interpretability on the probability scale.…”

Section: Discussionmentioning

confidence: 99%

“…SurvSHAP(t) is a SHAP-based explainer that allows for predictors to have a time-varying explanation. 20 The methodology extends SHAP to survival models by estimating an attribution for each predictor at each time point. Thus, each time-specific SHAP value describes the time-dependent influence of the predictor on the prediction.…”

Section: Explainable Machine Learning and Survivalmentioning

confidence: 99%

“…The consideration of a time‐dependent explanation is novel, and to our knowledge is only explored in one other recent work. SurvSHAP(t) is a SHAP‐based explainer that allows for predictors to have a time‐varying explanation 20 . The methodology extends SHAP to survival models by estimating an attribution for each predictor at each time point.…”

Section: Rationale and Relevant Literaturementioning

confidence: 99%

“…18,19 However, these methods assume that the covariate effects satisfy the proportional hazards assumption, which may not be reflective of the behavior of the predictive model. In recent literature, 20 an extension of SHAP has been presented that can capture time-varying covariate effects and takes advantage of its properties of local accuracy and consistency. However, this approach is limited by the interpretability of SHAP values and its long computation time.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Model‐agnostic explanations for survival prediction models

Suresh,

Görg,

Ghosh

2024

Statistics in Medicine

View full text Add to dashboard Cite

Advanced machine learning methods capable of capturing complex and nonlinear relationships can be used in biomedical research to accurately predict time‐to‐event outcomes. However, these methods have been criticized as “black boxes” that are not interpretable and thus are difficult to trust in making important clinical decisions. Explainable machine learning proposes the use of model‐agnostic explainers that can be applied to predictions from any complex model. These explainers describe how a patient's characteristics are contributing to their prediction, and thus provide insight into how the model is arriving at that prediction. The specific application of these explainers to survival prediction models can be used to obtain explanations for (i) survival predictions at particular follow‐up times, and (ii) a patient's overall predicted survival curve. Here, we present a model‐agnostic approach for obtaining these explanations from any survival prediction model. We extend the local interpretable model‐agnostic explainer framework for classification outcomes to survival prediction models. Using simulated data, we assess the performance of the proposed approaches under various settings. We illustrate application of the new methodology using prostate cancer data.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Explainable Machine Learning and Survivalmentioning

confidence: 99%

Section: Rationale and Relevant Literaturementioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Model‐agnostic explanations for survival prediction models

Suresh,

Görg,

Ghosh

2024

Statistics in Medicine

View full text Add to dashboard Cite

show abstract

“…Shapley values explain the extent to which each variable affects the model output relative to the baseline average. We used SurvSHAP(t) ( 26 ), which is capable of providing model explanations in the form of survival function rather than a single point or aggregation ( 27 ), to make time-dependent explanations for our models.…”

Section: Methodsmentioning

confidence: 99%

Individualized survival prediction and surgery recommendation for patients with glioblastoma

Zhu,

Wang,

Jing

et al. 2024

Front. Med.

View full text Add to dashboard Cite

BackgroundThere is a lack of individualized evidence on surgical choices for glioblastoma (GBM) patients.AimThis study aimed to make individualized treatment recommendations for patients with GBM and to determine the importance of demographic and tumor characteristic variables in the selection of extent of resection.MethodsWe proposed Balanced Decision Ensembles (BDE) to make survival predictions and individualized treatment recommendations. We developed several DL models to counterfactually predict the individual treatment effect (ITE) of patients with GBM. We divided the patients into the recommended (Rec.) and anti-recommended groups based on whether their actual treatment was consistent with the model recommendation.ResultsThe BDE achieved the best recommendation effects (difference in restricted mean survival time (dRMST): 5.90; 95% confidence interval (CI), 4.40–7.39; hazard ratio (HR): 0.71; 95% CI, 0.65–0.77), followed by BITES and DeepSurv. Inverse probability treatment weighting (IPTW)-adjusted HR, IPTW-adjusted OR, natural direct effect, and control direct effect demonstrated better survival outcomes of the Rec. group.ConclusionThe ITE calculation method is crucial, as it may result in better or worse recommendations. Furthermore, the significant protective effects of machine recommendations on survival time and mortality indicate the superiority of the model for application in patients with GBM. Overall, the model identifies patients with tumors located in the right and left frontal and middle temporal lobes, as well as those with larger tumor sizes, as optimal candidates for SpTR.

show abstract

Personalized surgical recommendations and quantitative therapeutic insights for patients with metastatic breast cancer: Insights from deep learning

Zhu,

Zhang,

Wang

et al. 2024

Cancer Innovation

View full text Add to dashboard Cite

BackgroundThe role of surgery in metastatic breast cancer (MBC) is currently controversial. Several novel statistical and deep learning (DL) methods promise to infer the suitability of surgery at the individual level.ObjectiveThe objective of this study was to identify the most applicable DL model for determining patients with MBC who could benefit from surgery and the type of surgery required.MethodsWe introduced the deep survival regression with mixture effects (DSME), a semi‐parametric DL model integrating three causal inference methods. Six models were trained to make individualized treatment recommendations. Patients who received treatments in line with the DL models' recommendations were compared with those who underwent treatments divergent from the recommendations. Inverse probability weighting (IPW) was used to minimize bias. The effects of various features on surgery selection were visualized and quantified using multivariate linear regression and causal inference.ResultsIn total, 5269 female patients with MBC were included. DSME was an independent protective factor, outperforming other models in recommending surgery (IPW‐adjusted hazard ratio [HR] = 0.39, 95% confidence interval [CI]: 0.19–0.78) and type of surgery (IPW‐adjusted HR = 0.66, 95% CI: 0.48–0.93). DSME was superior to other models and traditional guidelines, suggesting a higher proportion of patients benefiting from surgery, especially breast‐conserving surgery. The debiased effect of patient characteristics, including age, tumor size, metastatic sites, lymph node status, and breast cancer subtypes, on surgery decision was also quantified.ConclusionsOur findings suggested that DSME could effectively identify patients with MBC likely to benefit from surgery and the specific type of surgery needed. This method can facilitate the development of efficient, reliable treatment recommendation systems and provide quantifiable evidence for decision‐making.

show abstract

SurvSHAP(t): Time-dependent explanations of machine learning survival models

Cited by 40 publications

References 41 publications

Model‐agnostic explanations for survival prediction models

Model‐agnostic explanations for survival prediction models

Individualized survival prediction and surgery recommendation for patients with glioblastoma

Personalized surgical recommendations and quantitative therapeutic insights for patients with metastatic breast cancer: Insights from deep learning

Contact Info

Product

Resources

About