DeepSurv: personalized treatment recommender system using a Cox proportional hazards deep neural network

By Jared Katzman, Uri Shaham, A. Cloninger, Jonathan Bates, Tingting Jiang, Yuval Kluger. 2016

1,801 citations recorded by the source corpus.

BackgroundMedical practitioners use survival models to explore and understand the relationships between patients’ covariates (e.g. clinical and genetic features) and the effectiveness of various treatment options. Standard survival models like the linear Cox proportional hazards model require extensive feature engineering or prior medical knowledge to model treatment interaction at an individual level. While nonlinear survival methods, such as neural networks and survival forests, can inherently model these high-level interaction terms, they have yet to be shown as effective treatment recommender systems.MethodsWe introduce DeepSurv, a Cox proportional hazards deep neural network and state-of-the-art survival method for modeling interactions between a patient’s covariates and treatment effectiveness in order to provide personalized treatment recommendations.ResultsWe perform a number of experiments training DeepSurv on simulated and real survival data. We demonstrate that DeepSurv performs as well as or better than other state-of-the-art survival models and validate that DeepSurv successfully models increasingly complex relationships between a patient’s covariates and their risk of failure. We then show how DeepSurv models the relationship between a patient’s features and effectiveness of different treatment options to show how DeepSurv can be used to provide individual treatment recommendations. Finally, we train DeepSurv on real clinical studies to demonstrate how it’s personalized treatment recommendations would increase the survival time of a set of patients.ConclusionsThe predictive and modeling capabilities of DeepSurv will enable medical researchers to use deep neural networks as a tool in their exploration, understanding, and prediction of the effects of a patient’s characteristics on their risk of failure.

Subfields: Artificial Intelligence, Statistics and Probability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Statistics and Probability

Citations in the corpus

An overview of deep learning in medical imaging focusing on MRI (2018, background): “seizures [17] and cardiac arrests [18], to computer-aided detection [19] and diagnosis [20] supporting clinical decision making and survival analysis [21], to drug discovery [22] and as an aid in therapy selection and pharmacogenomics [23], to increased operational efficiency [24], stratified care d”
Knowledge Graphs: Opportunities and Challenges (2023, background): “Therefore, medical recommender systems, especially biomedical knowledge graph-based recommender systems (such as doctor recommender systems and medicine recommender systems), have been put forward to deal with this issue (Katzman et al. 2018).”
The Application of Deep Learning in Cancer Prognosis Prediction (2020, methodology): “DeepSurv is a feed forward NN that uses patient’s clinical data as input and applied dropout, learning rate decay, regularization, and other commonly used hyperparameters to optimize for different datasets [60].”
Predicting cancer outcomes from histology and genomics using convolutional networks (2017, methodology): “More advanced “deep” neural networks that are composed of many layers were recently adapted to optimize Cox proportional hazard likelihood and were shown to have equal or superior performance in predicting survival using genomic profiles containing hundreds to tens of thousands of features (29, 30) ”
Multimodal Co-Attention Transformer for Survival Prediction in Gigapixel Whole Slide Images (2021, methodology): “Our goal is to develop a set-based neural network architecture F that integrates {Wj}Kj=1 and gattr to estimate the hazard function fhazard(T = t | T t,X) 2 [0, 1], which measure the probability of the patient surviving after time point t (typically implemented as a Sigmoid activation after the last”
Time-to-Event Prediction with Neural Networks and Cox Regression (2019, methodology): “Our work distinguishes itself from this in the following way: The method by Katzman et al. (2018), denoted DeepSurv, is constrained by the proportionality assumption of the Cox model, while we propose an extension of the Cox model where proportionality is no longer a restriction.”
Explainable AI for Trees: From Local Explanations to Global Understanding (2019, methodology): “Neural Network - The neural network model was implemented using the DeepSurv Python package [30].”
Disease-Atlas: Navigating Disease Trajectories using Deep Learning (2018, background): “Deep learning has seen increasing use in medical applications, with successes in traditional survival analysis (Ranganath et al., 2016; Luck et al., 2017) survival analysis with competing risks (Lee et al., 2018; Alaa and van der Schaar, 2017) and treatment recommendations (Katzman et al., 2016).”
Deep neural network models for computational histopathology: A survey (2019, methodology): “A proportional hazards model is commonly used to model an individuals survival and can be implemented using a neural network (Katzman et al., 2018) and several groups have used this approach in digital pathology (Zhu et al.”
Machine Learning for Survival Analysis (2019, background): “Several deep survival analysis approaches (Katzman et al. 2016; Ranganath et al. 2016) are also proposed to assist physicians with their clinical decisions on patients by estimating their risk of diseases and providing personalized treatment recommendations.”
Language Models are Few-shot Learners for Prognostic Prediction (2023, methodology): “More speciﬁ-cally, in our experiments, we adopt CrossEntropy loss for BOR and CoxPH loss for OS and PFS prediction following DeepSurv (Katzman et al., 2018).”
SurvNAM: The machine learning survival model explanation (2021, methodology): “Many models have been developed to relax the linear relationship assumption accepted in the Cox model [57, 58, 59, 60, 61, 62].”
Deep learning in cancer diagnosis, prognosis and treatment selection (2021, methodology): “Prognosis prediction MLP Cox-nnet [59], DeepSurv [60], RankedDeepSurv [61] Cox regression used as the last layer of MLP models for prognosis prediction”