Random survival forests with multivariate longitudinal endogenous covariates

Abstract

Predicting the individual risk of clinical events using the complete patient history is a major challenge in personalized medicine. Analytical methods have to account for a possibly large number of time-dependent predictors, which are often characterized by irregular and error-prone measurements, and are truncated early by the event. In this work, we extended the competing-risk random survival forests to handle such endogenous longitudinal predictors when predicting event probabilities. The method, implemented in the R package DynForest, internally transforms the time-dependent predictors at each node of each tree into time-fixed features (using mixed models) that can then be used as splitting candidates. The final individual event probability is computed as the average of leaf-specific Aalen–Johansen estimators over the trees. Using simulations, we compared the performances of DynForest to accurately predict an event with (i) a joint modeling alternative when considering two longitudinal predictors only, and with (ii) a regression calibration method that ignores the informative truncation by the event when dealing with a large number of longitudinal predictors. Through an application in dementia research, we also illustrated how DynForest can be used to develop a dynamic prediction tool for dementia from multimodal repeated markers, and quantify the importance of each marker.

Dynamic Prediction of Excessive Daytime Sleepiness Through Random Survival Forest: An application of the PPMI data

“…We computed the individual dynamic predictions accounting for multiple longitudinal predictors through the extended random survival forest method in order to deal with time-dependent predictors. The DynForest R package, which is a userfriendly R package and easy-to-use random forest methodology, is utilized to achieve this [46]. Moreover, the importance of the variables using…”

“…Nevertheless, this method has been unable to incorporate time-dependent predictors. Thus, Devaux et al [30] proposed an alternative way to RSF with multivariate longitudinal time-dependent covariates.…”

Reasoning and causal inference regarding surgical options for patients with low‐grade gliomas using machine learning: A SEER‐based study

“…For the T‐learner, we trained two separate five‐layered DeepSurv 39 models on patients who received STR and GTR. We also trained dropouts that met multiple additive regression trees (DART), 45 regularized coxnet (Rcoxnet), 46 cox proportional hazards model (CPH), and random survival forest (RSF) 47 for comparison. All models, except BITES and BSL, were trained and used in terms of T‐learners.…”

“…To estimate the ATE of GTR compared to STR, we used doubly robust learning (DRL) 47 to derive a two‐stage adjusted logistic regression (LR); the odds ratios (OR) obtained was called OR d . First, an LR was used to fit the treatment and response, and another LR was used to predict the response residuals from the treatment residuals, thus making treatment as independent of other covariates as possible.…”