Evaluating Random Forests for Survival Analysis Using Prediction Error Curves

Mogensen, Ulla B; Ishwaran, Hemant; Gerds, Thomas Alexander

doi:10.18637/jss.v050.i11

Cited by 389 publications

(309 citation statements)

References 29 publications

(45 reference statements)

Supporting

Mentioning

302

Contrasting

Unclassified

Order By: Relevance

“…We evaluated the performance of the models obtained by following different strategies to combine clinical and molecular data in terms of Brier score (Graf et al, 1999), a time-dependent quadratic score for survival data. The results are summarized in the so called "prediction error curves" using the R packages pec (Mogensen et al, 2012). It is worth noting that all models are trained in the training set and their performance evaluated in the test set only.…”

Section: Colon Cancer Datamentioning

confidence: 99%

Boosting in Cox regression: a comparison between the likelihood-based and the model-based approaches with focus on the R-packages CoxBoost and mboost

Bin

2016

Comput Stat

View full text Add to dashboard Cite

Boosting in Cox regression: a comparison between the likelihood-based and the model-based approaches with focus on the R-packages CoxBoost and mboostAbstract Despite the limitations imposed by the proportional hazards assumption, the Cox model is probably the most popular statistical tool used to analyze survival data, thanks to its flexibility and ease of interpretation. For this reason, novel statistical/machine learning techniques are usually adapted to fit it, including boosting, an iterative technique originally developed in the machine learning community and later extended to the statistical field. The popularity of boosting has been further driven by the availability of user-friendly software such as the R packages mboost and CoxBoost, both of which allow the implementation of boosting in conjunction with the Cox model. Despite the common underlying boosting principles, these two packages use different techniques: the former is an adaption of the model-based boosting, while the latter adapts the likelihood-based boosting. Here we contrast these two boosting techniques as implemented in the R packages from an analytic point of view, and we examine the solutions there adopted to treat mandatory variables, i.e. variables that for some reasons must be included in the model. We explore the possibility of extending solutions currently only implemented in one package to the other. We illustrate the usefulness of these extensions through the application to two real data examples.

show abstract

Section: Colon Cancer Datamentioning

confidence: 99%

Boosting in Cox regression: a comparison between the likelihood-based and the model-based approaches with focus on the R-packages CoxBoost and mboost

Bin

2016

Comput Stat

View full text Add to dashboard Cite

show abstract

“…Statistical analyses were performed by R 3.0 with survey, mstate, rms, clinfun, and pec packages. [30][31][32][33][34][35][36] …”

Section: Statistical Analysesmentioning

confidence: 99%

Self-Reported Health Status Predicts Other-Cause Mortality in Men with Localized Prostate Cancer: Results from the Prostate Cancer Outcomes Study

et al. 2015

View full text Add to dashboard Cite

BACKGROUND: Guidelines recommend against treating localized prostate cancer (PCa) in men with a greater than10-year life expectancy. However, physicians have difficulty accurately estimating life expectancy. OBJECTIVE: We used data from a population-based observational study to develop a nomogram to estimate long-term other-cause mortality based on self-reported health status (SRHS), race/ethnicity, and age at diagnosis. DESIGN: This was an observational study. SUBJECTS: Men diagnosed with localized PCa from October 1994 through October 1995 participated in the study. MAIN MEASURES: Initial measures obtained 6 months after diagnosis included sociodemographic and tumor characteristics, treatment, and a single item on the SRHS, with response options ranging from excellent to poor. We used Surveillance, Epidemiology, and End-Results program data to determine date and cause of death through December 2010. We estimated other-cause mortality with proportional hazards survival analyses, accounting for competing risks. KEY RESULTS: We evaluated 2,695 men, of whom 74 % underwent aggressive therapy (surgery or radiotherapy). At the initial survey, 18 % reported excellent (E), 36 % very good (VG), 31 % good (G), and 15 % fair/poor (F/P) health. Healthier men were younger, and more likely to be white, better educated, and to undergo surgery. At follow-up, 44 % of the cohort had died; 78 % of deaths were from causes other than PCa. SRHS predicted other-cause mortality; for men reporting E, VG, G, F/P health, the cumulative incidences of other-cause mortality were 20 %, 29 %, 40 %, and 53 %, respectively, p<0.001. Compared to a reference of excellent SRHS, multivariable hazard ratios (95 % CI) for other-cause mortality for men reporting VG, G, and F/P health were 1.22 (0

show abstract

“…An alternative implementation of the time-dependent Brier score for assessing the prognostic performance of prognostic models for time-to-event endpoints can be found in the package pec (Mogensen, Ishwaran, and Gerds 2012). The basic approach is similar to peperr, i.e., one has to define a wrapper for each fitting procedure in order to determine the estimated survival probabilities.…”

Section: Prediction Error Curvesmentioning

confidence: 99%

c060: Extended Inference with Lasso and Elastic-Net Regularized Cox and Generalized Linear Models

Sill¹,

Hielscher²,

Becker³

et al. 2014

J. Stat. Soft.

View full text Add to dashboard Cite

We have developed the R package c060 with the aim of improving R software functionality for high-dimensional risk prediction modeling, e.g., for prognostic modeling of survival data using high-throughput genomic data. Penalized regression models provide a statistically appealing way of building risk prediction models from high-dimensional data. The popular CRAN package glmnet implements an efficient algorithm for fitting penalized Cox and generalized linear models. However, in practical applications the data analysis will typically not stop at the point where the model has been fitted. One is for example often interested in the stability of selected features and in assessing the prediction performance of a model and we provide functions to deal with both of these tasks. Our R functions are computationally efficient and offer the possibility of speeding up computing time through parallel computing. Another feature which can drastically reduce computing time is an efficient interval-search algorithm, which we have implemented for selecting the optimal parameter combination for elastic net penalties. These functions have been useful in our daily work at the Biostatistics department (C060) of the German Cancer Research Center where prognostic modeling of patient survival data is of particular interest. Although we focus on a survival data application of penalized Cox models in this article, the functions in our R package are in general applicable to all types of regression models implemented in the glmnet package, with the exception of prediction error curves, which are specific to time-to-event data.

show abstract

Evaluating Random Forests for Survival Analysis Using Prediction Error Curves

Cited by 389 publications

References 29 publications

Boosting in Cox regression: a comparison between the likelihood-based and the model-based approaches with focus on the R-packages CoxBoost and mboost

Boosting in Cox regression: a comparison between the likelihood-based and the model-based approaches with focus on the R-packages CoxBoost and mboost

Self-Reported Health Status Predicts Other-Cause Mortality in Men with Localized Prostate Cancer: Results from the Prostate Cancer Outcomes Study

c060: Extended Inference with Lasso and Elastic-Net Regularized Cox and Generalized Linear Models

Contact Info

Product

Resources

About