Abstract
AbstractDynamic prediction models provide predicted survival probabilities that can be updated over time for an individual as new measurements become available. Two techniques for dynamic survival prediction with longitudinal data dominate the statistical literature: joint modelling and landmarking. There is substantial interest in the use of machine learning methods for prediction; however, their use in the context of dynamic survival prediction has been limited. We show how landmarking can be combined with a machine learning ensemble—the Super Learner. The ensemble combines predictions from different machine learning and statistical algorithms with the goal of achieving improved performance. The proposed approach exploits discrete time survival analysis techniques to enable the use of machine learning algorithms for binary outcomes. We discuss practical and statistical considerations involved in implementing the ensemble. The methods are illustrated and compared using longitudinal data from the UK Cystic Fibrosis Registry. Standard landmarking and the landmark Super Learner approach resulted in similar cross-validated predictive performance, in this case, outperforming joint modelling.
Highlights
Predictive models for time-to-event outcomes are used widely in medicine to identify individuals at elevated risk, to inform treatment strategies and to update patients about their prognosis
Recent examples of dynamic prediction models focussed on 10-year cardiovascular disease risk based on electronic health records (Paige et al, 2018), survival for people with cystic fibrosis (CF) using registry data (Keogh et al, 2018), intervention-free survival for patients with aortic stenosis based on a cohort (Andrinopoulou et al, 2015) and survival based on breast cancer recurrence data in a French cohort (Lafourcade et al, 2018)
Because most censoring in the UK CF Registry is administrative censoring, the assumption that Ci is independent of both T∗i and the covariates, as required for inverse probability of censoring weighted (IPCW) methods, is reasonable for this dataset
Summary
Predictive models for time-to-event outcomes are used widely in medicine to identify individuals at elevated risk, to inform treatment strategies and to update patients about their prognosis. Joint models are flexible and provide consistent predictions when correctly specified (Jewell & Nielsen, 1993; Rizopoulos et al, 2017), and development of statistical software has made the analysis feasible (Hickey et al, 2016; Philipson et al, 2018; Rizopoulos, 2010, 2016). This approach can be computationally complex, for large datasets (Rizopoulos et al, 2017). Recent papers have compared the two approaches using simulation studies. Rizopoulos et al (2017) demonstrated that joint models tend to outperform landmarking when the effect of time is correctly specified in the longitudinal submodel but as misspecification increases, the differences become smaller and landmarking can even outperform joint modelling but, in contrast, a misspecified association structure between the longitudinal and survival processes did not substantially affect the relative performance of the two analysis methods. Ferrer et al (2018) agreed that joint models outperform landmarking for a correctly specified joint model but noted that landmarking is less sensitive to a misspecified longitudinal process. Suresh et al (2017) found that joint modelling provided better performance than landmarking for an illness-death model but the difference was quite small. Maziarz et al (2017) reported that partly conditional models (landmarking-style models) offered comparable performance to joint models and were more computationally efficient
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callMore From: Journal of the Royal Statistical Society Series A: Statistics in Society
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
