2021/10/21 by Annabel L. Davies, Davies, Annabel L., A C C Coolen +3
Computer Science · Mathematics · Social Sciences · #Bayesian Methods and Mixture Models #FOS: Computer and information sciences #Insurance, Mortality, Demography, Risk Management #Methodology (stat.ME) #Statistical Methods and Bayesian Inference #Statistical Methods and Inference
paper · pdf · doi:10.48550/arxiv.2110.11196
openalex publication_date 2021/10/21 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Predicting patient survival probabilities based on observed covariates is an\nimportant assessment in clinical practice. These patient-specific covariates\nare often measured over multiple follow-up appointments. It is then of interest\nto predict survival based on the history of these longitudinal measurements,\nand to update predictions as more observations become available. The standard\napproaches to these so-called `dynamic prediction' assessments are joint models\nand landmark analysis. Joint models involve high-dimensional parametrisations,\nand their computational complexity often prohibits including multiple\nlongitudinal covariates. Landmark analysis is simpler, but discards a\nproportion of the available data at each `landmark time'. In this work we\npropose a `retarded kernel' approach to dynamic prediction that sits somewhere\nin between the two standard methods in terms of complexity. By conditioning\nhazard rates directly on the covariate measurements over the observation time\nframe, we define a model that takes into account the full history of covariate\nmeasurements but is more practical and parsimonious than joint modelling.\nTime-dependent association kernels describe the impact of covariate changes at\nearlier times on the patient's hazard rate at later times. Under the\nconstraints that our model (i) reduces to the standard Cox model for\ntime-independent covariates, and (ii) contains the instantaneous Cox model as a\nspecial case, we derive two natural kernel parameterisations. Upon application\nto three clinical data sets, we find that the predictive accuracy of the\nretarded kernel approach is comparable to that of the two existing standard\nmethods.\n