B0296
Title: Smoothed Beran's estimator with bootstrap bandwidths: Application to COVID-19 hospital length-of-stay
Authors: Rebeca Pelaez - Universidade da Coruna (Spain) [presenting]
Ricardo Cao - Universidade da Coruna (Spain)
Juan Vilar Fernandez - Universidade da Coruna (Spain)
Abstract: In the biomedical field, estimating the probability of a patient surviving beyond a specified time t, given a covariate x, is a significant problem. This involves estimating the conditional survival function, $S(t|x)$. In many cases, the time variable is randomly censored, meaning that some survival times are unknown because the study concludes before all individuals have experienced the event. The most commonly used nonparametric estimator of $S(t|x)$ under censoring was introduced by a prior study. This and other usual estimators in the literature are based on covariate smoothing. In a recent study, the smoothed Beran's estimator, smoothed also in the time variable, is proposed. Asymptotic theory was proved and simulation studies showed its good performance. A resampling technique to approximate them is proposed. The approach combines the obvious bootstrap and the smoothed bootstrap for the covariate and the time variables. The construction of bootstrap-based confidence regions is also addressed. Simulation studies show a reasonable behaviour of the proposals. They are applied to obtain nonparametric estimations and confidence regions of the conditional survival function of length-of-stay in hospitals for COVID-19 patients. The study leads to deeper insights into differences in hospitalised virus patients based on their age, sex and pre-existing conditions such as obesity or COPD.
