The influx of deep learning (DL) techniques into the field of survival analysis in recent years has led to substantial methodological progress; for instance, learning from unstructured or high-dimensional data such as images, text or omics data. In this work, we conduct a comprehensive systematic review of DL-based methods for time-to-event analysis, characterizing them according to both survival- and DL-related attributes. In summary, the reviewed methods often address only a small subset of tasks relevant to time-to-event data—e.g., single-risk right-censored data—and neglect to incorporate more complex settings. Our findings are summarized in an editable, open-source, interactive table:https://survival-org.github.io/DL4Survival. As this research area is advancing rapidly, we encourage community contribution in order to keep this database up to date.

This tutorial article demonstrates how time-to-event data can be modelled in a very flexible way by taking advantage of advanced inference methods that have recently been developed for generalized additive mixed models. In particular, we describe the necessary pre-processing steps for transforming such data into a suitable format and show how a variety of effects, including a smooth nonlinear baseline hazard, and potentially nonlinear and nonlinearly time-varying effects, can be estimated and interpreted. We also present useful graphical tools for model evaluation and interpretation of the estimated effects. Throughout, we demonstrate this approach using various application examples. The article is accompanied by a new R -package called pammtools implementing all of the tools described here.

The use of random forests is increasingly common in genetic association studies. The variable importance measure (VIM) that is automatically calculated as a by-product of the algorithm is often used to rank polymorphisms with respect to their ability to predict the investigated phenotype. Here, we investigate a characteristic of this methodology that may be considered as an important pitfall, namely that common variants are systematically favoured by the widely used Gini VIM. As a consequence, researchers may overlook rare variants that contribute to the missing heritability. The goal of the present article is 3-fold: (i) to assess this effect quantitatively using simulation studies for different types of random forests (classical random forests and conditional inference forests, that employ unbiased variable selection criteria) as well as for different importance measures (Gini and permutation based); (ii) to explore the trees and to compare the behaviour of random forests and the standard logistic regression model in order to understand the statistical mechanisms behind the preference for common variants; and (iii) to summarize these results and previously investigated properties of random forest VIMs in the context of genetic association studies and to make practical recommendations regarding the choice of the random forest and variable importance type. All our analyses can be reproduced using R code available from the companion website: http://www.ibe.med.uni-muenchen.de/organisation/mitarbeiter/020_professuren/boulesteix/ginibias/.

To assess the current dynamics of an epidemic, it is central to collect information on the daily number of newly diseased cases. This is especially important in real‐time surveillance, where the aim is to gain situational awareness, for example, if cases are currently increasing or decreasing. Reporting delays between disease onset and case reporting hamper our ability to understand the dynamics of an epidemic close to now when looking at the number of daily reported cases only. Nowcasting can be used to adjust daily case counts for occurred‐but‐not‐yet‐reported events. Here, we present a novel application of nowcasting to data on the current COVID‐19 pandemic in Bavaria. It is based on a hierarchical Bayesian model that considers changes in the reporting delay distribution over time and associated with the weekday of reporting. Furthermore, we present a way to estimate the effective time‐varying case reproduction number based on predictions of the nowcast. The approaches are based on previously published work, that we considerably extended and adapted to the current task of nowcasting COVID‐19 cases. We provide methodological details of the developed approach, illustrate results based on data of the current pandemic, and evaluate the model based on synthetic and retrospective data on COVID‐19 in Bavaria. Results of our nowcasting are reported to the Bavarian health authority and published on a webpage on a daily basis (https://corona.stat.uni-muenchen.de/). Code and synthetic data for the analysis are available from https://github.com/FelixGuenther/nc_covid19_bavaria and can be used for adaption of our approach to different data.

Background: Proteins are an essential part of medical nutrition therapy in critically ill patients. Guidelines almost universally recommend a high protein intake without robust evidence supporting its use.Methods: Using a large international database, we modelled associations between the hazard rate of in-hospital death and live hospital discharge (competing risks) and three categories of protein intake (low: < 0.8 g/kg per day, standard: 0.8–1.2 g/kg per day, high: > 1.2 g/kg per day) during the first 11 days after ICU admission (acute phase). Time-varying cause-specific hazard ratios (HR) were calculated from piece-wise exponential additive mixed models. We used the estimated model to compare five different hypothetical protein diets (an exclusively low protein diet, a standard protein diet administered early (day 1 to 4) or late (day 5 to 11) after ICU admission, and an early or late high protein diet).Results: Of 21,100 critically ill patients in the database, 16,489 fulfilled inclusion criteria for the analysis. By day 60, 11,360 (68.9%) patients had been discharged from hospital, 4,192 patients (25.4%) had died in hospital, and 937 patients (5.7%) were still hospitalized. Median daily low protein intake was 0.49 g/kg [IQR 0.27–0.66], standard intake 0.99 g/kg [IQR 0.89– 1.09], and high intake 1.41 g/kg [IQR 1.29–1.60]. In comparison with an exclusively low protein diet, a late standard protein diet was associated with a lower hazard of in-hospital death: minimum 0.75 (95% CI 0.64, 0.87), and a higher hazard of live hospital discharge: maximum HR 1.98 (95% CI 1.72, 2.28). Results on hospital discharge, however, were qualitatively changed by a sensitivity analysis. There was no evidence that an early standard or a high protein intake during the acute phase was associated with a further improvement of outcome.Conclusions: Provision of a standard protein intake during the late acute phase may improve outcome compared to an exclusively low protein diet. In unselected critically ill patients, clinical outcome may not be improved by a high protein intake during the acute phase.Study registrationID numberISRCTN17829198