Abstract:Clinical data from hospital admissions are typically utilized to determine the prognostic capacity of Coronavirus disease 2019 (COVID-19) indices. However, as disease status and severity markers evolve over time, time-dependent receiver operating characteristic (ROC) curve analysis becomes more appropriate. The present analysis assessed predictive power for death at various time points throughout patient hospitalization. In a cohort study involving 515 hospitalized patients (General Hospital Number 1 of Mexican Social Security Institute, Colima, Mexico from February 2021 to December 2022) with COVID-19, seven severity indices [Pneumonia Severity Index (PSI) PaO2/FiO2 arterial oxygen pressure/fraction of inspired oxygen (Kirby index), the Critical Illness Risk Score (COVID-GRAM), the National Early Warning Score 2 (NEWS-2), the quick Sequential Organ Failure Assessment score (qSOFA), the Fibrosis-4 index (FIB-4) and the Viral Pneumonia Mortality Score (MuLBSTA were evaluated using time-dependent ROC curves. Clinical data were collected at admission and at 2, 4, 6 and 8 days into hospitalization. The study calculated the area under the curve (AUC), sensitivity, specificity, and predictive values for each index at these time points. Mortality was 43.9%. Throughout all time points, NEWS-2 demonstrated the highest predictive power for mortality, as indicated by its AUC values. PSI and COVID-GRAM followed, with predictive power increasing as hospitalization duration progressed. Additionally, NEWS-2 exhibited the highest sensitivity (>96% in all periods) but showed low specificity, which increased from 22.9% at admission to 58.1% by day 8. PSI displayed good predictive capacity from admission to day 6 and excellent predictive power at day 8 and its sensitivity remained >80% throughout all periods, with moderate specificity (70.6-77.3%). COVID-GRAM demonstrated good predictive capacity across all periods, with high sensitivity (84.2-87.3%) but low-to-moderate specificity (61.5-67.6%). The qSOFA index initially had poor predictive power upon admission but improved after 4 days. FIB-4 had a statistically significant predictive capacity in all periods (P=0.001), but with limited clinical value (AUC, 0.639-0.698), and with low sensitivity and specificity. MuLBSTA and IKIRBY exhibited low predictive power at admission and no power after 6 days. In conclusion, in COVID-19 patients with high mortality rates, NEWS-2 and PSI consistently exhibited predictive power for death during hospital stay, with PSI demonstrating the best balance between sensitivity and specificity.

Length-of-stay times in hospital for COVID-19 patients using the smoothed Beran's estimator with bootstrap bandwidth selection

Estimating lengths-of-stay of hospitalised COVID-19 patients using a non-parametric model: a case study in Galicia (Spain)

An Extended Weibull Regression for Censored Data: Application for COVID-19 in Campinas, Brazil

Learning accurate personalized survival models for predicting hospital discharge and mortality of COVID-19 patients

Hospital length of stay throughout bed pathways and factors affecting this time: A non-concurrent cohort study of Colombia COVID-19 patients and an unCoVer network project

Biased and unbiased estimation of the average length of stay in intensive care units in the Covid-19 pandemic

Real-time analysis of hospital length of stay in a mixed SARS-CoV-2 Omicron and Delta epidemic in New South Wales, Australia

Tracing In-Hospital COVID-19 Outcomes: A Multistate Model Exploration (TRACE)

Nonparametric conditional survival function estimation and plug-in bandwidth selection with multiple covariates

An statistical analysis of COVID-19 intensive care unit bed occupancy data

Personalized Prediction of Hospital Mortality in COVID-19-Positive Patients

Predicting special care during the COVID-19 pandemic: A machine learning approach

Predictors of mortality in hospitalised patients with COVID-19: a 1-year case-control study

Time‑dependent ROC curve analysis to determine the predictive capacity of seven clinical scales for mortality in patients with COVID‑19: Study of a hospital cohort with very high mortality

Personalized Assessment of Mortality Risk and Hospital Stay Duration in Hospitalized Patients with COVID-19 Treated with Remdesivir: A Machine Learning Approach

Continual Deterioration Prediction for Hospitalized COVID-19 Patients

Development of a Model for Predicting Mortality Among Patients Hospitalized with COVID-19 During Their Stay in a Clinical Centre

Seasonal and Periodic Patterns in US COVID-19 Mortality using the Variable Bandpass Periodic Block Bootstrap

Modelling hospital length of stay using convolutive mixtures distributions

Smoothed bootstrap for right-censored data

Machine-Learning Approaches in COVID-19 Survival Analysis and Discharge-Time Likelihood Prediction Using Clinical Data