Predicting the need for escalation of care or death from repeated daily clinical observations and laboratory results in patients with SARS-CoV-2 during 2020: a retrospective population-based cohort study from the United Kingdom

Abstract

Objectives: Currently used prognostic tools for patients with SARS-CoV-2 infection are based on clinical and laboratory parameters measured at a single point in time, usually on admission. We aimed to determine how dynamic changes in clinical and laboratory parameters relate to SARS-CoV-2 prognosis. Design: retrospective, observational cohort study using routinely collected clinical data to model the dynamic change in prognosis of SARS-CoV-2. Setting: a single, large hospital in England. Participants: all patients with confirmed SARS-CoV-2 admitted to Nottingham University Hospitals (NUH) NHS Trust, UK from 1st February 2020 until 30th November 2020. Main outcome measures: Intensive Care Unit (ICU) admission, death and discharge from hospital. Statistical Methods: We split patients into 1st (admissions until 30th June) and 2nd (admissions thereafter) waves. We incorporated all clinical observations, blood tests and other covariates from electronic patient records and follow up until death or 30 days from the point of hospital discharge. We modelled daily risk of admission to ICU or death with a time varying Cox proportional hazards model. Results: 2,964 patients with confirmed SARS-CoV-2 were included. Of 1,374 admitted during the 1st wave, 593 were eligible for ICU escalation, and 466 had near complete ascertainment of all covariates at admission. Our validation sample included 1,590 confirmed cases, of whom 958 were eligible for ICU admission. Our model had good discrimination of daily need for ICU admission or death (C statistic = 0.87 (IQR 0.85-0.90)) and predicted this daily prognosis better than previously published scores (NEWS2, ISCARIC 4C). In validation in the 2nd wave the score overestimated escalation (calibration slope 0.55), whilst retaining a linear relationship and good discrimination (C statistic = 0.88 (95% CI 0.81 -0.95)). Conclusions: A bespoke SARS-CoV-2 escalation risk prediction score can predict need for clinical escalation better than a generic early warning score or a single estimation of risk at admission.