Development and validation of a prognostic 40-day mortality risk model among hospitalized patients with COVID-19

Abstract

Objectives: The development of a prognostic mortality risk model for hospitalized COVID-19 patients may facilitate patient treatment planning, comparisons of therapeutic strategies, and public health preparations.

Methods: We retrospectively reviewed the electronic health records of patients hospitalized within a 13-hospital New Jersey USA network between March 1, 2020 and April 22, 2020 with positive polymerase chain reaction results for SARS-CoV-2, with follow-up through May 29, 2020. With death or hospital discharge by day 40 as the primary endpoint, we used univariate followed by stepwise multivariate proportional hazard models to develop a risk score on one-half the data set, validated on the remainder, and converted the risk score into a patient-level predictive probability of 40-day mortality based on the combined dataset.

Results: The study population consisted of 3123 hospitalized COVID-19 patients; median age 63 years; 60% were men; 42% had >3 coexisting conditions. 713 (23%) patients died within 40 days of hospitalization for COVID-19. From 22 potential candidate factors 6 were found to be independent predictors of mortality and were included in the risk score model: age, respiratory rate ≥25/minute upon hospital presentation, oxygenation <94% on hospital presentation, and pre-hospital comorbidities of hypertension, coronary artery disease, or chronic renal disease. The risk score was highly prognostic of mortality in a training set and confirmatory set yielding in the combined dataset a hazard ratio of 1.80 (95% CI, 1.72, 1.87) for one unit increases. Using observed mortality within 20 equally sized bins of risk scores, a predictive model for an individual's 40-day risk of mortality was generated as -14.258 + 13.460*RS + 1.585*(RS-2.524)^2-0.403*(RS-2.524)^3. An online calculator of this 40-day COVID-19 mortality risk score is available at www.HackensackMeridianHealth.org/CovidRS.

Conclusions: A risk score using six variables is able to prognosticate mortality within 40-days of hospitalization for COVID-19.

Trial registration: Clinicaltrials.gov Identifier: NCT04347993.

Conflict of interest statement

The authors declare no conflicts of interest. Authors DAB and SMB are co-owners and NSB is an employee of Berry Consultants, LLC, a company that designs and analyses clinical trials for pharmaceutical and medical device companies, NIH cooperative groups, patient advocacy groups, and international consortia. Berry Consultants received no funding related to this project. This does not alter our adherence to PLOS ONE policies on sharing data and materials.

Figures

Fig 1. Relationship between age cubed and 40-day mortality.

Fig 2. 40-day mortality based on risk score.

40-day mortality among COVID-19 patients in the training and confirmatory sets as divided into bins of 20 patients of ascending mean Risk Scores. Bins 6, 11, and 16 contain 157 patients each and all other bins have 156 patients.

Fig 3. Patient mortality risk by day 40 using a proportional hazards model.

Fig 4. Patient-specific risk of mortality by day 40 using actual mortality in datasets.