Derivation of a Contextually-Appropriate COVID-19 Mortality Scale for Low-Resource Settings

Abstract

Background: In many low- and middle-income countries, where vaccinations will be delayed and healthcare systems are underdeveloped, the COVID-19 pandemic will continue for the foreseeable future. Mortality scales can aid frontline providers in low-resource settings (LRS) in identifying those at greatest risk of death so that limited resources can be directed towards those in greatest need and unnecessary loss of life is prevented. While many prognostication tools have been developed for, or applied to, COVID-19 patients, no tools to date have been purpose-designed for, and validated in, LRS. Objectives: This study aimed to develop a pragmatic tool to assist LRS frontline providers in evaluating in-hospital mortality risk using only easy-to-obtain demographic and clinical inputs. Methods: Machine learning was used on data from a retrospective cohort of Sudanese COVID-19 patients at two government referral hospitals to derive contextually appropriate mortality indices for COVID-19, which were then assessed by C-indices. Findings: Data from 467 patients were used to derive two versions of the AFEM COVID-19 Mortality Scale (AFEM-CMS), which evaluates in-hospital mortality risk using demographic and clinical inputs that are readily obtainable in hospital receiving areas. Both versions of the tool include age, sex, number of comorbidities, Glasgow Coma Scale, respiratory rate, and systolic blood pressure; in settings 'with' pulse oximetry, oxygen saturation is included and in settings 'without' access, heart rate is included. The AFEM-CMS showed good discrimination: the model including pulse oximetry had a C-statistic of 0.775 (95% CI: 0.737–0.813) and the model excluding it had a C-statistic of 0.719 (95% CI: 0.678–0.760). Conclusions: In the face of an enduring pandemic in many LRS, the AFEM-CMS serves as a practical solution to aid frontline providers in effectively allocating healthcare resources. The tool’s generalisability is likely narrow outside of similar extremely LRS settings, and further validation studies are essential prior to broader use.