A Prospective Cohort Study Investigating the Prognostic Performance and Utility of 3 Severity of Illness Scores

Background: Mortality prediction is difficult in resource-constrained settings. Severity of illness scores have not been tested in hypoxemic adults in Africa. Research Question: How well do 3 severity of illness scores (Modified Early Warning Score [MEWS], quick Sequential Organ Failure Assessment [qSOFA], and Universal Vital Assessment [UVA]) and the ability to walk predict mortality in hypoxemic hospitalized adults in Africa? Study Design and Methods: All adults with hypoxemia on admission in 5 hospitals in Kenya, Malawi, and Rwanda between November 2022 and April 2023 were prospectively enrolled in the study. The capability of MEWS, qSOFA, UVA, and the ability to walk was evaluated to predict hospital mortality. In exploratory analyses, differences in disease severity and mortality were compared between sites. Results: A total of 24,724 admissions were screened; 1,732 of these were hypoxemic and had complete outcomes data. Median age was 52 years (interquartile range, 36-70 years), and hospital mortality was 35% (n = 615). Sites varied in the completeness of score variables (44%-99%). Increased odds of mortality were found using predefined thresholds for each score; UVA predicted best, with an OR of 3.40 (95% CI, 2.54-4.56). The area under the receiver-operating curves for MEWS, qSOFA, and UVA were 0.66 (95% CI, 0.62-0.69), 0.66 (95% CI, 0.63-0.69), and 0.69 (95% CI, 0.65-0.72), respectively, using complete case analysis; they were 0.61 (95% CI, 0.58-0.64), 0.65 (95% CI, 0.62-0.67), and 0.66 (95% CI, 0.64-0.69) with missing data imputed as normal. Inability to walk independently was also predictive of mortality (OR, 2.26; 95% CI, 1.62-3.15). UVA pairwise comparisons showed different mortality between 4 of 6 sites; these differences remained significant in 2 comparisons when adjusting for illness severity. Interpretation: In the largest prospective cohort of hypoxemic adults in Africa to date, MEWS, qSOFA, UVA, and ability to walk on admission had modest capability to predict hospital death. Missing data were common. Imputation of missing variables only slightly altered performance, and thus it is possible that scores could be simplified. UVA had the best predictive performance and may be cautiously used to aid clinical decision-making, quality improvement, research comparisons, and risk adjustment.