Admission vital signs as predictors of COVID-19 mortality: a retrospective cross-sectional study

Background COVID-19 remains a major healthcare concern. Vital signs are routinely measured on admission and may provide an early, cost-effective indicator of outcome - more so in developing countries where such data is scarce. We sought to describe the association between six routinely measured admission vital signs and COVID-19 mortality, and secondarily to derive potential applications for resource-limited settings. Methods Retrospective analysis of consecutive patients admitted to King Edward VIII Hospital, South Africa, with COVID-19 during June-September 2020 was undertaken. The sample was subdivided into survivors and non-survivors and comparisons made in terms of demographics and admission vital signs. Univariate and multivariate analysis of predictor variables identified associations with in-hospital mortality, with the resulting multivariate regression model evaluated for its predictive ability with receiver operating characteristic (ROC) curve analysis. Results The 236 participants enrolled comprised 153(77.54%) survivors and 53(22.46%) non-survivors. Most participants were Black African(87.71%) and female(59.75%) with a mean age of 53.08(16.96) years. The non-survivor group demonstrated a significantly lower median/mean for admission oxygen saturation (%) [87(78-95) vs. 96(90-98)] and diastolic BP (mmHg) [70.79(14.66) vs. 76.3(12.07)], and higher median for admission respiratory rate (breaths/minute) [24(20-28) vs. 20(20-23)] and glucose (mmol/l) [10.2(6.95-16.25) vs. 7.4(5.5-9.8)]. Age, oxygen saturation, respiratory rate, glucose and diastolic BP were found to be significantly associated with mortality on univariate analysis. A log rank test revealed significantly lower survival rates in patients with an admission oxygen saturation < 90% compared with >= 90% (p = 0.001). Multivariate logistic regression revealed a significant relationship between age and oxygen saturation with in-hospital mortality (OR 1.047; 95% CI 1.016-1.080; p = 0.003 and OR 0.922; 95% CI 0.880-0.965; p = 0.001 respectively). A ROC curve analysis generated an area under the curve (AUC) of 0.778 (p < 0.001) when evaluating the predictive ability of oxygen saturation, respiratory rate, glucose and diastolic BP for in-hospital death. This improved to an AUC of 0.832 (p < 0.001) with the inclusion of age. Conclusion A multivariate regression model comprising admission oxygen saturation, respiratory rate, glucose and diastolic BP (with/without age) demonstrated promising predictive capacity, and may provide a cost-effective means for early prognostication of patients admitted with COVID-19 in resource-limited settings.