Objectives: Sophisticated scores have been proposed for prognostication of mortality due to SARS-CoV-2 but perform inconsistently. We conducted these meta-analyses to uncover why and to pragmatically seek a single dependable biomarker for mortality. Design: We searched the PubMed database for the keywords SARS-CoV-2 with biomarker name and mortality. All studies published from 01st December 2019 to 30th June 2021 were surveyed. To aggregate the data, the meta library in R was used to report overall mean values and 95% confidence intervals. We fitted a random effects model to obtain pooled AUCs and associated 95% confidence intervals for the European/North American, Asian, and overall datasets. Setting and Participants: Data was collected from 131 studies on SARS-CoV-2 PCR-positive general hospital adult admissions (n=76,169 patients in total). Main Outcome Measures: We planned a comparison of pooled area under curves (AUCs) from Receiver Operator Characteristic curves plotted for admission D-dimer, CRP, urea, troponin and interleukin-6 levels. Main Results: Biomarker effectiveness varies significantly in different regions of the world. Admission CRP levels are a good prognostic marker for mortality due to SARS-CoV-2 in Asian countries, with a pooled area under curve (AUC) of 0.83 (95% CI 0.80-0.85), but only an average predictor of mortality in Europe/North America, with a pooled AUC of 0.67 (95% CI 0.63-0.71, P<0.0001). We observed the same pattern for D-dimer and IL-6. This variability explains why the proposed prognostic scores did not perform evenly. Notably, urea and troponin had pooled AUCs ≥ 0.78 regardless of location, implying that end-organ damage at presentation is a key prognostic factor. These differences might be due to age, genetic backgrounds, or different modes of death (younger patients in Asia dying of cytokine storm while older patients die of multi-organ failure). Conclusions: Biomarker effectiveness for prognosticating SARS-CoV-2 mortality varies significantly by geographical location. We propose that biomarkers and by extension prognostic scores need to be tailored for specific populations. This also implies that validation of commonly used prognostic scores for other conditions should occur before they are used in different populations.
