Novel laboratory-based risk scores predicts mortality in patients with COVID-19

Although many clinical risk scores exist internationally, few rely solely on routine clinical laboratory markers which is a rapid way for physicians to assess clinical conditions. Several studies have highlighted that clinical laboratory test abnormalities are associated with poor disease outcomes in COVID-19 patients. In the current study, the authors aimed to identify which commonly ordered blood tests were most strongly associated with mortality and to build a risk score that could be applied in hospital settings.

Researchers analyzed 33 clinical biochemical and hematological markers collected as part of routine clinical care from 324 adults hospitalized with COVID‑19. The study collected admission data and spanned multiple hospitals in Ontario. Logistic regression models were used to evaluate the association between laboratory markers and mortality, adjusting for age and sex. After considering data completeness, as well as statistical associations to mortality, six markers (creatinine, sodium, lactate, bicarbonate, base excess and pH) were identified. Base excess was removed due to it’s clinical correlation with bicarbonate. The five remaining markers entered into multivariable modeling. From this, two risk score models were constructed: one using single-marker associations and one using a multivariable approach. The cohort was randomly split into training and validation subsets to evaluate predictive performance.

Univariable findings showed elevated creatinine, WBC and lactate, as well as low pH, bicarbonate and base excess at admission were linked to higher odds of mortality. Both univariable and multivariable risk score models demonstrated similar predictive ability, achieving area‑under‑the‑curve values of 0.80-0.83 in training and validation cohorts, with higher sensitivity than specificity.

These results suggest that acid–base disturbances may play an important role in early COVID‑19 mortality risk and can be captured using routine laboratory tests. While the scoring models showed promising accuracy, limitations included inconsistent laboratory testing across institutions and small sample sizes for some analytes, restricting generalizability. The authors emphasize the need for validation in larger and external populations and propose future integration of such risk scores into electronic medical record systems to support rapid triage and clinical decision‑making in urgent care settings.