1169. Derivation And Validation of an International Clinical Prognostication Model for 28-day Sepsis Mortality.

Abstract

Abstract Background Survival prediction models have largely been derived and validated only in high-resource Western countries or in single center studies. We sought to create a prediction model for 28-day mortality using laboratory and physiologic parameters from 3 international sepsis cohorts and externally validated the model. Methods During 2014 to 2021, adult hospitalized patients with suspected infection were enrolled in Durham, United States (N=180) and those with suspected infection and ≥2 SIRS (Systemic Inflammatory Response Syndrome) criteria in Takeo, Cambodia (N=200), and Kumasi, Ghana (N=187). Twenty-five clinical laboratory and physiologic parameters were candidate covariates and sepsis screening scores included as comparators. First, bivariate Cox regression models were performed to determine risk of individual parameters. Then, a 10-fold cross-validated forward stepwise model selection technique was used to eliminate nonsignificant variables using a p-value < 0.10 and the cross-validated C-statistic was estimated. Lastly, this model was applied to an external cohort of hospitalized adults with suspected infection and ≥2 SIRS in Fort Portal, Uganda (N=331 with 9.3% 28-day mortality). Results Among 567 participants, overall mortality was 16.4% at 28-days. Mortality rate highest in Ghana (31.0%), followed by Cambodia (11.0%) and the United States (7.2%). Bivariate analyses identified hypernatremia ( >145 mEq/L) being associated with the highest risk of death (hazard ratio: 7.42; 95% CI: 3.65 to 15.10; Figure 1). On multivariable analysis, a 28-day mortality model including mean arterial pressure, Glasgow Coma score, blood sodium, lactate, and blood urea nitrogen (Table 1) resulted in a 10-fold cross-validated C-statistic of 0.80 (95% CI: 0.61 to 0.88). This model predicted mortality accurately in the validation cohort with a C-statistic of 0.74 (95%CI: 0.69 to 0.79). Figure 1.Forest plot for bivariate analyses for one month survival across United States, Cambodia, and Ghana cohorts. Conclusion Hypotension, altered mental status, serum sodium, serum BUN, and plasma lactate accurately identified risk of death by 28-days among those with suspected sepsis in 3 international derivation cohorts and in a validation cohort in Uganda. Our findings emphasize the importance of clinical laboratory results for sepsis risk stratification. Disclosures Ephraim L. Tsalik, MD PhD, Danaher Diagnostics, Predigen, and Biomeme: In the past 3 years, I have had held equity and consulted for Predigen and Biomeme. Currently, I am an employee of Danaher Diagnostics. Christopher W. Woods, MD MPH, Predigen, Inc: Co-founder.