Prediction of 48-h Intensive Care Unit Mortality in Hospitalized Cancer Patients: Performance of the EPIC...

PurposeThe EPIC deterioration index (EDI) and the Sequential Organ Failure Assessment (SOFA) scores are commonly used. However, no studies have examined their performance in hospitalized cancer patients just before intensive care unit (ICU) transfer. We examined the prognostic accuracy of the EDI and SOFA for 48-h ICU mortality.MethodsThis study evaluated consecutive hospitalized medical oncology patients transferred to the ICU from January 1, 2021 to December 31, 2023. Univariable and multivariable logistic regression models examined if EDI scores (EDI-now) and SOFA scores immediately prior to ICU transfer were predictive of 48-h ICU mortality. We assessed the performance of EDI-now, EDI at 24 h before ICU transfer (EDI-24 h), the difference between EDI-24 h and EDI-now (EDI-change) and SOFA using the area under the receiver operating characteristic curve (AUROC).ResultsThere were 1987 hospitalizations among 1907 unique patients (mean age = 61, male = 58%); 236 (11.9%) died within 48 hours of ICU transfer. In the multivariable analysis, EDI-Now (OR [95% CI]; 3.11 [2.26,4.27], p < 0.0001) and SOFA (OR [95% CI]; 1.78 [1.31,2.43], p = 0.0002) were significantly associated with 48-h ICU mortality. The AUROC for EDI-now, EDI-24 h, EDI-change and SOFA were 0.657, 0.563, 0.635, and 0.621, respectively. The optimal cut-off threshold for EDI-now was 55 with 73.7% sensitivity, 54.4% specificity, 17.9% positive predictive value (PPV) and 91.5% negative predictive value (NPV). The optimal cut-off threshold for SOFA score was 5 with 63.1% sensitivity, 53.6% specificity, 15.5% PPV and 91.5% NPV.ConclusionEDI-now ≥ 54.3 and SOFA ≥ 4 were associated with greater 48-h ICU mortality. High NPVs and low PPVs suggest EDI and SOFA scores could potentially be used to identify patients at lower risk of dying within 48 hours but cannot be used to predict higher risk of 48-h mortality prior to ICU transfer. Better prognostic scores are needed for the medical oncology population.