Scripties UMCG - Rijksuniversiteit Groningen
English | Nederlands

Assessment of the performance of 10 intensive care risk scores: validation study in a prospective cohort of 1072 patients

(2019) Doornbos, E.F.

Mortality risk scores are often used tools for prognostication in the intensive care unit (ICU). The number of mortality risk scores developed and published is significant, causing older risk scores to be disregarded and no longer considered for external validations. Moreover, many risk scores applied in clinical practice lack external validation, which hampers clinical implementation. We performed external validation on a selection of risk scores, found in a systematic overview, in a prospective cohort study in the ICU of the UMCG, and recalibrated the best performing risk score.
Risk score formulas were retrieved from the original manuscripts and converted into mortality risk score calculators using the validation platform Evidencio. Data of adult patients acutely admitted to a single-center ICU and included in the SICS-I was used as input variables. After validation, the risk scores were compared based on discrimination and calibration using an AUROC and calibration plots. For the recalibration, a multiple logistic regression analysis was rerun using the original risk score variables. The beta-coefficients and intercept obtained, were the basis for the updated risk score formula.
A total of 10 risk scores out of 41 were included for validation, using 1072 patients of the SICS-I. Risk scores with the highest discriminative power were MPM₂₄-II and SAPS II, with an AUROC of 0.79 (95% CI: 0.75-0.83) and 0.77 (95% CI: 0.73-0.81) respectively. SMS-ICU had the worst discriminative power with an AUROC of 0.67 (95% CI: 0.64-0.70). Risk scores with the highest calibration were SMS-ICU and SAPS_R (regression coefficients: 0.78 and 0.77, respectively). The SOFA score showed the lowest calibration (regression coefficient: 0.33).
External validation of 10 mortality risk scores in an adult ICU population showed that the discrimination was modest in all risk scores. However, the MPM24-II was the most accurate and showed an excellent calibration after recalibration.

To top