Hospital mortality prediction for intermediate care patients: Assessing the generalizability of the Intermediate Care Unit Severity Score (IMCUSS)

Abstract

Purpose: The Intermediate Care Unit Severity Score (IMCUSS) is an easy to calculate predictor of in-hospital death, and the only such tool developed for patients in the intermediate care setting. We sought to examine its external validity.

Materials and methods: Using data from patients admitted to the intermediate care unit (IMCU) of an urban academic medical center from July to December of 2012, model discrimination and calibration for predicting in-hospital death were assessed using the area under the receiver operating characteristic (AUROC) and the Hosmer-Lemeshow goodness-of-fit chi-squared (HL GOF X²) test, respectively. The standardized mortality ratio (SMR) with 95% confidence intervals (95% CI) was also calculated.

Results: The cohort included data from 628 unique admissions to the IMCU. Overall hospital mortality was 8.3%. The median IMCUSS was 10 (Interquartile Range: 0-16), with 229 (36%) patients having a score of zero. The AUROC for the IMCUSS was 0.72 (95% CI: 0.64-0.78), the HL GOF X² = 30.7 (P < 0.001), and the SMR was 1.22 (95% CI: 0.91-1.60).

Conclusions: The IMCUSS exhibited acceptable discrimination, poor calibration, and underestimated mortality. Other centers should assess the performance of the IMCUSS before adopting its use.

Keywords: Intermediate care; Mortality prediction; Outcome prediction score; Progressive care; Stepdown care.

Fig. 1.

Calibration of the IMCUSS in the overall population. The total number of patients (left axis) in each of seven subgroup is shown in grey bars. Calibration is characterized by a plot of observed (solid circles) and expected (open circles) in-hospital mortality (right axis) within each subgroup. With good calibration, the two curves would be superimposed. Subgroups are defined by ranking predicted risk of in-hospital death from lowest to highest and then dividing the population into subgroups (usually deciles) of increasing risk. Because there are many patients with identical risk (i.e.; ties) at the boundary between subgroups, they differ in size (i.e.; 229 patients receive no IMCUSS point and are therefore assigned the same predicted risk). Further, a smaller number of groups (seven) has been specified to avoid any one subgroup from being too small. Ranges of risk for in-hospital death for each of the seven subgroups are as follows: group 1 (1.6%–1.6%), group 2 (2.9%–2.9%), group 3 (3.8% to 4.2%), group 4 (4.6% to 62%), group 5 (6.7% to 9.7%), group 6 (10.6% to 15.0%), and group 7 (163% to 71.9%). Gaps in the range of predicted risk between subgroups occur because the IMCUSS did not assign any patients to an intervening predicted mortality risk.