Q3: An ICU Acuity Score for Exploring the Effects of Changing Acuity Throughout the Stay

IntroWe develop a straightforward ICU acuity score (Q3) that is calculated every 3 hours throughout the first 10 days of the ICU stay. Q3 uses components of the Oxford Acute Severity of Illness Score (OASIS) and incorporates a new component score for vasopressor use. In well-behaved models of ICU mortality, the marginal effects of Q3 are significant across the first 10 days of the ICU stay. In separate models, Q3 has significant effects on ICU remaining length of stay. The score has implications for work that seeks to explain modifiable mechanisms of changing acuity during the ICU stay. MethodsFrom the MIMIC-III database, select ICU stays from 5 adult ICUs were partitioned into consecutive 3-hour segments. For each segment, the number of vasopressors administered and all 10 OASIS component scores were computed. Models of ICU mortality were estimated. OASIS component effects were examined, and vasopressor count bins were weighted. Q3 was defined as the sum of 8 retained OASIS components and a new weighted vasopressor score. Models of ICU mortality quadratic in Q3 were estimated for each of the first 10 ICU days and were subjected to segment-level, location-specific tests of discrimination and calibration on newer ICU stays. Marginal effects of Q3 were computed at different levels of Q3 by ICU day, and average marginal effects of Q3 were computed at each location by ICU day. ICU remaining length of stay (LOS) models were also estimated and the effects of Q3 were similarly examined. ResultsDaily ICU mortality models using Q3 show no evidence of misspecification (Pearson-Windmeijer p>0.05, Stukel p>0.05), discriminate well in all ICUs over the first 10 days (AUROC [~] 0.72 - 0.85), and are generally well calibrated (Hosemer-Lemeshow p>0.05, Spiegelhalters z p>0.05). A one-unit increase in Q3 from typical levels (Q3=15) affects the odds of ICU mortality by a factor of 1.14 to 1.20, depending on ICU day (p<0.001), and the ICU remaining LOS by 5.8 to 9.6% (p<0.001). On average, a one-unit increase in Q3 increases the probability of ICU mortality by 1 to 2 percentage points depending on location and ICU day, and ICU remaining LOS by 5 to 10 hours depending on location and ICU day. ConclusionQ3 significantly affects ICU mortality and ICU remaining LOS in different ICUs across the first 10 days of the ICU stay. Depending on location and ICU day, a one-unit increase in Q3 increases the probability of ICU mortality by or 1-2 percentage points and ICU remaining LOS by 5 to 10 hours. Unlike static acuity scores or those updated infrequently, Q3 could be used in explanatory models to help elucidate mechanisms of changing ICU acuity.