THE CRISPE CLINICAL DECISION INSTRUMENT: A NOVEL TOOL TO PREDICT MORTALITY IN PATIENTS WITH CIRRHOSIS.

Background:
Emergency department (ED) patients with cirrhosis are clinically heterogenous and experience high rates of short-term mortality, making disposition decisions challenging. We derived and internally validated the “Cirrhosis Risk Instrument for Stratifying Post-Emergency department mortality” (CRISPE) clinical decision instrument (CDI) for predicting 30-day mortality after an initial ED visit.

Methods:
From a retrospective cohort of individuals with cirrhosis presenting to the ED between January 2021 and December 2021 (n=2124), 131 variables available to clinicians at time of ED disposition determination were considered for CDI inclusion. Cirrhosis diagnosis was confirmed by chart review. Least absolute shrinkage and selection operator (LASSO) was used to select features down to a target ≤ 10 events-per-variable ratio, fit as a logistic regression model. Internal validation, to penalize model performance characteristics for expected derivation optimism, was performed via Harrell’s bootstrap-based method for bias-estimation and correction.[NC1] Model fit was assessed using receiver operating characteristic area under the curve (AUROC). The model was then refined by similar methods for 14-day mortality using a reduced predictor set.

Results:
Patients had a mean age of 61.3 ± 12.9, 91% were white, 20% were Medicaid and 91% came from home. Cirrhosis was primarily due to alcohol (28%), metabolic dysfunction associated steatohepatitis (MASH, 27%), and viral hepatitis (21%). Median MELD-Na was 17 (IQR 11-23.6). At baseline, 49% had ascites and 32% had hepatic encephalopathy.

At 30-day follow-up, 179 (8.5%) patients had died. Variable included in the model with multivariable-adjusted odds ratios are summarized in Table 1. The fit model achieved an AUROCC of 0.848 for 30-day mortality at derivation and AUROCC = 0.818 after internal validation and bias-correction for expected optimism. Predictors of 30-day mortality were: tachypnea, hypoxia, hyponatremia, hypothermia, elevated shock-index (Systolic BP to HR ratio), hepatocellular carcinoma, hypoalbuminemia, hyperbilirubinemia, altered mental status, and age. Those presenting to the ED for one or more structural reasons (e.g. unable to get outpatient visit, need urgent but not emergent medication management, no primary care, etc.) were less likely to die. Three variables were dropped for the simplified 14-day mortality model, which achieved a bias-corrected 0.808 AUC (Figure 1A,1B).

Conclusion:
CRISPE, to our knowledge the first CDI for risk-stratifying short-term mortality in ED patients with cirrhosis, was highly accurate for predicting 30- and 14-day mortality. CRISPE fills the need for a CDI which supports the triage of a heterogenous population into high-risk patients needing hospitalization and low-risk patients who can potentially be discharged after ED management alone.