Patient risk profiling in acute medicine: the way forward?

BACKGROUND The identification of high-risk patients could form a basis for targetted intervention following an emergency medical admission. METHODS All emergency admissions to our institution over 12 years (2002-13) were included. An Illness Severity method based on admission laboratory parameters, previously developed between 2002 and 2007, was investigated for the 2008-13 cohort. We compared the area under the receiver operating characteristic (AUROC) to predict a 30-day in-hospital death between the original and validating cohorts using logistic multiple variable analyses. We defined six risk subgroups, based on admission laboratory data and examined the frequency of 30-day in-hospital mortality within these subgroups. RESULTS About 66 933 admissions were recorded in 36 271 patients. Between 2002 and 2007, the 30-day in-hospital mortality was 11.3% but between 2008 and 2013 was 6.7% (P < 0.001). This represented an absolute risk reduction (ARR) of 4.6%, a relative risk reduction (RRR) of 41.0%, and a number needed to treat of 21.6. The laboratory model was similarly predictive in both cohorts-for 2002-07, the AUROC was 0.82 (95% CI 0.81, 0.82) and for 2008-13 was 0.82 (95% CI 0.81, 0.83). Two high-risk subgroups were identified within each cohort; for 2002-07, these contained 15.0 and 30.2% of admitted patients but 95.5% of in-hospital deaths. For 2008-13, these two groups contained 15.7 and 31.0% of admitted patients but 97.0% of in-hospital deaths. CONCLUSION A previously described laboratory score method, based on admission biochemistry, identified patients at high risk for an in-hospital death. Risk profiling at admission is feasible for emergency medical admissions and could offer a means to outcome improvement.