A Clinical Deterioration Prediction Tool for Internal Medicine Patients

Abstract

Many early warning models for hospitalized patients use variables measured on admission to the hospital ward; few have been rigorously derived and validated. The objective was to create and validate a clinical deterioration prediction tool using routinely collected clinical and nursing measurements. Multivariate regression analysis was used to determine clinical variables statistically associated with clinical deterioration; subsequently, the model tool was retrospectively validated using a different cohort of medical inpatients. The Braden Scale (P = .01; odds ratio [OR] = 0.91; confidence interval [CI] = 0.84-0.98), respiratory rate (P < .01; OR = 1.08; CI = 1.04-1.13), oxygen saturation (P < .01; OR = 0.97; CI = 0.96-0.99), and shock index (P < .01; OR = 2.37; CI = 1.14-3.98) were predictive of clinical deterioration 2-12 hours in the future. When applied to the validation cohort, the tool demonstrated fair concordance with actual outcomes. This tool created using routinely collected clinical measurements can serve as a very early warning system for hospitalized medical patients.