Assessing ICU Severity of Illness With Readily Available Administrative Data

Abstract

SESSION TITLE: Critical Care - ICU Management SESSION TYPE: Original Investigation Poster PRESENTED ON: Wednesday, October 26, 2016 at 01:30 PM - 02:30 PM PURPOSE: The Acute Physiology and Chronic Health Evaluation (APACHE) and Simplified Acute Physiology Score (SAPS) tools are gold standard for evaluating severity of illness in ICU populations. These tools are resource intensive which many community hospitals cannot afford. Utilizing readily available administrative data may offer alternative methods for these hospitals. METHODS: A Retrospective chart review of 298 random patients at a large teaching-hospital with a mixed medical-surgical 29-bed ICU. APACHE-IV, SAPS, and administrative data were analyzed as predictors of actual mortality. All Patient Refined-Diagnosis Related Group Risk of Mortality (APR-DRG-ROM) and Severity of Illness (APR-DRG-SOI) are routinely reported measures of severity based on admission diagnosis and comorbidities. Case-Mix Index (CMI) represents resource utilization relative to national averages. Statistical analysis performed with JMP 12.2 and MedCalc 14.8. RESULTS: Summary statistics revealed 140 females, 158 males, mean age of 71, 75% medical cases and 25% surgical cases. ROC curves for APACHE-IV and SAPS showed Area Under the Curve (AUC) of 0.82 (95% CI 0.77 - 0.86) and 0.81 (95% CI 0.76 - 0.85) respectively, which were used as controls. APR-DRG-ROM AUC was 0.79 (95% CI 0.74 - 0.84) and APR-DRG-SOI was 0.72 (95% CI 0.66 - 0.77). CMI AUC was 0.59 (95% CI 0.53 - 0.65). Pairwise comparison of ROC curves showed no difference between APACHE-IV and SAPS (p=0.279) or APR-DRG-ROM (p=0.421). Differences existed in APACHE-IV vs APR-DRG-SOI (mean diff 0.103, p = 0.014) and APACHE-IV vs CMI (mean diff 0.228, p=<0.0001). A multiple-regression model combining APR-DRG-ROM, APR-DRG-SOI, CMI, age, gender, and medical vs surgical resulted in an ROC AUC of 0.85. CONCLUSIONS: Our APACHE-IV control data was consistent with prior reported AUCs (0.76 to 0.86). APR-DRG-ROM demonstrated good discrimination and significant correlation to the APACHE-IV and SAPS methods. Our multiple-regression model offered the highest AUC, demonstrating potential for a new model. CLINICAL IMPLICATIONS: For community hospitals without dedicated research support, obtaining APACHE-IV scores is difficult. APR-DRG-ROM is a good marker for severity of illness in ICU patients and can be used for research in these settings. DISCLOSURE: The following authors have nothing to disclose: Joseph Carrington, Malini Fonseka Chandra Serharan, Galo Sanchez Palacios, Jaime Barnes No Product/Research Disclosure Information