(1397) Deterioration Index: Are Trends or an Absolute Score Important to Improve Patient Outcomes?

Introduction: Early warning systems (EWS) for deterioration and specific syndromes like sepsis may improve patient outcomes through early identification and interventions. Epic Deterioration Index (EDI) is a vendor-supplied proprietary EWS that has been implemented in many US hospitals despite modest empirical evidence as to its value. EDI’s utility may be limited by false-positive and -negative predictions due to score fluctuations and miscalibration among intermediate-risk patients. We evaluated whether EDI trends are more predictive of clinical deterioration than absolute values.

Methods: We conducted a retrospective cohort study of hospitalized adults on the wards between January 1, 2019, and December 31, 2022, at Oregon Health & Science University, a quaternary academic hospital. Our EDI updates every 20 minutes and is passively displayed on electronic health record lists; values > 70 automatically trigger rapid response team consultation. Using discrete-time logistic regression in 6-hour time windows, we compared the EDI, an EDI trend (maximum slope in previous 6 hours), and a combination of these variables for predicting the composite of death on the wards or ICU transfer within the subsequent 24 hours. We calculated observation-level areas under the receiver-operator characteristic (AUROC) and threshold-based predictive values for these approaches.

Results: We evaluated 47,594 ward encounters from 36,580 unique patients; 3,123 encounters involved the primary outcome (2,956 ICU transfers, 167 ward deaths). Patients who died on the wards had higher maximum EDI scores (median 66, interquartile range [IQR] 57-76) than those admitted to the ICU (42, IQR 29-54), and those who did not experience the composite outcome (29, IQR 21-39). The EDI’s AUC was 0.686 (95% confidence interval 0.687-0.690) and the EDI slope’s AUC was 0.608 (0.607 - 0.610). Combining the EDI with its slope yielded an AUC of 0.692 (0.691-0.694). An EDI threshold of 40 showed sensitivity of 29.5% and specificity of 75.0%, while a threshold of 70 showed sensitivity of 2.8% and specificity of 99.8%.

Conclusions: Incorporating change over time to EDI scores slightly improves this tool’s discrimination for clinical deterioration. Future work should evaluate the extent to which more detailed trend assessments improve predictive accuracy.