Abstract
Objective: Acute Physiology and Chronic Health Evaluation (APACHE) is most widely used as a mortality prediction score in US intensive care units (ICUs), but its calculation is onerous. The authors aimed to develop and validate automatic mapping of physicians' admission diagnoses to structured concepts for automated APACHE IV calculation. Methods: This retrospective study was conducted in medical ICUs of a tertiary healthcare and academic centre. Boolean-logic text searches were used to map admission diagnoses, and these were compared with conventional APACHE database entry by bedside nurses and a gold-standard physician chart review. The primary outcome was APACHE IV predicted hospital mortality. The tool was developed in a larger cohort of ICU patients. Results: In a derivation cohort of 192 consecutive critically ill patients, the diagnosis coefficient coded by three different methods had a positive correlation, highest between manual and gold standard (r 2 = 0.95; mean square error (MSE)=0.040) and least between manual and automatic tool (r 2=0.88; MSE=0.066). The automatic tool had an area under the curve (95% CI) value of 0.82 (0.74 to 0.90) which was similar to the physician gold standard, 0.83 (0.75 to 0.91) and standard manual entry, 0.81 (0.73 to 0.89). The Hosmer-Lemeshow goodness-of-fit test demonstrated good calibration of automatically calculated APACHE IV score (χ 2=6.46; p=0.6). The automatic tool demonstrated excellent discrimination with an area under the curve value of 0.87 (95% CI 0.83 to 0.92) and good calibration (p=0.58) in the validation cohort of 593 patients. Conclusion: A Boolean-logic text search is an efficient alternative to manual database entry for mapping of ICU admission diagnosis to structured APACHE IV concepts.