Accuracy between ICU admission and discharge diagnoses in non-survivors: A retrospective cohort study

INTRODUCTION: Models predicting in-hospital mortality for intensive care unit (ICU) patients, like Acute Physiologic and Chronic Health Evaluation (APACHE) IV, depend on correct data about admission diagnoses. Incorrectly recording diagnoses or changes in diagnoses during admission may impact estimating mortality. METHODS: All deceased patients admitted to the ICU between Jan 1, 2018 and Dec 31, 2020 were included. Up to two discharge diagnoses were assigned using APACHE IV diagnostic codes. These were compared to the up to two diagnoses documented at admission. When differences occurred, these were classified as registration errors or diagnostic change. The APACHE IV predicted mortality was calculated using either admission or discharge diagnoses. The agreement between both methods was expressed as the mean difference and the intra-class correlation coefficient (ICC). RESULTS: 886 (16 %) of 5633 patients died. In 363 (41 %) patients, there was no difference between admission and discharge diagnoses. Registration errors occurred in 138 (16 %) patients. 416 (47 %) patients had diagnostic change. The mean difference between predictions was 0.019 (95 % CI: 0.015-0.024). The ICC was 0.97 (95 % CI: 0.97-0.98). DISCUSSION: Differences between ICU admission and discharge diagnoses occur frequently, this leads to only a small discrepancy in the overall predicted mortality of deceased ICU patients.