Abstract: TH-PO011
Kidney Outcome Comparisons of Clinician Evaluated Electronic AKI Alerts
Session Information
- AI, Digital Health, Data Science - I
November 02, 2023 | Location: Exhibit Hall, Pennsylvania Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Augmented Intelligence, Digital Health, and Data Science
- 300 Augmented Intelligence, Digital Health, and Data Science
Authors
- Aklilu, Abinet Mathias, Yale School of Medicine, New Haven, Connecticut, United States
- O'Connor, Kyle D., Yale School of Medicine, New Haven, Connecticut, United States
- Wilson, Francis Perry, Yale School of Medicine, New Haven, Connecticut, United States
Background
The personalized recommendations for hospitalized patients with Acute Kidney Injury (AKI) using a Kidney Action Team (KAT-AKI) trial is an ongoing multicenter randomized clinical trial evaluating early recommendations for AKI. It is unknown whether clinician adjudication of an electronic alert for AKI can improve AKI diagnosis.
Methods
We used data from KAT-AKI trial that utilizes an EHR alert for real-time diagnosis of AKI in hospitalized patients. The alert implements the KDIGO serum creatinine(sCr) based AKI criteria (≥50% ↑ or 0.3mg/dL ↑) in 7day and 48hour windows. A trained team of a physician and pharmacist receives an InBasket EHR alert. The team independently reviews sCr trend to adjudicate the AKI diagnosis then select agree or disagree. Patients with eGFR ≤15ml/min/1.73m2, end-stage kidney disease or admission sCr ≥4mg/dL were excluded. We compared the distribution of △sCr72 (a difference between sCr at AKI and the maximum sCr in the subsequent 72hrs), △sCrany (difference between AKI sCr and max sCr in the next hospital days), 14-day progression to a higher AKI stage, length of stay(LOS) between alerts flagged as AKIyes and AKIno. We used Wilcoxon-Rank Sum test to compare continuous variables and χ2 test for categorical variables.
Results
Between November 2021 and April 2023, 2414 alerts were screened. The team disagreed with 431(17.9%) of EHR alerts. Baseline sCr and AKI sCr were lower in the AKIno group(Table1). Median(IQR) ΔsCr72 was -0.02[-0.20, 0.21]mg/dL (AKIyes) vs 0.00 [-0.10, 0.12]mg/dL (AKIno), while ΔsCrany was 0.02[0.00,0.35]mg/dL vs 0.10[0.00, 0.34]mg/dL, respectively. 14-day AKI progression was 12.6%(AKIyes) vs 9.7%(AKIno), p~0.099. There was no difference in LOS.
Conclusion
Real-time clinician adjudication of an AKI alert did not improve AKI diagnosis as measured by prediction of AKI persistence or progression.
Table1
Baseline Characteristics | AKINo n = 413 | AKIYes n = 1983 | p-value |
Age, median (IQR) Female sex, n (%) Race/Ethnicity, n (%) Asian Black White Hispanic Baseline sCr, median (IQR), mg/dL sCr at AKI, median (IQR), mg/dL KDIGO AKI stage, n (%) 1 2 3 | 68.48 [56.85, 79.66] 244 (56.6) 7 (1.6) 77 (17.9) 297 (68.9) 59 (13.7) 0.70 (0.40, 1.30) 1.03 (0.60, 1.64) 415 (96.3) 13 (3.0) 3 (0.7) | 73.21 [61.93, 82.75] 493 (47.6) 26 (1.3) 393 (19.8) 1388 (70.0) 198 (10.0) 1.09 (0.80, 1.51) 1.58 (1.28, 2.07) 1847 (93.1) 112 (5.6) 24 (1.2) | < 0.001 0.001 0.473 . . . . < 0.001 < 0.001 0.051 |
Funding
- Other U.S. Government Support