Developing a Complex Rule-Based Clinical Decision Support System for Detection of Acute Kidney Injury after Pediatric Cardiac Surgery

Background: Acute kidney injury (AKI) is a frequent complication in children with congenital heart disease following open-heart surgery. In intensive care units, detecting AKI requires expertise of clinicians who have to compile data from various sources within a critical and time-sensitive setting. However, as electronic health records provide data in a machine-readable format, this process could be executed through computerized systems to support physicians. Therefore, we developed a time-aware, rule-based clinical decision support system (CDSS) to detect, stage, and track the temporal progression of AKI in children.

Methods: Retrospective clinical routine data from n = 256 children who underwent cardiac surgery with cardiopulmonary bypass (CPB) between 2015 and 2021 were integrated into a standardized dataset. Based on a previous power analysis for accuracy testing of the CDSS, this cohort was enriched with several patients known to have AKI 3. We adapted the criteria proposed by Kidney Disease Improving Global Outcome, including serum creatinine, urine output and estimated glomerular filtration rate, and translated them into computable rules for the CDSS. As reference standard, patients were manually labeled for stage, timing, and duration by blinded clinical experts. We assessed the sensitivity and specificity of the CDSS per AKI stage using a nonparametric approach that adjusted for the clustered data per patient. Episodes with an in-between episode interval of <48 hours were merged.

Results: In the reference cohort, incidences of AKI, diagnosed by blinded clinical experts, were 32.4% (AKI 1), 15.2% (AKI 2), and 7.8% (AKI 3). Compared with this reference, a preliminary version of the CDSS achieved sensitivities of 88.5% (IQR 82.3–92.7%) for AKI 1, 91.9% (80.8–96.8%) for AKI 2, and 99.4% (96.1–99.9%) for AKI 3. The specificities were 97.1% (94.7–98.5%), 98.5% (96.7–99.3%), and 99.4% (97.6–99.8%), respectively.

Conclusion: We demonstrate that a rule-based computerized system is able to perform a complex AKI detection and staging process, including 11 criteria in 3 stages. To ensure the proper function of the automated AKI detection system, standardized machine-readable data of high data quality are required. In the next step, a large dataset will be labeled by the CDSS to develop a prediction model for AKI 1–3 after cardiac surgery with CPB.

No conflict of interest has been declared by the author(s).

Publication History

Article published online:
13 February 2024