Development of Labeling Algorithm for Early Prediction of Acute Kidney Injury

Area of Science:

• Nephrology
• Clinical Informatics
• Healthcare Data Analysis

Background:

• Acute kidney injury (AKI) is a common and serious complication in hospitalized patients, leading to higher morbidity and mortality rates.
• Existing predictive models often identify AKI retrospectively, limiting timely intervention.

Purpose of the Study:

• To develop and validate a labeling algorithm for capturing the earliest onset time of in-hospital AKI.
• To utilize KDIGO 2012 criteria for precise AKI staging and timing.

Main Methods:

• Retrospective analysis of 143,512 in-hospital cases from 2015-2021.
• Application of a developed labeling algorithm using serum creatinine and urine output data.
• Identification of earliest AKI onset based on KDIGO 2012 criteria.

Main Results:

• 31.97% (45,882) of analyzed cases were diagnosed with AKI.
• Urine output criteria identified the earliest onset in 87.29% of AKI cases.
• The algorithm successfully captured early AKI onset indicators.

Conclusions:

• The developed algorithm effectively identifies the earliest onset of in-hospital AKI.
• Urine output measurements are crucial for early AKI detection.
• This approach can potentially improve patient management and outcomes for AKI.