ASN's Mission

To create a world without kidney diseases, the ASN Alliance for Kidney Health elevates care by educating and informing, driving breakthroughs and innovation, and advocating for policies that create transformative changes in kidney medicine throughout the world.

learn more

Contact ASN

1401 H St, NW, Ste 900, Washington, DC 20005

email@asn-online.org

202-640-4660

The Latest on Twitter

Kidney Week

Abstract: FR-PO089

Building a Prediction Model for Postoperative AKI Using Machine Learning: The CMC-AKIX Model

Session Information

Category: Acute Kidney Injury

  • 101 AKI: Epidemiology‚ Risk Factors‚ and Prevention

Authors

  • Min, Ji Won, Catholic University of Korea Bucheon Saint Mary's Hospital, Bucheon, Gyeonggi-do, Korea (the Republic of)
  • Yoon, Hye Eun, Catholic University of Korea Incheon Saint Mary's Hospital, Incheon, Incheon, Korea (the Republic of)
Background

Postoperative acute kidney injury (AKI) is associated with increased mortality and morbidity in patients undergoing surgeries performed under general anesthesia. There are several models that predict postoperative AKI risk, but most are single-center studies that need external validation.

Methods

In this retrospective cohort analysis, we included noncardiac surgeries performed between 2009 and 2019 at 7 university hospitals in South Korea. Postoperative AKI was defined as an increase of serum creatinine at least 1.5 times the baseline value or initiation of renal replacement therapy within 30 days of the postoperative period. We tested 6 machine learning prediction models: deep neural networks (DNN), logistic regression, decision tree, random forest, light gradient boosting machine (GBM), and naïve Bayes, and compared model performance using the area under the curve (AUC) of the receiver-operating characteristic.

Results

A total of 239,267 surgeries were included, and 7,935 postoperative AKI events (3.3%) occurred. The 6 different statistical analysis methods were run on various combinations of 40 independent preoperative predictors that we had selected (Table 1). Model 1 included all variables, Model 2 included variables that had been significantly associated with postoperative AKI in previous studies, and Model 3 included variables that were found significant in multivariate analysis. Among them, Model 1 run on DNN (AUC = 0.821) and light GBM (AUC = 0.823) and Model 3 run on DNN (AUC = 0.807) demonstrated the best prediction performance.

Conclusion

We have developed a high-performance risk prediction system for postoperative AKI that can be easily applied using preoperative patient characteristics and laboratory data.

Table 1. Performance metrics of postoperative AKI prediction models.

Funding

  • Private Foundation Support