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 X

Kidney Week

Abstract: TH-PO030

Deep Learning-Based Prediction of Postoperative AKI After Noncardiac Surgery Using Intraoperative Vital Sign Parameters in a Minute Scale

Session Information

Category: Augmented Intelligence, Digital Health, and Data Science

  • 300 Augmented Intelligence, Digital Health, and Data Science

Authors

  • Park, Sehoon, Seoul National University Hospital, Jongno-gu, Seoul, Korea (the Republic of)
  • Kim, Yong Chul, Seoul National University Hospital, Jongno-gu, Seoul, Korea (the Republic of)
  • Kim, Dong Ki, Seoul National University Hospital, Jongno-gu, Seoul, Korea (the Republic of)
  • Lee, Hajeong, Seoul National University Hospital, Jongno-gu, Seoul, Korea (the Republic of)
Background

Certain models were used to predict the risk of PO-AKI mainly with preoperative characteristics; however, complex intraoperative vital sign information was difficult to be combined to such strategy although intraoperative hemodynamic alteration is one of the major factor affecting PO-AKI risks. We aimed to construct an externally validated deep learning-based prediction model for PO-AKI, including complex, real-time collected hemodynamic information.

Methods

We collected systolic and diastolic pressure values and heart rate information collected from the real-time intraoperative monitoring in a tertiary university hospital (N = 51,345). The test and the internal validation set was split as 8:2 manner, and additional external validation was performed in two additional tertiary university hospitals (N = 47,093 and 12,259). Total collected measured points of blood pressure values were 17,816,251 (systolic) and 17,793,025 (diastolic) and of heart rate was 17,505,759. Deep-learning model was constructed using the EfficientNet based CNN model. The outcome was PO-AKI and critical AKI events, and the critical AKI was defined as high stage AKI or AKI associated with death or dialysis. We compared the model performances with AUC-ROC values, and the conventional SPARK classification was used as the reference model.

Results

The deep-learning model only including intraoperative variables showed moderate but tolerable discrimination power against PO-AKI [AUC-ROC 0.707 (development), 0.637 and 0.729 (validation)] or critical AKI [AUC-ROC 0.724 (development), 0.729 and 0.716 (validation)] events. When major PO-AKI risk factors were incorporated to the models, the powers outperformed the conventional SPARK model; for PO-AKI [0.765 (development), 0.716 and 0.761 (validation)] and for critical AKI [0.816 (development), 0.741 and 0.794 (validation)]. The model performances even improved in the ensemble model learning both preoperative tabular data and the intraoperative prediction models.

Conclusion

Complex intraoperative vital sign information, including blood pressure and heart rate, can be used to develop tolerable deep-learning based PO-AKI risk stratification model that can be used after non-cardiac surgeries.