Kidney Week

Abstract: TH-PO007

Construction of a Predictive Equation for CKD Exacerbation by Multifactorial Analysis Using Machine Learning and Analysis on Differences in CKD Exacerbation Factors in Each CKD Stage

Session Information

• Augmented Intelligence for Prediction and Image Analysis
  October 24, 2024 | Location: Exhibit Hall, 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

• Ogawa, Koki, Saitama Ika Daigaku Sogo Iryo Center, Kawagoe, Saitama, Japan

Koki Ogawa, MD

• Hara, Hiroaki, Saitama Ika Daigaku Sogo Iryo Center, Kawagoe, Saitama, Japan

Hiroaki Hara, MD

• Takahashi, Yasushi, NEC Solution Innovator Kabushiki Kaisha, Koto-ku, Tokyo, Japan

Yasushi Takahashi

• Oana, Seiko, NEC Solution Innovator Kabushiki Kaisha, Koto-ku, Tokyo, Japan

Seiko Oana

• Nakamura, Yumiko, Saitama Ika Daigaku Sogo Iryo Center, Kawagoe, Saitama, Japan

Yumiko Nakamura, MD

• Maeshima, Akito, Saitama Ika Daigaku Sogo Iryo Center, Kawagoe, Saitama, Japan

Akito Maeshima, MD, PhD

• Hasegawa, Hajime, Saitama Ika Daigaku Sogo Iryo Center, Kawagoe, Saitama, Japan

Hajime Hasegawa, MD, PhD

Background

Multiple factors such as comcomitant disorders, prescribed medications and physical characteristics are involved in CKD exacerbation. In recent years, machine-learning has enabled us to analyze the CKD exacerbation in relation to the multiple factors. We aimed to construct a predictive equation for the CKD exacerbation by analyzing changes in the multiple factors by machine-learning. Additionally, the differences in exacerbation factors per CKD stage were also examined.

Methods

The analysis included 3098 of CKD patients at our institution between Apr. 2006 and Mar. 2021 for which data analysis was possible more than once a year apart. For each case, 209 items such as blood test results, blood pressure, body weight, age and medications were learned and analyzed using machine learning. We defined the CKD exacerbation as "change in CKD heat map" (eg. green to yellow) because final renal outcome involves both decreased eGFR and worsening proteinuria during follow-up period. For the case in which CKD heat map exacerbated, the item of which the exacerbation contribution was large was extracted, and the difference of exacerbation factors (EF) by CKD stage was examined.

Results

The prediction equation was constructed using 85% of population, and its accuracy was verified by adapting the constructed equation to the remaining 15% of the population. The accuracy rate was 76.8% in Green group, 75.6% in Yellow group, and 70.8% in Orange group, respectively. The largest EF was the proteinuria throughout all CKD stage, and it was baseline eGFR next. Except for these two factors, EF differed according to CKD stage, and EF next to the above two factors were concomitant hypertension in Green group, elevation of serum uric acid in Yellow group, and the presence or progression of anaemia in Orange group.

Conclusion

The prediction equation constructed in this study showed sufficient accuracy for clinical use. And, that principal CKD exacerbation factors greatly differed for each stage except for proteinuria and baseline eGFR seemed to be the key knowledge for the management of CKD patients.