Development and Validation of a Deep Learning Based Diabetes Prediction System Using a Nationwide Population-Based Cohort
- 주제(키워드) Diabetes mellitus , type 2 , Mass screening , Prediabetic state , Prediction
- 주제(기타) Endocrinology & Metabolism
- 설명문(일반) [Rhee, Sang Youl] Kyung Hee Univ, Dept Endocrinol & Metab, Sch Med, Seoul, South Korea; [Sung, Ji Min] Yonsei Univ, Yonsei Univ Hlth Syst, Integrat Res Ctr Cerebrovasc & Cardiovasc Dis, Coll Med, Seoul, South Korea; [Kim, Sunhee] Yonsei Univ Hlth Syst, Yonsei Univ, Coll Med, Seoul, South Korea; [Cho, In-Jeong] Ewha Womans Univ, Div Cardiol, Sch Med, Seoul, South Korea; [Lee, Sang-Eun; Chang, Hyuk-Jae] Yonsei Univ, Yonsei Univ Hlth Syst, Severance Cardiovasc Hosp, Div Cardiol,Coll Med, Seoul, South Korea
- 등재 SCIE, SCOPUS, KCI등재
- 발행기관 KOREAN DIABETES ASSOC
- 발행년도 2021
- 총서유형 Journal
- URI http://www.dcollection.net/handler/ewha/000000183505
- 본문언어 영어
- Published As http://dx.doi.org/10.4093/dmj.2020.0081
초록/요약
Background: Previously developed prediction models for type 2 diabetes mellitus (T2DM) have limited performance. We devel-oped a deep learning (DL) based model using a cohort representative of the Korean population. Methods: This study was conducted on the basis of the National Health Insurance Service-Health Screening (NHIS-HEALS) co-hort of Korea. Overall, 335,302 subjects without T2DM at baseline were included. We developed the model based on 80% of the subjects, and verified the power in the remainder. Predictive models for T2DM were constructed using the recurrent neural net-work long short-term memory (RNN-LSTM) network and the Cox longitudinal summary model. The performance of both models over a 10-year period was compared using a time dependent area under the curve. Results: During a mean follow-up of 10.4 +/- 1.7 years, the mean frequency of periodic health check-ups was 2.9 +/- 1.0 per subject. During the observation period, T2DM was newly observed in 8.7% of the subjects. The annual performance of the model created using the RNN-LSTM network was superior to that of the Cox model, and the risk factors for T2DM, derived using the two mod-els were similar; however, certain results differed. Conclusion: The DL-based T2DM prediction model, constructed using a cohort representative of the population, performs bet-ter than the conventional model. After pilot tests, this model will be provided to all Korean national health screening recipients in the future.
more