Identifying progressive CKD from healthy population using Bayesian network and artificial intelligence: A worksite-based cohort study

Sci Rep. 2019 Mar 25;9(1):5082. doi: 10.1038/s41598-019-41663-7.

Abstract

Identifying progressive early chronic kidney disease (CKD) patients at a health checkup is a good opportunity to improve their prognosis. However, it is difficult to identify them using common health tests. This worksite-based cohort study for 7 years in Japan (n = 7465) was conducted to evaluate the progression of CKD. The outcome was aggravation of the KDIGO prognostic category of CKD 7 years later. The subjects were male, 59.1%; age, 50.1 ± 6.3 years; and eGFR, 79 ± 14.4 mL/min/1.73 m2. The number of subjects showing CKD progression started to increase from 3 years later. Vector analysis showed that CKD stage G1 A1 was more progressive than CKD stage G2 A1. Bayesian networks showed that the time-series changes in the prognostic category of CKD were related to the outcome. Support vector machines including time-series data of the prognostic category of CKD from 3 years later detected the high possibility of the outcome not only in subjects at very high risks but also in those at low risks at baseline. In conclusion, after the evaluation of kidney function at a health checkup, it is necessary to follow up not only patients at high risks but also patients at low risks at baseline for 3 years and longer.

MeSH terms

  • Adult
  • Artificial Intelligence*
  • Bayes Theorem*
  • Disease Progression
  • Female
  • Glomerular Filtration Rate / physiology
  • Humans
  • Kidney Function Tests
  • Male
  • Middle Aged
  • Prognosis
  • Renal Insufficiency, Chronic / pathology*