Prediction of blood test values under different lifestyle scenarios using time-series electronic health record

• Published in: PloS one Vol. 15; no. 3; p. e0230172
• Main Authors: Hasegawa, Takanori, Yamaguchi, Rui, Kakuta, Masanori, Sawada, Kaori, Kawatani, Kenichi, Murashita, Koichi, Nakaji, Shigeyuki, Imoto, Seiya
• Format: Journal Article
• Language: English
• Published: United States Public Library of Science 20.03.2020; Public Library of Science (PLoS)
• Subjects: Aged; Analysis; Biology and Life Sciences; Blood; Blood tests; Body composition; Body Composition - physiology; Clinical Decision Rules; Cohort analysis; Cohort Studies; Computer simulation; Diabetes; Diseases; Elderly; Electronic health records; Electronic Health Records - statistics & numerical data; Electronic medical records; Electronic records; Environmental effects; Environmental factors; Female; Genetic diversity; Genetic research; Genomes; Geriatrics; Health; Hematologic Tests - statistics & numerical data; Humans; Japan; Life Style; Lifestyles; Male; Medical records; Medical research; Medicine; Medicine and Health Sciences; Metabolic disorders; Middle Aged; Noise; Older people; Physical Sciences; Physiological aspects; Research and Analysis Methods; Social Sciences; State space models; Time; Time series; Variables; Japan
• ISSN: 1932-6203; 1932-6203
• DOI: 10.1371/journal.pone.0230172

Owing to increasing medical expenses, researchers have attempted to detect clinical signs and preventive measures of diseases using electronic health record (EHR). In particular, time-series EHRs collected by periodic medical check-up enable us to clarify the relevance among check-up results and individual environmental factors such as lifestyle. However, usually such time-series data have many missing observations and some results are strongly correlated to each other. These problems make the analysis difficult and there exists strong demand to detect clinical findings beyond them. We focus on blood test values in medical check-up results and apply a time-series analysis methodology using a state space model. It can infer the internal medical states emerged in blood test values and handle missing observations. The estimated models enable us to predict one's blood test values under specified condition and predict the effect of intervention, such as changes of body composition and lifestyle. We use time-series data of EHRs periodically collected in the Hirosaki cohort study in Japan and elucidate the effect of 17 environmental factors to 38 blood test values in elderly people. Using the estimated model, we then simulate and compare time-transitions of participant's blood test values under several lifestyle scenarios. It visualizes the impact of lifestyle changes for the prevention of diseases. Finally, we exemplify that prediction errors under participant's actual lifestyle can be partially explained by genetic variations, and some of their effects have not been investigated by traditional association studies.