Identifying Glucose Metabolism Status in Nondiabetic Japanese Adults Using Machine Learning Model with Simple Questionnaire

Table 1

Review of the recent and important studies on prediabetes screening.

Ref. no.	Screening target	Factors	Models	Tool challenges
[14]	FPG 100–125 mg/dL, 120 mPG 100–125 mg/dL, or HbA1c 5.7–6.4%	25 of socioeconomic, clinical, and biochemical factors	RF, GBM, LR, and ANN	Invasive measurement factors were required for screening
[16]	FPG ≥100 mg/dL	Global diet quality score, age, smoking, alcohol drinking, unable to walk, use of rations card, time spent in sedentary activities	RF, GLMM, LASSO, and EN	Well-trained interviewers were needed to obtain dietary information
[17]	HbA1c 5.7–6.4%	Age, sex, BMI, waist circumference, and blood pressure	RF, GBM, XGB, LR, and DL	Lack of individuals with high blood glucose levels from screening targets; Some of factors could not be answered on the spot and may require the linkage of the laboratory data
[18]	FPG 110–125 mg/dL or HbA1c 5.7–6.4%	Age, BMI, waist-to-hip ratio, systolic blood pressure, waist circumference, sleep duration, smoking status, and vigorous recreational activity time per week	XGB and LR	Lack of individuals with hyperglycemia after glucose loading from screening targets; Some of factors could not be answered on the spot and may require the linkage of the laboratory data
[19]	FPG ≥110 mg/dL or 120 mPG ≥140 mg/dL	Age, sex, BMI, smoking, FPG, fasting plasma triglyceride level, and history of high FPG	LR	Research participants were limited to staffs in an oil field in China invasive measurement factors were required for screening
[20]	FPG 100–125 mg/dL, HbA1c 5.7–6.4%, or 120 mPG 140–199 mg/dL	Semiquantitative food frequency questionnaire answers and clinical and anthropometric measurements scores	LR	Well-trained interviewers were needed to obtain dietary information; Invasive measurement factors were required for screening

Abbreviation: FPG: fasting plasma glucose level; 120 mPG: 120-min postload plasma glucose level during OGTT; HbA1c: hemoglobin A1c; BMI: body mass index; RF: random forest; GBM: gradient boosting machine; LR: logistic regression; ANN: artificial neural network; GLMM: generalized linear mixed model; LASSO: least absolute shrinkage and selection operator; EN: elastic net; XGB: XGBoost; DL: deep learning.