TABLE 2.

Performance metrics of select models for predicting prediabetes¹

Algorithm	Predictors	Train AUC (95% CI)	AUC (95% CI)	Sensitivity	Specificity
Random guessing	NA	0.531 (0.512, 0.551)	0.556 (0.473, 0.639)	0.811	0.246
GLM	Age	0.635 (0.616, 0.653)	0.702 (0.629, 0.774)	0.774	0.570
GLM	GDQS	0.515 (0.495, 0.534)	0.511 (0.423, 0.598)	0.547	0.488
GLM	Age and GDQS	0.636 (0.617, 0.654)	0.709 (0.640, 0.779)	0.774	0.575
GLM	Age, GDQS food groups, hours sedentary, alcoholic beverage consumption, unable to walk, use of rations card, sex, tobacco use	0.654 (0.635, 0.672)	0.716 (0.645, 0.787)	0.755	0.600
GLMM	Age and GDQS, family random intercept	0.878 (0.867, 0.889)	0.710 (0.640, 0.779)	0.755	0.599
GLMM	Age and GDQS food groups, family random intercept	0.873 (0.861, 0.884)	0.711 (0.642, 0.781)	0.793	0.594
GLMM	Age, GDQS food groups, hours sedentary, alcoholic beverage consumption, unable to walk, use of rations card, sex, tobacco use, family random intercept	0.872 (0.861, 0.883)	0.722 (0.652, 0.792)	0.717	0.662
LASSO	Age, GDQS food groups, hours sedentary, alcoholic beverage consumption, unable to walk, use of rations card, sex, tobacco use	0.644 (0.625, 0.663)	0.705 (0.633, 0.776)	0.774	0.580
Elastic net (α = 1)	Age, GDQS food groups, hours sedentary, alcoholic beverage consumption, unable to walk, use of rations card, sex, tobacco use	0.641 (0.626, 0.659)	0.700 (0.627, 0.772)	0.774	0.570
Random forest	Age, GDQS food groups, hours sedentary, alcoholic beverage consumption, unable to walk, use of rations card, sex, tobacco use	1.000 (1.000, 1.000)	0.705 (0.633, 0.776)	0.774	0.517

¹AUC, area under the receiver operating characteristic curve; GDQS, Global Diet Quality Score; GLM, generalized linear model; GLMM, generalized linear mixed model; LASSO, least absolute shrinkage and selection operator; NA, not applicable.