Table 3.

Predictors selected to form prediction models by at least one of the 50 training datasets on top of the initial model^*, frequency of each prediction model, the final winning models^** and the c-index of each winning model based on the holdout validation datasets.

Model number	Training			Validation
	Predictors	Frequency (%), n=50	Winning model**	Mean c-index (SD)
1	Age, weight change, HbA1c change, HgA1c change rate	20 (40)	x	0.8124 (0.01774)
2	Age, weight change, HbA1c change, HgA1c in prior 6 mos	8 (16)	x	0.8079 (0.01687)
3	Age, weight change, HbA1c change, HgA1c in prior 18 mos	6 (12)	x	0.8220 (0.01035)
4	Age, weight change, HbA1c change, HgA1c at index	3 (6)
5	Age, weight change, HbA1c change, abdominal pain	3 (6)
6	Age, weight change, HbA1c change, BMI	2 (4)
7	Age, weight change, HbA1c change, alcohol consumption	2 (4)
8	Age, weight change, HbA1c change, sodium value	1 (2)
9	Age, weight change, HbA1c change, chest pain	1 (2)
10	Age, weight change, HbA1c change, ALT change	1 (2)
11	Age, weight change, HbA1c change, Hgb value	1 (2)
12	Age, weight change, HbA1c change, race/ ethnicity	1 (2)
13	Age, weight change, HbA1c change, years of health plan enrollment	1 (2)

^*The initial model included age, weight change in 1 year and HbA1c change in 1 year.

^**The three most frequent models.