Table 3.

Cross-validated risks of the 7 candidate learners that define the 11 Super Learners for estimating the denominators of the IPW weights. Each candidate learner is defined by a type of algorithm (’glm’ or ’polyclass’) and a subset of explanatory variables identified by univariate regressions based on a particular significance level (p-value).

Learner	glm	glm	glm	glm	glm	polyclass	polyclass
Subset of explanatory variables	p≤1e-30	p≤1e-10	p≤1e-5	p≤0.1	p≤1	p≤1e-30	p≤1
met initiation at t=0	0.13159	0.1305	0.13046	0.1303	0.13027	0.12903	0.12759
sul initiation at t=0	0.02852	0.0285	0.02847	0.02839	0.02834	0.0279	0.02781
met+sul initiation at t=0	0.02663	0.02641	0.02629	0.02627	0.02627	0.02661	0.02617
other therapy initiation at t=0	0.01971	0.01837	0.01839	0.01756	0.01741	0.01962	0.01691
met initiation at t>0	0.04005	0.04001	0.03996	0.03981	0.03982	0.03179	0.0316
sul initiation at t>0	7.86e-03	7.87e-03	7.86e-03	7.86e-03	7.85e-03	7.15e-03	7.2e-03
met+sul initiation at t>0	5.33e-03	5.34e-03	5.27e-03	5.27e-03	5.24e-03	4.65e-03	4.6e-03
other therapy initiation at t>0	2.29e-03	2.29e-03	2.28e-03	2.28e-03	2.29e-03	2.28e-03	2.28e-03
censoring by health plan disenrollment	0.04918	0.04917	0.04916	0.04914	0.04914	0.04909	0.04908
censoring by death	3.86e-03	3.84e-03	3.84e-03	3.85e-03	3.86e-03	3.9e-03	3.88e-03
insufficient GFR monitoring	0.02717	0.02631	0.02623	0.02608	0.02601	0.02589	0.02088