Statistical analyses were performed using SAS version 9.3 (SAS Institute, Cary, North Carolina). We used the SURVEYLOGISTIC procedure to statistically control for potentially confounding variables via logistic regression modeling.27 Control variables for all models included patient sex, patient age group, patient race/ethnicity, specialty/clinic type, clinic ownership/size, geographical region, and data set of origin. Clinical characteristics were not included in these models under the argument that quality indicators should be applicable to all patients, except where specific complicating diagnoses are defined. All hypothesis testing was 2-tailed, and P < .05 was considered statistically significant. We did not adjust statistical significance for multiple comparisons, but interpretation of the findings reflects recognition of this issue. Our statistical power to detect differences in quality associated with EHR and CDS varied widely across indicators and was largely dependent on visit sample size. In comparing EHR with no EHR, the statistical power to detect a 5% absolute change in indicator performance (eg, 66% vs 71%) at a statistical significance level of P = .05 (2-tailed) was 90% or greater for 11 indicators, 75% to 89.9% for 3 indicators, 50% to 74.9% for 5 indicators, and less than 50% for 1 indicator. Similarly, for CDS vs no CDS among EHR visits, the statistical power was 90% or greater for 7 indicators, 75% to 89.9% for 2 indicators, 50% to 74.9% for 3 indicators, and less than 50% for 8 indicators.