phase-in (months 15-21), and postimplementation (months 22-36). We excluded the phase-in from the main analysis. Covariates included age, sex, race, ethnicity, tumor type, and a monthly time trend.We estimated the average marginal effect of each exposure as of December 2018 as the absolute percentage-point difference between preimplementation and postimplementation predicted probabilities of zoledronate prescription. An interaction between nudge exposure and study phase facilitated comparisons of each nudge's marginal effect vs control. All tests were 2-sided with an α of 0.025 (corrected for 2 comparisons), using Stata statistical software (version 16, StataCorp).| Across 7 practice sites, 220 clinicians prescribed 14 701 zoledronate or denosumab prescriptions to 2595 patients (1926 [74.2%] women) with breast (1528 [58.9%]), lung (730 [28.1%]), or prostate (337 [13.0%]) cancer. The mean (SD) age was 62.7 (12.3), 67.0 (12.7), and 66.8 (11.9) years at sites A, B, and other, respectively.The Figure shows unadjusted trends in monthly zoledronate prescription by nudge exposure. As of December 2018, nudges were associated with statistically significant increases in the probability of zoledronate prescription compared with the control: site A, 26.0 percentage-point increase (95% CI, 18.5-33.4; P < .001 with leadership endorsement and performance feedback), and site B, 44.9 percentage-point increase (95% CI, 18.5-71.4; P = .001 with additionally accountable justification) (Table ). Results were robust to exclusion of clinicians practicing at multiple sites (n = 22).