Cummings 1989b.

Study characteristics
Methods	Design: Cluster‐randomized controlled trial Setting: 4 health maintenance organization medical centres, USA Recruitment: Conducted in clinic waiting rooms
Participants	81 providers 2056 English‐speaking people who made a visit to any doctor participating in the study were eligible for inclusion. (1032 control, 1024 intervention), av.age 45, 55% F, 17 cpd
Interventions	Intervention: • Physicians received training on smoking cessation counseling in 3 x 1‐hour seminars • Practices were provided with free self‐help materials, stickers, quit date prescription pads, and posters. Nurses and office staff were coached on the program and supporting materials by a member of the research staff Control: usual care. No further description on what the usual care entailed was reported
Outcomes	Continuous 9m abstinence at 12m follow‐up Validation: CO levels (cut‐off not defined), salivary cotinine < 30 ng/ml Quit attempts Measures of provider implementation: Ask, Assist‐Self‐help, Assist‐Quit date, Assist‐Prescribe, Arrange
Funding Source	Partial support by grant CA38337 from the National Cancer Institute. Dr. Cummings' work was supported in part by the Henry J. Kaiser Foundation Faculty Fellowship in General Internal Medicine
Author's declarations of interest	Not reported
Notes	Strategy: Provider training + Outreach facilitation Level: Provider + Practice Comparison type: Multi‐component vs. standard care
Risk of bias
Bias	Authors' judgement	Support for judgement
Sequence Generation	Low risk	QUOTE: "A computer randomly assigned the units to either the experimental or control group"
Allocation concealment	Low risk	QUOTE: "A computer randomly assigned the units to either the experimental or control group"
Blinding of outcome assessors All outcomes	Low risk	Smoking status was biochemically validated
Incomplete outcome data All outcomes	Low risk	Attrition rates were under 50% and similar between groups. The overall loss to follow‐up was 24.7% (n = 507/2056); 23.5% (n = 241/1024) in the intervention group and 25.8% (n = 266/1032) in the control group were lost to follow‐up at 1 year
Recruitment bias (cluster RCTs only)	Low risk	Participants were affiliated with the practice before randomization
Balanced baseline characteristics? (cluster RCTs only)	Low risk	Similar between experimental and control groups
Adjustment for clustering in analysis? (cluster RCTs only)	Low risk	QUOTE: "We tested the effect of this two‐stage cluster sampling design by estimating logistic regression models with random effects terms representing the groupings by physician. These adjustments had no substantial effect on comparisons between the experimental and control groups. Therefore, for simplicity, we present the results with the patient as the unit of analysis"