Statistical Analysis Methods in the REACH Trial
The REACH randomized controlled trial used generalized estimating equation (GEE) models as the primary statistical analysis method to test intervention effects on HbA1c and secondary outcomes, with additional analyses including Cox proportional hazards models, regression models for engagement analysis, and mediation analysis to examine intervention mechanisms 1, 2.
Primary Analysis Approach
The trial employed generalized estimating equation (GEE) models to evaluate:
- Treatment effects on HbA1c relative to the control group across multiple time points (3,6,12, and 15 months)
- Potential additive effects of the FAMS (Family-focused Add-on for Motivating Self-care) component
- Effects on adherence to self-care behaviors and diabetes self-efficacy 1
This approach was appropriate for the repeated measures design, allowing analysis of outcomes collected at baseline and multiple follow-up points while accounting for within-subject correlation.
Secondary and Exploratory Analyses
Engagement Analysis
Multiple analytic approaches were used to understand the relationship between participant engagement and clinical outcomes 3:
- Modified instrumental variable (IV) analysis - This was the most informative approach, suggesting differences in treatment effects between engagers and non-engagers of -0.29% to -0.51% for HbA1c
- Standard regression models - Used to analyze associations between response rates to interactive texts and patient characteristics
- Mediation analysis - Examined whether barrier reduction mediated outcome improvements (though this did not show significant mediation effects) 2
- Moderation analysis - Tested whether treatment effects varied by baseline characteristics
Time-to-Event Analysis
Cox proportional hazards models stratified by wave were used to analyze:
- Time to HbA1c test completion
- Piecewise hazard ratios for different time intervals (weeks 1-4,5-12, and 13-24) 4
Subgroup and Effect Modification Analyses
The trial examined whether treatment effects differed by:
- Baseline HbA1c levels (particularly ≥8.5% vs <8.5%)
- Race/ethnicity
- Socioeconomic disadvantage indicators
- Other patient characteristics (age, gender, education, income, health literacy) 2
Key Methodological Strengths
The analysis plan demonstrated several rigorous features:
- Intention-to-treat analysis maintained randomization integrity
- High retention rates (>90%) minimized attrition bias 2
- Multiple time points allowed assessment of both short-term and sustained effects
- Leave-one-out sensitivity analyses assessed robustness of findings 3
- Stratified randomization by wave controlled for temporal effects
Important Analytical Considerations
The modified IV analysis proved more powerful than standard regression or mediation for detecting engagement effects, though it required careful specification of assumptions 3. The trial found that while REACH improved medication adherence and diet through 12 months, barrier reduction did not mediate these improvements, suggesting the intervention worked through mechanisms beyond simple barrier removal 2.
The analysis revealed that treatment effects were greatest at 6 months (-0.31% overall; -0.74% for those with baseline HbA1c ≥8.5%) but were not significant at 15 months, indicating the need for ongoing intervention to sustain benefits 2.