A systematic approach to subgroup analyses in a smoking cessation trial

Area of Science:

• Psychiatry and Behavioral Science
• Clinical Trial Methodology
• Statistical Modeling

Background:

• Traditional subgroup analyses in clinical trials face challenges including multiple comparisons, model misspecification, and multicollinearity, potentially leading to erroneous conclusions.
• These limitations hinder the accurate identification of patient subgroups that may respond differently to treatments.

Purpose of the Study:

• To introduce and demonstrate a novel, systematic Best Approximating Model (BAM) approach for subgroup analyses.
• To overcome the pitfalls associated with traditional hypothesis-driven subgroup testing.

Main Methods:

• Applied the Best Approximating Model (BAM) to a randomized, controlled trial of OROS-methylphenidate versus placebo with nicotine patch for smoking cessation in adults with attention-deficit/hyperactivity disorder (ADHD).
• The trial involved 255 adult smokers over 11 weeks, measuring prolonged and point prevalence smoking abstinence.
• BAM identifies multiple moderators and estimates their simultaneous impact on treatment effect sizes.

Main Results:

• While the original analysis found no overall treatment effect on smoking cessation, BAM identified significant subgroup effects.
• Key moderators for prolonged smoking abstinence included lifetime history of substance use disorders and symptom severity of ADHD.
• Significant subgroup effects were also observed for point prevalence smoking abstinence in younger adults (18-29 years).

Conclusions:

• The BAM approach yielded different conclusions regarding subgroup effects compared to hypothesis-driven methods.
• BAM's examination of moderator independence and avoidance of multiple testing offers a more robust method for identifying treatment effect variations across patient subgroups.
• This approach has the potential to enhance personalized medicine by better guiding the matching of individual patients to specific treatments.