Mean corrected generalized estimating equations for longitudinal binary outcomes with report bias

Area of Science:

• Public Health
• Addiction Medicine
• Biostatistics

Background:

• Cocaine addiction presents a significant global public health challenge.
• Cognitive-behavioral therapy (CBT) is a key intervention for cocaine dependence recovery and relapse prevention.
• Existing statistical methods may be invalidated by self-report bias in cocaine use data.

Purpose of the Study:

• To evaluate the effectiveness of cognitive-behavioral therapy in treating cocaine dependence.
• To address the challenge of self-report bias in longitudinal binary outcome data.
• To propose and validate a novel statistical approach for analyzing biased self-reported data.

Main Methods:

• Collected self-reported cocaine use and urine test data from patients at Yale-New Haven Hospital.
• Utilized generalized estimating equations (GEE) as a baseline method.
• Developed and applied a novel mean corrected generalized estimating equations (MCGEE) approach to account for reporting bias.

Main Results:

• The proposed MCGEE method provides consistent and asymptotically normal estimators even with reporting bias.
• Simulation studies demonstrated the superior performance of MCGEE compared to standard GEE.
• Real data analysis confirmed the practical utility of the MCGEE approach in a cocaine use study.

Conclusions:

• The MCGEE approach offers a more valid statistical framework for analyzing longitudinal binary outcomes affected by self-report bias.
• Accurate urine test results are crucial for identifying and correcting for bias in self-reported drug use.
• This methodology enhances the reliable assessment of interventions like CBT for cocaine addiction.