The impact of substance abuse, including tobacco, alcohol, illicit drugs, and prescription opioids, in the U.S. is immense, with the combined cost of its effects on health care, related crime, and lost productivity exceeding $740 billion each year. Critical to addressing this national crisis is the need for improved data analysis and statistical tools that enable substance abuse researchers to more effectively exploit their datasets. Such tools provide improved detection and estimation of statistical regularities, thus offering the opportunity for new clinical findings that support the accelerated development of targeted treatments for individuals or groups suffering from substance use disorders (SUD). In particular, improving statistical methods and tools for subgroup analysis on clinical trial data would be invaluable to SUD research. Subgroup analyses are utilized in efficacy and effectiveness studies and randomized controlled trials to identify heterogeneity in treatment effects (HTE) by assessing how clinical factors impact estimates of treatment effect sizes. This allows researchers to better understand and explain how patients respond differently when clinical, patient, and system factors moderate treatment effect sizes. Such analyses are also utilized, especially in the absence of valid clinical theory, by practitioners, administrators, and policy makers to make decisions that impact patients, their quality of life, and health care costs. This study will extend and evaluate the Best Approximating Model (BAM) technology as an advanced statistical modeling approach that offers improvements over conventional subgroup data analysis methods. First, BAM uses a systematic model search and selection strategy to incorporate known or posited factors while simultaneously identifying subgroups and estimating their impact on treatment effect sizes. Second, BAM includes robust estimation, specification analysis, stochastic/exhaustive model search, and validation within the single model selection framework of a generalized additive model. Third, a BAM is designed to handle common problems encountered in subgroup analyses including possible model misspecification and overfitting as well as multicollinearity, small sample size bias, and Type I error inflation due to multiple comparisons. Phase I research will investigate extending BAM technology as an advanced statistical modeling tool BAM- HTE for robust subgroup analysis by: (i) incorporating new model selection functionality, (ii) implementing novel subgroup specification testing and validation methods, (iii) evaluating its reliability and validity for robust subgroup analysis, and (iv) demonstrating its robust subgroup analysis capability on a NIDA-funded (CTN-0037) dataset. Phase I results will establish the essential feasibility for Phase II BAM-HTE prototype development, evaluation, and findings dissemination, which in turn will provide the foundation for subsequent Phase III commercialization.
The impact of substance abuse, including tobacco, alcohol, illicit drugs, and prescription opioids, in the U.S. is immense, with the combined cost of its effects on health care, related crime, and lost productivity exceeding $740 billion each year. Critical to addressing this national crisis is the need for improved data analysis and statistical tools that enable substance abuse researchers to more effectively exploit their datasets. Such tools provide improved detection and estimation of statistical regularities, thus offering the opportunity for new research findings that support the accelerated development of targeted treatments for individuals or groups suffering from substance use disorders (SUD). In particular, improving statistical methods and tools for subgroup analysis on clinical trial data would be invaluable to SUD research. This Phase I feasibility study will investigate extending Best Approximating Model (BAM) technology for subgroup analysis by: (i) incorporating new model selection functionality, (ii) implementing novel subgroup specification testing and validation methods, (iii) evaluating its reliability and validity for robust subgroup analysis, and (iv) demonstrating its robust subgroup analysis capability on a NIDA-funded (CTN-0037) dataset.
