A Moyer1, J W Finney, J T Elworth, H C Kraemer. 1. Center for Health Care Evaluation, Veterans Affairs Palo Alto Health Care System, Menlo Park, California 94025, USA.
Abstract
OBJECTIVE: Despite enthusiasm for the potential of matching patients to alcohol treatments to improve outcomes, consistent findings have not emerged. This review considers the extent to which methodological factors may account for the pattern of findings from research on Patient x Alcohol Treatment interactions. METHOD: We focused on 55 studies that compared more than one type of alcohol treatment and included formal statistical tests for interactions. We examined four predictors of the number of significant interactions found in the 55 studies: (1) the number of statistical tests for interactions conducted, (2) the average number of participants, (3) whether or not participants were randomized to treatment and (4) the proportion of tested interactions that were hypothesis- or rationale-driven, as opposed to exploratory. RESULTS: Only the number of statistical tests for interactions predicted the number of patient-treatment interactions identified per study (zero-order r = 0.47; r2 = 0.22). A substantial number of tests for interactions (43) was conducted, on average, per study. Only a minority of the studies (33%) included enough participants to have a reasonable probability (0.80) of identifying a medium-sized matching effect. CONCLUSIONS: Drawing general conclusions regarding matching patients to alcohol treatments is hampered because Type I error has contributed to the matches identified, studies in this area are often underpowered, and the combinations of patient and treatment variables that have been tested are few relative to the numerous possible combinations. To be productive, future research will need to focus on patients at the extremes of matching dimensions and on distinct treatments. (J Stud. Alcohol 62: 62-73, 2001)
OBJECTIVE: Despite enthusiasm for the potential of matching patients to alcohol treatments to improve outcomes, consistent findings have not emerged. This review considers the extent to which methodological factors may account for the pattern of findings from research on Patient x Alcohol Treatment interactions. METHOD: We focused on 55 studies that compared more than one type of alcohol treatment and included formal statistical tests for interactions. We examined four predictors of the number of significant interactions found in the 55 studies: (1) the number of statistical tests for interactions conducted, (2) the average number of participants, (3) whether or not participants were randomized to treatment and (4) the proportion of tested interactions that were hypothesis- or rationale-driven, as opposed to exploratory. RESULTS: Only the number of statistical tests for interactions predicted the number of patient-treatment interactions identified per study (zero-order r = 0.47; r2 = 0.22). A substantial number of tests for interactions (43) was conducted, on average, per study. Only a minority of the studies (33%) included enough participants to have a reasonable probability (0.80) of identifying a medium-sized matching effect. CONCLUSIONS: Drawing general conclusions regarding matching patients to alcohol treatments is hampered because Type I error has contributed to the matches identified, studies in this area are often underpowered, and the combinations of patient and treatment variables that have been tested are few relative to the numerous possible combinations. To be productive, future research will need to focus on patients at the extremes of matching dimensions and on distinct treatments. (J Stud. Alcohol 62: 62-73, 2001)