American Statistical Association
Objective: In patients seeking treatment for substance dependence, key targets of treatment such as drug abuse and HIV risk behaviors tend to vary widely in their level of severity at baseline, particularly for “practical” or effectiveness trials where exclusionary criteria are minimized. The purpose of this analysis is to explore the importance of including baseline severity in the analysis of treatment effects on HIV risk behavior in a clinical trial of a drug abuse treatment.
Methods: In a randomized, community-based multisite trial sponsored by the National Institute on Drug Abuse, Clinical Trials Network, among 353 women with substance dependence and post-traumatic stress disorder, Seeking Safety (SS), a cognitive behavioral intervention, reduced HIV risk behavior more than the control condition (Women’s Health Education (WHE), among patients with higher levels of HIV risk behavior at baseline (Hien et al., in press). The primary outcome of HIV risk was the count of unprotected sexual occasions (USO). For the present analysis, the distribution of USO at baseline is examined, and generalized linear mixed models to test the effect of treatment on USO are compared with and without inclusion of baseline USO and the baseline by treatment interaction in the model.
Results: The counts of USO vary widely and conform best to a zero-inflated negative binomial distribution. Zero-inflated negative binomial regression models were fit. Compared to the model that includes baseline USO, where a baseline by treatment interaction was detected, no treatment effect is detected in the model that does not include baseline as a factor.
Conclusions: Inclusion of the baseline level of the outcome variable in the model may increase power to detect effects of treatment to the extent that baseline and outcome scores are correlated, or may detect treatment effects in the form of a baseline by treatment interaction. Like the interaction observed in this study, experimental treatment may be more effective among patients with higher levels of baseline severity, a ‘symptom reduction’ effect. Or, experimental treatment may be more effective among those with low severity at baseline, a ‘relapse prevention’ effect. Such effects may be important to developing treatment guidelines and to understanding the mechanisms of action of treatments. Statistical models that do not account for baseline severity will miss such effects. Baseline score of the outcome measure and the baseline by treatment interaction should be considered for inclusion in the plans for primary outcome analysis and exploratory analyses of intervention trials for substance abuse, HIV risk, and other behavioral health outcomes.
|Date:||Tuesday, September 22, 2009|
|Time:||3:00 - 4:00 P.M.|
New York State Psychiatric Institute
1051 Riverside Drive
6th Floor Multipurpose Room (6602)
New York, New York