This Behavioral Science Track Award for Rapid Transition (B/START) application proposes to investigate the use of an innovative analysis to improve outcomes and efficient delivery of interventions that reduce crack use. As part of NIDA's Cooperative Agreement for AIDS Community-Based Outreach, St. Louis crack users were randomly assigned to either: NIDA's standard educational intervention or the EachOneTeachOne (EOTO) enhanced education/counseling, peer-led intervention. The interventions reduced the overall sample's mean number of days of crack use in a month by 5 days (outcomes) from baseline to the three month follow-up. However, regression analysis of NIDA/EOTO data revealed that different types of crack users were helped most by the NIDA intervention or the EOTO intervention (an attribute-treatment interaction; ATI). If future crack users were assigned to either the NIDA or EOTO intervention based on which intervention was predicted to yield the best outcome for each individual, the number of days per month that crack was consumed would decrease from baseline to follow-up by 6.2 days- a 24 percent improvement in outcomes compared to randomly assigned interventions. It has been argued that regression techniques do not provide enough statistical power to detect ATIs. Indeed, several participant attributes correlated nearly significantly (p=.06 to .08) with outcomes in one intervention and not the other, so a more powerful analytical technique should improve the precision of predicting crack users' intervention outcomes. Not only could interventions be assigned to individuals with greater precision because of the more accurate estimates of future crack users' outcomes, reduction in crack use might be greater. One analytical technique, artificial neural network (ANN) analysis (used by engineers and economists), has evidenced better specificity and sensitivity than clinicians' diagnoses and regression techniques for medical diagnoses and outcomes. ANN may be more powerful than regression because ANN: assumes no particular data distribution (bell-shaped vs dichotomous), accounts for high-order interactions among variables without a-priori specification, and accounts for multicolinearity. In this study, ANN will be compared to linear regression in terms of their a) degree of error in predicting observed outcomes, b) power to detect ATIs, and c) clinical utility for ATI research using. The study will produce data regarding which intervention is more effective for reducing crack use among different types of crack users and will attempt to demonstrate whether ANN is a useful technique for detecting ATIs in this population. The study's implications are far-reaching as ATI research is used in many treatment, educational, and industrial settings.
| Ridenour, Ty A; Cottler, Linda B; Compton, Wilson M et al. (2003) Is there a progression from abuse disorders to dependence disorders? Addiction 98:635-44 |
| Ridenour, Ty A; Cottler, Linda B; Robins, Lee N et al. (2002) Test of the plausibility of adolescent substance use playing a causal role in developing adulthood antisocial behavior. J Abnorm Psychol 111:144-55 |
