Multimodal predictive modeling of individual treatment outcome in cocaine dependence with combined neuroimaging and behavioral predictors

Background: Developing personalized treatments for cocaine dependence remains a significant clinical challenge. Positron emission tomography (PET) has shown that the [¹¹C]raclopride signal in the ventral striatum is associated with treatment success in a positively reinforced contingency management program. The present study investigates whether this signal can be used to predict treatment outcome at an individual level. Methods: Predictive models were developed using PET signals from 5 regions of the striatum and follow-up data in 24 patients, and evaluated using cross-validation. Results: The ventral striatal PET signal alone can predict individual treatment response with a substantial degree of accuracy (cross-validated correct rate. = 82%). Incorporating information from other regions-of-interest (ROIs) in the striatum does not improve predictive performance, except for a small improvement with adding the posterior caudate. The addition of baseline demographic variables, including baseline severity measures, does not improve predictive performance. On the other hand, early treatment response and motivation, reflected by cumulative clinic attendance, performs as well as the PET signal (83%) by week 3 in the 24-week study. The combined model with both PET signals and cumulative clinic attendance demonstrates a significant improvement of performance, peaking at 96% during week 3 of the trial. Conclusions: These results suggest that a multimodal model can predict treatment success in cocaine dependence at an individual level, and pose hypotheses for the underlying neural circuitry mechanisms responsible for individual variations in treatment outcome.