Bridging the gap: predicting early contingency management outcomes

Background: Contingency management (CM) is a psychosocial treatment used to improve socially significant behaviors. Its efficacy has been demonstrated many times over in a variety of treatment contexts. Development of predictive modeling has lagged behind other technological advancements in treatment and no studies have relied on predictive models to provide additional intervention supports to those who may not be successful under default treatment parameters.Objectives: This study validates a framework for predicting early CM treatment success and leveraging predictions from the early success framework as predictors in an end-of-treatment outcome model.Methods: Classification models predicted outcomes on the first two urine toxicology and breathalyzer tests of combined clinical trials consisting of more than 800 participants (426 female). Probabilities from that model were used as predictors in a secondary model that predicted continuous abstinence of at least 4 weeks (LDA-4) and at least 6 weeks (LDA-6) at the end of treatment. Two models were used because assessment results are only predictive of tests in close temporal proximity, and previous literature has already revealed a link between early tests and end of trial outcomes.Results: The best performing early treatment success model achieved a receiver operating characteristic area under the curve (ROC-AUC) score of 0.83. Models classifying LDA-4 and LDA-6 achieved ROC-AUC scores of 0.81 and 0.77 respectively.Conclusion: Baseline assessments can be used to predict treatment outcomes, allowing clinicians and researchers the opportunity to identify participants who may only benefit from modified CM, such as by offering higher magnitude incentives.