Validation of a Bayesian learning model to predict the risk for cannabis use disorder.

BACKGROUND: Cannabis use disorder (CUD) is a growing public health problem. Early identification of adolescents and young adults at risk of developing CUD in the future may help stem this trend. A logistic regression model fitted using a Bayesian learning approach was developed recently to predict the risk of future CUD based on seven risk factors in adolescence and youth. A nationally representative longitudinal dataset, Add Health was used to train the model (henceforth referred as Add Health model). METHODS: We validated the Add Health model on two cohorts, namely, Michigan Longitudinal Study (MLS) and Christchurch Health and Development Study (CHDS) using longitudinal data from participants until they were approximately 30 years old (to be consistent with the training data from Add Health). If a participant was diagnosed with CUD at any age during this period, they were considered a case. We calculated the area under the curve (AUC) and the ratio of expected and observed number of cases (E/O). We also explored recalibrating the model to account for differences in population prevalence. RESULTS: The cohort sizes used for validation were 424 (53 cases) for MLS and 637 (105 cases) for CHDS. AUCs for the two cohorts were 0.66 (MLS) and 0.73 (CHDS) and the corresponding E/O ratios (after recalibration) were 0.995 and 0.999. CONCLUSION: The external validation of the Add Health model on two different cohorts lends confidence to the model's ability to identify adolescent or young adult cannabis users at high risk of developing CUD in later life.

A Bayesian learning model to predict the risk for cannabis use disorder.

BACKGROUND: The prevalence of cannabis use disorder (CUD) has been increasing recently and is expected to increase further due to the rising trend of cannabis legalization. To help stem this public health concern, a model is needed that predicts for an adolescent or young adult cannabis user their personalized risk of developing CUD in adulthood. However, there exists no such model that is built using nationally representative longitudinal data. METHODS: We use a novel Bayesian learning approach and data from Add Health (n = 8712), a nationally representative longitudinal study, to build logistic regression models using four different regularization priors: lasso, ridge, horseshoe, and t. The models are compared by their prediction performance on unseen data via 5-fold-cross-validation (CV). We assess model discrimination using the area under the curve (AUC) and calibration by comparing the expected (E) and observed (O) number of CUD cases. We also externally validate the final model on independent test data from Add Health (n = 570). RESULTS: Our final model is based on lasso prior and has seven predictors: biological sex; scores on personality traits of neuroticism, openness, and conscientiousness; and measures of adverse childhood experiences, delinquency, and peer cannabis use. It has good discrimination and calibration performance as reflected by its respective AUC and E/O of 0.69 and 0.95 based on 5-fold CV and 0.71 and 1.10 on validation data. CONCLUSION: This externally validated model may help in identifying adolescent or young adult cannabis users at high risk of developing CUD in adulthood.

A preliminary risk prediction model for cannabis use disorder.

The ongoing trend toward legalization of cannabis for medicinal/recreational purposes is expected to increase the prevalence of cannabis use disorder (CUD). Thus, it is imperative to be able to predict the quantitative risk of developing CUD for a cannabis user based on their personal risk factors. Yet no such model currently exists. In this study, we perform preliminary analysis toward building such a model. The data come from n = 94 regular cannabis users recruited from Albuquerque, New Mexico during 2007-2010. As the data are cross-sectional, we only consider risk factors that remain relatively stable over time. We apply statistical and machine learning classification techniques that allow n to be small relative to the number of predictors. We use predictive accuracy estimated using leave-one-out-cross-validation to evaluate model performance. The final model is a LASSO logistic regression model consisting of the following seven risk factors: age; level of enjoyment from initial cigarette smoking; total score on Impulsive Sensation-Seeking Scale questionnaire; score on cognitive instability factor of Barratt Impulsivity Scale questionnaire; and scores on neuroticism, openness, and conscientiousness personality traits of Neuroticism, Extraversion, and Openness inventory. This model has an overall accuracy of 0.66 and the area under its receiver operating characteristic curve is 0.65. In summary, a preliminary relative risk model for predicting the quantitative risk of CUD is developed. It can be employed to identify users at high risk of CUD who may be provided with early intervention.