A preliminary risk prediction model for cannabis use disorder

Abstract

Highlights Substance use disorders are currently a major public health crisis in the US. The prevalence of cannabis use disorder is rising due to legalization of cannabis. This study built models to predict the risk of cannabis use disorder for a user. Risk factors include personality traits, impulsivity and initial smoking enjoyment.

Classical and Neural Network Machine Learning to Determine the Risk of Marijuana Use

“…Comparing this LR model to those surveyed in the literature, it outperforms the model implemented by Rajapaksha et al, which achieved an overall accuracy of 0.66 and AUC score of 0.65 [ 5 ]. The present LR model is at least comparable to Spechler’s model, which achieved an AUC of 0.71 for males and 0.81 for females [ 6 ].…”

“…In an annual survey conducted by the University of Michigan’s Institute for Social Research, marijuana has consistently been the most commonly used illicit drug [ 2 ]. This trend has led researchers to study the extent to which certain personality traits can identify risk factors for THC use in individuals [ 3 , 4 , 5 , 6 , 7 , 8 ]. Subsequently, intervention methods focusing on identified risk-prone personality traits to reduce the prevalence of THC use, particularly among adolescents, have also been studied [ 9 , 10 ].…”

“…Neural networks have been successful, and often superior to other algorithms, when making predictions in situations such as the success rates in a smoking cessation program [ 18 ], incidence of metabolic syndrome [ 19 ] and infectious disease surveillance [ 20 ]. Although it is becoming more common to use machine learning techniques to predict the risk of drug use or abuse [ 3 , 4 , 5 , 6 , 7 , 8 , 21 , 22 , 23 ], few researchers have employed NN models [ 21 , 23 ]. In a literature search, while four studies specifically modeled THC prediction using machine learning techniques, none employed a NN model.…”

“…In a literature search, while four studies specifically modeled THC prediction using machine learning techniques, none employed a NN model. Three of these studies utilized a form of LR [ 5 , 6 , 7 ] and one used a DT [ 8 ], with moderate to high predictive accuracy. These studies incorporated personality traits along with additional risk factors.…”

“…Rajapaksha et al utilized LASSO LR to predict cannabis use disorder (CUD) in known, regular THC users, achieving an overall accuracy of 0.66 and AUC of 0.65. Considering personality predictors, they found that CUD risk was higher in younger individuals prone to sensation seeking, having higher openness to experience and lower conscientiousness scores [ 5 ]. Fehrman et al found that a DT model produced the best predictive results for the risk of 18 different drugs in their study, reporting a sensitivity of 79% and specificity of 80% for the risk of THC use [ 8 ].…”