Related Experiment Video
Updated: Jul 12, 2026
A Machine Learning Approach to Design an Efficient Selective Screening of Mild Cognitive Impairment
Published on: January 11, 2020
Using Machine Learning with the Brief Symptom Inventory to Screen for Comorbid Addictions: A Proof-of-Principle Study
Maor Daniel Levitin1,2, Dvora Shmulewitz3, Roi Eliashar4
1School of Psychology, Tel-Aviv University, Tel Aviv, Israel, maorl@ica.org.il.
Introduction:
Mental health disorders are leading causes of morbidity worldwide and often co-occur with substance use and other addictive behaviors. However, many individuals in mental healthcare settings are not screened for addictions. General psychopathology measures, such as the Brief Symptom Inventory (BSI), may offer valuable insights into addiction risk. This study explores whether machine learning models can utilize BSI responses to predict problematic substance use (alcohol, drugs) and addictive behaviors (gambling, gaming, hypersexual behavior, and pornography use).
Methods:
A population sample of Jewish adults in Israel (N = 2,451) was assessed for mental health, including the BSI, problematic substance use (alcohol, drugs), and potentially addictive behaviors: gambling, gaming, compulsive sexual behavior, and pornography use. Machine learning models - including decision trees, random forest, boosting, LASSO, subset selection, and elastic net regression - were employed to predict addiction outcomes based on BSI responses.
Results:
The BSI demonstrated varying degrees of predictive ability across different addictions and models. Test set explained variance (R2) on the continuous track ranged from 1% (gambling, decision tree) to 25% (compulsive sexual behavior, random forest); area under the precision-recall curve (AUPRC) ranged from 0.04 (pornography, LASSO) to 0.36 (gaming, decision tree), in each case exceeding the prevalence baseline. Behavioral addictions generally showed descriptively higher predictability than substance use disorders.
Conclusions:
The BSI contains substantial information on addiction risk, providing preliminary, proof-of-principle evidence that an existing general psychopathology measure carries item-level signal relevant to comorbid addiction screening, with the strongest signal for behavioral addictions. These findings serve as proof of principle for the notion that existing psychopathology measures can be leveraged to inform about potential comorbidity. Being able to efficiently identify which mental health patients must be screened for addictions will help to determine the most appropriate interventions.
