**1. Introduction**

Substance use disorder (SUD) frequently occurs among people with bipolar disorder (BD), worsening their clinical trajectories [1,2]. A comorbid diagnosis of BD and SUD occurs in up to 30–60% of people with SUD, depending on the substance used, including alcohol [3], cannabis [4], tobacco [5], or others [3,6,7], with men having higher lifetime risks of SUD

**Citation:** Oliva, V.; De Prisco, M.; Pons-Cabrera, M.T.; Guzmán, P.; Anmella, G.; Hidalgo-Mazzei, D.; Grande, I.; Fanelli, G.; Fabbri, C.; Serretti, A.; et al. Machine Learning Prediction of Comorbid Substance Use Disorders among People with Bipolar Disorder. *J. Clin. Med.* **2022**, *11*, 3935. https://doi.org/10.3390/ jcm11143935

Academic Editors: Ana Adan and Marta Torrens

Received: 9 June 2022 Accepted: 4 July 2022 Published: 6 July 2022

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

than women [8]. The presence of SUD accounts for a higher number of lifetime mood episodes and hospitalizations [9]; lifetime medical comorbidities [10]; reduced cognitive and psychosocial functioning [11]; and an increased risk for suicide [12], impulsive and aggressive behavior [13], or mortality [14]. Substance use may also attenuate the efficacy or compliance to psychopharmacological treatments, further worsening BD course [15,16].

The strongest comorbid associations of SUD among individuals with BD are found with alcohol use disorder (AUD), followed by cannabis and other illicit drugs [8]. Interestingly, the most current report by the National Epidemiological Survey on alcohol and related conditions [17] suggests that the presence of both alcohol use and having a psychiatric diagnosis, including BD, are associated with higher utilization rates of lifetime poly-substance abuse [18] compared with individuals without these clinical characteristics. Patients with BD with multiple SUDs have even more severe outcomes, including the risk of overdose, criminal conviction, low adherence to treatments, and reduced global functioning [10,19,20].

Despite its burden, the relationship between SUDs and BD has been minimally studied. Indeed, a few longitudinal studies have examined the predictors of SUD onset in BD, reporting that alcohol use disorder (AUD) might be predicted by psychotic symptoms [21], while cannabis use disorder might be predicted by younger age, lower education, and previous substance use [22]. In addition, the generalizability of much-published research on this issue is problematic, given that individuals who exclusively meet the criteria for a single SUD do not represent the naturalistic population in clinical settings [19]. This more significant symptomatic burden of comorbid SUDs in adults with BD points out the necessity of identifying the risk factors of co-occurrence in order to implement appropriate preventative strategies.

Evidence has suggested the feasibility of developing predictive models in psychiatry through machine-learning algorithms [23,24]. Several studies have used data mining and machine learning techniques to predict patient outcomes, including SUD [25]. However, to the best of our knowledge, no study has applied machine-learning techniques to date to predict the presence of comorbid SUD in individuals with BD. In addition, no studies on the topic have analyzed to what extent BD phenotypes differ according to the type of SUD.

The current study aims at identifying the most meaningful variables associated with SUD, AUD, and AUD in comorbidity with any other SUD in a large sample of patients with BD through the use of a random forest (RF) model. These variables will then be used in a regression model to provide further information on the associations between BD and specific types of SUD.

#### **2. Materials and Methods**
