Machine-Learning Prediction of Comorbid Substance Use Disorders in ADHD Youth Using Swedish Registry Data

Abstract

Background: Children with attention-deficit/hyperactivity disorder (ADHD) have a high risk for substance use disorders (SUDs). Early identification of at-risk youth would help allocate scarce resources for prevention programs.Methods: Psychiatric and somatic diagnoses, family history of these disorders, measures of socioeconomic distress and information about birth complications were obtained from the national registers in Sweden for 19,787 children with ADHD born between 1989-1993. We trained 1) crosssectional machine learning models using data available by age 17 to predict SUD diagnosis between ages 18-19; and 2) a longitudinal model to predict new diagnoses at each age.Results: The area under the receiver operating characteristic curve (AUC) was 0.73 and 0.71 for the random forest and multilayer perceptron cross-sectional models. A prior diagnosis of SUD was the most important predictor, accounting for 25% of correct predictions. However, after excluding this predictor, our model still significantly predicted the first-time diagnosis of SUD during age 18-19 with an AUC of 0.67. The average of the AUCs from longitudinal models predicting new diagnoses one, two, five and ten years in the future was 0.63.Conclusions: Significant predictions of at-risk co-morbid SUDs in individuals with ADHD can be achieved using population registry data, even many years prior to the first diagnosis. Longitudinal models can potentially monitor their risks over time. More work is needed to create prediction models based on electronic health records or linked population-registers that are sufficiently accurate for use in the clinic.

Social and Behavioral Determinants of Health in the Era of Artificial Intelligence with Electronic Health Records: A Scoping Review

“…(3) Neural Networks. Eight studies utilized neural networks [27,39,44,52,53,65,87,100]. Shi et al [87] implement bidirectional RNN to predict pediatric diagnosis whereas Vrbaški et al [100] develop predictors for lipid profile prediction.…”

“…Overall, 29.5% of the studies reviewed focused on SBDH factors associated with healthcare access and quality, 24.7% focused on economic stability, 20% focused on social and community context, 16.3% focused on neighborhood and built environment, and 9.5% focused on education access and quality. Widely studied SBDH factors include substance use/abuse (9%) [14,16,[20][21][22][23][24][25][26][27][28][29][30][31][32][33][34][35], education (7.3%) [16,21,[36][37][38][39][40][41][42][43][44][45][46], employment status (6.3%) [16,20,21,23,31,36,38,39,[45][46][47][48], socioeconomic status (6.3%) [29,39,44,46,[49]…”

“…(2.1%) [89][90][91][92][93]. Other significant SBDH factors include geographic location [24,31,44,70], health literacy [43,47,88,94], social patterns (sexual health, adverse experiences and behavioral attitudes) [33,52,73,95], social environment [41,56,77], health access [21,54,88], living condition [20,31,35], social behavior [54,63,82], and financial insecurity [81,86].…”

“…The association between SBDH factors such as socioeconomic factors, employment status, SES and mental, behavioral, and neurodevelopmental disorders (opioid misuse, attention deficit hyperactivity disorder (ADHD), substance use disorders (SUDs)) and injury, poisoning, and certain other consequences of external causes (suicide, postdeployment stress) was widely studied. Zhang-James et al and Afshar et al studied the impact of low socioeconomic status and socioeconomic distress in individuals with at-risk comorbid SUDs [44] and opioid misuse [46], respectively. Zheng et al developed an earlywarning system to identify patients at risk of suicide attempts [39], while a few studies stated that the predictors like SES [53] and socioeconomic factors [64] can be used to predict suicide risk.…”

“…Zhou et al demonstrated the viability of NLP techniques for the automated evaluation of a large number of lifestyle habits in patients with Alzheimer's disease using free text EHR data [22]. Zhang-James et al used machine learning algorithms to construct prediction models to identify ADHD youth at risk for SUDs by taking socioeconomic, educational, and geographic data into account [44].…”