Automated Substance Use Detection from Electronic Health Records in the Pediatric Setting

点击翻译按钮获取中文摘要

Project Summary The majority of adults with substance use disorders (SUD) report beginning to use substances as adolescents; thus, adolescence represents a critical time for screening of substance use initiation and implementing interventions to prevent or reduce use. The healthcare system prioritizes substance use screening, including during adolescence, with the goal of identifying when substance use is occurring and monitoring use to determine necessary intervention. Unfortunately, the majority of this information is documented in unstructured clinical notes, making it difficult for providers to monitor change in substance use for an adolescent over encounters. Published studies also suggest bias around substance use screening such as laboratory tests exists. Both limitations prevent electronic health record (EHR) data from being used to study the contexts and consequences of substance use in populations of adolescents. Rather than changing clinician behavior, which can be challenging, this study utilizes automated artificial intelligence algorithms to detect substance use screening occurrences and results in the EHRs. Our work could allow current provider-led preferences and practices in substance use documentation to continue while simultaneously increasing access to documented information and mitigating screening bias to avoid perpetuating racism and inequity in healthcare. As a result, the study has the potential to aid in long-term efforts to target prevention, intervention, and referral for treatment in adolescence and ultimately reduce risk of SUD across the lifespan. Our work will be completed through accomplishing the following aims: Aim 1: Examine the generalizability of an automated substance use detection system in a sample of ~5,000 adolescent patients who receive well child and/or outpatient specialty visits, maximizing contexts where substance use screening is most likely to occur; and Aim 2: Assess differences in substance use screening and positive screening results by gender, insurance type, minoritized race and ethnicity status, and clinical context, evaluating whether bias is detected in structured data, unstructured data, or both data sources. In addition, participatory research principles will be used to solicit feedback from clinicians and researchers about the application of findings to clinical care. By the end of the funding period, we will have validated the performance of the automated system, assessed bias in identifying substance use screening results, and gained insights from clinician feedback about application to clinical care.