(136) A systematic review of algorithms used to identify neurodevelopmental disorders in electronic healthcare databases

Background: Investigating neurodevelopmental outcomes in safety studies based on secondary use of data is often of interest but challenging given heterogeneous clinical definitions and medical coding systems. Validation of endpoints is critical for robust pharmacoepidemiology research on potential risks of drugs/vaccines during pregnancy.

Objectives: To perform a systematic literature review to summarize algorithms that have been used in electronic healthcare databases to identify neurodevelopmental disorders (NDD) in studies conducted in Europe, Canada, and the USA.

Methods: We developed a search strategy to identify studies on NDD describing phenotype algorithms or code sets, from January 1st, 2010 to April 1st, 2023. Our study adhered to the Preferred Reporting Items for Systematic Reviews (PRISMA) guidelines of 2021, and PROSPERO registration number CRD42023489703. The search strategy included key terms to identify NDD (e.g., intellectual developmental disorders, communication disorders, autism spectrum disorders (ASD), attention deficit hyperactivity disorders (ADHD), specific learning disorders, motor disorders, tic disorders), electronic healthcare data, timing of exposure (pregnancy) and terms to identify studies likely to contain algorithms. Extractions used a standardized data collection form that included NDD definition, algorithm/codes, prevalence in unexposed, and if available validation method and performance metrics. Eligibility criteria assessment and data extraction were performed by two independent reviewers; inconsistencies were checked by a third reviewer. Appropriate summary descriptive statistics were used for categorical and continuous variables.

Results: 113 studies were identified, most used unvalidated algorithms, and required only one ICD-9 or ICD-10 code. 34.5% of studies were conducted in USA, 9.7% in Canada, and the rest in Nordic countries, France, and the UK. 56% of the studies used registries, 23% electronic medical records, and 16.8% insurance claims. ASD (35.7%), followed by ADHD (34.6%), and mental retardation/intellectual disability (15.1%) were most commonly studied. We identified 8 studies that validated NDD algorithms. The prevalence of NDD in unexposed populations varied across studies (e.g., ASD: 0.5-2.6%; ADHD: 1.1-15.5%), which in part varied mainly due to sampling variation.

Conclusions: Few studies used validated algorithms to identify NDD. This presentation will review the different NDD definitions used, describe the databases that can be used to study NDD, discuss assessment timing, and present the algorithms focusing on validated algorithms.