Electronic Health Record Based Algorithm to Identify Patients with Autism Spectrum Disorder

Area of Science:

• Medical Informatics
• Neurodevelopmental Disorders
• Clinical Epidemiology

Background:

• Electronic health record (EHR) data present challenges for large-scale Autism Spectrum Disorder (ASD) cohort identification due to unreliable diagnostic codes.
• International Classification of Diseases 9th edition (ICD-9) codes have limited accuracy in predicting disease status.

Purpose of the Study:

• To develop, evaluate, and validate an automated algorithm for precise ASD patient cohort selection from EHR data.
• To investigate the co-occurrence patterns of medical comorbidities in a large ASD cohort using the developed algorithm.

Main Methods:

• Extracted ICD-9 codes and clinical note concepts from EHR.
• Developed a rule-based algorithm and a predictive classifier for ASD cohort identification.
• Compared algorithm performance against ICD-9 codes and validated results across multiple institutions.

Main Results:

• The rule-based algorithm demonstrated significantly higher positive predictive values (PPV) for ASD cohort identification compared to baseline ICD-9 codes (e.g., 0.864 vs. 0.460 combined).
• Validation across three institutions confirmed the algorithm's robust performance (PPV of 0.848 at Children's Hospital of Philadelphia).
• Clustering analysis of comorbidities in over 20,000 ASD patients revealed distinct groups: psychiatric, developmental, and seizure disorders.

Conclusions:

• Automated algorithms enhance the accuracy of ASD cohort selection from EHR, facilitating large-scale research.
• Identified comorbidity patterns suggest distinct clinical trajectories within the ASD population.
• This work supports future EHR-based studies and the development of individualized ASD treatments.