[Washington DC] – [11/6/2024] – A new study reveals significant inconsistencies in the assignment of International Classification of Diseases (ICD) codes, raising concerns about the reliability of using ICD codes for creating patient cohorts and defining phenotypes in observational studies, which is a widely spread practice.
The research, titled "Are ICD Codes Reliable for Observational Studies? Assessing Coding Consistency for Data Quality," was conducted by a team from the Biomedical Informatics Center of George Washington University, led by Dr. Stuart J. Nelson. The study examined the transition from ICD-9CM to ICD-10CM coding within the VA system using advanced deep learning and statistical models to evaluate the consistency of ICD code assignments over time and across different care locations.
The key findings of study include: Among the 687 most-used ICD code clusters, a substantial number experienced significant changes during the transition to ICD-10CM, with 66% showing problematic inconsistencies; 37% of these changes had no clear explanation, indicating potential issues in the application of codes; and Variability in coding was especially pronounced across different care locations, pointing to challenges in creating consistent and reliable cohorts for research purposes.
The study highlights the potential risks of relying solely on ICD codes to establish patient cohorts for observational studies. Inconsistencies in coding, particularly during major transitions like the shift from ICD-9CM to ICD-10CM, can compromise data quality and lead to inaccurate research results. "The variations in coding we observed, particularly during the transition to ICD-10CM, raise significant concerns about the reliability of this data for research, especially in multi-site studies. We recommend researchers proceed with caution when using ICD codes to define patient cohorts."said Dr. Stuart J. Nelson, first author of the study.
This work was supported by the United States Veterans Administration Health Services Research Department grant 1I21HX003278-01A1; and by the Agency for Healthcare Research and Quality grant R01 HS28450-01A1. For more information about the study or to request interviews, please contact: Stuart J. Nelson, stunelson@gwu.edu. View the full paper.
About the Biomedical Informatics Center at George Washington University:
The Biomedical Informatics Center at George Washington University is a leading research institution dedicated to advancing medical knowledge and improving healthcare through innovative studies and data analysis.