Assessing the validity of ICD-10 administrative data in coding comorbidities.

Area of Science:

• Health Informatics
• Medical Coding Accuracy
• Administrative Data Analysis

Background:

• Administrative data are crucial for chronic disease prevalence estimation and health informatics research.
• The International Classification of Diseases, 10th Revision (ICD-10) is widely used for coding comorbidities in administrative datasets.
• Ensuring the accuracy of ICD-10 coded comorbidities is vital for reliable health data.

Purpose of the Study:

• To assess the validity of coding comorbidities in administrative data using the ICD-10 system.
• To evaluate changes in ICD-10 coding accuracy over a 20-year period.
• To determine the impact of undercoding on predicting in-hospital mortality.

Main Methods:

• Analysis of three chart review cohorts from Alberta, Canada (2003, 2015, 2022) involving 4008, 3045, and 9024 patients, respectively.
• Nurse reviewers assessed 17 clinical conditions using a standardized protocol, with chart review data serving as the reference standard.
• Comparison of comorbidity coding accuracy between chart reviews and ICD-10 administrative data.

Main Results:

• A mean difference in prevalence between chart reviews and ICD-10 coding was observed: 2.1% (2003), 7.6% (2015), and 6.3% (2022).
• Coding sensitivity for 17 conditions varied significantly, ranging from 39.6-85.1% in 2003 to 3.0-89.7% in 2022.
• While ICD-10 validity decreased over time, its impact on predicting in-hospital mortality remained minimal, with C-statistics of 0.84 (2003), 0.81 (2015), and 0.78 (2022).

Conclusions:

• Comorbidities were increasingly undercoded in administrative data over two decades.
• The validity of ICD-10 coding for comorbidities has decreased, though it remained relatively stable for certain mandated conditions.
• Potential exists for artificial intelligence (AI) methods using electronic health records to enhance coding practices and data quality.