Data consistency in the English Hospital Episodes Statistics database

Area of Science:

• Health Informatics
• Data Quality Assessment
• Clinical Coding

Background:

• Large administrative healthcare datasets are crucial for insights but require data quality assessment.
• Internal consistency evaluation is vital when a gold standard for criterion validity is unavailable.
• This study focuses on International Statistical Classification of Diseases and Related Health Problems, tenth revision (ICD-10) code recording inconsistencies in England's Hospital Episodes Statistics.

Purpose of the Study:

• To identify and quantify recording inconsistencies of mandatory ICD-10 codes.
• To analyze factors associated with coding inconsistencies in administrative healthcare data.
• To assess the potential impact of data quality issues on patient care.

Main Methods:

• Selected three exemplar conditions with mandatory ICD-10 coding: autism, type II diabetes mellitus, and Parkinson's disease dementia.
• Utilized random forest classifiers to identify variables linked to coding inconsistencies in hospital spells (April 2013-March 2021).
• Analyzed coding patterns, focusing on first occurrences, subsequent spells, and spells without overnight stays.

Main Results:

• Significant inconsistencies were found: 43.7% for autism, 8.6% for diabetes, and 31.2% for Parkinson's disease dementia in subsequent spells.
• Inconsistencies correlated with non-coding of underlying conditions, hospital trust changes, and time gaps between spells.
• Non-overnight stay spells showed higher inconsistency rates for diabetes and Parkinson's disease dementia.

Conclusions:

• Data inconsistencies for mandatory diagnoses are prevalent in administrative datasets.
• Inaccurate coding can potentially impact clinical decision-making and patient care.
• Emphasizes the need for robust data quality checks in healthcare administrative systems.