Undercounting diagnoses in Australian general practice: a data quality study with implications for population health...

Area of Science:

• Health Informatics
• Clinical Epidemiology
• Data Quality

Background:

• Electronic medical records (EMRs) often contain diagnoses as free-text or coded terms.
• Current healthcare planning and quality improvement initiatives frequently exclude free-text data, focusing only on coded diagnoses.
• This reliance on coded data may lead to an underestimation of disease prevalence.

Purpose of the Study:

• To assess if using only coded diagnosis data leads to under-reporting of disease prevalence.
• To quantify the extent of under-reporting for six common chronic diseases.

Main Methods:

• A cross-sectional data quality study analyzed de-identified EMR data from 84 general practices in Victoria, Australia.
• The study included 456,125 patients with at least three EMR visits between January 2021 and December 2022.
• Compared patient counts from coded diagnoses versus clinically validated free-text entries for asthma, COPD, dementia, and diabetes types.

Main Results:

• All six chronic diseases showed undercounts when using coded diagnoses alone, ranging from 2.57% to 36.72%.
• Five of these undercounts were statistically significant.
• Overall, 26.4% of patient diagnoses were not coded, with significant practice-level variation in coding, except for type 2 diabetes.

Conclusions:

• Diagnosis data relying solely on coded entries in Australian EMRs significantly underreports prevalence compared to including validated free-text data.
• This underreporting impacts population health, healthcare planning, resource allocation, and patient care.
• Utilizing phenotypes from validated free-text entries can enhance diagnostic accuracy and reporting reliability.