OBJECTIVES: The quality of real-world data (RWD) is highly dependent on the nature of routine clinical practice. General practitioners (GPs) in the Netherlands are not obliged to enter a diagnosis in the electronic health records (EHR) by a coded diagnosis, which can consequently result in missed cases when conducting observational studies using RWD. We aimed to quantify the extent to which diagnoses are registered by means of a diagnosis codes or a proxy using three use cases.

METHODS: The PHARMO GP data was used to select all patients with a coded or proxy-based diagnosis of hypercholesterolemia, type 2 diabetes (T2D) and dementia. International Classification of Primary Care (ICPC) coded diagnoses were compared with a proxy based diagnosis, defined as an total cholesterol (TC) measurement ≥6.5 mmol/L for hypercholesterolemia, ≥2 non-insulin drug prescriptions used in diabetes for T2D, and free text notes for dementia, respectively. We calculated the proportion of patients with a coded diagnosis out of those with a coded or proxy based diagnosis.

RESULTS: The study included 750,509 patients with hypercholesterolemia (150,897 with an ICPC coded diagnosis, 368,821 with a proxy based diagnosis and 230,791 with both), 318,491 patients with T2D (76,953 with an ICPC coded diagnosis, 23,261 with a proxy and 218,277 with both) and 69,036 patients with dementia (60,366 with an ICPC coded diagnosis, 6,781 with a proxy and 1,889 with both). A coded diagnosis was observed in 50.9%, 92.7%, and 90.2% of the patients with hypercholesterolemia, T2D and dementia, respectively.

CONCLUSIONS: The results of this study suggest that additional information available in GP data, such as measurements, treatment and free text notes, should be considered to complement coded diagnoses when selecting diseases/symptoms. The added value depends on the diagnosis of interest. In the future, automation of structuring this additional information can improve data preparation and enhance the quality.