Different data, different outcomes: balancing accuracy, effort, and availability for policy needs.

Objectives
To assess the fitness for purpose of reimbursement-oriented and care-oriented electronic health record (EHR) data, identifying which type or combination is most appropriate for informing health policy decisions.

Methods
This detailed case study used reimbursement-oriented and care-oriented EHR data to select a cohort of patients treated for colon carcinoma at Jeroen Bosch Hospital in the Netherlands in 2019 and assessed the possibility to determine the date of diagnosis of these patients. Patients were selected through an iterative, multidisciplinary process using structured and unstructured EHR data, followed by manual validation. Multiple inclusion-exclusion strategies were tested, with patient counts and precision and recall percentages reported for each.

Results
The combination of structured and unstructured data initially resulted in a patient group of 548 patients. Various combinations of EHR data types and exclusion filters resulted in reduced, but different, numbers of patients. After manual validation, a patient group of 177 patients remained. Assessment of both structured and unstructured data resulted in the determination of the date of diagnosis for all 177 validated patients. Precision and recall were calculated for eight different strategies, resulting in a wide range of precision and recall percentages.

Conclusions
Reimbursement-related EHR data offered restricted accuracy for identifying colon carcinoma patients and the start of their care trajectory, but adding clinical, unstructured EHR data improved results. This, however, increases cost, effort, and privacy risks and requires careful balancing. Policymakers should be aware of these limitations and carefully interpret results when using routine health data for secondary purposes.