(173) Identifying and profiling patients with heart failure in a population-based cohort through linkage of primary and secondary care data

Background: Coding of diagnoses in primary care may lack granularity, posing challenges to determining the clinical profile of patients.

Objectives: To investigate whether hospital data may improve the profiling of patients with heart failure (HF) diagnosed in primary care.

Methods: The PHARMO Data Network links anonymous, patient-level electronic healthcare data from primary and secondary healthcare settings in the Netherlands. GP data were obtained from 2015 to 2022. Patients with HF were identified by ICPC code K77 (and subcodes) or free text using a text mining algorithm. In the linked hospital data, ICD-10 codes were subsequently used to further differentiate HF subtypes. Patients were followed from the first HF diagnosis in the study period until the end of the study period or death. Differences between HF patients identified in GP and hospital data versus those only known in GP data were tested using chi-squared tests, t-tests, standardised mean differences (SMD), and hazard ratios (HR).

Results: The source population included 4,279,786 GP patients. After applying all inclusion and exclusion criteria, 66,525 patients with a recorded HF diagnosis in GP data were included, of which 48% (N=32,086) also had an HF diagnosis in the linked hospital data (hospitalisation or ambulatory visit). 52% were female, median age at diagnosis was 80 years (IQR: 72-87), and median follow-up time from CED was 2.1 years (IQR: 0.7-4.2). 44% of patients died during the follow-up, with a median time to death of approximately 5 years.
Of patients identified in both GP and hospital data, 80% had an unspecified HF diagnosis in primary care. Of these, 64% also had an unspecified diagnosis in hospital data, whereas 23% were diagnosed with left ventricular failure (LVF), 13% with congestive heart disease (CHD), and 1% had HF with hypertension. In GP data, it was also possible to distinguish between HF patients with preserved (N=361) or mildly reduced and reduced left ventricular ejection fraction (N=600), whereas ICD-10 codes in hospital data only enabled the identification of CHD and LVF.
Compared to patients identified in primary care only, those identified in both GP and hospital data had a higher number of cardiovascular comorbidities (3.5 vs 2.2, SMD=0.811), greater use of HF-related medications (4.4 vs 3.4, SMD=0.694), and greater risk of death (47% vs 40%, HR=1.13, CI: 1.10-1.16).

Conclusions: Linking GP and hospital data can provide more detailed diagnosis information and improve the identification of HF subtypes in real-world data, compared to using GP data alone. Patients identified in both GP and hospital data had a more severe clinical profile compared to those identified in GP data only.