Heart failure in the information Age data-driven strategies for personalised medicine in primary care

This thesis addresses improving heart failure (HF) diagnosis in primary care, focusing on the role of general practitioners (GPs) in the Netherlands. HF often goes undetected due to its non-specific symptoms, leading to delayed intervention and poor outcomes. The study employs a two-pronged approach—descriptive analysis and predictive modeling—using electronic health records (EHRs) from GPs in the Amsterdam area.
The descriptive analysis uncovered inaccuracies in HF diagnoses recorded in EHRs, with both over- and underregistration of cases. These gaps in medical coding can complicate reliable data interpretation in retrospective studies if they are not addressed. Using consultation notes in conjunction with the aforementioned codes allowed us to compensate for these shortcomings. Further analyses revealed demographic differences in HF incidence, suggesting that risk prediction models should be tailored to specific regions and strata.
In the predictive stage, two new time-to-event models were developed using readily available EHR data from high-risk individuals. One model relied on diagnostic codes, while the other used unstructured data from consultation notes. The text-based model performed better in early HF detection, demonstrating the value of unstructured data for improving predictions. Additionally, the study evaluated the use of natriuretic peptide biomarkers for triaging HF patients, showing that individualized thresholds could reduce unnecessary referrals.
Overall, this thesis demonstrates the potential of using information contained in EHRs for personalized HF diagnosis in primary care. Our findings provide a basis for developing decision-making tools and strategies to assist GPs in identifying HF.