When COVID-19 struck in early 2020, health systems faced the dual challenge of rapidly identifying high-risk patients and optimizing limited vaccine supplies.
Research challenge
Early COVID-19 prediction tools often relied on hospitalized patients, missing the vast majority of infections managed in primary care. Moreover, policy shifts, testing availability, and vaccination rollouts introduced time-dependent biases that could quickly erode model accuracy. The pressing need was for a living, general population-based risk prediction algorithm—one that could be updated daily, applied seamlessly in GP practices, and guide vaccination or shielding strategies to prevent severe complications such as hospitalization, institutionalization, or death. The challenge was compounded by the complex interplay of socio-demographic, clinical, and temporal factors in COVID-19 outcomes, and the scarcity of datasets linking these dimensions in a privacy-compliant, population-representative way.
Our approach
The NL-COVID database—developed through an unprecedented collaboration between general practitioners (GPs), public health specialists, epidemiologists, data scientists, and ICT providers — became a cornerstone for actionable insights. Covering 95% of Dutch GP practices’ electronic health records (EHRs) and partially enriched with daily questionnaire inputs from frontline clinicians, it offered a uniquely complete and clinically detailed view of COVID-19 in the community.
Read the full case study
"*" indicates required fields
