Developing a digital phenotype to subdivide adult immunosuppressed COVID-19 outcomes within the English Primary Care Sentinel Network.

Background: Adults classified as immunosuppressed have been disproportionately affected by the COVID-19 pandemic. Compared to the immunocompetent, certain patients are at increased risk of suboptimal vaccine response and adverse health outcomes if infected. However, there has been insufficient work to pinpoint where these risks concentrate within the immunosuppressed spectrum; surveillance efforts typically treat the immunosuppressed as a single entity, leading to wide confidence intervals. A clinically meaningful and computerised medical record (CMR) compatible method to subdivide immunosuppressed COVID-19 data is urgently needed. Methods: We conducted a rapid scoping review into COVID-19 mortality across UK immunosuppressed categories to assess if differential mortality risk was a viable means of subdivision. We converted the risk hierarchy that surfaced into a pilot digital phenotype-a valueset and series of ontological rules ready to extract immunosuppressed patients from CMR data and stratify outcomes of interest in COVID-19 surveillance dataflows. Results: The rapid scoping review returned COVID-19 mortality data for all immunosuppressed subgroups assessed and revealed significant heterogeneity across the spectrum. There was a clear distinction between heightened COVID-19 mortality in haematological malignancy and transplant patients and mortality that approached the immunocompetent baseline amongst cancer therapy recipients, autoimmune patients, and those with HIV. This process, complemented by expert clinical input, informed the curation of the five-part digital phenotype now ready for testing in real-world data; its ontological rules will enable mutually exclusive, hierarchical extraction with nuanced time and treatment conditions. Unique categorisations have been introduced, including 'Bone Marrow Compromised' and those dedicated to differentiating prescriptions related and unrelated to cancer. Codification was supported by existing reference sets of medical codes; absent or redundant codes had to be resolved manually. Discussion: Although this work is in its earliest phases, the development process we report has been highly informative. Systematic review, clinical consensus building, and implementation studies will test the validity of our results and address criticisms of the rapid scoping exercise they are predicated on. Conclusion: Comprehensive testing for COVID-19 has differentiated mortality risks across the immunosuppressed spectrum. This risk hierarchy has been codified into a digital phenotype for differentiated COVID-19 surveillance; this marks a step towards the needs-based management of these patients that is urgently required.