A nationwide deep learning pipeline to predict stroke and COVID-19 death in atrial fibrillation (preprint)

Deep learning (DL) and machine learning (ML) models trained on long-term patient trajectories held as medical codes in electronic health records (EHR) have the potential to improve disease prediction. Anticoagulant prescribing decisions in atrial fibrillation (AF) offer a use case where the benchmark stroke risk prediction tool (CHA2DS2-VASc) could be meaningfully improved by including more information from a patient's medical history. In this study, we design and build the first DL and ML pipeline that uses the routinely updated, linked EHR data for 56 million people in England accessed via NHS Digital to predict first ischaemic stroke in people with AF, and as a secondary outcome, COVID-19 death. Our pipeline improves first stroke prediction in AF by 17% compared to CHA2DS2-VASc (0.61 (0.57-0.65) vs 0.52 (0.52-0.52) area under the receiver operating characteristics curves, 95% confidence interval) and provides a generalisable, opensource framework that other researchers and developers can build on.

Understanding COVID-19 trajectories from a nationwide linked electronic health record cohort of 56 million people: phenotypes, severity, waves & vaccination (preprint)

Background: Updatable understanding of the onset and progression of individuals COVID-19 trajectories underpins pandemic mitigation efforts. In order to identify and characterize individual trajectories, we defined and validated ten COVID-19 phenotypes from linked electronic health records (EHR) on a nationwide scale using an extensible framework. Methods: Cohort study of 56.6 million people in England alive on 23/01/2020, followed until 31/05/2021, using eight linked national datasets spanning COVID-19 testing, vaccination, primary & secondary care and death registrations data. We defined ten COVID-19 phenotypes reflecting clinically relevant stages of disease severity using a combination of international clinical terminologies (e.g. SNOMED-CT, ICD-10) and bespoke data fields; positive test, primary care diagnosis, hospitalisation, critical care (four phenotypes), and death (three phenotypes). Using these phenotypes, we constructed patient trajectories illustrating the transition frequency and duration between phenotypes. Analyses were stratified by pandemic waves and vaccination status. Findings: We identified 3,469,528 infected individuals (6.1%) with 8,825,738 recorded COVID-19 phenotypes. Of these, 364,260 (11%) were hospitalised and 140,908 (4%) died. Of those hospitalised, 38,072 (10%) were admitted to intensive care (ICU), 54,026 (15%) received non-invasive ventilation and 21,404 (6%) invasive ventilation. Amongst hospitalised patients, first wave mortality (30%) was higher than the second (23%) in non-ICU settings, but remained unchanged for ICU patients. The highest mortality was for patients receiving critical care outside of ICU in wave 1 (51%). 13,083 (9%) COVID-19 related deaths occurred without diagnoses on the death certificate, but within 30 days of a positive test while 10,403 (7%) of cases were identified from mortality data alone with no prior phenotypes recorded. We observed longer patient trajectories in the second pandemic wave compared to the first. Interpretation: Our analyses illustrate the wide spectrum of severity that COVID-19 displays and significant differences in incidence, survival and pathways across pandemic waves. We provide an adaptable framework to answer questions of clinical and policy relevance; new variant impact, booster dose efficacy and a way of maximising existing data to understand individuals progression through disease states.

Association of COVID-19 vaccines ChAdOx1 and BNT162b2 with major venous, arterial, and thrombocytopenic events: whole population cohort study in 46 million adults in England (preprint)

BackgroundThromboses in unusual locations after the COVID-19 vaccine ChAdOx1-S have been reported. Better understanding of population-level thrombotic risks after COVID-19 vaccination is needed. MethodsWe analysed linked electronic health records from adults living in England, from 8th December 2020 to 18th March 2021. We estimated incidence rates and hazard ratios (HRs) for major arterial, venous and thrombocytopenic outcomes 1-28 and >28 days after first vaccination dose for ChAdOx1-S and BNT162b2 vaccines. Analyses were performed separately for ages <70 and [≥]70 years, and adjusted for age, sex, comorbidities, and social and demographic factors. ResultsOf 46,162,942 adults, 21,193,814 (46%) had their first vaccination during follow-up. Adjusted HRs 1-28 days after ChAdOx1-S, compared with unvaccinated rates, at ages <70 and [≥]70 respectively, were 0.97 (95% CI: 0.90-1.05) and 0.58 (0.53-0.63) for venous thromboses, and 0.90 (0.86-0.95) and 0.76 (0.73-0.79) for arterial thromboses. Corresponding HRs for BNT162b2 were 0.81 (0.74-0.88) and 0.57 (0.53-0.62) for venous thromboses, and 0.94 (0.90-0.99) and 0.72 (0.70-0.75) for arterial thromboses. HRs for thrombotic events were higher at younger ages for venous thromboses after ChAdOx1-S, and for arterial thromboses after both vaccines. Rates of intracranial venous thrombosis (ICVT) and thrombocytopenia in adults aged <70 years were higher 1-28 days after ChAdOx1-S (adjusted HRs 2.27, 95% CI:1.33- 3.88 and 1.71, 1.35-2.16 respectively), but not after BNT162b2 (0.59, 0.24-1.45 and 1.00, 0.75-1.34) compared with unvaccinated. The corresponding absolute excess risks of ICVT 1-28 days after ChAdOx1-S were 0.9-3 per million, varying by age and sex. ConclusionsIncreases in ICVT and thrombocytopenia after ChAdOx1-S vaccination in adults aged <70 years were small compared with its effect in reducing COVID-19 morbidity and mortality, although more precise estimates for adults <40 years are needed. For people aged [≥]70 years, rates of arterial or venous thrombotic, events were generally lower after either vaccine.