RECAP (remote COVID-19 Assessment in Primary Care): A risk score to predict hospital admission.

Talk Code: 
Brendan Delaney
Brendan Delaney, Trish Greenhalgh, Simon de Lusignan, Erik Mayer, Francesca Fiorentino, Ana Luisa Neves, Ana Espinosa-Gonzales, Denys Prociuk, Ella Mi, Emma Mi.
Author institutions: 
Imperial College London Institute of Global Health Innovation, University of Oxford Nuffield Dept of Primary Care


During the COVID-19 pandemic approximately 10% of patients will become sufficiently ill to require hospital admission. Early in the first wave it became clear that existing ‘early warning scores’ such as NEWS2 were unsuitable, being neither sensitive nor specific. Oxygen saturation alone is no more predictive than NEWS2, and that other clinical features need to be taken into account. QCOVID gives a prediction of patient risk based on existing conditions, age, BMI ethnicity etc, but does not contain any data about presenting symptoms and signs. We aimed to develop and validate a model for risk of hospital admission, based on acute features, to guide triage and management of patients with acute COVID-19 in the community.


A Delphi panel of 50 clinicians was used to suggest a minimum set of clinical data items, including potential severity levels as outcome sets for stem questions. Templates were prepared for use in SystmOne and EMIS using appropriate and available SNOMED clinical terms. Practices recruited to the study were required to complete the template on contact with a patient aged 18 and over with a clinical diagnosis of COVID-19. Verbal consent supported by a web-based information sheet was recorded by using the SNOMED term for ‘consent for research study obtained' and the study Central Portfolio Management System (CPMS) number. We used two existing networks (North West London Whole Systems Integrated Care and The Royal College of General Practitioner’s Research and Surveillance Centre at Oxford University) where data from records with the consent code and CPMS number were extracted from EHR systems and linked with relevant outcome data, hospital admissions (the primary outcome), ICU admission and death (the secondary outcomes), all within 28 days. A sample size of 1317 was required to derive a model of 24 predictors for an admission rate of 10%. An additional 1400 subjects were required to validate the model with a desired performance of 85% specificity and a precision of 0.05. Allowing for 6 loss to follow up the sample size was 2880. The study was supported by NIHR CRN as an Urgent Public Health Study for the pandemic.


Data collection commenced in October 2020. The study closed to recruitment on Feb 18th 2021, in NW London 1976 subjects were recruited by 87 practices and in the RSC, 1180 subjects from 61 practices, (total 3,156). Data are currently being extracted and linked for analysis. A model will be developed and validated against the remaining 1440 subjects. Suitable cut-points on risk will be defined to enable patients to be divided into green, amber, red groups for monitoring or admission.


The next step is to deploy the model algorithm as an automatic calculation when the RECAP template is completed. As COVID-19 is an ongoing health problem the RECAP score should prove invaluable in supporting safe clinical management of COVID-19 in the community.

Submitted by: 
Brendan C. Delaney
Funding acknowledgement: 
The study was supported by funding independently obtained by the two CI’s Greenhalgh and Delaney but conducted jointly by Oxford and Imperial with Imperial as Sponsor. Funders: Community Jameel Imperial College COVID-19 Excellence Fund (Delaney) Economic and Social Research Council (Greenhalgh) NIHR Oxford Biomedical Research Centre NIHR Imperial Biomedical Research Centre NIHR Imperial Patient Safety Translational Research Centre