During 2020, it became clear that assessment of the severity of COVID-19 infection required clinical tools specific to the condition and that repurposing tools such as the National Early Warning Score (NEWS2), designed for the early diagnosis of sepsis, would not be safe clinical practice [1]. The management of COVID-19 by clinicians is challenged by uncertainty about the disease progression [2]. There is evidence that a small percentage of patients present a dramatic deterioration of clinical status around the 8th to 10th day of disease, often associated with unperceived low oxygen saturations (known as “silent hypoxia”) that may require hospital and intensive care unit (ICU) admissions [3-5]. The inability to predict which patients will experience clinical deterioration adds an additional level of complexity to the clinical challenge and diagnostic uncertainty that general practitioners (GPs) have faced during the pandemic, particularly as most of the consultations are carried out remotely (usually by telephone and occasionally by video) to minimize the risk of transmission [6].

It was initially suggested that NEWS2 could be used to assess severity of patients with COVID-19 [7]. NEWS2 is calculated from patient’s temperature, pulse rate, respiratory rate, systolic blood pressure, pulse oximetry reading, and presence of new onset of acute confusion [8]. It is commonly used in hospital settings and ambulance service prior to transfer to hospital to assess the risk of deterioration of a patient [9]. However, NEWS2 seems to be a late indicator of decompensation, typically triggering within the last 12 hours before a transfer to ICU is considered necessary and, therefore, this limits its application and validity in a primary care or community care setting where an earlier warning would be preferred [9,10].

The Roth score (originally developed as a measure of breathlessness in cardiopulmonary disease [11]) was briefly considered by the Royal College of General Practitioners (RCGP) as possibly useful in the assessment of breathlessness when assessing patients with signs and symptoms of COVID-19 [12]. However, a rapid literature review concluded that the Roth score might have a low sensitivity (ie, a normal score in patients with “silent hypoxia”), and therefore should not be used by GPs when assessing patients over the phone or in video consultations [13].

This new condition and the forced shift toward remote consultations during the pandemic have increased the challenges and uncertainty commonly faced in general practice [6]. Primary care clinicians need a tool to guide the management of patients with suspected COVID-19 to be able to identify those whom they can reassure, those that need monitoring, and those that require urgent further assessment or referral to hospital. Even though the validity of NEWS2 for this purpose was a subject of intense debate during the height of the first COVID-19 wave, the score is still being used by primary care clinicians to assess patients prior to transfer to hospital [9]. The use of NEWS2 outside the hospital setting has not been validated, and it was not designed to capture the idiosyncrasies of COVID-19 infection. Therefore, there is need to develop an early warning score that incorporates key features of acute COVID-19 and that can be safely used by GPs when assessing patients remotely [14].

We reviewed the literature on COVID-19 early warning scores, then conducted a series of focus groups with 72 primary care clinicians (mostly GPs and including advanced nurse practitioners and paramedics) to derive elements that might form part of a suitable score, value sets, and appropriate SNOMED terms [15]. This paper describes the process of quantitative development and validation of the Remote COVID-19 Assessment in Primary Care (RECAP) score. The objective was to produce a multivariate risk prediction tool to facilitate the early identification, by primary care physicians and other clinicians working in the community, of those patients with COVID-19 that are at higher risk of becoming severely ill and inform the early escalation of their treatment, while also reducing unnecessary referrals in low-risk patients, with the aim of improving patient outcomes.

This primary care data linkage study follows a prospective cohort observational design, whereby patients presenting in primary or community care with signs and symptoms suggestive of COVID-19 will be followed and their data linked with hospital outcomes, particularly focusing on hospital admission, ICU admission, and death. For data collection purposes, the initial set of items identified in earlier qualitative work [15], known as RECAP-V0, will be integrated into an electronic template to be used by primary care physicians (see Figure 1 for a summary of items included in RECAP-V0). This will enable the standardized recording of patients’ signs and symptoms and subsequent linkage with hospital and mortality data. Data collected will be used to develop and validate a multivariate regression model to predict hospital admission, ICU admission, and death.

A total of 2880 participants will be necessary to develop a model with a minimum 85% specificity, assuming 10% prevalence of hospital admission and 6% missing data. We will split the sample into two consecutive groups, taking the first 50% of participants’ data for model development and the last 50% for model validation. CCAS will also collect 2880 participants as we wish to explore the hypothesis that, on account of case mix and spectrum bias, patients already triaged to the national service may require a separate model. We will then separately develop and validate a model for CCAS. Doctaly will provide an additional validation data set for RECAP-V1 score.

Recruitment started in October 2020. Initially, only the NWL CCGs and RCGP RSC arms were actively recruiting. In order to engage clinicians with the study and facilitate participation, we have run three webinars or training workshops where the study team described the study objectives and deadlines, and provided a detailed description of the RECAP template and how it would be used at the clinical front line. These webinars took place in October 2020 and January 2021. The study team has been in close contact with participating practices through the Imperial College London arm office, which has overall responsibility for the project and is directly in charge of data collection in North West London, and the University of Oxford arm office, which has direct responsibility for data collection from RCGP RSC practices. The CCAS and Doctaly platform arms joined the study in February 2021.

The initial data set to be used for model development will consist of RCGP RSC and NWL arms data and will be complete by the end of March 2021; this includes the primary care data on recruited patients’ signs and symptoms collected by these two arms linked to outcomes 28 days later. Two stages of data extraction and analysis of this integrated data set have been identified: first, RECAP-V1 development using the first half of the data set will start in April 2021, and second, model validation using the second half of the data set will follow. As of March 24, 2021, we have recruited a combined sample of 3827 participants (173 active primary care practices enrolled) in these two arms. The CCAS arm started the recruitment process on March 15, 2021, and we expect to reach the desired sample size in this arm (2880 participants) by the end of May 2021. Data sharing agreements are being developed to access data that have already been collected from around 1400 participants by clinicians using the Doctaly platform. The CCAS and Doctaly platform data sets will be used to validate the RECAP-V1 model and will be analyzed independently. Once we have produced the model, and subject to findings, the RECAP-V1 score will be ready to be deployed and used by clinicians to guide the management of patients with suspected COVID-19 according to their predicted risk.

The study is sponsored by Imperial College London and ethical approval was granted by the North West-Greater Manchester East Research Ethics Committee and Health Research Authority on May 27, 2020 (Integrated Research Application System number: 283024, Research Ethics Committee reference number: 20/NW/0266). An amendment to include the CCAS and Doctaly arms was approved on February 1, 2021. Due to the low risk associated with participation in this study and the remote nature (telephone/video consultation) of most patient encounters, the review committee agreed that obtaining verbal consent for data linkage was acceptable.

To access and link retrospective data collected by the NHS 111 CCAS and Doctaly platforms in South East London (ie, data that have already been collected by the services prior to study participation) we requested the last ethics amendment submitted to be assessed under the Control of Patient Information (COPI) notice, data sharing provisions that allow public authorities and research bodies the use of COVID-19–relevant patient-level data without the need for patients’ explicit consent [25]. For NHS 111 CCAS prospective data—that is, data from patients seeking medical care after the RECAP template has been installed in Adastra—we will apply the same mechanism to seek consent that has been followed in the NWL and RCGP RSC arms, and patients in the clinical queue will receive an SMS text message with information on the study and how to participate.

Data and all appropriate documentation will be stored in accordance with General Data Protection Regulation (Data Protection Act 2018) for a minimum of 10 years after the completion of the study, including the follow-up period. Participants can withdraw from the study at any point by informing their GP or a member of the study team. They will be asked whether the data obtained before withdrawal can be retained for analysis or they would like their data to be destroyed instead.

The study was included in the National Institute of Health Research (NIHR) Clinical Research Network Portfolio (CPMS number: 45890) on September 25, 2020, and badged as NIHR Urgent Public Health Study on October 14, 2020. These measures facilitate the rapid mobilization of resources from NIHR and clinical research networks toward study dissemination and participant recruitment and help ensure that high-quality data can be collected on a timely basis (trial registration number: ISRCTN13953727).

The RECAP-V0 template has already been disseminated nationally through CRNs facilitating the standardization of clinical records of patients with COVID-19. Its use has been encouraged through webinars and invited talks arranged by CRNs. Once the risk prediction tool has been developed and validated, we will seek endorsement for it to be incorporated into the electronic template to support clinical decision making when assessing patients with COVID-19. We expect to reach wide national and international dissemination of the RECAP-V1 risk prediction tool through submission to academic journals and international conferences.

Patient and public participation has been incorporated at different stages of the project. Patients were involved, along with primary care clinicians, in the qualitative study carried out to identify the set of elements to be included in the RECAP-V0 [15]. Once the RECAP-V1 tool has been developed and validated, public participation will be sought to coproduce project lay summaries, which will be valuable to disseminate the study findings to a wider audience.