Published in The British Medical Journal - 11th November 2020

The Delphi panel was conducted in April and May 2020. Round 1 had a sample of 69 clinicians; three more joined for rounds 2–4. Of these, 62, 53, 56 and 51 completed the surveys, respectively—response rates of 90%, 74%, 78% and 71%. Progression towards quantitative consensus across the four rounds is shown in table 1.

In summary, at least 80% agreement on relevance, wording and value sets for severity was eventually achieved for eight items: pulse, temperature, symptoms of fever (for use particularly if the patient does not have a reliable thermometer), respiratory rate, oxygen saturation level, tiredness (which, if severe, may indicate hypoxia and hence potentially substitute for an oximeter reading), muscle aches and known risk factors (comorbidities). Additional items with moderate agreement included demographic risk factors (77% agreement), oxygen saturation level after a 40-step exercise test (75%), trajectory of breathlessness (73%) and duration of temperature (63%).

Our qualitative data set from the Delphi panel included over 200 pages of comments. Key themes are summarised and illustrated in online supplemental file 2 on These, along with rapid review, allowed us to characterise the clinical features of the deteriorating COVID-19 patient in primary care (box 1).

Box 1

The clinical course of the deteriorating COVID-19 patient in primary care

Synthesised from our qualitative data, supplemented from published sources

COVID-19 may present in primary care as a viral upper respiratory tract infection (eg, sore throat), lower respiratory tract infection (eg, cough, fever and mild dyspnoea), influenza-like illness (with fever, chills, headache and myalgia) or gastrointestinal illness (with abdominal pains, nausea and diarrhoea).2 6 Most patients have a relatively mild, self-limiting illness, but an unknown proportion (perhaps 10%) deteriorate, usually in week 2. Certain symptoms common in week 1, such as cough, mild fatigue and anosmia, do not appear to have prognostic significance.5 44

It is important to date-stamp the onset of first symptoms.6 Severe dyspnoea, especially at rest, may indicate progression of lung involvement. The trajectory of dyspnoea is important, as acute respiratory distress syndrome occasionally follows quickly from the onset of breathlessness.45 Formal scores for assessing dyspnoea severity appear to have a significant false negative rate and should not be used.20 A careful history, noting what the patient is able to do and what they cannot do today that they could do yesterday, is likely to be more important.20 A patient’s or carer’s concern about the severity of breathlessness may be significant and should not be dismissed as ‘anxiety’.

Pulse oximeter readings are extremely useful in assessing unwell patients with COVID-19, so long as the device is reliable (smartphone apps are inherently inaccurate and should not be used)22 and the patient or a relative is capable and confident to use it. The finger must be warm. While a low oximeter reading is concerning, a normal one should not necessarily reassure, as young fit patients in particular can compensate well in the early stages of deterioration.

So-called silent hypoxia, defined as the development of respiratory failure without the subjective perception of dyspnoea, is a recently described feature of severe COVID-19 and appears to have a poor prognosis.44 46–49 Anecdotal accounts suggest that in some patients, silent hypoxia may manifest as profound tiredness, but we could not find published research on this association. New confusion (especially in older patients with comorbidity) was considered by clinicians in our sample to be a poor prognostic sign in COVID-19, but at the time of writing, evidence for this is limited.50

COVID-19 lung damage tends to be manifest as a perfusion defect (ie, difficulty transferring oxygen across the alveolar membrane) rather than a ventilatory defect (difficulty getting air to the alveoli, as in asthma).45 This may explain why COVID-19 can behave similarly to pneumocystis pneumonia in producing a fall in pulse oximetry reading on exertion (or in the minutes following exertion).21 51 Because of this, patients with suspected COVID-19 should not be subject to exercise testing unless there is a clinician present if their resting pulse oximetry reading is abnormal (below 96%).

An unwell patient may or may not have COVID-19. An overall assessment is needed using questions relating to (eg) hydration status, dizziness, falls, central chest pain, fall in blood pressure (if the patient has equipment at home), change in mental status (including lethargy, new confusion, difficulty in rousing), central cyanosis (eg, blue lips) and severe reduction in urine output. For this reason, a standard ‘red flag’ checklist should be quickly reviewed in all unwell patients.

There are some well-established risk factors for developing COVID-19 and worse outcome (eg, age, non-white ethnicity, high body mass index and comorbidities including cardiovascular disease, hypertension and active cancer).52–54 The extent to which these risk factors should be applied to ‘load the score’ of a patient who appears to have a mild form of the disease is not yet known, especially since shielded patients are the ones for whom a hospital or clinic visit carries most risk.

In addition to prompting new rapid reviews, the Delphi qualitative data shaped the development of the items in several ways. Particular forms of words (eg, to question patients about the severity and rate of deterioration of breathlessness) enabled us to refine our items and value sets. Participants alerted us to existing guidance and protocols used either nationally or locally (there was strong consensus that any new instrument should complement rather than replace these). Comments about missing or untrustworthy data when a patient was being assessed at home via telephone or video link prompted us to develop default value sets or alternative questions to compensate for such deficiencies. The free-text comments included rich data, based on real clinical experiences, from which we were able to construct the vignettes used in discussions. Numerous comments on the practicalities of applying the potential instrument prompted us to set up focus groups to explore these operational challenges further.

The five fictional vignettes and the results of the simulated scoring exercise on these are reproduced in online supplemental file 3 on In summary, while there was considerable variation in the number of points given (eg, in whether clinicians judged an ill-defined set of symptoms as ‘moderately’ or ‘severely’ tired), in all five vignettes, the simulated RECAP-V0 score as calculated appeared to prompt an appropriate and cautious response. For example, in a case of an elderly South Asian patient who spoke no English and with no equipment at home for the family to take measurements, all but one participant were prompted by the RECAP-V0 simulated result to assess the patient in a face-to-face encounter. In a case of an African Caribbean man with profound tiredness and rapidly worsening breathlessness in the second week of his illness, all participants were prompted to arrange urgent transfer to hospital. Free-text comments in several vignettes indicated that some respondents’ level of clinical concern had not been especially high and that they were surprised that the RECAP-V0 simulated score was so high.