Methodological quality of included studies
The results of the quality assessment are summarised in Figure 2 and Figure 3. We rated participant selection as introducing high risk of bias in seven studies. In five studies this was because a CT scan or other imaging was used to diagnose patients with pneumonia prior to inclusion in the study, leading to a highly selected patient population (Ai 2020a; Chen X 2020; Cheng 2020a; Liang 2020; Yang 2020d); RT‐PCR results were subsequently used to distinguish between COVID‐19 pneumonia and pneumonia from other causes. For all studies, testing was highly dependent on the local case definition and testing criteria that were in effect at the time of the study, meaning all patients that were included in studies had already gone through a referral/selection filter, which was not always described. The most extreme example of this is the study by Liang 2020, in which patients with radiological evidence of pneumonia and a clinical presentation compatible with COVID‐19 were only tested for SARS‐CoV‐2 after a panel discussion.
2  Risk of bias and applicability concerns graph: review authors' judgements about each domain presented as percentages across included studies
3  Risk of bias and applicability concerns summary: review authors' judgements about each domain for each included study Of the 16 studies included in this first version of the review, five studies did not use a cross‐sectional design. Three studies were diagnostic case‐control studies (Nobel 2020; Yang 2020d; Zhao 2020a), one study selected cases cross‐sectionally in five hospitals but only selected cases in one hospital (Chen X 2020), and one study emailed patients who had undergone testing for SARS‐CoV‐2 about olfactory symptoms prior to the SARS‐CoV‐2 test, with a response rate of 58% in SARS‐CoV‐2 positive cases and 15% in negative cases (Yan 2020a).
We rated all studies except two as carrying a high risk of bias for the index tests because there was little to no detail on how, by whom, and when the signs and symptoms were measured. In addition, there is considerable uncertainty around the reference standard, with some studies providing little detail on the RT‐PCR tests that they used or lack of clarity on blinding.
Participant flow was unclear in four studies (Yan 2020a; Yang 2020d; Zhao 2020a; Zhu 2020b), either because the timing of recording signs and symptoms and conduct of the reference standard was unclear, or because some tests received a second or third reference standard at unclear time points during hospital admission.
We rated applicability for participant selection as high risk when there was a risk of selection bias or studies did not describe selection. As for the applicability of the index tests and reference standard, we always scored this as low risk except for Chen X 2020, because blinding of the index tests was unclear, and Yang 2020d, because blinding and sample of the reference standard were unclear.