Diagnostic accuracy of clinical signs and symptoms of COVID-19: A systematic review and meta-analysis to investigate the different estimates in a different stage of the pandemic outbreak

Abstract

Background: The coronavirus (COVID-19) pandemic caused enormous adverse socioeconomic impacts worldwide. Evidence suggests that the diagnostic accuracy of clinical features of COVID-19 may vary among different populations.

Methods: We conducted a systematic review and meta-analysis of studies from PubMed, Embase, Cochrane Library, Google Scholar, and the WHO Global Health Library for studies evaluating the accuracy of clinical features to predict and prognosticate COVID-19. We used the National Institutes of Health Quality Assessment Tool to evaluate the risk of bias, and the random-effects approach to obtain pooled prevalence, sensitivity, specificity, and likelihood ratios.

Results: Among the 189 included studies (53 659 patients), fever, cough, diarrhoea, dyspnoea, and fatigue were the most reported predictors. In the later stage of the pandemic, the sensitivity in predicting COVID-19 of fever and cough decreased, while the sensitivity of other symptoms, including sputum production, sore throat, myalgia, fatigue, dyspnoea, headache, and diarrhoea, increased. A combination of fever, cough, fatigue, hypertension, and diabetes mellitus increases the odds of having a COVID-19 diagnosis in patients with a positive test (positive likelihood ratio (PLR) = 3.06)) and decreases the odds in those with a negative test (negative likelihood ratio (NLR) = 0.59)). A combination of fever, cough, sputum production, myalgia, fatigue, and dyspnea had a PLR = 10.44 and an NLR = 0.16 in predicting severe COVID-19. Further updating the umbrella review (1092 studies, including 3 342 969 patients) revealed the different prevalence of symptoms in different stages of the pandemic.

Conclusions: Understanding the possible different distributions of predictors is essential for screening for potential COVID-19 infection and severe outcomes. Understanding that the prevalence of symptoms may change with time is important to developing a prediction model.

Figure 1

Flowchart of study identification, screening, inclusion, and exclusion in the systematic review, and further updating umbrella review. According to The PRISMA 2020 statement, the definitions of report, record, and study were as following: Report – a document supplying information about a particular study. Record – the title or abstract of a report indexed in a database or website. Study – an investigation, such as a clinical trial, that includes a defined group of participants, interventions, and outcomes.

Figure 2

Panel A. Sensitivity of fever, cough, fatigue and dyspnoea in COVID-19 decreases with the stage of the outbreak. Panel B. Sensitivity of fever, and cough mildly increase with age in diagnosis. Other figures are presented in Figure S3 in the Online Supplementary Document.

Figure 3

Forest plot for the proportion of symptoms and outcomes in patients with COVID-19 among the overall and different age groups. The definitions of the threshold for fever vary in different studies included. Fever of different thresholds (37.3°C, 38.0°C, 39.0°C) are displayed here.

Figure 4

Panel A. Weighted mean differences of white blood cell counts between severe and non-severe patients with COVID-19. Panel B. Weighted mean differences of lymphocyte count between severe and non-severe patients with COVID-19.

Figure 5

Forest plots. Panel A. Performance of predictors in detecting COVID-19 infection. Panel B. Performance of predictors in prognosticating severe COVID-19 infection. DOR – diagnostic odds ratio, PLR – positive likelihood ratio, NLR – negative likelihood ratio, AUROC – area under the receiver operating characteristic. Forest plots.