Diagnostic performances of common nucleic acid tests for SARS-CoV-2 in hospitals and clinics: a systematic review and meta-analysis

BACKGROUND: An optimised standard experimental setup across different hospitals is urgently needed to ensure consistency in nucleic acid test results for SARS-CoV-2 detection. A standard comparison across different nucleic acid tests and their optimal experimental setups is not present. We assessed the performance of three common nucleic acid tests, namely digital PCR (dPCR), quantitative PCR (qPCR), and loop-mediated isothermal amplification (LAMP), to detect SARS-CoV-2 in clinical settings. METHODS: In this systematic review and meta-analysis we compared sensitivity and specificity of qPCR, dPCR, and LAMP and their performances when different experimental setups (namely specimen type used, use of RNA extraction, primer–probe sets, and RNA extraction methods) are applied. We searched PubMed, BioRxiv, MedRxiv, SciFinder, and ScienceDirect for studies and preprints published between Feb 29 and Dec 15, 2020. Included dPCR, qPCR, and LAMP studies using any type of human specimens should report the number of true-positive, true-negative, false-positive, and false-negative cases with Emergency Use Authorization (EUA)-approved PCR assays as the comparator. Studies with a sample size of less than ten, descriptive studies, case studies, reviews, and duplicated studies were excluded. Pooled sensitivity and specificity were computed from the true and false positive and negative cases using Reitsma's bivariate random-effects and bivariate latent class models. Test performance reported in area under the curve (AUC) of the three nucleic acid tests was further compared by pooling studies with similar experimental setups (eg, tests that used RNA extracted pharyngeal swabs but with either the open reading frame 1ab or the N primer). Heterogeneity was assessed and reported in I(2) and τ(2). FINDINGS: Our search identified 1277 studies of which we included 66 studies (11 dPCR, 32 qPCR, and 23 LAMP) with 15 017 clinical samples in total in our systematic review and 52 studies in our meta-analysis. dPCR had the highest pooled diagnostic sensitivity (94·1%, 95% CI 88·9–96·6, by Reitsma's model and 95·8%, 54·9–100·0, by latent class model), followed by qPCR (92·7%, 88·3–95·6, and 93·4%, 60·9–99·9) and LAMP (83·3%, 76·9–88·2, and 86·2%, 20·7–99·9), using EUA-approved PCR kits as the reference standard. LAMP was the most specific with a pooled estimate of 96·3% (93·8–97·8) by Reitsma's model and 94·3% (49·1–100·0) by latent class model, followed by qPCR (92·9%, 87·2–96·2, and 93·1%, 47·1–100·0) and dPCR (78·5%, 57·4–90·8, and 73·8%, 0·9–100·0). The overall heterogeneity was I(2) 0·5% (τ(2) 2·79) for dPCR studies, 0% (4·60) for qPCR studies, and 0% (3·96) for LAMP studies. AUCs of the three nucleic acid tests were the highest and differed the least between tests (ie, AUC>0·98 for all tests) when performed with RNA extracted pharyngeal swabs using SARS-CoV-2 open reading frame 1ab primer. INTERPRETATION: All three nucleic acid tests consistently perform better with pharyngeal swabs using SARS-CoV-2 open reading frame 1ab primer with RNA extraction. dPCR was shown to be the most sensitive, followed by qPCR and LAMP. However, their accuracy does not differ significantly. Instead, accuracy depends on specific experimental conditions, implying that more efforts should be directed to optimising the experimental setups for the nucleic acid tests. Hence, our results could be a reference for optimising and establishing a standard nucleic acid test protocol that is applicable in laboratories worldwide. FUNDING: University Grants Committee and The Chinese University of Hong Kong.