This meta-analysis is reported in accordance with the Preferred Reporting Items for Systematic reviews and Meta-Analyses of Diagnostic Test Accuracy Studies statement8. A comprehensive search of three electronic databases (MEDLINE, EMBASE, and the Cochrane Central Register) up to July of 2020 was performed. Search terms for influenza included “Influenza, Human” [MeSh] OR “Influenza A virus” OR “Influenza B virus” OR “influenza” OR “flu”. Search terms for the tests included “rapid test*” OR “rapid diagnos*” OR “point-of-care test*” OR “immunoassay*” OR “immunochromatographic test*” OR “influenza FIA” OR “Quidel Sofia Influenza” OR “Rapid Detection Flu”. As this study was a systematic review of published articles, neither informed consent nor ethics approval was required. A manual search of references listed in relevant review articles was also conducted.

We included studies that met the following inclusion criteria: (1) full-length reports published in peer-reviewed English language journals; (2) studies that evaluated the performance of the Quidel Sofia rapid influenza FIA, compared to a reference standard; (3) studies that included patients with influenza-like illness; and (4) studies that provided sufficient data to calculate absolute numbers of true-positive, false-positive, false-negative, and true-negative results. Review articles, case reports, commentaries, and studies reporting outcomes without raw data or peer review were excluded. Demographics and underlying diseases of participants were not restricted.

Influenza-like illness was defined as fever ≥38°C and any signs/symptoms of respiratory tract infection (e.g., cough, sputum, sore throat, wheezing, etc.). We allowed the followings as specimens: nasopharyngeal aspirates, swabs, or washes; nasal aspirates, swabs, or washes; saliva; and throat swabs. A reference standard was either a commercial reverse transcriptase-polymerase chain reaction (RT-PCR) or a laboratory-developed RT-PCR.

Two authors independently reviewed potentially relevant studies and each study according to predefined eligibility criteria, after which data were extracted. Any disagreements that arose during the process of study selection or data extraction were resolved by discussion. A predefined form was used to extract data from each study. The following data from each study included in the meta-analysis were extracted: author, year of publication, study design, place of study, number of participants, age, proportion of children, sex, study period, type of reference standard, type of specimens, and type of population. As recommended by the Cochrane Collaboration, the Quality Assessment of Diagnostic Accuracy Studies (QUADAS)-2 tool was used to assess the risk of bias in diagnostic test accuracy9. Discrepancies were resolved by consensus between the two authors.

For diagnostic meta-analysis, random effects meta-analyses were performed to generate pooled estimates with 95% confidence intervals (CIs). Numbers of patients were extracted with true-positive, false-positive, false-negative, and true negative test results either directly or indirectly through a recalculation based on reported measures of accuracy in combination with the prevalence and sample size of the included study. Pooled sensitivity and specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), diagnostic odds ratio (DOR), and area under the receiver-operating characteristic curve (AUC) were calculated as pooled estimates with a 95% CI10. Hierarchical summary receiver-operating characteristic curves (HSROCs) were also constructed. The heterogeneity was evaluated using the Higgins I² statistics on a scale of 0-1. An I²>0.5 indicated a substantial level of between-study heterogeneity. To assess effects of potential sources of heterogeneity, subgroup analyses were performed using the following covariates to the model: study design (single vs. multicenter), number of participants (≥250 vs. <250), study period (influenza vs. non-specific season), and study population (children vs. adult). Pooled sensitivity and specificity estimates were calculated for each covariate. To investigate the effect of study quality, sensitivity analyses were performed. A p-value <0.05 was considered statistically significant. All statistical analyses were performed with Stata statistical software version 14.2 (Stata Corp LP, College Station, TX, USA) and Review Manager version 5.3 (Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, Denmark).

The literature search process is shown in Figure 1. Initially, 432 articles from PubMed, 1,310 articles from EMBASE, 356 articles from the Cochrane library, and an additional article from hand-searching were identified. After removing duplicate articles, 1,664 potentially eligible articles were screened. After reviewing titles and abstracts, 1,629 search records were removed and the remaining 35 articles were eligible for full text reading. Fifteen articles were excluded for reasons shown in Figure 1. With quantitative synthesis, seventeen studies were included in our final analysis11-27.

Features of included studies are summarized in Table 1. For influenza A, we identified 17 studies involving 8,334 participants. For influenza B, sixteen studies involving 7,909 subjects met the defined inclusion criteria. One study assessed influenza A infection only21. The number of patients in each trial ranged from 68 to 1,649. All studies were published between 2012 and 2018. Most studies evaluated combined populations of adults and children.

Results of the QUADAS-2 assessment are shown in Figure 2. For patient selection, the index test, and the reference standard domain, more than 50% of included studies were judged to have a low risk of bias. However, for the study flow and the timing domain, the risk of bias was high or unclear for 52.9% of the included studies because of unclear intervals between the index test and the reference standard. Considering our inclusion criteria, we had little concern for the applicability of results from selected studies for each domain.

Figures 3 and 4 show paired forest plots of sensitivity and specificity of the Quidel Sofia rapid influenza FIA for detecting influenza A and B. The pooled sensitivity across studies of the Quidel Sofia rapid influenza FIA for identifying influenza A was 0.78 (95% CI, 0.71-0.83), with a pooled specificity of 0.99 (95% CI, 0.98-0.99). The pooled PLR and NLR were 56.99 (95% CI, 31.87-101.90) and 0.23 (95% CI, 0.18-0.29), respectively. The DOR for influenza A was 251.26 (95% CI, 139.39-452.89).

The pooled sensitivity across studies for identifying influenza B was 0.72 (95% CI, 0.60-0.82), with a pooled specificity of 0.98 (95% CI, 0.96-0.99). The pooled PLR and NLR were 40.08 (95% CI, 17.26-93.07) and 0.29 (95% CI, 0.19-0.42), respectively. The DOR for influenza B was 140.20 (95% CI, 55.92-351.54). The pooled sensitivity and specificity of the Quidel Sofia rapid influenza FIA for identifying influenza A and B were not significantly different (p=0.341 for sensitivity and p=0.206 for specificity). Figure 5 shows HSROCs for the index test. AUCs of the Quidel Sofia rapid influenza FIA for identifying influenza A and influenza B were similar (0.96 with 95% CI of 0.94-0.98 for influenza A vs. 0.95 with 95% CI of 0.92-0.96 for influenza B, p=0.166).

The Higgins I² statistics proved significant heterogeneity for both the sensitivity (0.91 with 95% CI of 0.88-0.94 for influenza A and 0.87 with 95% CI of 0.82-0.92 for influenza B) (Figure 3) and specificity (0.92 with 95% CI of 0.90-0.95 for influenza A and 0.98 with 95% CI of 0.97-0.98 for influenza B) (Figure 4). Subgroup analyses were also performed to investigate potential sources of heterogeneity (Table 2). The sensitivity of the Quidel Sofia rapid influenza FIA was significantly increased when tests were performed in children (0.86 with 95% CI of 0.78-0.92 for influenza A and 0.79 with 95% CI of 0.71-0.85 for influenza B) than when tests were performed in adults (0.74 with 95% CI of 0.67-0.79 for influenza A and 0.33 with 95% CI of 0.10-0.65 for influenza B).

In sensitivity analysis to investigate the influence of each individual study on the overall analysis estimate, one study had a significantly different sensitivity than other studies on influenza A24. Even after exclusion of that study24, the pooled sensitivity across studies on influenza A was similar to that of overall studies (0.79 with 95% CI of 0.75-0.82). Instead, the heterogeneity decreased (0.51 with 95% CI of 0.23-0.79).