Search Strategy and Selection Criteria

A systematic review and meta-analysis was conducted in accordance with PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines (19). Literature searches were performed on the PubMed and Web of Science databases for the period from the inception of these databases to the beginning of 2015 to identify studies that reported laboratory-confirmed influenza infection (i.e., by culture, PCR, or serologic testing) and the proportion of symptomatic versus asymptomatic presentation. Search terms were chosen to ensure maximum coverage of possible literature and included the terms “influenza,” “carrier,” “carriage,” “shedding,” “asymptomatic,” “influenza AND prophylaxis NOT vaccine” (filtered for randomized control trials), “influenza AND (travel OR migration OR immigra*) AND (screening OR test OR testing OR detection),” “subclinical,” “serosurvey OR seroprevalence OR seroepidemiology.” Other keywords and connectors were also used (Technical Appendix 1).

To be eligible for inclusion, studies needed to 1) be peer-reviewed and 2) report the prevalence of asymptomatic influenza virus infections in humans or present the appropriate data from which that prevalence could be calculated. Laboratory confirmation of influenza was a requirement, and it had to be possible to correlate these data to the number of symptomatic patients. We did not impose limitations in terms of study design, influenza virus type, or exposure type (community or experimental inoculation). According to current World Health Organization guidelines, laboratory confirmation consisted of 1) conventional PCR (referred to here as PCR) or real-time reverse transcription PCR (rRT-PCR); 2) virus antigen detection by immunofluorescence or enzyme immunoassay methods; 3) serologic detection of antibodies (hemagglutination inhibition); or 4) virus culture (20). Studies were excluded when the use of antiviral agents without a placebo group was reported. In cases in which a placebo group was used and an asymptomatic proportion could be determined, only this subset was used; otherwise, the study was excluded. Results were restricted to studies published in English; however, no restriction was placed on the publication date of studies that fit these criteria.

Study Selection and Data Extraction

Two authors (L.F.-K. and M.C.) independently screened the publications for eligibility in a stepwise fashion. Search results were initially screened based on article titles and abstracts. Then, full-text analysis was performed to identify all studies which either reported asymptomatic prevalence or from which asymptomatic prevalence could be calculated. Any discrepancies that might have affected inclusion or exclusion of a study were resolved through discussion and consensus after independent evaluation by another author (L.Y.). The same 2 authors (L.F.-K. and M.C.) assessed the risk for bias of the studies included by using a modified version of the tool developed by Hoy et al. (21) for prevalence studies (Technical Appendix 2).

The definitions of asymptomatic influenza infection varied considerably between studies. Definitions ranged from a total absence of symptoms to a lack of influenza-like illness (ILI) or acute respiratory illness (ARI). For the sake of clarity, we used the term “asymptomatic” when there was a total absence of symptoms and “subclinical” when the patient did not meet the authors’ criteria for ILI or ARI. Asymptomatic influenza prevalence was considered to be the proportion of all persons with laboratory-confirmed influenza who had no symptoms, whereas subclinical influenza prevalence was the proportion of persons with laboratory-confirmed influenza who failed to meet the study’s definition of symptomatic infection. In addition to collecting data on asymptomatic and subclinical infection prevalence, we collected data on influenza virus type/subtype and study characteristics (e.g., study design, sample size, diagnostic test used to detect influenza virus infection, and the working definition of “symptomatic”).

Statistical Analysis

We used prevalence of asymptomatic versus subclinical carriers among persons with laboratory-confirmed influenza as primary endpoints of interest. We pooled the prevalence estimates of asymptomatic and subclinical influenza across studies by using 2 meta-analytical models, the inverse variance heterogeneity model (22) and the random effects model.

We observed considerable heterogeneity across studies. This heterogeneity was unlikely to be attributable only to random or systematic errors, and actual clinical heterogeneity was deemed to exist. Therefore, we created subgroups by influenza virus type/subtype with the aim of generating more homogeneous groups within which we could anticipate that the differences indeed reflected variability caused by random or systematic error rather than actual clinical heterogeneity. In addition, we built a linear model to examine the variance explained by the influenza virus type/subtype, laboratory test used to detect the virus, year of the study, and geographic location of the study to gain insight into the considerable heterogeneity observed in the prevalence of asymptomatic and subclinical infections. We conducted the meta-analyses by using MetaXL version 2.0 (EpiGear Int Pty Ltd, Brisbane, QLD, Australia), which also included the inverse variance heterogeneity method, and the generalized linear model by using Stata version 12 (StataCorp LP, College Station, TX, USA). All tests were 2-tailed, and a p value <0.05 was deemed statistically significant.

Yield of Search Strategy

Figure

Figure. PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analysis) flowchart of literature search for systematic review and meta-analysis of asymptomatic and subclinical influenza infection prevalence.

A total of 13,219 records were identified from literature searches of the 2 databases. This number was reduced to 9,900 after removal of publications that were either duplicates or not original research papers (e.g., review papers). An additional 3,663 papers were removed based on the title and 5,652 papers more based on the abstract. The full texts of the remaining 585 studies were examined, and 55 articles met the inclusion criteria and were included in the final analysis (Figure; Technical Appendix 1).

Characteristics of the Studies Included

The 55 articles provided 59 data points because 4 papers reported the prevalence of asymptomatic and subclinical carriers for influenza A and B viruses separately. Overall, 19 studies (22 data points) defined asymptomatic infection as cases in persons lacking symptoms, and 44 studies (46 data points) reported subclinical influenza virus infections.

Infection was confirmed by serologic testing, rRT-PCR, or viral culture; 28 studies reported use of serologic testing alone to confirm infection, 18 used rRT-PCR alone, and the remaining 9 used a combination of methods (5 serologic testing and rRT-PCR, 3 serologic testing and culture, and 1 rRT-PCR and culture). Among the 55 studies, influenza A virus (predominantly H1N1) was the most common type of infection; 5 studies reported influenza B virus infections, and 1 study reported influenza C infections (Technical Appendix 1 Table 2). Most studies reported on pandemic influenza virus types; 32 of these studies related to the 2009 pandemic influenza A/Mexico/4108/2009 strain. The risk for bias was moderate in 32% of the studies and low in the remaining 68%; no study was found to have a high risk for bias.

Quantitative Synthesis

The overall pooled prevalence for asymptomatic carriers was 19.1% (95% CI 5.2%–35.5%) for any type of influenza, 21.0% (95% CI 4.2%–41.0%) for influenza A, and 22.7% (95% CI 7.7%–39.8%) for influenza A(H1N1) (Table 1; Technical Appendix 2 Figure 1). For subclinical carriers, the overall pooled prevalence was 43.4% (95% CI 25.4%–61.8%) for any type of influenza, 42.8% (95% CI 22.3%–63.9%) for influenza A, and 39.8% (95% CI 16.4%–64.5%) for influenza A(H1N1) (Table 1; Technical Appendix 2 Figure 2). However, extensive heterogeneity was immediately evident for reported asymptomatic prevalence (τ² = 0.31) and subclinical prevalence (τ² = 0.45) that could not be explained by the influenza type/subtype alone. Similar results were obtained with the random effects model (Technical Appendix 2 Figures 3, 4).

Investigation of Heterogeneity

The considerable heterogeneity observed within asymptomatic and subclinical influenza prevalence could not be explained by the type/subtype of influenza, the laboratory tests used to detect the virus, the location where the study was conducted, or the year of the study. The multivariate regression models could only explain 16.8% and 14.8% of the observed variance for the asymptomatic and subclinical prevalence, respectively. Influenza type/subtype as an independent predictor was found to account for almost the entire variance (16%) found for the prevalence of asymptomatic carriers (Table 2).

Publication Bias

For both asymptomatic and subclinical carrier prevalence, the funnel plots showed no indication of publication bias. This result was confirmed by Doi plots (data not shown).