The review included 159 datasets in 119 studies. The number of included participants was unclear. The inclusion criteria largely ruled out partial verification, differential verification and incorporation bias. Reference standards were considered to be appropriate. Approximately 15% of studies reported blinding of interpreters to reference standard and 20% reported blinding to other clinical data. Approximately 15% to 20% of studies reported on withdrawals and/or uninterpretable results. Approximately one third of studies were clear about the basis for patient or specimen inclusion.
Pooled sensitivity was 62.3% (95% CI 57.9% to 66.6%). Pooled specificity was 98.2% (95% CI 97.5% to 98.7%). Positive likelihood ratio was 34.5 (95% CI 23.8% to 45.2%). Negative likelihood ratio was 0.38 (95% CI 0.34 to 0.43). The HSROC demonstrated variation in sensitivity (range 4.4% to 100%), but only 11% of studies showed less than 85% specificity.
Subgroup analyses showed that RIDTs had a significantly higher pooled sensitivity in children (66%, 95% CI 61.6% to 71.7%; 60 datasets) than in adults (53.9%, 95% CI 47.9% to 59.8%; 33 datasets); specificities were similar between the age groups. RIDTs were associated with significantly higher sensitivity for detecting influenza A (64.6%, 95% CI 59.0% to 70.1%; 72 datasets) than for detecting influenza B (52.2%, 95% CI 45% to 59.3%; 27 datasets). Results were unchanged when RIDT brand, specimen type and reference standard were taken into account. Comparing reference standards, RIDTs had a significantly higher pooled sensitivity when compared with viral culture (72.3%, 95% CI 66.8% to 77.9%; 48 datasets) than RT-PCR (53.9%, 95% CI 48.2% to 59.6%; 67 datasets). Further results are reported in the paper.