Reader-Based Influenza A/B Antigen Test

Of about 1,050 subjects enrolled, a few dozen specimens were excluded for eligibility, sample handling or inconclusive reference results, and a handful more gave a final invalid result, leaving about 1,000 specimens with valid results. Against the composite reference, influenza A sensitivity was about 96% and specificity about 96%. Influenza B sensitivity was about 82% and specificity about 98%.

By specimen type, nasal swabs (n = about 500) gave influenza A sensitivity of about 96% and influenza B sensitivity of about 81%. Nasopharyngeal swabs (n = about 500) gave about 96% and 83%. In read-now mode (n = about 500), influenza A and B sensitivity were about 94% and 81%. In walk-away mode (n = about 500) they were about 98% and 82%. Performance was similar across the study sites. The initial invalid rate was about 1%; about half of those samples gave valid results on retest and the rest were not retested.

In the reproducibility study, agreement with expected results was 100% for influenza A moderate and low positives, influenza B low positive, influenza B high negative and the negative sample. It was about 98% for influenza A high negative and influenza B moderate positive. The initial failure rate of external quality control materials, invalid and failed results combined, was about 1%, and every one gave the expected result on repeat testing.

FDA found the device substantially equivalent under 21 CFR 866.3328 (Class II) through the dual-track pathway, and FDA's CLIA database lists the test as waived.

Because the composite comparator is not viral culture alone, the special-control criteria for comparators other than culture appear to be the relevant benchmark: agreement with positives of at least 80% (lower bound at least 70%) and agreement with negatives of at least 95% (lower bound at least 90%). Influenza A cleared those bounds easily for both sensitivity and specificity. Influenza B is the informative case. The overall point estimate of about 82% clears the 70% floor, but the per-specimen-type estimates of about 81% (nasal) and 83% (nasopharyngeal) clear it by only modest margins. Splitting by specimen type nearly halves each denominator, so the intervals widen. Sponsors should power each specimen arm on its own.

Subgroups show why sample size matters. In subjects 22 years and older, influenza B sensitivity was about 70%, compared with about 96% in children 5 and under. Age subgroups are not held to the special-control criteria, but they show how much a point estimate can move with a few dozen positives. The specificity estimates count composite-negative specimens that the device called positive as false positives. With a two-of-three rule, a specimen positive by only one reference method is scored negative, so some apparent false positives may reflect low-level infection rather than device error.

The test is an aid in diagnosing influenza A and B in symptomatic patients. It distinguishes the two types from one processed nasal or nasopharyngeal swab on a single test cassette, and the reader interprets the result. The intended users include clinical laboratory staff and users in CLIA-waived settings, such as nurses and medical assistants in emergency rooms, urgent care clinics and outpatient clinics, including physician offices. The reader has two workflows: a walk-away mode in which the reader times the 15-minute development, and a read-now mode in which the operator times development externally and then inserts the cassette.

Under 21 CFR 866.3328, sensitivity and specificity for influenza A and B had to meet the published minimum criteria for each claimed specimen type, so nasal and nasopharyngeal swabs each needed enough positives to produce usable confidence intervals. The waiver claim added a second demand: the same accuracy had to come from untrained operators working within the daily routine of CLIA-waived sites. Because the device had two read modes, the evidence also had to show that neither workflow lowered accuracy.

Reproducibility also had to be shown in the hands of intended users near the decision point, and the labeling had to state the season in which performance was established and the influenza A subtypes that predominated during it.

Rapid influenza antigen tests are regulated under 21 CFR 866.3328, influenza virus antigen detection test system (devices detecting influenza A, B, and C virus antigens). A final order placed these tests in Class II with special controls. The regulation's own identification explains why: the virus mutates, new strains emerge that may affect performance, and because influenza can cause severe illness, the accuracy of these devices has serious public health implications.

The special controls set published minimum clinical performance criteria that apply to each specimen type claimed. Against an FDA-cleared nucleic acid-based test or another FDA-accepted comparator other than viral culture, positive percent agreement for influenza A and B must have a point estimate of at least 80 percent with a 95% confidence interval lower bound of at least 70 percent. Negative percent agreement must reach at least 95 percent with a lower bound of at least 90 percent. Against correctly performed viral culture, influenza A sensitivity must reach at least 90 percent (lower bound at least 80 percent), influenza B sensitivity at least 80 percent (lower bound at least 70 percent), and specificity at least 95 percent (lower bound at least 90 percent). The special controls also require annual analytical reactivity testing with contemporary strains identified by FDA in consultation with CDC, with the last three years of results in labeling by July 31 each year, and expedited testing when an emergency involving an influenza strain is declared.

Many of these tests are also meant for physician offices, urgent care and other CLIA-waived settings, so sponsors often use FDA's dual 510(k) and CLIA waiver by application pathway (described in final guidance). That pathway lets one set of comparison and reproducibility studies, run by untrained operators at intended-use sites, support both substantial equivalence and waiver. The companion CLIA waiver guidance recommends at least three intended-use sites, one to three untrained operators per site and at least nine overall. Reproducibility should be run at a minimum of three of the same sites with panels near the detection limit. What makes the category hard is that the clinical lower bounds must be met during a single respiratory season whose strain mix and prevalence the sponsor cannot control.

• Randomizing subjects to nasal or nasopharyngeal pairs produced separate performance estimates for each specimen type, at about 500 evaluable specimens each. The special controls apply their minimum criteria per claimed specimen type, so pooled data alone would not have been enough.
• Supplying every site with two readers, one per mode, let read-now and walk-away performance be reported separately. The workflow then could not hide a mode-specific error such as mistimed development.
• Using a two-of-three composite of two molecular assays and culture lowered the chance that one comparator's error would be counted as a device false positive or false negative. Same-day courier shipment kept the reference arm within each method's allowable specimen age.
• About fifteen CLIA-waived sites with one to three untrained operators each is well above the minimum of three sites and nine operators recommended in FDA's CLIA waiver guidance. It also put the operators, patients and workflows the waiver would cover into the accuracy estimate itself.
• The reproducibility panel included high-negative and low-positive levels and used non-laboratory staff, including administrative assistants. That tests the reader's decision threshold where operator technique is most likely to change a call.