Point-of-Care Molecular Flu A/B Test

Against the FDA-cleared molecular comparator, influenza A sensitivity was about 97% and specificity about 94%. Influenza B sensitivity was about 94% and specificity about 99%. The test was positive for influenza A in about 23% and for influenza B in about 10% of evaluable specimens.

Discordant testing with an alternative FDA-cleared molecular assay detected influenza A in most of the influenza A false positives and did not detect it in some of the false negatives. For influenza B, the alternative assay detected virus in more than half of the false positives and did not detect it in more than half of the false negatives. The initial invalid rate was about 9%, falling to about 1% after repeat testing per the instructions.

In the near cut-off study, untrained intended operators agreed with expected results in about 97% of low-positive influenza A samples, about 97% of low-positive influenza B samples and 100% of negatives. One negative ended as an unresolved invalid. In the reproducibility study, agreement with expected results was 100% across all sites, operators and days. FDA found the device substantially equivalent under the dual-track review, and FDA's CLIA database lists the test as waived.

Measured against the 2011 influenza guidance's recommendations for nucleic acid tests, both analytes meet every benchmark, including the recommended confidence-interval lower bounds for sensitivity and specificity. Influenza A specificity (about 94%) is the lowest point estimate, but the discordant analysis puts it in context: the alternative molecular assay found influenza A in most of the specimens counted as false positives. That points to differences in analytical sensitivity between methods rather than spurious signals. The decision summary labels the results sensitivity and specificity even though the comparator was a molecular assay. The guidance's convention for a non-culture comparator would be percent agreement.

Confidence interval width follows the number of comparator positives, not total enrollment: with about 300 influenza A positives and about 150 influenza B positives, the influenza B interval was roughly twice as wide. The 510(k) summary's clinical section reports slightly different limits for two of the estimates than the decision summary does, so the interval method is worth stating explicitly in the statistical analysis plan.

The test was submitted as a combined 510(k) and CLIA waiver package for dual review. Its intended use is as an aid in diagnosing influenza in patients with signs and symptoms of respiratory infection, together with clinical and epidemiological risk factors. A nasal swab is eluted and tested directly, with no separate nucleic acid extraction, in either clinical laboratory or point-of-care settings. The operator reads results visually as colored lines on a test strip.

The clinical evidence therefore had to show accuracy against a high-performance FDA-cleared molecular influenza assay on fresh, prospectively collected specimens from children and adults, with the investigational test run within one hour of collection at intended-use sites by the kind of people who would run it in practice. Both analytes needed enough comparator positives for usable confidence intervals.

For the waiver, it also had to show that untrained intended operators could detect samples at the limit of detection, where a visual read is most likely to be wrong, and that results were reproducible across sites, operators and days.

Molecular tests that detect and differentiate influenza A and B are classified under 21 CFR 866.3980, respiratory viral panel multiplex nucleic acid assay, as Class II devices subject to special controls. The special control is FDA's respiratory viral panel special controls guidance. FDA's 2011 influenza IVD guidance describes the clinical study: prospective, fresh specimens from sequentially enrolled patients for each claimed specimen type, at least three geographically diverse facilities, and, for point-of-care devices, sites representing the non-laboratory settings of intended use.

That guidance also publishes numeric expectations. Nucleic acid-based tests should show at least 90% sensitivity for each analyte and specimen type, with the lower bound of the two-sided 95% confidence interval above 80%. All influenza devices should show specificity with a 95% CI lower bound above 90%. When the comparator is an FDA-cleared nucleic acid test rather than culture, the guidance says results should be expressed as positive and negative percent agreement. These are nonbinding recommendations. They differ from the codified minimum performance criteria in 21 CFR 866.3328, which apply to antigen-based rapid influenza tests and not to nucleic acid assays.

A molecular test meant for physician offices and clinics also needs a CLIA waiver, which requires evidence that the test is simple and has an insignificant risk of an erroneous result in the hands of untrained operators. Since MDUFA III (2012), sponsors have been able to file a Dual 510(k) and CLIA Waiver by Application, in which one set of comparison and reproducibility studies run by intended waived users supports both decisions. FDA's current written recommendations for that pathway were finalized after this clearance. The difficulty is keeping laboratory-level molecular sensitivity when non-laboratory staff run the test between patients, in a season whose timing the sponsor cannot control.

• Non-laboratory personnel ran the investigational test at intended-use sites while trained staff ran the comparator at central laboratories. That lets one dataset answer both questions: substantial equivalence for the 510(k), and accuracy in waived hands for the CLIA waiver.
• Testing fresh swabs within one hour of collection follows the influenza guidance's preference for fresh specimens from sequentially enrolled patients over frozen archives for clinical sensitivity and specificity, and it matches how the test is used at the point of care.
• Collecting both swabs from the same nostril and sending the comparator swab to a high-performance FDA-cleared molecular assay keeps the reference as close as possible to the investigational sample, so disagreements more likely reflect the assays than sampling differences.
• Enrolling across about fifteen sites through a full season protected the influenza B arm. Influenza B was the less common analyte (about 150 comparator positives against about 300 for influenza A), and its sensitivity interval is the widest in the study.
• Low-positive samples at about 1X LoD were masked as patient samples in the daily workflow. This tests the hardest visual read, a faint line, under realistic conditions.