A Rapid Influenza Assay Moves from the Emergency Department to Employee Screening
A hospital laboratory develops a rapid antigen assay for influenza. In a prospective study, 400 adults who present to the emergency department with fever and cough are tested with the new assay and with RT-PCR as the reference standard. Of the 120 PCR-positive patients, 108 are antigen-positive. Of the 280 PCR-negative patients, 14 are antigen-positive. Later, the same assay is proposed for screening all hospital employees on day 1 of a respiratory outbreak, in which the anticipated prevalence is 2%. Employees would self-collect nasal swabs without supervision and would be excluded from work on the basis of a positive result.
1. Calculate the sensitivity, specificity, and positive predictive value of the assay in the emergency department cohort.
Read the worked answer
Set up the standard 2×2 table (reference standard = RT-PCR):
| PCR + | PCR − | Total | |
|---|---|---|---|
| Antigen + | 108 | 14 | 122 |
| Antigen − | 12 | 266 | 278 |
| Total | 120 | 280 | 400 |
- Sensitivity = TP/(TP+FN) = 108/120 = 0.90 (90%)
- Specificity = TN/(TN+FP) = 266/280 = 0.95 (95%)
- Prevalence in this cohort = 120/400 = 0.30
- PPV = TP/(TP+FP) = 108/122 = 0.885 (≈89%)
- NPV = 266/278 = 0.957 (≈96%)
The performance looks excellent - but that judgment is conditional on a 30% pretest probability. Sensitivity and specificity are properties of the test; predictive values are properties of the situation.
2. What would the positive predictive value be at 2% prevalence, and what does that mean for the proposed screening program?
Read the worked answer
Use Bayes' theorem or a hypothetical 1,000 employees:
- Diseased: 1,000 × 0.02 = 20 → true positives = 20 × 0.90 = 18
- Nondiseased: 1,000 × 0.98 = 980 → false positives = 980 × 0.05 = 49
- PPV = 18/(18 + 49) = 18/67 ≈ 0.27 (27%)
- NPV = 931/(931 + 2) ≈ 99.8%
Interpretation:
- Roughly three of every four employees excluded from work would not have influenza - an enormous operational and morale cost for a test that looked superb in the ED.
- The NPV is near-perfect, so a negative result is genuinely reassuring, which is the more defensible use of the assay (rule-out in low-prevalence settings, with confirmation of positives).
- Any program that acts punitively on a positive result must reflex to PCR confirmation. Better still, choose the threshold deliberately and understand that lowering the cut-point raises sensitivity at the cost of even more false positives.
3. The laboratory director cites the 400-patient study as proof the assay is validated for the screening use. What design limitation most undermines that claim, and what additional problem arises from unsupervised self-collection?
Read the worked answer
Spectrum bias. The validation cohort was symptomatic ED patients with high viral shedding and a high pretest probability. Sensitivity estimated in that group will not transfer to asymptomatic or paucisymptomatic employees, whose viral loads are lower and in whom antigen tests are known to perform substantially worse. The study population must resemble the population in which the test will be used.
Other limitations worth naming:
- Verification/reference-standard bias if PCR was obtained preferentially in those with a positive antigen result.
- Imperfect reference standard: RT-PCR is not a true gold standard; discordant results were not adjudicated.
- Small numbers in the positive cell produce wide confidence intervals.
Self-collection adds a second, independent problem: the study measured sensitivity and specificity under trained collection. Self-administered nasal swabs introduce a variable, generally lower adequacy of sampling, which will further reduce sensitivity in exactly the low-prevalence setting where every point of sensitivity matters. A bridging study in the actual intended population and collection method would be required.