A Point-of-Care Antigen Test Across Two Prevalences
A point-of-care antigen test for a respiratory infection is evaluated in 1,200 symptomatic adults at a tertiary hospital. Using a validated PCR as the reference, 240 patients are infected. The antigen test is positive in 204 of the infected patients and in 96 of the uninfected patients. The manufacturer's brochure states: 'Sensitivity 85%, specificity 90%; negative result reliably rules out infection.' A primary-care clinic with an estimated prevalence of 2% plans to use the same test. In a post hoc analysis, sensitivity is 95% in patients admitted to intensive care and 70% in outpatients with mild symptoms; the manufacturer proposes a 'severe-case' indication. A clinician says, 'A specificity of 90% means a 10% false-positive rate, so one in ten positives is wrong.'
1. Calculate sensitivity, specificity, PPV, and NPV in the tertiary-hospital cohort. Explain why the brochure's statement about a negative result is context-dependent.
Read the worked answer
Sensitivity = 204/240 = 0.85, or 85%. Specificity = (960 − 96)/960 = 864/960 = 0.90, or 90%. The positive predictive value is 204/(204 + 96) = 204/300 = 0.68, or 68%. The negative predictive value is 864/(864 + 36) = 864/900 = 0.96, or 96%. The brochure's statement that a negative result 'reliably rules out infection' is only true in a context with a pre-test probability similar to the trial's 20% prevalence, and even there 4% of negative results are false negatives. In a lower-prevalence setting, the NPV rises further, but that does not mean the test is universally reliable: the PPV falls, so a positive result becomes less trustworthy. The statement also ignores the consequences of a false-negative result, which depend on the disease and the patient. Predictive values are prevalence-dependent; sensitivity and specificity are not.
2. Recalculate the PPV and NPV at 2% prevalence, assuming sensitivity and specificity are unchanged. What is the clinical implication, and what is the flaw in the clinician's statement?
Read the worked answer
At 2% prevalence in 1,000 patients: 20 infected, 17 true positives and 3 false negatives; 980 uninfected, 98 false positives and 882 true negatives. PPV = 17/(17 + 98) = 17/115 = 0.148, or about 15%. NPV = 882/(882 + 3) = 882/885 = 0.997, or 99.7%. Thus a positive result in primary care is more likely to be a false positive than a true positive, and a confirmatory PCR is needed before treatment or isolation. A negative result is very reassuring, but the absolute number of missed cases is small only because the disease is rare. The clinician's statement confuses specificity with 1 − PPV. Specificity of 90% means that 10% of uninfected people test positive (the false-positive rate conditional on being uninfected), not that 10% of positive results are wrong. The false-positive fraction among positives is 1 − PPV, which depends on prevalence and here is 32% at 20% prevalence and 85% at 2% prevalence.
3. Interpret the subgroup sensitivity difference. Does it justify a 'severe-case' indication? What design would you require before accepting the claim?
Read the worked answer
The 95% versus 70% sensitivity difference is a spectrum effect. Intensive-care patients likely have higher viral loads, more severe disease, and better specimens, while mild outpatients may be earlier in infection or have lower pathogen burden. A post hoc subgroup finding, however, is weak evidence: the subgroups were not randomized, multiple comparisons were likely, and the confidence intervals may be wide. A 'severe-case' indication would also be odd clinically, because patients with severe disease already have a high pre-test probability and often need a definitive PCR; the test's value would be greater in those with uncertain status. Before accepting the claim, I would want a prospective, pre-specified study in consecutive patients, with blinded test readers, a consistent reference standard, and reporting of sensitivity and specificity by subgroup with confidence intervals. The study should also show that using the test in that subgroup changes management or improves outcomes, not merely that sensitivity is higher.