Screening ELISA: Predictive Values and a Negative Result in Advanced Disease
A hospital laboratory is evaluating a one-step sandwich ELISA for a tumour marker. In a validation panel of 1000 samples with disease status confirmed by imaging, the assay detected 92 of the 100 true cases and gave 108 false positives among the 900 disease-free samples. The planned use is first-line screening of an asymptomatic population in which prevalence is estimated at 1.5%. The laboratory also receives a sample from a patient with a 9 cm tumour whose ELISA is negative at the standard 1:1 dilution but strongly positive (about 40,000 ng/mL) at 1:100 and 1:1000 dilutions. A confirmatory immunoblot (sensitivity 90%, specificity 99.5%) is available and is applied only to samples that are ELISA-positive.
1. Calculate the sensitivity, specificity, PPV and NPV of the ELISA at 1.5% prevalence, working per 1000 screened individuals.
Read the worked answer
From the panel: sensitivity = 92/100 = 92%; specificity = (900 − 108)/900 = 792/900 = 88%.
Apply these to 1000 screened people at 1.5% prevalence (15 diseased, 985 healthy):
- True positives = 0.92 × 15 = 13.8
- False negatives = 15 − 13.8 = 1.2
- False positives = 0.12 × 985 = 118.2
- True negatives = 985 − 118.2 = 866.8
PPV = TP/(TP + FP) = 13.8/(13.8 + 118.2) = 13.8/132.0 = 10.5%. NPV = TN/(TN + FN) = 866.8/(866.8 + 1.2) = 866.8/868.0 = 99.9%.
The key point is that a test with 92% sensitivity still has a PPV near 10% when prevalence is low: nine out of ten ELISA-positive screenees are false positives, and 118 people are recalled per true case detected. Sensitivity and specificity are properties of the test; PPV is a property of the test plus the population.
2. Calculate the PPV after the confirmatory immunoblot is applied to ELISA-positive samples, and compare with the PPV if the test were used in a 20% prevalence group.
Read the worked answer
Feed the 132 ELISA-positives (13.8 true, 118.2 false) into the immunoblot:
- True positives after blot = 0.90 × 13.8 = 12.4
- False positives after blot = 0.005 × 118.2 = 0.59
- PPV = 12.4/(12.4 + 0.59) = 12.4/13.0 = 95.5%
So a two-step algorithm changes a 10.5% PPV into 95.5%, at the cost of about 1.4 missed cases per 1000 screened. Because specificity is applied to the large false-positive pool, its effect on PPV is far greater than a matching gain in sensitivity - a point worth stating explicitly.
At 20% prevalence: TP = 0.92 × 200 = 184; FP = 0.12 × 800 = 96; PPV = 184/280 = 65.7%. The same ELISA is far more informative in a high-risk or symptomatic population, which is the argument for restricting it to that setting rather than asymptomatic screening.
3. Explain the negative ELISA in the patient with the 9 cm tumour and propose two design changes that would prevent this failure.
Read the worked answer
This is the high-dose hook effect. In a one-step sandwich assay, sample, capture antibody and detection antibody are mixed together. At very high antigen concentration, both antibodies become saturated with separate antigen molecules, so no capture–detection sandwich can form on the solid phase. Signal therefore falls as concentration rises, producing a false negative in exactly the patients with the largest tumour burden. Diluting the sample lowers antigen into the assay's working range, which is why 1:100 and 1:1000 are strongly positive (40,000 ng/mL); the true value must be reported from the dilution, corrected by the dilution factor.
Design fixes:
- Convert to a two-step (sequential) assay: capture and wash first, then add detection antibody, so antigen excess cannot block sandwich formation to the same degree.
- Add a routine antigen-excess check - assay each sample at two dilutions and flag any sample where the diluted result exceeds the neat result, or spike a known amount of antigen and check recovery.
Procedurally, any result discordant with the clinical picture should be repeated at 1:10 and 1:100 before being reported as negative.