Validating HPV primary screening in a 22,000-smear laboratory
The director of a laboratory serving a 400-bed hospital and its screening programme is considering replacing cervical cytology with HPV primary screening on a new platform. The laboratory currently processes 22,000 cervical samples per year. A validation study of 1,500 specimens with complete colposcopy and histology follow-up gives a sensitivity of 94% and a specificity of 88% for CIN2 or worse. The prevalence of CIN2+ in the screening population is estimated at 1.2%. The laboratory has three cytotechnologists and one cytopathologist, and the colposcopy clinic can absorb about 15 new referrals per week. The platform is being marketed with the claim that it 'detects almost all disease'.
1. Calculate the positive and negative predictive values in this population and interpret them for the laboratory director. How many colposcopies are needed to find one case of CIN2+?
Read the worked answer
Working, per 1,000 women screened (prevalence 1.2% = 12 true cases, 988 unaffected):
- True positives = 0.94 × 12 = 11.28
- False negatives = 12 − 11.28 = 0.72
- False positives = 0.12 × 988 = 118.56
- True negatives = 988 − 118.56 = 869.44
PPV = 11.28 / (11.28 + 118.56) = 11.28 / 129.84 ≈ 8.7% NPV = 869.44 / (869.44 + 0.72) ≈ 99.9%
Interpretation: the test is excellent at reassuring - a negative result is genuinely reassuring, and only about 1 in 1,400 screened women with a negative result has an undetected CIN2+ lesion. It is poor at predicting: roughly 11–12 women must go to colposcopy for every one case of CIN2+ found, because nine out of ten positives are false positives at this prevalence.
Scale for this laboratory: with 22,000 samples per year, about 264 CIN2+ cases are expected. Positive results would be approximately 0.94 × 264 + 0.12 × 21,736 ≈ 248 + 2,608 ≈ 2,856 per year - about 55 colposcopies per week, far above the clinic's capacity of 15. That mismatch, not the sensitivity figure, is the decisive operational finding. Sensitivity and specificity are properties of the test; PPV and the downstream workload are properties of the population and the service.
2. On the basis of this data, should the laboratory switch? Set out the decision criteria and what additional information you would require.
Read the worked answer
The marketing claim is true but irrelevant. 'Detects almost all disease' describes sensitivity, which is high by design; the operational and human cost sits in the 8.7% PPV.
Criteria that must be satisfied before switching:
- Triage strategy. HPV primary screening is only deliverable with a triage step - genotyping for HPV16/18, p16/Ki-67 dual stain, or cytology - to reduce the colposcopy load to what the service can absorb. Presenting the validation data without a triage arm is an incomplete proposal.
- Capacity. Model the annual positive and referral volumes against colposcopy slots, histology capacity and clinic waiting times; a screening programme that cannot deliver colposcopy creates harm through delay.
- Validation adequacy. Confirm the study followed the accepted validation framework for the platform (sample size, disease prevalence in the validation set, reproducibility, and comparison against the existing method), not merely 1,500 convenient specimens.
- Workforce and workflow. Three cytotechnologists and one cytopathologist - who reads the triage cytology and the dual stains, and what is the turnaround time?
- Quality metrics, cost and equity. HPV-positive rate, HPV-negative CIN3+ audit, cost per CIN2+ detected, and whether harder-to-reach groups will return for repeated testing.
Recommendation: do not switch on these data alone. Run a phased pilot with a defined triage algorithm, pre-agreed referral caps, and a stopping rule. The correct answer is a managed implementation plan, not a binary yes.
3. Twelve months into the proposed service, an audit finds that 2 of 20 reviewed HPV-negative women with subsequent CIN3 were screen-negative on the same platform. What quality-assurance framework do you apply?
Read the worked answer
Recognise what this is: not a statistical curiosity but a possible false-negative signal. A 10% false-negative rate among reviewed CIN3 cases would trigger a formal investigation and, in most frameworks, a lookback.
Immediate actions:
- Verify the denominator - were the 20 cases selected because of symptoms or as a random audit? Selection bias changes the interpretation entirely.
- Review the original specimens and the pre-analytical record: collection device, transport time, storage, DNA yield, inhibition, and whether the sample was adequate (cellularity, presence of the internal control).
- Distinguish assay failure from biological false negativity - low viral load, HPV types not covered by the assay, or integration/deletion of the L1 target region used by many PCR tests.
- Check operator and lot-level performance: batch effects, reagent lot numbers, maintenance and calibration records.
Systemic response:
- Re-audit against a defined sampling frame; report the false-negative rate with confidence intervals rather than a bare proportion.
- Review the laboratory's participation in external quality assessment and its internal concordance testing.
- Re-examine the triage pathway: were these women triaged correctly, or was the triage test the weak link rather than the primary screen?
- If the problem is assay-related, notify the manufacturer, consider a lookback of affected batches, and document the corrective action.
Governance: present to the quality committee with a corrective action plan and a re-audit date. Two cases in twenty is enough to open the question - not enough to close it.