RE: p16/Ki-67 Dual Stain Cytology for Detection of Cervical Precancer in HPV-Positive Women: Table 1.

I read with great interest the article by Wentzensen et al. (1). The analysis is biased in that it is constrained to high-risk human papillomavirus (HPV)–positive (hrHPV+) cases only. Although this is made clear by the authors, readers unfamiliar with the consequences of this bias may misinterpret the performance indicators, especially the reported specificity and negative predictive value (NPV). This study compares two binary test combinations: 1) hrHPV plus cytology vs 2) hrHPV plus the dual stain (DS). “Binary tests” means that thresholds are applied to each test to limit the test results to only either positive or negative. Both test combinations are applied as the logical “and” of positive test results, meaning that the combination test result is positive if and only if both component test results are positive. It can be shown that for the “and” test combination, the combined test specificity must be at least as high as that of the most specific component test and could be 100% while the combined test sensitivity will be no more than that of the least sensitive component test and could be 0% (2). The other nontrivial combination of binary tests is the logical “or” of positive test results, which has opposite effects on sensitivity and specificity (2). Two other interesting properties of combining binary tests are: 1) the order in which the individual tests are performed does not matter (they are commutative); and 2) for either “and” or “or” combination, it is not necessary to perform both tests on all persons. For the “and” combination, if the first test result is negative then the combined test result will be negative regardless of the second test result, which is the case for this paper. While the positive predictive value (PPV) reported by Wentzensen et al. is the net performance indicator of the combined tests, the sensitivity, specificity, and NPV are not because the cases negative for hrHPV are not included in the analysis. To illustrate the effect of this, assume that the hrHPV positivity rate was 15% and hrHPV sensitivity was 90%. Then, while the 1509 subjects of this study were hrHPV+, about another 8500 were hrHPV-, plus the component test sensitivities are multiplicative. Table 1 shows the effect for DS triage when all screen cases are included in the analysis, ignoring the effects of verification bias (which are very small for NPV and specificity [2]). Clinical performance indicators for primary hrHPV testing with dual stain triage based on hrHPV+ cases only versus based on all cases screened.*,† * Assumes 8500 hrHPV- cases. † hrHPV test sensitivity of 90%. Clinical performance indicators for primary hrHPV testing with dual stain triage based on hrHPV+ cases only versus based on all cases screened.*,† * Assumes 8500 hrHPV- cases. † hrHPV test sensitivity of 90%. A common clinical interpretation of NPV (2) is via its complement (1-NPV), which is the probability that a woman told she is negative for significant disease actually is not. For CIN2+, the current analysis applied to 10 000 hrHPV+ women who are told they are negative, 356 would not be; whereas for all women screened only 31 of 10 000 told they are negative would not be. Both are correct, but it seems possible that some readers might interpret the reported NPV in the usual way, as being applied to all women screened, especially because the PPV can be correctly interpreted in this way.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.