A population-based observational study comparing Cervista and Hybrid Capture 2 methods: improved relative specificity of the Cervista assay by increasing its cut-off

Background High-risk human papillomavirus (HR HPV) testing has been shown to be a valuable tool in cervical cancer screening for the detection of cervical pre-cancer and cancer. Methods We report a purely observational study evaluating HR HPV prevalences in residual liquid-based cytology (LBC) samples using both the Cervista™ HPV HR Test and the Digene Hybrid Capture 2 High-Risk HPV DNA Test (HC2) in a sample of 1,741 women aged ≥30 years of a German routine screening population of 13,372 women. Test characteristics were calculated and a novel method for measuring test performances was applied by calculating ratios of sensitivity or specificity. Results The overall agreement of both tests for detection of HR HPV was excellent (κ = 0.8). Relative sensitivities for the detection of histologically confirmed severe cervical intraepithelial dysplasia (CIN3+) were similar for both HPV-tests, which was confirmed by the ratio analysis. However, discrepancy analysis between the Cervista HPV HR test and HC2 revealed a high false positive rate of the Cervista HPV HR test in the cytology normal category. Conclusions Performance of the Cervista HPV test in cervical specimens with abnormal cytology is comparable to HC2 as both tests were highly sensitive and specific for the detection of high grade cervical disease. We also demonstrate evidence that modification of the cut-off values drastically reduces the false positive rate in the cytology normal category without affecting the detection of CIN3+, which ultimately improved specificity of the Cervista HPV HR assay. Electronic supplementary material The online version of this article (doi:10.1186/s12879-014-0674-1) contains supplementary material, which is available to authorized users.

The purpose of the study is the comparison of the sensitivity and specificity criteria between Cervista HPV Assay (CER) and HR HPV HC2 Assay (HC2) in identifying women with abnormal cervical cellular histology. In order to perform a comparison of the tests, one must have a test that verifies the performance of the two tests in question, CER and HC2. In this case, the "gold standard" or verification method used is histology.
However, due to the method of data collection and the ethics of obtaining a histology sample for every patient, there exist histology results for only the patients who have tested positive in the liquid based cytology screening (LBC); a cytology result of greater than or equal to PapIII is considered positive. Furthermore, the likelihood of having a histology result for the patient depends on the Pap screening result, the higher the number for a positive test, the more likely there exists a histology test. Patients with relatively weak but positive LBC results were also unlikely to have a histology result. Thus, it is important to note that there exists very few gold standard data points for patients of negative cytology and varying degrees of incompleteness for patients with positive cytology. In general, patients with < Pap III+ cytology results had a large portion if not all of their Histology results missing.
In the following 3 tables, one can see that there was almost no information regarding false negative rates for CER (11 values out a possible of 1123), and 128 histology values out of a total of 484.
(1 − P revalence) * N total = T N + F P By substituting the equations into the first statement, we have the following: If we work with the proportions, thus percentage of patients with characteristic, "True Negative" for example, then we have the following equations: T rueN egative = AllN egative − F alseN egative = AllN egative − (P revalence − T rueP ositive) = AllN egative − P revalence + T rueP ositive = AllN egative + T rueP ositive − P revalence Now, using Cervista as an example, note that "All Negative" are all patients that tested negative by the test Cervista, and "True Positive" are all patients that tested positive by both Cervista and Histology.

Ratio of Specificity
Please see the Appendix for the methods concerning confidence interval calculation.

Conclusion
The calculations did not result in any statistically significant findings due to the amount of variance and unknown within the system. Thus, one cannot say that there exists a difference between Cervista relative to HC2. However, it should be noted here that calculations were made with only the true and false positives known. Especially, in our analysis, we assumed that the availability of the gold standard is not related to the occurrence of wrong test results.
From the methodological point of view, it is of interest that it was possible to calculate a measure of relative sensitivity and specificity even while the performance of the test for negative results is unverified. This was possible because both tests were evaluated in the same subjects, and thus the unknown prevalences were exactly equal for both tests.

Appendix
Confidence Intervals. It is reasonable to view the count data of TP or other categories are events from a Poisson distribution. Due to the large sample size, we can also assume normal approximation. Thus, if X ∼ P oisson(λ), then under large sample sizes with a small event rate, we can say X ∼ N ormal(λ, λ). Thus, applying the δ-method on the ratios, we can calculate the variance of the ratios: Cov(x, y) Using the above equations, we apply the δ-method to the sensitivity and specificity ratios, with the equations simplifying to: V ar(T P HC2 ) 3Ĉ ov(T P CER , T P HC2 ) doing so, we can convert the numbers to count data by renormalizing with the population size: