Modeling Cervical Cancer Screening Strategies With Varying Levels of Human Papillomavirus Vaccination

Key Points Question As human papillomavirus (HPV)–based cervical screening modalities change and HPV vaccination becomes more widespread, what are the likely outcomes for the interpretation of screening modalities? Findings In this decision analytical model with a simulated population of women aged 25 years and older, HPV-based screening modalities detected more abnormal cervical cells than traditional liquid-based cytology (LBC) approaches, but they did so at the cost of increased false positives. However, as HPV vaccination increased, HPV-based modalities resulted in fewer unnecessary colposcopies than LBC methods. Meaning These findings suggest that the ideal screening modality for a given population should account for HPV vaccination status to maximize the efficiency of screening.

As depicted in table 1, most of the parameters required for this work could be taken directly from the literature. While the Cochrane report 1 and the investigations it reports upon provide an estimate for the sensitivity and specificity of HPV testing for CIN2/3 cases, it does not intrinsically yield the sensitivity and specificity of HPV testing to HPV infection, which is a required parameter for this analysis. From literature [3][4][5] , we estimate the proportion of the population with a detectable HPV infection to be about 0.08. We further note that a recent analysis 2 estimated that half of CIN2/3 HPV-results would have been recorded as CIN2/3, HPV+ if a more sensitive test had been employed. Factoring this into Cochrane data 1 , this is equivalent to detecting an extra 1 of the 2 CIN2/3 cases missed. This yields a estimate for the proportion of CIN2/3 cases due to detectable HPV types as (19/20) or ν = 0.95, and yields a sensitivity estimate for detectable HPV with a 14 targeted type test as sn = 18/19 = 0.947 (1) and accordingly, a true detectable HPV prevalence rate in the population as h = 0.08 = 0.084.
(2) sn Finally, we can then estimate the specificity of the HPV test. The number of excess colposcopy, E, is given by This can be readily re-arranged and solved to obtain sp for HPV testing, as given in the main paper table 1.

False positives and false negatives with screening frequency
Under the assumption of independence of screening tests, we apply Poisson statistics and find the cumulative probability of at least one false positive after n tests is where sp is specificity of the test. An equally important consideration is the probability of missing disease in successive tests. For a woman with non-regressing CIN2/3, the probability it will be missed with each test is the complement of the sensitivity (sn) raised to the power of n, explicitly stated as As frequency is an important consideration in determining the lifetimes probability that one gets an incorrect screening result, we can model cumulative probability of a false positive result over a screening lifetime (assuming screening begins at 25 and ceases at 70) for a CIN2/3 negative woman and the probability of missing successive true, persistent CIN 2/3 for different modalities and intervals (1 year, 3 year, and 5 year). This is shown in figure 4.

Mathematical analysis of Triage testing
It is straightforward if rather tedious to establish formula for the outcome of triage tests. For N patients, we define sn H is the sensitivity of HPV testing to HPV infection and sp H the sensitivity. We further define sn L as LBC sensitivity to CIN2/3 and sp L as LBC CIN2/3 specificity. It can be shown that regardless of order of triage, the total of CIN2/3 missed, ω, and the excess colposcopy (false positives, ε) are the same for both approaches, given respectively by   In terms of excess colposcopy and positives missed, order (HPV primary with LBC reflex or LBC primary with HPV reflex) does not matter. But there is some variation with respect to the total number of tests performed. For HPV primary and LBC primary respectively, total tests performed are Ideally, we wish to select modality so that total tests is minimised. This is highly dependent on prevalence of CIN2/3, itself a function of HPV infection rate. In the case of HPV primary testing, we can state  and as the right-hand side of the identity is a positive constant, we can thus state that the number of tests required is a growing linear function of h, and define this constant C1 for brevity. We assume that CIN2/3 infection is a linear function of HPV infection rate, so that p = φh + λ . We can substitute this into tlp, and write and again, the right-hand side is a positive constant, designated C2. From the prior section, we know that in a typical population, about 1 in 1000 CIN2/3 cases is not due to detectable HPV strains, allowing us to get λ = 0.001. Thus we can infer as previously discussed that φ 0.225. It also follows that C1 >> C2, which means that the number of tests required grows more rapidly with HPV infection rate for HPV primary testing than primary LBC testing. One can show that below a threshold h infection rate ht, primary HPV testing will result in fewer tests required. Above this threshold, thp > tlp. This threshold is given by and for the parameters here, one can estimate ht = 7.60% (7.29 7.63% to 95% confidence interval). This is actually lower a threshold than the h estimated in the paper, but as can be seen from supplementary figure 1, the number of tests is relatively similar in both cases for typical h.

Mathematical analysis of Co-testing
Co-testing can be modelled as the process of primary testing with one modality (LBC/HPV), followed by a single reflex test on constants are derived previously, we can again show differences in test numbers vary little, as in the figure below.
negative results with the complementary modality (LBC for HPV primary, HPV for LBC primary). It can be readily shown thatregardless of order of testing, the number of cases detected (d), number of excess colposcopies / false positives (ε), and number of cases missed (ω) are given by The total number of tests, however, will differ very slightly, depending on test order. For LBC and HPV tests initially performed, the total number of tests are given by As with Triage testing, this will vary with prevalence and HPV infection rate. Using the same function (p = φh + λ ) and where po is initial prevalence, sn is test sensitivity and sp test specificity. The final consideration is the number of extra tests required if a retest mandate was initialised. For an initial population to be screened of No, we define a function g(n) = sp + p(n)(1 sp sn), we can find the number of retests r(n) at each iteration and the total number of tests after n iterations, t(n), respectively by n 1 The total extra precancers detected after n tests, φ (n) and total false positives after n tests, ε(n) are given respectively by Repeat HPV testing Repeat HPV tests can be quantified neatly with the following relationship. For n tests, the number of missed positives (ω(n)), false positives (excess colposcopy) (ε(n)) and total tests required per iteration, t(n), are given by pH pH H

Regional / Country comparison
We can also simulate the outcome of different modalities employed by existent regions and countries, in a similar manner to that shown in table 2 of the main paper. Three different regimes are outlined below, and exact transition probabilities were taken from the Markov chains outlined at the end of this document, with parameters as given in main paper and supplementary material.

LBC testing -Ontario, Canada
The current Ontario Cervical Screening Program (OCSP) screening recommendations outline an LBC testing regime, similar to that described in main paper figure 1. Specifically, any high grade abnormalities are recommended for immediate colposcopy in women greater than 25 years of age 6 . Likely outcomes are shown in table 2.
Co-testing -USA The USA: Practice in the USA varies markedly across the country, and co-testing regimes have in recent years been common in many jurisdictions. To see likely outcomes from recommended guidelines, we can look to the 2012 7 guidelines to see likely outcomes of screening with co-testing. Results of this analysis are shown in table 3. However, these results should be interpreted with the caveat that these algorithms are no longer recommended in the USA, whose screening protocols are now completely different from any other country.

Triage testing -Ireland
Ireland has a national screening programme, and has moved to HPV primary screening with LBC reflex. This triage schema recommends testing at 3 year intervals for women 25-30, and at 5 year intervals for those aged over 30 8 . Likely outcomes are shown in table 4.