The National Cancer Institute (NCI) is cooperating with Glaxo Smith Kline in a randomized trial of prophylactic human papillomavirus (HPV) vaccination in Costa Rica. The company is providing the vaccines and funding for the regulatory aspects of the trial; however, data management, data analysis, and publications are entirely controlled by the NCI. The NCI collaborates with a large number of companies working on cervical cancer screening, including cytology, HPV testing, and biomarker development. All collaboration with industry colleagues is overseen by the NCI Ethics Office and is unpaid; the author has no relevant financial holdings or personal conflicts of interest.
These views are personal and do not reflect the official opinions of the NCI.
There is justifiable excitement regarding the recent introduction of prophylactic vaccines against human papillomavirus (HPV) types 16 (HPV-16) and HPV-18. Preventing these infections theoretically could avert approximately 70% of cervical cancer cases worldwide. In the U.S., numerous influential advocates are calling for universal vaccination of adolescent females. For example, mandatory vaccination of preadolescent girls linked to school attendance soon may become law in several states and the District of Columbia. Given the promise of the vaccines, perhaps it is inevitable that vaccine introduction is proceeding before full consideration of how universal vaccination would affect existing, successful cervical cancer prevention programs.1
Determining the impact and cost effectiveness of the vaccines unavoidably will require the passage of time. Nevertheless, it is worth describing in broad terms for the readers of Cancer Cytopathology how successful, broad HPV vaccination of adolescent girls may affect cytology and HPV testing.
HPV and cervical cancer
Predicting the impact of vaccination on cytology and HPV testing requires a basic, up-to-date understanding of HPV natural history and cervical cancer development. Interested readers are referred to a recent, excellent update.2 Only a few references are given here.
Cervical cancer arises in a few steps, as shown in Figure 1. Previously, the prevailing conception of cervical cancer and its precursors followed a histologic model that was formulated by Richart and Barron3 in which it was believed that cancer arose by a stepwise increase in severity from mild (grade 1 cervical intraepithelial neoplasia [CIN1]), to moderate (CIN2), to severe (CIN3) intraepithelial neoplasia, and to cancer. Histologic classifications still are useful for clinical management; but, for scientific purposes, it is more accurate to think of broad steps that include sexual transmission of 1 or more of the 15 to 20 carcinogenic, anogenital types of HPV infections; persistence rather than the more common alternative of clearance; progression of infected cervical cells to precancer; and invasion. Histologic CIN3 provides a good approximation of precancer, whereas CIN2 is more equivocal, heterogeneous, and prone to regression. This model is useful, because it matches epidemiologic and laboratory data and does not require morphologic distinctions that are unreliable like cytologic atypical squamous cells of undetermined significance (ASCUS) or histologic CIN1.4
Approximately 15 to 20 types of HPV can cause cervical cancer. Acute infections with these types and their cytologic/colposcopic signs are extremely common,5 particularly at younger ages. However, cervical cancers are relatively rare and typically occur many years later.6 The typical age of cervical HPV infection is similar to the age of other sexually transmitted infections, with a large peak after sexual initiation.7 Greater than 90% of HPV infections and accompanying cytopathic signs clear within 2 years, with frequent acquisition of other types that clear in turn.8 It is persistence of a carcinogenic type of infection that poses a major risk of CIN3 diagnosis within 5 to 10 years and a risk of invasion in the subsequent decades. The time of diagnosis is not the same as the time of first development. Increasing evidence suggests that the time between HPV infection and invasive cancer is mainly a function of slow, horizontal expansion of initially miniscule CIN3 lesions that can develop within a few years of HPV persistence.
The 15 to 20 carcinogenic HPV types convey greatly different risks.9 Frequently transmitted during sexual intercourse, the different HPV types clear, persist, and cause precancer independent of each other. In other words, each type runs its race without interference or help from the others.8 Clearance of an HPV type leaves the woman partly or fully immune to that type, at least to the extent that we have been able to observe to date in ongoing prospective cohort studies.
HPV-16 predicts the greatest risk, whatever the cytologic appearance.10 Conversely, controlling for HPV type, the absolute risk for subsequent CIN3 of HPV-positive ASCUS is equivalent to that for low-grade squamous intraepithelial lesion (LSIL).11 It is worth noting that even noncarcinogenic HPV infections can cause microscopic and/or macroscopic abnormalities consistent with CIN1 or even CIN2 without implying a high risk of cancer. For unknown reasons, only approximately one-third of infections defined by DNA tests have concurrent cytologic abnormalities of at least ASCUS.12 There is a strong cross-sectional association between the size of lesions and viral load13, 14; however, viral load of most HPV types does not have strong, prospective predictive value for subsequent CIN3 diagnosis (HPV-16 is the major exception).15, 16
In summary, it is very important to keep in mind that HPV infections are common and variable. Acute infections with their inconsistently accompanying, equivocal, and minor microscopic abnormalities greatly outnumber significant lesions particularly at young ages (Fig. 2).17 Optimally, we would hope that screening and management of screening-detected abnormalities would permit recognition and treatment of precancers without over-treatment of women whose findings would resolve without treatment. This balancing act will be made more difficult after the widespread use of vaccines, as demonstrated below.
Figure 344 shows why vaccine makers and DNA diagnostics companies all have focused initially on HPV-16 and HPV-18. Persistent HPV-16 infection is one of the most carcinogenic agents to which humans commonly are exposed. HPV-18 is associated with a particularly strong risk of glandular lesions and occult cancers. The fractions of invasive cancer caused by specific HPV types (reflecting their carcinogenic potency) permit a prioritization of types by relative burden of disease. The major vaccine makers already plan to extend their current virus-like particle vaccines to approximately 6 carcinogenic types that, together, cause >90% of cervical cancer worldwide,18 but we do not know when these extended vaccine formulations will be available.
Structure of current cervical cancer prevention programs
Current cervical cancer prevention programs in the U.S. are designed around the natural history schema described above and include the familiar components listed in Table 1.19 In effect, these programs rely mainly on the microscopic (cytologic and histologic) and visual (colposcopic) correlates of HPV infection to predict the cervical cancer risk of women with different screening or diagnostic results. Increasingly, HPV testing is being used with the other techniques to provide better risk prediction and the possibility of fewer cycles of screening.20 However, especially among young women, a single positive HPV test has low specificity and poor positive predictive value. Expanded use of HPV testing will require reliable and accurate evaluation of type-specific HPV persistence, the true predictor of cancer risk, which is beyond the scope of this article. (U.S. Food and Administration [FDA]-approved, type-specific HPV tests will be introduced to the clinical setting over the next few years.) It is important to note that HPV testing using the pooled probe test that currently is approved by the FDA is valuable mainly when the test is negative. Because it has high sensitivity for concurrent and incipient precancer and cancer, as a corollary, the test provides excellent reassurance when none of the carcinogenic types of HPV included as a group in the test are found.
Table 1. Parts of a Cervical Cancer Screening Program in the U.S.*
This process is repeated frequently to detect precancerous lesions before they become invasive. The objective is to provide as much reassurance against cancer development as is possible without excessive intervention among women who are not at risk.
1. Screening (normal women)
2. Triage of equivocal screening results with another independent test
3. Histologic diagnosis of abnormal screening results
4. Postcolposcopy follow-up if no lesion is found to treat
5. Posttreatment confirmation of cure
The key concept of risk prediction
When discussing risk prediction, as we do below, it is critical to clarify the meaning of positive predictive value and negative predictive value of prevention techniques and strategies. Cervical cancer prevention specialists typically study the risk of histologic CIN3 as the best surrogate of invasive cancer. Alternatively, they study the risk of CIN2/CIN3, in line with typical treatment practices. Surrogate endpoints are needed, because no one willingly would permit a woman to develop invasive cancer under observation. Figure 4 shows that the positive predictive value of any abnormal result (Papanicolaou test, HPV test, colposcopic impression, biopsy diagnosis, or any combination) is the risk of cervical cancer (eg, the percentage risk for CIN3) among women with that result. The negative predictive value is the reassurance that a woman with a normal test result is not at risk of cancer, at least until the next expected visit whenever that may be. To be satisfactory, negative predictive must be very high; a negative predictive value from >98% to 99% against CIN3 (and a much higher reassurance against interval cancer) until the next routine visit would be typical for an acceptable screening method.
Predicting the Impact of Reducing HPV-16 and HPV-18 on Screening and Management
With the preceding background and the following simplifying assumptions, we can predict the impact on cytology and HPV testing of a successful vaccination program that prevents HPV-16 and HPV-18 infections. The discussion is based on the current American Cancer Society cytology and HPV screening guidelines,21 the American Society of Colposcopy and Cervical Pathology guidelines on management of cytologic and histologic abnormalities,22, 23 and the documented performance of available cervical cytology and HPV-DNA tests.24 For the analysis, it was assumed that combined vaccine uptake of the approved Merck Gardasil quadrivalent vaccine25, 26 (and the analogous, pending GlaxoSmithKline bivalent vaccine27 against HPV-16 and HPV-18) will be rapid and nearly complete among young adolescent females. It is plausible that this may happen given the enormous marketing efforts that are occurring and will continue and the great, positive interest of advisory groups28 and lawmakers in many jurisdictions. It is worth debating elsewhere whether immediate, universal coverage is a greater public health priority in the U.S. than other needs that are competing for the same resources; however, for the purposes of this discussion, it is projected that the current momentum will continue. Slower uptake, which also is possible given substantial controversy over the role of mandatory vaccination,29, 30 would delay the effects of vaccination on screening programs but ultimately would not change them.
The following characteristics of the vaccines are assumed. The Merck Gardasil vaccine protects against HPV-6 and HPV-11, which cause genital warts, as well as against HPV-16 and HPV-18, but the first 2 types are not important for the prevention of cervical cancer. Partial cross-protection against a few HPV-16–related or HPV-18–related HPV types has been postulated for the Glaxo Smith Kline vaccine because of its adjuvant; however, the cross-protection, if confirmed, would not change the conclusions of this simplified analysis. Although it is the most uncertain assumption, let us guess that the durability of virtually perfect vaccine prophylaxis against HPV-16 and HPV-18 infections will last for decades or that boosting will be highly effective, permitting lifelong protection by vaccination. Shorter durability of vaccine protection would lessen the impact of vaccination on screening programs.
Vaccination would reduce the predictive values of cytology
We recently completed an extensive examination of HPV type (determined by polymerase chain reaction analysis), age, and cytologic abnormality within a random sample of the entire adult female population of Guanacaste, Costa Rica,12 which is a province at historically high risk of cervical cancer. To summarize the core of the many complex interrelations, the major carcinogenic species all demonstrated cytologic abnormalities in approximately one-third of the carcinogenic infections that were detected at the same time by HPV-DNA testing. However, by far, HPV-16 caused the most obviously severe cytologic abnormalities (Fig. 5). Other carcinogenic types were equally likely to cause ASCUS and LSIL but were much less likely to cause high-grade squamous intraepithelial lesion (HSIL). Observed from a different angle, a large fraction of HSIL (and cancer, which is always rare to begin with) was caused by HPV-16. Therefore, removing HPV-16 from the viral population by vaccination would leave many ASCUS and LSIL diagnoses but a depleted number of HSIL diagnoses. The positive predictive value of an abnormal cytology for predicting CIN3 and cancer would decrease (Fig. 6), because the already small proportion of abnormal tests that are diagnosed as HSIL or cancer would decrease further. This would be both a quantitative effect and a qualitative effect; there would be fewer definite abnormal slides, and the remaining abnormalities would be more equivocal on average.
To make a more complicated but equally important point, the marginal (added) reassurance (negative predictive value) of cytology would be altered as follows: Even without any kind of cervical test, only a small percentage of women in the general population have or are about to get CIN3 or cancer. Thus, the intrinsic prior probability that they are harboring incipient CIN3/cancer is <5% (ie, the negative predictive value is >95%). Currently, routine cytology can reduce that baseline risk safety level to approximately 1% or 2% by detecting approximately 66% of the prevalent underlying CIN3/cancer. However, if HPV-16 and HPV-18 were removed from the population, the baseline prevalence of CIN3/cancer would drop substantially. In effect, vaccination competes with cytology, doing a large part of the same job of raising safety and reducing the most lethal HPV types, including HPV-16, which also is the easiest to detect cytologically. In short, vaccination effectively would eliminate some of the underlying value of cervical cytology programs.
Reduction in the predictive values of colposcopic impression
Successful vaccination would affect colposcopy as strongly if not more than it would affect cytology. The visual appearance of cervical HPV infections, as evaluated by colposcopy or derivative techniques, is even more highly variable and difficult to classify than cytopathic effects. In fact, the lack of reproducibility and accuracy of colposcopy represents a considerable clinical challenge that is insufficiently appreciated.31, 32 A group of National Institutes of Health and American Society for Colposcopy and Cervical Pathology researchers recently reached this conclusion partly based on a large intercolposcopist comparison of digitized cervical images (a main table from that comparison is reproduced here as Figure 7). We noted the following sobering observations: 1) The colposcopists agreed poorly on whether any lesion was present, whether or not they included metaplasia as a lesion; 2) they disagreed on the number of acetowhite lesions; and 3) none of them showed a strong association between the presence of HPV DNA of carcinogenic, noncarcinogenic, or all types and numbers of lesions or area of acetowhitening. Surprisingly, many women who were infected with multiple HPV types had no visually apparent abnormality, and vice versa. However, as shown in Figure 7, HPV-16 was associated with the highest probability of a clearly recognized lesion, including those lesions that led to a histologic diagnosis of precancer. Therefore, the removal of HPV-16 by vaccination would leave an even more difficult task for colposcopists approaching the already difficult task of targeting lesions for biopsy diagnosis.
Reduction in the predictive values of a pooled-probe HPV test
The effect of vaccination on testing protocols relying on HPV tests that pool all carcinogenic types would parallel the issues raised for cytology and colposcopy. The power of HPV testing derives in large part from detection of HPV-16 and HPV-18, as shown for screening in Figure 810 and for triage of ASCUS.33 The elimination of HPV-16 and HPV-18 would make cervical cancer and CIN3 much rarer. Although the sensitivity of HPV testing for the remaining cancers caused by other types would remain high, the absolute number of true positive results would drop by approximately 66%. Meanwhile, the prevalence of other carcinogenic infections that are less likely to cause cancer would not decrease by nearly two-thirds. Consequently, the number of HPV infections found that would predict cancer or CIN3 would diminish relative to the number of HPV-positive women destined not to get cervical cancer or CIN3. The positive predictive value of HPV testing using pooled tests would drop along with the cost-effectiveness of each HPV screen. The marginal additional reassurance afforded by a negative HPV test would diminish for the same reasons given for cytology.
Conjectures regarding how programs might be affected
It is impossible to predict exactly what the real-world reaction would be if the predictions described above come true, and the performance of our current prevention tools is reduced noticeably after vaccination. The best blend of vaccination and screening/management tools is not obvious, and no one could claim reasonably that they have the answer without rigorous cost-effectiveness work still to come. The following preliminary projections seem rational based on current evidence, but they are conjectural and are offered only to promote more formal analyses.
First, vaccination of younger girls (ages 11–12 years) is likely to proceed more systematically than catch-up vaccination among older girls, implying that the older group will continue to require screening without modification for their lifetimes. If vaccine durability is long, then vaccination is much more effective and cost-effective when girls are vaccinated before the onset of sexual life. The prophylactic vaccines have no known therapeutic effect for existent infections. At a certain age, most women who will get infected already have been exposed, and those who are destined to develop CIN3 or cancer would not be affected. That age has not been established to date; however, in terms of population cost-effectiveness, it may turn out to be younger than the upper bound of FDA approval and the initial recommendation of the American Committee on Immunization Practices.34 Thus, catch-up vaccination may not be as effective practically as hoped, or it may not even be implemented systematically when younger girls are vaccinated. Prevention programs may choose to rely mainly on screening and treatment for the cohort of women that is already sexually experienced, explicitly pushing any changes judged necessary for the vaccine era further into the future.
If and when vaccination programs do gain wide coverage among younger adolescent girls, we will need to reconsider the age of first cervical screening. If the data are supportive, then we might want to raise the age, because the risk of early cancers is linked to HPV-16 and HPV-18. The other carcinogenic types are less threatening, and cost-effectiveness analyses support raising the age of first cervical screening to approximately 24 years.35
In addition to initiating screening later, we may want to stretch out screening intervals, if durability proves truly long-term, if it is demonstrated that boosters are cost-effective, and especially if HPV testing is added routinely to cervical cytology. I personally guess that the best screening fit in the U.S., granted successful vaccination coverage, initially would be a combination of a molecular test, such as type-specific HPV-DNA testing, and cytology (possibly computer-assisted) performed at approximately 3- to 5-year intervals. For those interested in projections of cytology volume, an excellent analysis36 projected that this kind of strategy may reduce cervical cytology volume by approximately 50%. I predict that, as confidence in HPV testing grows, it eventually will replace cytology as the first-line screening test in the U.S.
I project that HPV typing soon will be added to joint detection of the pool of carcinogenic types, because it will prove worth knowing whether HPV-16 or HPV-18 is present. These types, when they are found even once, predict much higher risks than other types.10, 33, 37 Also, testing separately for these types permits clear knowledge of whether they are clearing or persisting, which is important because persistent HPV-16 (and, to a lesser extent, HPV-18) predicts a substantial risk of CIN3 diagnosis within 5 years.38, 39 Typing for all 15 to 20 carcinogenic HPV types may be feasible technically but would raise the possibility of information overload for the clinician, especially because HPV infections often include multiple types with discordant times of acquisition, persistence, and clearance. Any widespread use of HPV typing will require robust, highly validated assays40 and clear clinical guidelines, neither of which are available currently available. I personally imagine that typing for at least HPV-16 and HPV-18 will emerge as a fully validated strategy within approximately 5 years.
These guesses are meant to apply to the U.S. Although it may be expensive, combining molecular tests and cytology can achieve very high sensitivity and negative predictive value for any CIN3 or early cancer caused by a type that is not included in the improving vaccines. Repeat testing after 6 to 12 months of screening abnormalities, when precancer is not found immediately, can offer specificity and good positive predictive value. However, considering regions of greater need and often lower resources outside the U.S., it is my own belief that we are moving worldwide from a morphologic prevention model based on cytology/colposcopy/histology to a model based on a virologic view of HPV and its molecular interaction with the human host.41, 42 New, inexpensive, and rapid HPV tests will permit the use of molecular screening, sometimes in a screen-and-treat strategy, in many areas where it is not currently feasible.
By removing the most evident and threatening cytologic, colposcopic, and HPV testing results from cervical cancer prevention programs, vaccination will leave behind more equivocal and less predictive abnormalities. All parts of the prevention program, from screening through posttreatment follow-up, will be affected. Overall, this is happy news with regard to further reducing suffering and death from cervical cancer. However, it would not be cost-effective to add HPV vaccination without adjusting the rest of the cervical cancer prevention program. It has been estimated that the current cervical cancer prevention effort in the U.S. costs approximately $6 billion per year. HPV vaccines are expensive and would cost billions of dollars to implement. At least for the public sector, health resources are limited and competitive.
Research and time will tell which prevention strategies are best for specific regions, tailored to their resources and societal priorities. The important task will be to make sure that all women have access to excellent strategies based on the latest knowledge and reliable prevention tools.