Background. Cervical cancer is caused by persistent infection with human papillomavirus (HPV). Most infections and associated lesions clear spontaneously. It is important to define the determinants and timing of clearance, so that viral persistence can be recognized and managed.
Methods. We investigated HPV natural history among 4504 subjects from ALTS (Atypical Squamous Cells of Undetermined Significance/Low-Grade Squamous Intraepithelial Lesions Triage Study). A discrete-time Markov model was used to simultaneously describe the prevalence, incidence, and persistence of type-specific HPV infection over 24 months in women with equivocal or mildly abnormal cytological results. Interactions between multiple HPV types infecting the same woman were examined for incidence of new infection (after an HPV-16 infection) and persistence of a current infection within groups defined by phylogenetic relatedness or by carcinogenicity.
Results. Ninety-one percent (95% credible interval [CI], 90%–92%) of prevalent HPV infections at enrollment cleared within 24 months. The probability that an infection would persist for a further 6 months increased with the duration of infection, from 37% (95% CI, 35%–39%) for a newly observed infection to 65% (95% CI, 61%–70%) for an infection that had already persisted for ⩾ 18 months. No consistent evidence of interactions was found between multiple HPV types regarding the incidence of new infection after an HPV-16 infection or regarding persistence of current HPV infection.
Conclusion. Although virtually all HPV infections clear within 2 years, the remaining infections have a high potential for persistence and, by implication, progression to precancer and cancer. Once biological and behavioral determinants are controlled for, HPV infections with different types seem to be independent of each other.
Cervical cancer arises via the following steps: infection of the cervical epithelium with 1 or more of the ∼ 15 carcinogenic human papillomavirus (HPV) genotypes, persistence of the infection, neoplastic progression to cervical precancer, and invasive cancer . Women may have multiple concurrent and sequential infections with different types, and most infections clear within 1–2 years . Cancer is a relatively uncommon and late outcome of persistent infection with a carcinogenic type .
The reasons why most HPV infections clear whereas some persist with an elevated risk of precancer are unknown. Large international case series can be used to estimate the percentage of cancers due to each type [4, 5]. However, prospective studies are required to study the early natural history of HPV infection. The power of such studies may be increased by targeting women at high risk of HPV infection. Approximately half of women with equivocal cytological results (called “atypical squamous cells of undetermined significance” [ASCUS] in the 1991 Bethesda System of nomenclature) are carcinogenic HPV DNA positive . Most women with a low-grade squamous intra-epithelial lesion (LSIL) are carcinogenic HPV positive, as LSIL represent cytologic evidence of HPV infection . Enrollment of women with ASCUS (∼ 5% of screening Pap tests) and LSIL (2%–3%) effectively includes one-quarter of all women with concurrently detectable HPV DNA, although the exact proportion depends on thresholds for cytological abnormality [8, 9]. HPV-infected women with ASCUS or LSIL are likely to have higher viral loads, more multiple HPV types (∼ 30%–40% of women), and greater probability of underlying precancer than HPV-infected women with normal cytological results and very low viral load [8, 10].
With recognition of these advantages and limitations, we conducted a natural history study of individual HPV types as part of the NCI-sponsored ALTS (ASCUS/LSIL Triage Study). ALTS was a randomized trial designed to compare different strategies for the initial management of ASCUS and LSIL [6, 7]. The aim of the present article was to study the prevalence, incidence, and persistence of HPV infection and the impact of selected factors (age, region, cytological results, and number of sex partners) on these indicators among women referred for ASCUS or LSIL and followed-up for 2 years. We also studied the interaction between HPV types regarding the occurrence of a new HPV type after an HPV-16 infection and regarding the duration of a current infection. A companion article concentrates on the risk of diagnosis of precancer, found as expected to be linked to HPV persistence .
Population. The design of ALTS has been described in detail elsewhere [6, 7]. Briefly, between January 1997 and November 1998, 4 clinical centers across the United States enrolled 5060 women referred because of a community, conventional Pap smear interpretation of ASCUS (3488 women) or LSIL (1572 women) in the previous 6 months. After giving informed consent, women were randomized to 1 of 3 trial arms: immediate colposcopy, HPV triage, or conservative management. At the enrollment visit, a study nurse administered a standard questionnaire and conducted a pelvic examination including a cervical specimen collected using a broom device and placed in a liquid-based cytological medium (ThinPrep; Cytyc). An aliquot was used to test for a pool of carcinogenic HPV types using Hybrid Capture 2 (HC2; Digene); the results were blinded except for the enrollment phase of the HPV triage arm and for all women at exit colposcopy (see below). A separate Dacron swab specimen was placed into Digene Specimen Transport Medium for the investigational HPV typing (Roche Molecular Systems) used in this analysis.
Referral to colposcopy after enrollment was determined by trial arm. Women in the immediate colposcopy arm were referred on the same day if practical or within 3 weeks, regardless of the enrollment test results. Women in the HPV triage arm were referred to colposcopy if the enrollment HPV result by HC2 was positive or missing or if the enrollment cytological result was high-grade squamous intraepithelial lesion (HSIL). Women in the conservative management arm were referred only if the cytological diagnosis was HSIL.
Follow-up visits were scheduled every 6 months for 24 months. At each follow-up visit, specimens were collected as at enrollment, and management was uniform; in all arms, only cytologic HSIL triggered referral to colposcopy. All women were scheduled for colposcopy as part of their exit visit at 24 months.
Women with a histological diagnosis of cervical intraepithelial neoplasia grade 2 (CIN2) or CIN3 were treated by a loop electrosurgical excision procedure (LEEP). At the exit visit, women with persistent low-grade lesions, including HC2-positive ASCUS, were also offered a LEEP.
The current analysis is focused on HPV persistence as an early event in natural history and excludes all women diagnosed with CIN3 or cancer at any point during follow-up. Because current diagnostic methods are not sufficiently sensitive to rule out the presence of small CIN3 lesions at study entry, we conservatively considered all cases of CIN3 or cancer that were detected during the 24-month trial to be prevalent at enrollment. Women with CIN2 were not excluded, because it is more transient on average than CIN3 and more difficult to distinguish from the cytopathic effect of recently acquired infections [6, 7].
HPV DNA testing. HPV DNA genotyping was performed using L1 consensus primer PGMY09/11 polymerase chain reaction (PCR) amplification and reverse-line blot hybridization. At enrollment, HPV testing was performed on 4915 women (97.1% of the 5060 women recruited into ALTS) to detect 27 HPV types (6, 11, 16, 18, 26, 31, 33, 35, 39, 40, 42, 45, 51–59, 66, 68, 73 [PAP238A], 82 subtype [W13b], 83 [PAP291], and 84 [PAP155]). For 2833 patients, HPV testing included detection of 11 additional HPV types (61, 62, 64, 67, 69–72, 81, 82 subtype [IS39], and 89 [CP6108]). Follow-up specimens were tested for all 38 types.
Statistical methods. Type-specific HPV DNA test results at enrollment, during follow-up, and at exit were analyzed in a unified discrete-time Markov model, which considers the prevalence, incidence, and persistence of HPV infection as part of a single ongoing process. This process is illustrated in figure 1 for a single HPV type in a single subject. The circles show HPV infection states that may be observed during a scheduled ALTS visit, and the arrows denote possible transitions between these states from one visit to the next. The states can be divided into 3 groups labeled “P” (prevalent infections), “I” (incident infections), and “U” (a single state representing an uninfected individual). At study enrollment, the status is either U (uninfected) or P1 (prevalent infection at baseline). At any point during follow-up, there are only 2 possible choices for a prevalent infection: either the HPV infection clears, moving the subject to state U, or it persists, moving the subject from state P1 to state P2 at the second visit, then to P3, to P4, and so on, at subsequent visits. Each increase in the subscript number denotes a longer lasting infection. An uninfected subject may remain uninfected (indicated by the circular arrow from state U to itself) or may experience an incident infection, moving the subject to state I1. Incident infections may persist (moving the subject from state I1 to state I2, then to I3, etc.) or clear in the same way as prevalent infections.
The model is completed by specifying the probability of being HPV DNA positive at the next visit. These probabilities are described using a collection of logistic regression models. A separate intercept term was used for each of the 9 states in figure 1 that represent a current HPV infection, allowing persistence to depend on duration of infection and on prevalent/ incident status. In principle, each covariate in the model—such as age, study center, or number of sex partners—could have a separate odds ratio (OR) for each of these 9 states, but, to simplify the model, a common OR is assumed for any given covariate within 3 submodels: (1) a prevalence model, which describes the probability of being infected on the first visit (P1); (2) an incidence model, which describes the probability of incident infection (I1); and (3) a persistence model, which describes the probability of an existing infection, whether prevalent or incident, continuing at the next visit (P2-P5 and I2-I4). In the basic model, different HPV types are assumed to be independent. However, HPV infections of different types may follow a similar course in the same woman. Correlations between different HPV types were accounted for by 2 different extensions to the basic model. Viral interactions were investigated by including the presence or absence of another HPV type at the current visit as a predictor in the logistic regression model for HPV positivity at the next visit, generating ORs for incidence or for persistence given the presence versus absence of another type. HPV interaction analyses were done in subgroups of types determined by phylogenetic relatedness (i.e., species) or by carcinogenicity. HPV types 16, 18, 31, 33, 35, 39, 45, 51, 52, 56, 58, 59, 66, and 68 were considered carcinogenic types . Unmeasured host or environmental factors may also create correlations in the incidence of different HPV types. Such factors were accounted for by including individual-level random effect or “frailty” terms in the model .
The model is a Bayesian full probability model, in which all unobserved quantities are random variables . In this framework, missing HPV data are imputed automatically, and the uncertainty in the imputation is propagated throughout the model. Markov chain Monte Carlo simulation was used to estimate the posterior distribution of all the parameters. Computations were carried out on the high-performance Helix Systems at the National Institutes of Health (available at: http://helix.nih.gov).
Of the 5060 women recruited into ALTS, 542 women with a histologic diagnosis of CIN3, adenocarcinoma in situ, or cancer were excluded. A further 14 were excluded because they had no type-specific HPV DNA data for any visit, leaving 4504 women.
Risk factors for prevalence, incidence, and persistence.Table 1 displays mutually adjusted ORs for the basic covariates in the 3 submodels: prevalence, incidence, and persistence. Both the prevalence and incidence of HPV decreased with age, whereas persistence increased. For women ⩾ 50 years of age, compared with women <20 years of age, the OR for prevalent infection was 0.42 (95% credible interval [CI], 0.33–0.54) and the OR for a newly incident infection was 0.39 (95% CI, 0.31–0.48), whereas, for an existing HPV infection, the OR for persistence to the next visit was 1.47 (95% CI, 1.11–1.94). Comparing women referred with a community cytologic interpretation of LSIL to those with ASCUS, no difference was observed in persistence (OR, 0.98 [95% CI, 0.92–1.05]). The subtle differences by study arm reflect less censoring of women with CIN2 in the conservative management arm. Women with multiple sex partners were at higher risk of prevalent infection (OR, 1.38 [95% CI, 1.31–1.46]) and incident infection (OR, 1.84 [95% CI, 1.74–1.95]) than those with 1 partner. Women reporting no male sex partners in the previous year were at lower risk of prevalent HPV infection than those with a single partner (OR, 0.77 [95% CI, 0.63–0.93]). Incidence was also slightly lower among women reporting no recent sex partners, compared with women with a single, steady partner (OR, 0.90 [95% CI, 0.82–1.00]). Persistence was not affected by sexual behavior.
Duration of infection.Figure 2 shows the population average clearance of prevalent HPV infections by length of follow-up, for the 10 most prevalent HPV types. Averaging over all types, 36% (95% CI, 34%–37%) of prevalent infections persisted to 6 months, 20% (95% CI, 18%–21%) persisted to 12 months, 13% (95% CI, 11%–14%) to 18 months, and 9% (95% CI, 8%–10%) to the end of follow-up at 24 months.
Figure 3 gives an alternative presentation of the persistence data. It shows the probability of persistence to the next visit, given that a prevalent infection has already been observed to persist for 0, 6, 12, and 18 months. The results are standardized to the baseline values of all covariates listed in table 1. The probability that a prevalent infection would persist to the next visit increased from 37% (95% CI, 35%–39%) to 52% (95% CI, 49%–55%) after 12 months, to 59% (95% CI, 55%–63%) after 18 months, and to 65% (95% CI, 61%–70%) after 24 months. Very similar results were found for incident infections. The probability that a newly observed incident infection would persist to the next visit was 37% (95% CI, 35%–39%), increasing to 52% (95% CI, 48%–55%) after 12 months and to 60% (95% CI, 55%–66%) after 18 months (data not shown).
HPV interactions for persistent infection.Table 2 shows ORs for persistence of HPV types in species α-9 (HPV types 16, 31, 33, 35, 52, 58, and 67) given concurrent presence versus absence of any other α-9 type. All the results are consistent with an OR of 1. The pooled OR, assuming a common effect for all HPV types, is 0.89 (95% CI, 0.78–1.01).
Table 2 also shows results for the same interaction model applied to HPV types in species α-7 (HPV types 18, 39, 45, 59, 68, 70, and 85). HPV-85 was removed from the analysis because of small numbers (n = 15). HPV-59 was marginally more likely to persist when there was coinfection with another HPV type in species α-7 (OR, 1.46 [95% CI, 0.98–2.21]), but the pooled OR was consistent with no interaction (OR, 1.12 [95% CI, 0.94–1.34]).
The same interaction model was also applied to carcinogenic HPV types (table 2). HPV-66 was possibly more likely to persist in the presence of other carcinogenic types (OR, 1.45 [95% CI, 0.99–2.20]). The pooled OR was borderline elevated (1.09 [95% CI, 1.00–1.19]), a result driven by the elevated persistence of HPV-66.
HPV interactions for incident infection. A second interaction model was used to investigate the effect of current infection with HPV-16 on the subsequent incidence of other types. A current infection with HPV-16 was used as a predictor of incident infection, at the next visit, with other types in the 3 groups previously studied (species α-9, species α-7, and carcinogenic HC2 types).
Table 3 shows the effect of a current HPV-16 infection on the risk of future infections for other types in species α-9. Current infection with HPV-16 was a risk factor for future infection with HPV-31 (OR, 1.95 [95% CI, 1.32–2.89]), and there was a borderline increased risk of HPV-67 (OR, 1.65 [95% CI, 0.98–2.80]). These associations may not be causal, however. A current infection with HPV-16 may be a marker of high risk of infection with any type due to unobserved behavioral, environmental, or immunological factors. To account for this, a frailty term was also included in the incidence model, and the results are also shown in table 3. Adding the frailty terms attenuated associations between incidence of HPV-16 and other types in α-9. In particular, the OR for future infection with HPV-31 and HPV-67 were reduced to 1.38 and 1.16, respectively (the CIs were consistent with no association).
The same model was applied to types in species α-7 (table 3). A negative association with HPV-68 was made stronger by the inclusion of frailty terms so that a woman with HPV-16 infection was at substantially lower risk of future HPV-68 infection (OR, 0.37 [95% CI, 0.18–0.77]). Similarly, HPV-59 showed a borderline decreased risk (OR, 0.55 [95% CI, 0.32–0.96]). No associations were observed for HPV-51 and HPV56 (the remaining types in the HC2 high-risk probe set) or for HPV-66 (data not shown).
The SD of the frailty term quantifies the residual variation in risk of incident HPV infection, after accounting for the risk factors listed in table 1. In a model including all carcinogenic types, the SD of the frailties was 0.91 (95% CI, 0.85–0.97), implying a 20-fold variation in risk between top and bottom 10th percentiles among women with the same age, center, referral diagnosis, and sexual behavior.
For current vaccine and diagnostics development efforts that will soon affect millions of women [15, 16], it is important to understand the natural history of individual HPV genotypes. Using a unified statistical model, we were able to describe the prevalence, incidence, and clearance of type-specific HPV infection among women with ASCUS/LSIL. The behavior of prevalent and incident HPV infections was very similar. They showed the same risk factors with very similar ORs and had the same pattern of persistence with time. This similarity may be largely due to the short duration of HPV infection. If the vast majority of HPV infections clear within 2 years, then most of the infections observed at enrollment in a group of mainly young women may be newly incident.
An important finding of this study is that infections are more likely to persist the longer they have been observed. Although virtually all infections are cleared within the first 2 years, the subset of infections that remain have high persistence potential. It now appears that the previous, widely held conception of the natural history of HPV-associated minor lesions (that one-third regress, one-third persist, and one-third progress) was an artefact of (1) failing to appreciate the large fraction of infections that are highly transient and (2) mistaking sequential, rapidly clearing infections for persistence. It might be that the long-term risk posed by HPV infection is established within a few years. This would provide a clinical motivation to treat persistent HPV infections once clearance is unlikely and the risk of CIN3 diagnosis is sufficiently high, particularly given the suboptimal sensitivity of a single colposcopic biopsy to rule out CIN3 [17, 18]. However, the choice of when to treat will require a weighing of concerns over safety and overtreatment, particularly among young women [19, 20]. We do not attempt a formal risk-benefit analysis here but make the following observations. The ALTS data are most applicable to women referred to colposcopy for LSIL or HPV DNA–positive ASCUS and followed by repeated HPV testing. Our data suggest that after 12–18 months of persistence, we can expect the infection to continue for at least another 6 months, even among young women who have the highest capacity for HPV clearance.
Any comparison of our findings on persistence with previous studies must be informed by the fact that subjects in this study had a community diagnosis of ASCUS or LSIL before recruitment, whereas most previous studies predominantly included cytologically normal women. We tried to mitigate any effect of clinically advanced precancerous lesions on persistence behavior by removing women diagnosed with CIN3 during follow up. We also conducted a sensitivity analysis in which women with CIN3 were included but were censored as persistent with whatever HPV types were present at the time of CIN3 diagnosis. In this sensitivity analysis, HPV-16 emerged clearly as the most persistent type, in agreement with previous findings . A tentative but possibly important conclusion from this sensitivity analysis is that the greater persistence of HPV-16 may be due to an increased ability to induce CIN3, compared with other types.
Previous studies of HPV infections have noted that infection with multiple HPV types occurs more often than would be expected if they were completely independent [22, 23, 24, 25]. A novel feature of our analysis is an attempt to model the correlation between different HPV types in terms of individual frailties and thus separate type-specific virus effects from host effects. The large SD of the frailty term implies substantial unexplained variation in the risk of incident HPV infection among women with the same observable risk factors. The frailty model cannot distinguish between individual immunological susceptibility and unmeasured environmental factors that vary between women, such as the sexual behavior of a woman' partner. The causes of this variation therefore remain unknown.
We also extended the model to incorporate interactions between different HPV types. We concentrated on the 13 most clinically important high-risk types defined by the HC2 high-risk probe set (plus HPV-66, a carcinogenic type that is detected well by the assay) and types closely related to HPV-16 and HPV-18, the most common types in cervical cancer worldwide. The theoretical possibility of biological interaction between carcinogenic HPV types is of interest because current HPV vaccines offer mainly type-specific protection. Any biological interaction between HPV types in the same women could either decrease or increase the efficacy of such a vaccine. Our definition of interaction in this short-term natural history study is limited to measurements of current infection. The examination of any long-term effect of infection with one type on incidence or persistence of another type would require either long-term follow-up or a complete history of prior HPV infection. With these limitations, the results from this large and statistically powerful study suggest that HPV infections are fundamentally independent of each other, with minimal interactions at the viral-viral level regarding persistence/clearance or acquisition. In particular, there is no evidence of interaction between different carcinogenic HPV types, with the exception of HPV-68, for which risk of future infection was lower for women with a current HPV-16 infection. The absence of viral interactions should help to alleviate concerns that the removal of common carcinogenic types from the population, particularly HPV-16 and HPV-18, by an effective vaccination program will eventually lead to their replacement by other, currently less important carcinogenic types [23, 24, 25].
ALTS GROUP MEMBERS
National Cancer Institute, Bethesda, MD. D. Solomon, project officer; M. Schiffman, co-project officer.
University of Alabama at Birmingham. E. E. Partridge, principal investigator; L. Kilgore, co-principal investigator; S. Hester, study manager.
University of Oklahoma, Oklahoma City. J. L. Walker, principal investigator; G. A. Johnson, co-principal investigator; A. Yadack, study manager.
Magee–Womens Hospital of the University of Pittsburgh Medical Center Health System, Pittsburgh, PA. R. S. Guido, principal investigator; K. McIntyre-Seltman, co-principal investigator; R. P. Edwards, investigator; J. Gruss, study manager.
University of Washington, Seattle. N. B. Kiviat, co-principal investigator; L. Koutsky, co-principal investigator; C. Mao, investigator.
Colposcopy Quality Control Group. D. Ferris (principal investigator), Medical College of Georgia, Augusta; J. T. Cox (coinvestigator), University of California, Santa Barbara; L. Burke (coinvestigator), Beth Israel Deaconess Medical Center Hospital, Boston, MA.
HPV Quality Control Group. C. M. Wheeler (principal investigator) and C. Peyton-Goodall (lab manager), University of New Mexico Health Sciences Center, Albuquerque; M. M. Manos (coinvestigator), Kaiser Permanente, Oakland, CA.
HPV Testing Advisor. A. Lorincz (senior scientific officer), Digene Corporation, Gaithersburg, MD.
Pathology Quality Control Group. R. J. Kurman (principal investigator), D. L. Rosenthal (coinvestigator), and M. E. Sherman (coinvestigator), Johns Hopkins Hospital, Baltimore, MD; M. H. Stoler (coinvestigator), University of Virginia Health Science Center, Charlottesville.
Quality of Life Group. D. M. Harper (investigator), Dartmouth Hitchcock Medical Center, Lebanon, NH.
Coordinating Unit and Data Analysis. J. Rosenthal (project director), M. Dunn (data management team leader), J. Quarantillo (senior systems analyst), D. Robinson (clinical center coordinator), Westat, Rockville, MD; B. Kramer (senior programmer/analyst), Information Management Services, Inc., Silver Spring, MD.
We thank Dr. S. Franceschi for her valuable discussions of the manuscript.