Predictors of human papillomavirus persistence among women with equivocal or mildly abnormal cytology


  • Delphine Maucort-Boulch,

    1. Hospices Civils de Lyon, Service de Biostatistique, Lyon, France
    2. Université de Lyon, Université Lyon I, Villeurbanne, France
    3. CNRS, UMR 5558, Laboratoire Biostatistique Santé, Pierre-Bénite, France
    Search for more papers by this author
  • Martyn Plummer,

    Corresponding author
    1. International Agency for Research on Cancer, Lyon, France
    • International Agency for Research on Cancer, 150 cours Albert Thomas, 69372 Lyon cedex 08, France
    Search for more papers by this author
  • Philip E. Castle,

    1. Hormonal and Reproductive Epidemiology Branch, Division of Cancer Epidemiology and Genetics, National Cancer Institute, NIH, DHHS, Rockville, MD
    Search for more papers by this author
  • Franklin Demuth,

    1. Information Management Services, Silver Spring, MD
    Search for more papers by this author
  • Mahboobeh Safaeian,

    1. Hormonal and Reproductive Epidemiology Branch, Division of Cancer Epidemiology and Genetics, National Cancer Institute, NIH, DHHS, Rockville, MD
    Search for more papers by this author
  • Cosette M. Wheeler,

    1. Departments of Molecular Genetics and Microbiology and Obstetrics and Gynecology, University of New Mexico Health Sciences Center, School of Medicine, Albuquerque, NM
    Search for more papers by this author
  • Mark Schiffman

    1. Hormonal and Reproductive Epidemiology Branch, Division of Cancer Epidemiology and Genetics, National Cancer Institute, NIH, DHHS, Rockville, MD
    Search for more papers by this author

  • Dr Cosette M. Wheeler has received support through her institution from Roche Molecular Systems for HPV genotyping studies and to conduct HPV vaccine studies for Merck and Co. Ltd. and GlaxoSmithKline. The other named authors do not have a commercial or other association that might pose a conflict of interest.


We investigated short-term persistence of human papillomavirus (HPV) infection among 2,408 women with low-grade or equivocal cytological abnormalities followed for 24 months. Odds ratios (ORs) for persistence to the next 6-month visit were estimated by a discrete time survival model. Prevalent HPV infections persisted longer in older women, but no association with age was found for incident HPV infections. Increased likelihood of persistence was found among current smokers of >20 cigarettes per day compared with smokers of ≤10 cigarettes per day (OR=1.43; 95% confidence interval [CI]: 1.02–2.01) and among current injectable contraceptive users (OR=1.15; 95% CI: 1.01–1.32). Persistence was more likely among infections with higher viral load (OR=2.05; 95% CI: 1.65–2.53) or with concurrent cytological abnormalities (OR=1.19; 95% CI: 1.03–1.39 and 1.29; 95% CI: 0.99–1.70 for ASCUS/LSIL and ASC-H/HSIL, respectively). We conclude that new HPV infections in older women are not riskier by the metric of viral persistence than those in younger women. Other risk factors such as oral contraceptive use and multiparity that have been associated with cervical cancer or cervical intraepithelial neoplasia grade 3 were not associated with short-term HPV persistence.

Human papillomaviruses (HPVs) are common sexually transmitted infections. Approximately 15 carcinogenic HPV types cause virtually all cases of cervical cancer and its immediate precursor, cervical intraepithelial neoplasia grade 3 (CIN3).1–4 However, most infections clear quickly and are no longer detectable by sensitive DNA tests two years after initial detection.5–9

Approximately one-third of women with carcinogenic HPV infections are found to have cytologic abnormalities, mainly equivocal (atypical squamous cells of undetermined significance, ASCUS) or minor (low-grade squamous intraepithelial lesions, LSIL) lesions.10 These abnormalities come and go along with the infections that cause them.

If a carcinogenic HPV infection fails to clear, the risk of CIN3 diagnosis increases rapidly, particularly for HPV16.4 Initially, CIN3 lesions are extremely small and easily missed by currently-available diagnostic methods based on colposcopic biopsy.11 Both CIN3 and persistent carcinogenic HPV infection are used as surrogate outcomes for cervical cancer risk in vaccine trials and screening programs. From an etiologic perspective, the identification of determinants of HPV persistence and CIN3 diagnosis is a research priority.

Risk factors for HPV viral persistence and/or CIN3 among HPV-infected women can be subdivided into host factors, environmental/behavioral factors, and viral factors. Age is an obvious host factor.12, 13 Among immunocompetent women, the environmental/behavioral risk factors studied include the established risk factors for cervical cancer (apart from lack of screening), e.g., smoking,14, 15 prolonged hormonal contraceptive use,16 and multiparity.17 Less studied environmental risk factors include coinfection with Chlamydia trachomatis,18, 19 condom use by partners19 and nutritional factors.20–22 HPV viral factors that have been studied include viral type and variant, viral load, the effect of multiple concurrent HPV infections, and detection of HPV E6 and E7 transcripts.7, 23–26

HPV persistence must be studied on a type-specific basis, because each type is an independent agent.6, 27, 28 To study the impact of cofactors on duration of viral infection requires a very large study of HPV-positive women with good clinical follow-up. One practical approach is to follow women with ASCUS (43% containing at least one carcinogenic HPV type)29 and LSIL (76% containing at least one carcinogenic HPV type).30

In an earlier analysis in a large prospective study of women with ASCUS and LSIL, we found that approximately half of infections did not persist from initial detection to the subsequent 6-month follow-up visit.6 Here, we report determinants of type-specific HPV persistence from visit to visit, in 6-month intervals, over the 24 months of follow-up.


ALTS: The ASCUS LSIL Triage Study; ASCUS: atypical squamous cells of undetermined significance; CI: confidence interval; CIN2: cervical intraepithelial neoplasia grade 2; CIN3: cervical intraepithelial neoplasia grade 3; HC2: Hybrid Capture 2; HPV: human papillomavirus; HSIL: high-grade squamous intraepithelial lesion; LEEP: loop electrosurgical excision procedure; LSIL: low-grade squamous intraepithelial lesions; OR: Odds ratio

Material and Methods

Study design and population

The ASCUS LSIL Triage Study (ALTS) was a randomized trial designed to determine the optimal clinical management of women with a Pap smear interpretation of ASCUS or LSIL according to the 1991 Bethesda System. ALTS has already been described extensively elsewhere.31 Briefly, between 1997 and 1999, 5,060 women with a cytology interpretation of ASCUS (3,488 women) or LSIL (1,572 women) in the previous 6 months were referred from clinics associated with one of the four clinical centers. At enrollment, women were assigned randomly to one of three arms: immediate colposcopic examination, referral for colposcopy based on cytology at a threshold of high-grade squamous intraepithelial lesion (HSIL), or referral based on either cytologic HSIL or HPV detection by Hybrid Capture 2 (HC2, Qiagen, Gaithersburg, MD, formerly Digene Corporation). The ALTS protocol was reviewed and approved by institutional review boards at the NCI and at each of the four clinical centers. All participants gave informed consent.

At enrollment, women underwent a pelvic examination with collection of two cervical specimens, the first in PreservCyt for ThinPrep cytology (Cytyc-Hologic, Marlborough, MA) and the second in specimen transport medium (STM, Qiagen) for HPV genotyping. Each participant was also administered a standardized questionnaire by trained study staff to capture demographic, gynecologic, sexual, reproductive, medical and other information.

A 24-month follow-up was scheduled including visits every 6 months. At each follow-up visit, specimens were collected as at enrollment, and clinical management was uniform: only cytologic HSIL triggered referral (or re-referral) to colposcopy. Women with a histological diagnosis of cervical intraepithelial neoplasia grade 2 (CIN2) or CIN3 according to community center pathology (or, in some cases, Pathology Quality Control Group review) were offered treatment by loop electrosurgical excision procedure (LEEP).

HPV persistence was studied by dividing the 24-month follow-up into 6-month intervals between visits: 0–6 months, 6–12 months, 12–18 months, and 18–24 months. Infections were followed from the visit at which they were first detected until the first missing visit, until clearance, or until diagnosis of CIN2 (when the natural history of the infection was interrupted by treatment).

Women with a diagnosis of CIN3, adenocarcinoma in situ, or cervical cancer at any point during follow-up, as determined by the Pathology Quality Control Group, were considered to have had CIN3 at enrollment, whether or not it was detected by colposcopic biopsy at that time,11, 32 and were excluded from the analysis. This was done because our interest lies in HPV persistence as an early event in natural history rather than as a consequence of an established CIN2-3 lesion. Infections were excluded from the analysis if the woman did not return for a visit 6 months after the first appearance of the infection.

HPV DNA testing

Clinical triage was based on HC2 using probe B, a pooled-probe DNA test that targets 13 carcinogenic HPV types. However, the present analysis of HPV persistence is based on type-specific detection of HPV using L1 consensus primer PGMY09/11 PCR amplification and reverse-line blot hybridization of the STM specimens33 for the detection of 27 HPV types (HPV6, 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]) and a β-globin internal control. For 2,833 patients at enrollment, 11 additional HPV types were tested for (HPV61, 62, 64, 67, 69–72, 81, 82 subtype [IS39] and 89 [CP6108]). Follow-up specimens were tested for all 38 types.

Statistical methods

Statistical analysis was carried out using a previously developed model for the natural history of HPV infection.6 Briefly, a discrete-time survival model was used to estimate the probability that an infection would persist to the next visit, based on the current value of potential risk factors for persistence. Covariate effects were expressed as odds ratios (ORs) for persistence to the next visit. The probability of persistence was also allowed to depend on observed duration of infection.

All models were adjusted for study center (Birmingham, AL; Pittsburgh, PA; Oklahoma City, OK; Seattle, WA), randomization arm in the trial, and community Pap smear interpretation (ASCUS or LSIL) that determined entry into the study. These variables did not meaningfully influence risk of persistence,6 and so are not presented here.

All statistical analyses were conducted at the level of type-specific infections, nested within women. A woman could contribute more than one HPV type. Thus a multilevel model was used34 with random effects, or frailties, to account for correlation between infections in the same woman due to unobserved host or environmental factors. Distinct HPV types had different baseline clearance rates, which were modeled as random intercepts. All models were fitted using the lme4 package for R.35

Presentation of results

The tables enumerate the number of infections in each category of potential risk factors. For time-varying factors the tables show the risk factor classification when the infection was initially observed and also the number of infections for which the risk factor classification changed before the infection cleared. This second count gives an indication of the stability of time-varying risk factors.

When more than two groups were compared, floating standard errors were calculated for the log-odds ratios, and the figures show floating confidence intervals (CIs).36 Floating CIs show the uncertainty in the dose response relationship in a way that is independent of the choice of reference category.

Hypothesis tests for interaction were systematically carried out on all cofactors presented in the tables in order to investigate whether the ORs for persistence differed between carcinogenic and non-carcinogenic infections. When a significant interaction was found, stratified results are presented. Otherwise pooled results are presented for all HPV types.


Of the 5,060 women recruited into ALTS, 542 with a histologic diagnosis of CIN3, adenocarcinoma in situ, or cervical cancer were excluded. Among the remaining women, 2,019 had prevalent infections at enrollment and returned at 6 months, allowing an assessment of HPV persistence. During the follow-up, 389 additional women were found to have incident infections and returned 6 months later. Thus overall 2,408 women were eligible for analyses. This represented 7,542 HPV infections, comprising 3,456 prevalent infections and 4,086 incident infections. Carcinogenic HPV types (HPV16, 18, 31, 33, 35, 39, 45, 51, 52, 56, 58, 59, 66 and 68) comprised 52% of infections.

Age and other demographic factors

Figure 1 shows clearance curves for HPV infection by age at enrollment separately for prevalent infections, first detected at enrollment, and incident infections, first detected at a follow-up visit. Among prevalent infections, persistence was higher in older women, whereas for incident infections there was no clear trend with age. These findings were confirmed by the model, which showed significantly different effects of age for prevalent versus incident infections (p=0.008). Figure 2 therefore shows ORs for persistence stratified by prevalent versus incident detection. Further stratification by carcinogenic and non-carcinogenic HPV types showed similar patterns (data not shown).

Figure 1.

Persistence of human papillomavirus infections by age at enrollment.

Figure 2.

Odds ratios (ORs) for persistence of human papillomavirus infections of any type, by age and prevalent versus incident infection.

CI: confidence interval, FSE: floating standard error, FCI: floating confidence interval.

We did not observe any association between education or access to care (as measured by insurance) with persistent HPV infection (data not shown). The two most common classifications of race and ethnicity in the study population were White, non-Hispanic (59%) and African-American (32%). Comparing these two groups, African-American women had a higher risk of persistence of HPV infections (OR=1.16; 95% CI: 1.01–1.32). The numbers in the other racial/ethnic groups were too small to reliably assess differences in risk.

Contraceptive use and number of live births

Table 1 shows results for contraceptive use and number of live births. Current injectable contraceptive users were at marginally increased risk of persistent infection (OR=1.15; 95% CI: 1.01–1.32). Any contraceptive use was associated with persistence of carcinogenic types (OR=1.20; 95% CI: 1.03–1.40) but not with non-carcinogenic types (OR=0.99; 95% CI: 0.86–1.14) (p for interaction=0.03).

Table 1. Odds ratios (OR) for 6-month persistence of prevalent and incident human papillomavirus (HPV) infections of any type, by contraceptive use and number of live births
inline image

Smoking habits

Table 2 shows the association of persistence with smoking habits. There was no difference in persistence between current smokers and current non-smokers. However, among current smokers an intake of >20 cigarettes per day was associated with higher risk of persistence to the next visit (OR=1.43; 95% CI: 1.02–2.01) compared to ≤10 cigarettes per day.

Table 2. Odds ratios (OR) for 6-month persistence of prevalent and incident human papillomavirus (HPV) infections of any type, by smoking habits
inline image

Observed duration of infection

Figure 3 shows ORs for persistence by observed duration of infection. The pattern was significantly different for carcinogenic and non-carcinogenic types (test for interaction, p=0.016). Thus, Figure 3 presents ORs separately for carcinogenic types and non-carcinogenic types. In both cases, infections observed for two consecutive visits, which had therefore persisted for at least 6 months, were more likely to continue to persist to the subsequent 6-month interval. However, for non-carcinogenic types, the excess risk of persistence diminished and was no longer significant after 18 months duration (OR=1.09, 95%CI 0.79–1.50), whereas among carcinogenic types it remained high (OR=2.00, 95%CI 1.54–2.61). Adjusting the model, restricted to single carcinogenic infections, for semi-quantitative viral load, did not meaningfully change the results for persistence by observed duration of infection.

Figure 3.

Odds ratios (OR) for 6-month persistence of prevalent and incident human papillomavirus infections by observed duration of infection.

CI: confidence interval, FSE: floating standard error, FCI: floating confidence interval.

HPV viral load and cytologic interpretation

Figure 4 summarizes influences of viral load measured by HC2, cytologic interpretation, and cervicography visual impression on risk of persistence. The semi-quantitative HC2 measurement has been shown to correlate well with viral load measured by real-time PCR.37 However, in the presence of multiple HPV types, the cumulative viral load measured by HC2 is confounded with the multiplicity of infection. Therefore, this analysis was restricted to women with a single carcinogenic HPV type. Persistence was higher among infections with a semi-quantitative viral load of ≥1.0 pg/mL (the positive cutoff for HC2) versus <1.0 pg/mL (OR=2.05; 95% CI: 1.65–2.53), but there was no evidence of a dose-response effect above 1.0 pg/mL (p = 0.14). Women with abnormal cytology were at higher risk of persistence of a current carcinogenic HPV infection.

Figure 4.

Odds ratios (OR) for 6-month persistence of human papillomavirus infections with a single carcinogenic type, by semi-quantitative viral load (as determined by Hybrid Capture 2), cytologic interpretation, and cervicography.

CI: confidence interval, FSE: floating standard error, FCI: floating confidence interval, ASCUS: atypical squamous cells of undetermined significance, LSIL: low-grade squamousintraepithelial lesions.

Because viral load is highly correlated with cytologic abnormality,10 we investigated the association of combined viral load and cytology with persistence. This revealed an increasing risk of persistence in women with any cytological abnormality, or with viral load ≥1 pg/mL. Any abnormality by cervicography, particularly minor abnormalities called “atypical”, increased the risk of persistence of carcinogenic types (and also non-carcinogenic types, data not shown).

Self-reported history of other sexually transmitted infections

Limited numbers of cases of Chlamydia, Trichomonas, and Herpes Simplex Virus 2 were reported by participants, particularly during follow-up. None of these sexually transmitted infections influenced persistence of HPV infections (data not shown).


Using a previously developed model for the short-term natural history of HPV infection,6 we were able to quantify the determinants of short-term persistence among women with ASCUS/LSIL. The major finding of this study was that HPV persistence increased with age for prevalent but not for incident infections. This helps to resolve the conflicting findings in the literature regarding the association of age and persistent HPV infection. Previous short-term follow-up studies, largely confined to prevalent infections detected at baseline screening, found HPV persistence increases with age,24, 38 in agreement with our present findings. Persistent prevalent infections have been most clearly shown to increase the risk of developing CIN3.39 However, without explicitly comparing prevalent and incident infections, the one study of incident infections did not find an effect of age on duration of infection.40

Prevalent HPV infections can arise from a variety of possible natural histories with different meanings. In the youngest women, we can presume that most of these infections are new, as time since first intercourse is short. However, among older women, prevalent infections represent a complicated pool, because we do not know the antecedent natural history. On the one hand, it is possible that these are longstanding, steadily-detectable infections that would have been detected if previous screening had been performed. Such infections can signal already-established underlying CIN3. On the other hand, prevalent infections can also represent new infections, which were transmitted shortly before first detection. Finally, they can represent infections that re-emerged from latency, a still poorly-understood phenomenon, shortly before screening. In general, follow-up of this mixed group of natural histories has indicated that persistent prevalent infections confer a high risk of developing cervical cancer and that the risk of persistence increases with age, probably because many of the infections signal already established underlying CIN3.

We found few behavioral determinants of HPV persistence, in line with several other previous investigations.24, 38, 41 In general, with exception of intensive smoking and possibly contraceptive use, established risk factors for invasive cervical cancer (and CIN3) among HPV-positive women were not shown to influence the risk of 6-month persistence.

Several investigators have found that carcinogenic HPV types, particularly HPV16, persist longer on average than non-carcinogenic types, with a strong association of multi-year persistence with risk of diagnosis of CIN2 or worse lesions.6, 7, 23, 42 We also found that the impact of already-observed infection of 18 months on risk of continued persistence was stronger for carcinogenic types than non-carcinogenic types.

We found that both higher viral load of carcinogenic HPV types and its correlate, cytologic abnormality, predict an approximately 1.5- to 2-fold elevation in risk of persistence. This corroborates a recent report based on real time PCR24 and a large follow-up study of HPV-typed women with stratification on cytology.7 Women with a semi-quantitative viral load of ≥1.0 pg/mL (the positive cutoff for HC2) were at high risk for HPV persistence, but there was no increasing effect beyond this value. All grades of cytologic abnormalities showed a similarly elevated risk of persistence compared with cytologically normal women with low viral load. Similar associations of increased viral load with cross-sectional risk of CIN3/cancer43 and prospective risk of CIN3/cancer have been reported.44 It appears that a low viral-load infection without cytological abnormality represents a distinctly lower risk for viral persistence than infections of greater viral load associated with cytologic abnormality.

One possible limitation of this analysis is the short duration of follow-up. An analysis of predictors of long-term persistence may have yielded different results. A substantial proportion of HPV clearance occurs early6, 45 but it is long-term persistence that is linked to risk of CIN3 and cervical cancer.

Another limitation of this study is that the population was restricted to women referred for ASCUS or LSIL, which account for only about 5% of cytological interpretations from screening in the United States.46 However, approximately one-third of women with carcinogenic HPV infections are found to have HPV-related cytologic abnormalities mainly ASCUS or, to a lesser extent, LSIL.10 Although the ALTS population excluded women without cytologic abnormalities (who probably tended to have lower viral load and fewer HPV types on average) as well as women with community cytology showing HSIL, the population still represents a large proportion of infected women. Among women with invasive cervical cancer in United States screening programs, 69% have ASCUS/LSIL cytology, not a clear-cut high-grade or cancer interpretation.47 Moreover, there was an average delay of approximately 2 months between the time of the referral cytology and the enrollment examination in ALTS, during which time many lesions apparently cleared and possibly a few worsened. This permitted an examination of viral load and enrollment cytology with sufficient variability to find women that were cytologically normal with very low viral loads (and many who were HPV-negative).

In conclusion, this study established an important difference between incident and prevalent HPV infections. Older age of the woman is important only for prevalent infections. This finding has important public health implications for prophylactic HPV vaccines and for the use of multiple rounds of HPV testing in screening. Current HPV vaccines are designed to prevent true incident infection. Risk of new infections decreases with age, leading to a decreasing value of vaccination. We found that risk of persistence of an incident infection does not increase with age. In other words, infections among older women are not riskier by the metric of viral persistence than those found among younger women. With regard to cervical cancer screening among older women using HPV testing, we predict that a positive HPV test the first time a woman is screened would imply a higher risk of cervical cancer than a positive HPV test in subsequent rounds because the initial HPV test detects higher risk prevalent infections. Programs that incorporate longitudinal HPV testing at various intervals will need to recognize this difference.


The authors thank Dr. Silvia Franceschi for her valuable discussions of the manuscript and Mr. Didier Colin for preparing the figures.