Moscicki A, Hills N, Shiboski S, Powell K, Jay N, Hanson E, Miller S, Clayton L, Farhat S, Broering J, Darragh T, Palefsky J. Risks for Incident Human Papillomavirus Infection and Low-Grade Squamous Intraepithelial Lesion Development in Young Females. JAMA. 2001;285(23):2995–3002. doi:10.1001/jama.285.23.2995
Author Affiliations: Departments of Pediatrics (Dr Moscicki and Mss Hills, Powell, Jay, Hanson, Miller, Clayton, Farhat, and Broering), Epidemiology and Biostatistics (Dr Shiboski), Anatomic Pathology (Dr Darragh), and Stomatology (Dr Palefsky), University of California, San Francisco.
Context Low-grade squamous intraepithelial lesions (LSILs) have been described
as a benign cytological consequence of active human papillomavirus (HPV) replication.
Several studies have reported that certain behavioral and biological risks
exist for LSIL, suggesting that HPV alone is not sufficient for the development
of LSIL. However, because most of these studies have been cross-sectional,
it is not known whether behavioral and biological risks are simply risks for
HPV infection itself.
Objective To prospectively examine risks of incident HPV infection in HPV-negative
females and of incident LSIL development in females with HPV infection.
Design Prospective cohort study conducted between 1990-2000, with a median
follow-up of 50 months.
Setting and Participants Females aged 13 to 21 years who attended 2 family planning clinics in
the San Francisco bay area; 496 had prevalent HPV infection and 105 were HPV-negative.
Main Outcome Measure Incident development of HPV infection and LSIL, analyzed by various
demographic, behavioral, and clinical risk factors.
Results Fifty-four incident HPV infections occurred in the 105 females who were
HPV-negative at study entry (median duration of follow-up for those who remained
HPV-negative was 26 months). Multivariable analysis showed that risks of HPV
included sexual behavior (relative hazard [RH], 10.10; 95% confidence interval
[CI], 3.24-31.50 per new partner per month), history of herpes simplex virus
(RH, 3.54; 95% CI, 1.37-9.10), and history of vulvar warts (RH, 2.73; 95%
CI, 1.27-5.87). Current use of oral contraceptives had a significantly protective
effect (RH, 0.49; 95% CI, 0.28-0.86). Among the 496 individuals who were HPV-positive
at baseline or in follow-up, there were 109 incident cases of LSIL during
the follow-up interval, with a median follow-up time of 60 months for those
who never developed LSIL. Human papillomavirus infection was the most significant
risk factor for development of LSIL. The multivariable model showed the following
risks for LSIL: HPV infection for less than 1 year (RH, 7.40; 95% CI, 4.74-11.57);
HPV infection for 1 to 2 years (RH, 10.27; 95% CI, 5.64-18.69); HPV infection
for 2 to 3 years (RH, 6.11; 95% CI, 1.86-20.06); and daily cigarette smoking
(RH, 1.67; 95% CI, 1.12-2.48).
Conclusion Our results indicate distinct risks for HPV and LSIL. In addition, most
women with HPV infection in our study did not develop LSIL within a median
follow-up period of 60 months. These findings underscore the hypothesis that
certain biological risks thought to be associated with LSIL are, in fact,
risks for acquisition of HPV. Cigarette smoking was a risk specific to LSIL,
supporting the role of tobacco in neoplastic development.
Low-grade squamous intraepithelial lesions (LSILs) are relatively common
manifestations of cervical human papillomavirus (HPV) infections. Most pathologists
agree that LSIL is a cellular response to HPV infection1,2
resulting from the expression of specific viral proteins that interact with
the host cytoskeletal structures and induce cell proliferation.3- 5
Some researchers suggest that all HPV infections result in pathologic changes
of LSIL at some point in the natural history of infection; however, the discrepancy
between rates of infection with HPV and development of LSIL is quite striking,
even in adolescent and young women in whom the prevalence of both is high.
Rates of HPV infection in sexually active young females have been consistently
reported to range from 19% to 46%; these are significantly higher than reported
rates of LSIL, which range from 1.1% to 7.0% in the same groups.6- 8
Differences in the length of HPV and LSIL persistence do not completely explain
the observed discordance because recent studies have reported that most females
who have HPV infection clear it within 12 to 24 months,9,10
a rate that is similar to the regression rate for LSIL in young women.11,12 Consistent discrepancies suggest
that the development of LSIL is a distinct event in the natural history of
HPV infection. Consequently, risks associated with LSIL should be distinct
from risks for infection with HPV.
While studies have shown that risks for HPV and LSIL, specifically those
associated with sexual behavior, significantly overlap,13,14
many studies have found that certain behavioral and biological risks that
seem to be relevant to the development of LSIL are not significant risks for
infection with HPV,6,15- 18
suggesting that HPV alone is not sufficient for the development of LSIL. From
an epidemiological perspective, behavioral and biological risks for LSIL are
difficult to distinguish from risks for HPV infection in cross-sectional studies
because all females with LSIL have HPV, and females without LSIL are less
likely to have HPV infections.
The purpose of the current investigation was to determine the risk factors
for incident cervical LSIL development in a cohort of adolescent and young
women with HPV infections, and the risk factors associated with acquiring
an HPV infection for females who tested negative for HPV DNA at entry into
The females in this report were part of an active, ongoing natural history
study of HPV infection that began in 1990.6,10
Females gave consent according to the guidelines set forth and approved by
the Committee for Human Research, University of California, San Francisco.
We screened females aged 13 to 20 years who were attending 1 of 2 family planning
clinics between 1990 and 1994 for detection of cervical HPV DNA using a commercial
test (HPV Profile, Digene Diagnostics, Silver Spring, Md). We asked those
who tested positive to participate in the study. The characterization of this
cohort recruited using the commercial test is described in detail elsewhere.6,10 We replaced testing for HPV DNA by
the polymerase chain reaction (PCR) technique in 1995. In addition, we retested
all samples retroactive to baseline using PCR. All data in this article refer
to HPV DNA using PCR amplification.
For the HPV incidence study we included only females who entered the
study negative for HPV using PCR. Because of the possible false negatives,
which have been associated with HPV testing,19
we used conservative criteria for defining initial HPV negative tests: only
those from the cohort who at baseline and first follow-up had negative results
for HPV using PCR were included for the incident HPV study. For the LSIL incidence
study we included only females who entered the study negative for LSIL. Because
of the known false-negative rate (approximately 50%-70%) of cytology for squamous
intraepithelial lesions, we defined similar conservative criteria for the
LSIL incidence study by including only females from the cohort who at baseline
and first follow-up had normal cytologic findings. In addition, individuals
in the HPV incidence study who developed an incident HPV infection were included
in the LSIL analysis. However, we excluded from the LSIL incidence study those
who did not test positive for HPV by PCR at any time during the study, because
most scientists agree that HPV is necessary for LSIL development.2 Including HPV-negative subjects in the LSIL analysis
may have diluted the risk factor analysis. The median number of visits was
9 (interquartile range [IQR], 4-15) and the median time in the study was 50
months (IQR, 23-79 months) for females who were followed up under the above
Visit protocols are detailed elsewhere.10
Baseline and interval visits on all HPV-positive (every 4 months) and HPV-negative
(every 6 months until they turned HPV-positive, and then every 4 months) participants
included cervical cytology, testing of cervical samples for HPV DNA collected
using sterile Dacron swabs, and face-to-face interviews to obtain information
on demographics and sexual and substance use behaviors. Colposcopic examination
of the vulva, vagina, cervix, and perianal areas was also performed on all
subjects with the aid of 3% acetic acid at each visit. Cervical samples were
obtained for Chlamydia trachomatis and Neisseria gonorrhoeae, and vaginal samples were collected, measured
for pH, and examined microscopically for the presence of clue cells and Trichomonas vaginalis. These latter cervical and vaginal
samples were taken at the baseline and annual examinations, as well as at
interval visits for those females who were currently symptomatic for lower
genital tract infections. C trachomatis was identified
initially using standard culture techniques.20
In 1996, we replaced the culture test by the ligase chain reaction technique
(LCX STD system, Abbott Laboratories, Abbott Park, Ill), which did not significantly
alter the incidence of C trachomatis (annual incidence
of <3% for each). For purposes of this study, bacterial vaginosis was diagnosed
if 2 of the following 3 criteria were met: vaginal pH greater than 5.0, 20%
or greater of cells on wet mount clue cells, and a positive "whiff" test using
potassium hydroxide.21 Herpes serology22 using Western blot was performed on the majority
of subjects at baseline. During follow-up, viral cultures were performed only
when signs or symptoms (ulcers or severe cervicitis) were present at any given
visit. We diagnosed herpes simplex infections as follows: if a subject tested
positive for serology at baseline or at any other visit prior to LSIL diagnosis;
if a laboratory test for herpes simplex virus (HSV) yielded a positive result;
or if a subject reported a history of HSV at baseline or between visits.
At baseline, all females with areas of atypia suggestive of high-grade
squamous intraepithelial lesions (HSILs) as defined on colposcopy23 were randomly assigned to biopsy or no biopsy. A
total of 53 females received biopsy per this protocol. In addition, females
with lesions consistent with carcinoma in situ by colposcopy23
(n = 15) also underwent biopsy. This method allowed for a more accurate diagnosis
and characterization of lesions present at entry. A subset of this group and
results of histology based on this protocol were published earlier.24 An analysis of the potential effect of these biopsies
on the natural history of HPV was conducted and no effects were found.
We tested cervical samples at baseline for HPV DNA using HPV Profile
(Digene Diagnostics, Silver Spring, Md). All samples have been subsequently
retested using 1 of 2 PCR techniques. Methods for the initial amplification
technique have been described previously.10
Our initial method sought the presence of amplified HPV material in a dot
blot format using an enhanced chemiluminescent method. A generic probe mix
was used that determined the presence of 1 or more of 25 different HPV types
nonspecifically and β-globin for internal consistency. Samples were also
specifically probed for low-risk HPV types 6/11/42/44, and high-risk HPV types
16, 18, 31/33/35, 39, 45, 51, 52, 56, and 58.
Our second method for amplification and detection used Roche reverse
blot "strip" method (Roche Molecular Systems, Inc, Alameda, Calif).25 A denatured biotin-labeled PCR product was hybridized
to an array of immobilized oligonucleotides: HPV types 16, 18, 26, 31, 33,
35, 39, 45, 51, 52, 55, 56, 58, 59, 68, MM4 (W13B), MM7 (P291), MM9 (P238A),
6, 11, 40, 42, 53, 54, 57, 66, MM8 (P155), and 2 β-globin controls. Of
the 7114 HPV samples tested, we tested 1957 using the strip method and 5157
using the dot blot test. A total of 156 were tested by both methods. We found
agreement between the 2 tests to be 87.8%, yielding a κ statistic of
0.63, which indicates substantial agreement statistically (P<.001).
Since we did not use the strip method for all samples, the type-specific
data used for this analysis included only HPV types for which all subjects
were screened at all visits. We classified all other positive HPV types by
the strip method as HPV type "other" and HPV generic probe positives by dot
blot with no type identified were also classified as HPV type "other."
When the study began in 1990, we used standard World Health Organization
cytologic nomenclature for condyloma and cervical intraepithelial neoplasia
(CIN) grades 1-3. We asked the females who were diagnosed with CIN 1-3 by
cytology at any given visit to return for biopsy. For purposes of this analysis,
we categorized diagnoses per the Bethesda system for rating cytology using
the nomenclature LSIL and HSIL.26 We diagnosed
LSIL cases in this study based on cytology; HSIL cases were classified based
on a diagnosis of HSIL on either cytology or histology.
We first made estimates of the distribution of time to HPV-positive
results for females screened as HPV-negative by PCR at entry using flexible
parametric regression methods for interval-censored survival outcomes.27 The hazard function of this distribution was modeled
as a polynomial (spline) function of duration of follow-up and likelihood
ratio tests were used to select the best-fitting model among fitted alternatives.
We made estimates of the distribution of time to initial diagnosis of LSIL
from the time of initial HPV infection using a generalization of the method
described above to account for additional uncertainty about the time of initial
HPV infection. For the females observed to be HPV-positive at entry, initial
infection is known only to fall in the interval between initiation of sexual
activity and enrollment. For the females screened as initially HPV-negative,
we assumed the time of initial HPV infection to be uniformly distributed in
the interval between the last observed negative and first positive test results.
We used a discrete-time version of the proportional hazards model allowing
for interval censored outcomes and incorporated time-dependent covariates
to examine risks for the development of both LSIL for females with preceding
positive results for either high- or low-risk HPV types, and for the first
incidence of HPV for those females who entered the study with HPV-negative
In the incident LSIL analysis, we treated variables pertaining to HPV
infection in several ways. The risk associated with current HPV status at
each visit (including high-risk, low-risk, and any HPV type) was examined.
In addition, we examined the risk related to length (months) of persistent
infection with the same HPV type (ie, the same type within each individual;
types were allowed to vary among individuals). We also subdivided this latter
variable into categories of positive (with the same type) for less than 1
year, 1 to 2 years, 2 to 3 years, and more than 3 years. We then entered these
categorical variables into the analysis, using HPV negativity as the baseline
to which each duration category was compared.
We first examined each variable in a univariate analysis and, using
a significance level of .05, entered those variables found to be significant
into a multivariable analysis for each of the 2 outcomes, incident HPV infection
and diagnosis of LSIL. Variables that lost significance at the .05 level in
the multivariable analysis were excluded from the final models. P values were calculated using the Wilcoxon nonparametric test for
continuous variables and the χ2 test for categorical variables.
Statistical software used was SAS version 6.12 (SAS Instititute, Cary, NC)
and S-PLUS version 5.0 (Insightful Corp, Seattle, Wash).
Of the 897 females who consented to participate in the study, 81 females
had incomplete data at baseline (missing questionnaires, missing Papanicolaou
test data, or inadequate PCR samples) and were not included in the study.
A total of 155 had a prevalent LSIL at entry (baseline or first follow-up
visit) and were excluded. In addition, we excluded 17 females who developed
HSIL prior to LSIL. Of the remaining 644 females, 105 were HPV-negative using
PCR at entry (baseline and first follow-up visit) and eligible for the HPV
incidence study; 51 of these never developed an HPV infection and were excluded
from the LSIL analysis. In addition, 97 females had fewer than 3 visits and
were considered lost to follow-up for the LSIL analysis, resulting in a total
of 496 females for the LSIL analysis.
Analysis comparing the latter females lost to follow-up (n = 97) with
those included in the LSIL study revealed no significant differences in terms
of race, age at entry into the study, years sexually active, number of recent
partners, or sexually transmitted disease (STD) history. However, these females
had significantly fewer sexual partners at baseline (median, 4; IQR, 2-7 vs
median, 5; IQR, 3-9 for those in the study, P = .03)
and were of lower socioeconomic status, as measured by the education levels
of their mothers (40.7% had completed some college education in the excluded
group vs 53.9% in the analysis group, P = .02).
Demographic characteristics of and comparisons between the 2 groups
(HPV and LSIL incidence cohorts) are given in Table 1. The group HPV-negative at entry had fewer lifetime sexual
partners, but recent sexual behavior as defined by the number of partners
reported in the last 2 months was not significantly different for the 2 groups.
No other significant differences between the 2 groups were found.
There were 54 incident cases of HPV infection. Forty-six percent of
the HPV infections were with high-risk types only, 9% were with low-risk HPV
types only, 13% were with mixed high- and low-risk HPV types, and 32% were
with a type other than those in the standard mix. Figure 1 shows the estimated distribution of time for subjects remaining
free of HPV infection. The estimated proportions of females infected at 12
and 36 months following enrollment were 0.17 (95% confidence interval [CI],
0.14-0.19) and 0.55 (95% CI, 0.31-0.79), respectively. (These estimates are
obtained from the corresponding estimates and CIs for the probability of remaining
uninfected in Figure 1 by subtraction
from 1.) The curve predominantly reflects high-risk HPV types since the majority
of the group had high-risk infection. The median time of follow-up for the
females remaining HPV-free was 26 months (IQR, 10-54 months).
The univariate analysis of risks for cervical HPV infection adjusted
for time in study is shown in Table 2.
Multivariable analysis is summarized in Table 3. We found significant risks for infection with HPV to be
sexual behavior, history of HSV, and a history of vulvar warts. The current
use of oral contraceptives when included in the multivariable analysis continued
to have a significantly protective effect. Two variables that had been significant
in univariate analysis, history of marijuana use and lifetime sex partners,
lost significance in the multivariable analysis and were not included in the
Ten (19%) of the 54 females with incident HPV developed LSIL. Among
them, 7 had a high-risk HPV type, 1 had a low-risk type, and the remaining
2 had unknown HPV types. There was no difference between the proportion of
those with incident high-risk types who developed LSIL and those with incident
low-risk HPV types. One participant tested positive for HPV after the LSIL
diagnosis. No difference was found between those positive for low- and high-risk
types for time from HPV infection to LSIL diagnosis.
The HPV type distribution at baseline or first incidence in the LSIL
study (n = 496) was as follows: 75 were positive for types 6/11/42/44; 134
for type 16; 62 for type 18; 57 for types 31/33/35; 23 for type 39; 27 for
type 45; 44 for type 51; 54 for type 52; 15 for type 56; 32 for type 58; and
282 were positive for other HPV types. These numbers do not sum to the sample
size of 496 because some individuals were positive for multiple types: 96
for 2 types and 81 for 3 or more types.
There were 109 incident cases of LSIL. Figure 2 shows the estimated distribution of time for subjects remaining
free of LSIL. The estimated proportions of individuals developing LSIL at
36 and 60 months following initial HPV infection were 0.15 (95% CI, 0.13-0.17)
and 0.21 (95% CI, 0.17-0.25), respectively. After 60 months, the estimated
proportion developing LSIL remained relatively flat, reflecting few observed
infections among females with longer follow-up. The median time of follow-up
after HPV detection for those remaining LSIL-free was 60 months (IQR, 28-82
months). No differences assessed by overlapping pointwise 95% CIs between
high- and low-risk HPV types were found for time to LSIL (curves not shown).
Table 2 shows the univariate
analyses of risk factors for LSIL adjusted for time in study. Those females
who were HPV-positive at the time of visit were approximately 7 times more
likely to develop LSIL than those who were negative. The relative hazard for
current infection with a high-risk HPV was 5.39 (95% CI, 3.57-8.15) vs a relative
hazard for current infection with a low-risk type of 10.68 (95% CI, 4.80-23.78);
however, this latter estimate was far less precise, because infections with
low-risk types alone made up only 4% of the total positive HPV tests and were,
therefore, infrequent vs infection with high-risk types (63% were positive
for only high-risk types, 2% were simultaneously positive for both high- and
low-risk types, and 31% were positive for HPV type "other").
In order to examine the association between HPV persistence and the
development of incident LSIL, we also examined the duration of type-specific
HPV infections as a risk factor for LSIL. We found that each additional year
of HPV type-specific positivity increased the relative hazard by slightly
more than 2-fold (2.27; 95% CI, 1.86-2.78). To investigate whether this relationship
was strictly linear over time, a quadratic term in time was added to the model
and was found to be significant at the .07 level. We also examined a categorized
version of our duration variable, in which HPV persistence was divided into
intervals of positive for less than 1 year, 1 to 2 years, 2 to 3 years, and
more than 3 years. While there was some indication of a leveling off of risk
after 2 years of infection, the small size of the last category prevented
us from reaching any firm conclusions regarding nonlinearity.
All factors that were significant in the univariate analysis at the
.05 level were included in a multivariable analysis. We used the categorized
HPV infection duration variables rather than those indicating high- and low-risk
status at visit to characterize HPV infection in the multivariable analysis,
because so few subjects were positive for low-risk HPV types at the time of
LSIL diagnosis. Therefore, estimates for the categorized variables were more
precise. Presence of HPV and cigarette smoking remained significant. Marijuana
use and number of lifetime partners lost significance in this analysis and
were not included in the final model (Table
4). Interaction terms were tested in the model, but were excluded
from the final model since they were not significant.
Although LSIL is considered a common benign manifestation of HPV infection,
the findings from this study underscore the differences in the biological
and behavioral risks associated with the acquisition of HPV as compared with
the development of LSIL. Our results contrast with many cross-sectional studies.13,14,28 Since HPV infection
and LSIL are strongly associated, incident studies are better epidemiological
tools to address differences in risk. Not surprisingly, a current positive
HPV test result with either low- or high-risk HPV types was the strongest
risk factor for LSIL. Sexual behavior, specifically exposure to new partners,
represented the strongest risk factor for incident HPV infection.
The association between sexual behavior and incident HPV is quite notable
if the actual risk is taken into account. That is, risk increased nearly 10-fold
for each new partner per month reported. The strong association between sexual
behavior and HPV infection supports the hypothesis that HPV is a sexually
transmitted virus and that most "new" infections in young females are predominantly
due to exposure, rather than the "transient" up-regulation of latent infections
in nonimmunocompromised hosts.
The association with vulvar warts is not surprising because the virus
can easily be transmitted from the vulvar area to the cervix via intercourse
or from other sources, including tampons. The association between incident
HPV and the history of HSV is most interesting because HSV has frequently
been associated with the development of cervical cancer.29
It is well known that primary and secondary HSV infections result in breaks
of the cervical mucosal barrier through inflammation and ulceration, thereby
granting easy access to the basal epithelial cells for viruses such as HPV.
The lack of association with other risk factors supports the strong role of
sexual exposure in HPV acquisition and diminishes support for other behaviors
that most likely reflect risky sexual behavior of the female or her partner.
The independent protective role of oral contraceptives in acquiring
HPV is also interesting since prolonged use of oral contraceptives has been
associated with development of squamous intraepithelial lesions in several
studies.16,30 Since condom use
was not significant in the univariate analysis, we postulate that this relationship
may be due to the influence of estrogen or progesterone on immune regulation.31
Lack of association between sexual behavior and LSIL also underscores
the fact that in addition to HPV, other risks are necessary for LSIL development.
Indeed, in this study we found that daily cigarette smoking had a deleterious
effect, which contributed to the development of LSIL. The mechanism for this
is not understood; however, several epidemiologic studies have found a role
for cigarette smoking in invasive cervical cancer.32
Nicotine and other cigarette metabolites have been found in the cervical mucus.33 We suspect that the relationship of cigarette smoking
to LSIL may result from its ability to cause immune dysregulation.34 As discussed, LSIL is a histological result of active
HPV replication that may involve up-regulated protein transcription. Several
studies have implicated the immune system in viral regulation.35- 39
The lack of association between LSIL and other risk factors such as
contraceptive use or C trachomatis infection as found
in other studies13,40 may be due
to several factors. First, the cohort falls into quite a young age group,
where the squamous intraepithelial lesion is predominantly LSIL. That is,
many of the other studies using older populations have defined LSIL by cytology
alone, potentially resulting in the underdiagnosis of HSIL.41
Consequently, these studies have identified risk factors for a group with
mixed HSIL and LSIL. The majority of our participants with LSIL underwent
biopsy to rule out HSIL. Second, HPV infection is a required risk for LSIL.
Our population appears to frequently contract new HPV infections, whereas
in older females with less frequent rates of sexual activity, HPV DNA detection
more often reflects a persistent infection.9
Third, risk factors such as oral contraceptive use may require years of exposure
before the "risk effect" is seen. Again, our participants were relatively
young with less than 4 years of oral contraceptive exposure. The lack of association
between STD infections and LSIL may have been due to the low prevalence of
STDs in this population.13 However, we emphasize
again that STDs may be a measure of risk behavior. Risk behavior connotes
HPV exposure, which in turn increases the risk of the occurrence of LSIL.
Although LSIL was strongly associated with HPV, the majority of young
females in our study with HPV infection never developed LSIL, suggesting that
certain biological risks or conditions are necessary in order for the LSIL
to occur. In addition, the interval between HPV infection and LSIL varied,
implying that in some females a period of latency occurs before LSIL develops.
The reason for this latency and later up-regulation is not clear. Our findings
suggest that the risk of developing LSIL in young females is only present
within the first 3 years after detection of HPV DNA. There was no evidence
that the risk continued after 3 years; however, the analysis was limited due
to small numbers in this group. Since this study was performed on young females,
the risk cannot be generalized to older women.
In summary, there were clear differences in risks for incident HPV infection
and LSIL development, suggesting that many of the risks that have been associated
with LSIL in cross-sectional studies were either risk factors for or surrogate
markers of HPV infection. The strongest risk for HPV detection in young female
is sexual exposure. However, the majority of females in this study with a
positive HPV result never developed LSIL within a median follow-up time of
60 months, which suggests that certain biological risks or conditions are
required for the LSIL to occur. One such risk in our group was the use of
tobacco, a known carcinogen.