Factors That Influence the Rate of Epithelial Maturation in the Cervix in Healthy Young Women

Participants 

Study participants were selected from a larger cohort of women participating in an on-going longitudinal study of HPV that has been described elsewhere [22]. Briefly, beginning in 2000, women were voluntarily enrolled from a college health center and a family planning clinic. The inclusion criteria to the larger cohort were: ages 13-21 years, history of less than 5 years of sexual activity, and intent to attend interval visits for at least 3 years. The exclusion criteria were: history of treatment for squamous intraepithelial lesions (SIL) of the cervix, any immunosuppressive condition, or pregnancy. The only exclusion during follow-up was histology-proven incident cervical intra-epithelial neoplasia (CIN) 2/3. The study was approved by the Committee on Human Subject Research at the University of California, San Francisco and the San Francisco State University Committee on Human Research.

Study design 

Each participant had a baseline visit and was asked to return for 4-month interval visits. Each visit included a face-to-face interview to collect information concerning socio-demographic characteristics, behaviors, and medical history. A physical examination was performed by research staff clinicians to collect cervical and vaginal samples as described previously [23]. Briefly, bacterial vaginosis, Candida, and T. vaginalis infections were diagnosed on routine normal saline and KOH wet mounts [24]. HPV DNA was detected from cervical vaginal lavage (CVL) samples using the Roche Reverse Line Blot assay [25]. Commercial PCR assays were applied to cervical swabs to test for C. trachomatis and N. gonorrhoeae. The assays for C. trachomatis and N. gonorrhoeae were performed at annual and any symptomatic visits. If a positive result was found, the cervical samples stored from a minimum of two prior visits were also tested to more closely characterize the onset of infection. After these samples were obtained, colposcopy was performed using 3% acetic acid applied to the cervix to accentuate the epithelial appearance [26]. After centering the cervix as much as possible, colpophotographs were taken at 10x and 16x magnifications using a Nikon camera (N2020 or N90S) attached to the colposcope and 35-mm film. In addition to the study visits, participants who reported genitourinary symptoms or concerns at any other time were seen by research staff for diagnosis and treatment of any infections.

Colpophotographic measurements 

The visible areas of interest on the epithelium were measured using computerized planimetry techniques that were published previously [10], [27] and modified for this study. Colpophotographs were viewed as digital images (1024 × 1536 pixels) using Adobe Photoshop CS2. Nondestructive adjustment layers, sharpening tools, and shadow brightening tools were custom applied to each colpophotograph to improve image clarity. Only those Photoshop tools that adjust all pixels consistently and thus preserve the integrity of the original image data were applied. Areas of immature epithelium, defined using standard colposcopic features for columnar epithelium and early-mid squamous metaplasia [26], were outlined with digital freehand markings. Discontinuous areas required multiple discrete markings. The area of the total cervical face, defined as the cervical portio visible on clinical examination, was outlined with a combination of automated curves and freehand markings. Pixel counts were calculated to represent each marked area. “Percent-immaturity” was calculated as the area of immature epithelium divided by the area of total cervical face.

Use of the percent-immaturity measure improved the ability to compare colpophotographs since absolute pixel counts do not account for variations in photographic conditions (angle of view) and inconsistent cervical size portrayed on the photographs taken at different visits for a single woman. Furthermore, we purposely viewed the colpophotographs from all study visits on the screen simultaneously. This approach allowed a consistency in the definition of the total cervical face and the identification of any photographs that may have been incorrectly labeled. We performed a small analysis to compare our simultaneous approach to that of measuring single colpophotographs in random order. The variability of the calculated percent-immaturity ranged from 15-20% when colpophotographs were measured in random order by two reviewers. However, this variability was reduced to 5-10% using the simultaneous approach and appears consistent with previous work indicating an error rate of 5-10% upon repeat measurements of a single photo because of the inherent practical limitations of gross freehand outlining [10], [27]. Additionally to minimize error, if the cervical epithelial topography was deemed to show no change compared with the previous colpophotograph, the identical value of percent immaturity was entered into the dataset. Otherwise, complete measurements and pixel calculations were performed. Given the inherent coarseness of using colpophotographs, these approaches allowed us to focus on gross cervical changes rather than small subtle changes that would be difficult to reproduce and indistinguishable from measurement error. The reviewer (L.H.) was blinded to all participants’ characteristics and study data. A subset (10%) of colpophotographs was measured by a second blinded reviewer (A-B.M.) and measurements were reconciled.

Other measures and statistical analyses 

Because our specific aim was to examine maturational changes in the cervical epithelium, women from the larger cohort were selected for this analysis according to these additional inclusion criteria: a minimum of 10% immaturity on the first available colpophotograph, and minimum of 2 years of follow-up during which the colpophotographs were adequate for image analysis. Among the larger cohort of 616 women, approximately one-third of women were eligible for this analysis. The first consecutive 145 eligible women (enrolled October 2000 to September 2002) were selected.

All of the variables in this analysis were time dependent except for the static predictor variable, ethnicity. Each visit with an available colpophotograph was analyzed. “Visit (x)” refers to the visit currently under analysis; “visit (x-1)” is the nearest prior visit; and “visit (x-2)” is the second nearest prior visit. The outcome variable was the rate of cervical maturation, calculated from the percent-immaturity at visit (x) minus the percent-immaturity at visit (x-1). A negative value indicated a decrease in cervical immaturity and an assumption of interim squamous metaplasia. In contrast, a positive value indicated an increase in cervical immaturity and an assumption of interim reversal to more immaturity. The time-dependent predictors assessed at visit (x) were: chronological age, time since menarche, time since first intercourse, number of lifetime sexual partners, number of new sexual partners since the previous visit, frequency of sexual intercourse (number of episodes in the prior 2 months), condom use (in the prior 2 months: never, rarely, half the time, most of the time, or always), and the following dichotomous variables (currently yes or no): cigarette smoking, pregnancy, and any hormonal contraceptive use. The time-dependent variables for the genital infections were also dichotomous (yes or no). Bacterial vaginosis, candida, and HPV were each defined as any infection detected between 2-4 months before visit (x) and were inclusive of all diagnoses made at study and non-study visits. C. trachomatis, N. gonorrhoeae, and T. vaginalis infection were each defined as any infection detected between 4 and 8 months before visit (x) to allow more time for the treatment and resolution of acute inflammation and cellular repair. To address the possibility of other timeframes of longitudinal influence, we also examined alternative definitions that included infections up to 18 months before visit (x) or that included infections from visit (x) to (x-1), or visit (x-1) to visit (x-2); but all of these results were unchanged and thus are not presented here.

Data analyses included bivariate and multivariate generalized linear models with repeated measures. Once an individual woman's cervical epithelium appeared largely mature (<10% immaturity) and stable, her subsequent visits were excluded to avoid misleading errors in the repeated-measures models. Each model was adjusted for time that was entered a priori in 4-month increments, and for percent-immaturity measured at baseline since greater immaturity at baseline was significantly associated with an accelerated rate of maturation during follow-up (correlation coefficient = .83, p < .0001). This finding was expected, because a woman who begins the study with a greater area of percent-immaturity would inherently have a greater likelihood to demonstrate cervical maturational activity during follow-up. A separate bivariate model was created for each predictor variable discussed above. Then an initial multivariate model included all predictor variables, and the least significant variable was removed in each subsequent iteration. Two-way interaction terms between time and the independent variables were examined. Goodness of fit was evaluated using the Akaike information criterion (AIC) or the Bayesian information criterion (BIC) when we performed model comparisons and selections. The final most parsimonious model retained the time variable, percent-immaturity measured at baseline, and all predictor variables found to be significant at p < .05. Because we observed that some women experienced short time periods of increasing cervical immaturity, a small subanalysis was conducted to examine factors associated with this reversal towards more immaturity. These analyses compared the time intervals of reversal to those of no change as a dichotomous outcome variable. Bivariate and multivariate generalized estimating equation models were constructed for the predictor variables above.

Sociodemographic characteristics and behaviors 

Colpophotographs were adequate for measurement from 815 visits for the 145 women, yielding 465.1 woman-years of follow-up and means of 5.6 visits (SD, 1.8) and 2.7 years (SD, 1.0) of follow-up per woman. The mean time between these visits was 5.9 months (SD, 2.2 months). The demographic characteristics at baseline, and the behaviors and infections over the entire study period are shown in Table 1. This young cohort with a mean baseline age of 17.8 years was ethnically diverse and had been sexually active for a mean of 1.8 years at baseline. Vaginal intercourse was common during the study, with a mean of seven episodes per month. Consistent condom use was reported at only 17% of visits, and hormonal contraception use at 28%. The hormonal contraception consisted largely of oral contraceptive pills containing 30–35 μg of estrogen (74%) or 20 μg of estrogen (12%). No progestin-only pills were reported. The remaining hormonal methods were injections of medroxyprogesterone acetate (21%) and medroxygrogesterone acetate/estradiol cypionate (.9%). C. trachomatis infections were diagnosed 27 times in 19 distinct women; and N. gonorrhoeae and T. vaginalis were rare. Within the study, women presented with current pregnancies at 2% of visits.

Table 1. Demographic characteristics, behaviors, and test results (N = 145 women)

	Characteristic	Mean (SD) or n (%) a
	Age at baseline, years	17.8 (2.1)
	Ethnicity
	African-American	10 (7%)
	Asian	24 (16%)
	Latina/Hispanic	48 (33%)
	White	43 (30%)
	Mixed/other	20 (14%)
	Age of menarche, years	12.3 (1.3)
	Age of first intercourse, years	16.0 (2.1)
	Number of lifetime sexual partners at study entry	4.5 (6.3)
	Number of new sexual partners since previous visit	.4 (.8)
	Episodes of sexual activity in past 2 months	14 (16)
	Condom use in past 2 months
	Never/rarely	479 (61%)
	Half/most of the time	173 (22%)
	Always	136 (17%)
	Current pregnancy	20 (2%)
	Current hormonal contraceptive use	221 (28%)
	Current smoking	196 (26%)
	Chlamydia trachomatis infectionb	27 (3%)
	Neisseria gonorrhoeae infectionb	1 (.1%)
	Trichomonas vaginalis infectionb	0 (0%)
	Bacterial vaginosisc	12 (1%)
	Candida infectionc	15 (2%)
	HPV infection by any of 32 typesc	234 (31%)
	HPV infection by high risk typesc, d	179 (23%)
	HPV infection by low risk typesc, e	46 (6%)

HPV = human papillomavirus.

a Behaviors that were repeatedly measured are reported for all 815 visits by the 145 women. Variations in the denominators are due to missing data. b Defined as any infection during the 4-8 months prior to each visit. c Defined as any infection during the 2-4 months prior to each visit. d High-risk types: 16, 18, 26, 31, 33, 35, 39, 45, 51, 52, 53, 56, 58, 59, 66, 68, 73, 82. e Low-risk types: 2, 6, 11, 40, 42, 54, 55, 57, 67, 69, 70, 72, 83, 84.

Cervical maturation 

Our sample of 145 women began the study with a median area of cervical immaturity of 39% (interquartile range, 20, 52%) and ended the study with a median of 8% (IQ 3, 14%) immaturity. From the 815 visits, 670 short intervals were derived between each visit (x) and visit (x-1) as shown in Table 2. Among these short intervals, 388 (58%) demonstrated a decrease in percent-immaturity (median 8%, range −64 to −.6%), suggesting maturational change; 234 (35%) had no change; and 48 (7%) had an increase in percent-immaturity (median +12%, range +.1 to +62%), suggesting reversal toward more immaturity.

Table 2. Changes in percent-immaturity of the cervical epithelium during entire study and during discrete intervals

	Direction of net change over entire study (N = 145 women)	Women, n (%)	Amount of change in percent-immaturity, median	Total follow-up time, mean years (SD)
	Maturation	137 (95%)	−25%	2.7 (1.0)
	No change	2 (1%)	NA	2.6 (.8)
	Reversal	6 (4%)	+8%	3.1 (1.1)
	Direction of change during each short interval (n = 670 intervals)	Intervals, n (%)	Amount of change in percent-immaturity, median	Duration of interval, mean months (SD)
	Maturation	388 (58%)	−8%	7.5 (5.6)
	No change	234 (35%)	NA	6.5 (4.7)
	Reversal	48 (7%)	+12%	9.1 (7.2)

By study-end compared with baseline, 137 (95%) women had experienced a net maturational change of −25% immaturity (median) observed during a mean of 2.7 years of follow-up. In these 137 women, the incremental patterns of change varied. For the majority (n = 91) of women, the net maturation was produced from a mixture of short intervals of maturation and no change, as in Figure 1. A smaller number (n = 46) of women exhibited a net maturation via a mixture of short intervals that included reversal as well as maturation and no change, as in Figure 2. In addition, in our sample, there were two (1%) women who experienced no net change during a mean 2.6 years of follow-up time; and six (4%) women who had a net reversal change of +8% immaturity (median) during a mean of 3.1 years of follow-up.

View Large Image Download to PowerPoint

Figure 1 Selected colpophotographs for one woman with short intervals of maturation or no change: (A) 21% immaturity at Time 0; (B) 8% immaturity at 11 months of follow-up; (C) 4% immaturity at 27 months of follow-up; and (D) 4% immaturity at 31 months of follow-up.

View Large Image Download to PowerPoint

Figure 2 Selected colpophotographs for one woman with short intervals of reversal or maturational change: (A) 44% immaturity at time 0; (B) 61% immaturity at 3 months of follow-up; (C) 37% immaturity at 11 months follow-up; (D) 22% immaturity at 14 months follow-up; (E) 19% immaturity at 29 months follow-up.

Factors that influence cervical change 

The bivariate analyses to examine associations between each predictor variable and the rate of cervical maturation are shown in Table 3. Three factors were significantly associated with an increased rate of maturation: oral contraceptive pill use (p = .05), smoking (p = .01), and frequency of sexual activity (p = .03). Table 4 demonstrates the final multivariate model in which two factors, oral contraceptive use and smoking, remained significant. Oral contraceptive use was associated with an additional 2.3% maturational change for each time increment of 4 months, compared with nonuse of oral contraception (p = .04). For example, an average acceleration of 14% maturational change over 2 years would be gained from continuous use of oral contraception compared with nonuse of oral contraception. Similarly, smoking was associated with an additional 3.9% maturational change for each time increment compared with nonsmoking (p = .01). This would translate to an acceleration of 23% maturational change during 2 years of smoking compared with nonsmoking. Two-way interactions between time and each predictor variable, time-by-oral-contraception, and time-by-smoking were not significant (data not shown). Because the oral contraceptive use may be a proxy for new sexual partners, another multivariate model was created to adjust for the number of new partners, but the results were unchanged (data not shown).

Table 3. Bivariate analyses of the associations between demographic, behavioral, and infectious factors and the rate of cervical epithelial maturationa

	Characteristic	Parameter estimate	Standard error	P Value
	Age	.003	.003	.37
	Ethnicityb
	African-American	–.011	.035	.76
	Asian	.003	.025	.91
	Latina/Hispanic	–.014	.020	.48
	Mixed	–.025	.027	.35
	Age of menarche	–.004	.006	.53
	Age of first intercourse	.002	.004	.65
	Number of lifetime sexual partners at study entry	<.001	.001	.98
	Number of new sexual partners since the previous visit	–.006	.006	.31
	Frequency of sexual activity	–.001	<.001	.03
	Condom usec
	Always	–.009	.014	.50
	Most of the time	–.015	.015	.33
	Half the time	–.011	.018	.53
	Rarely	–.026	.016	.10
	Current pregnancy	.035	.031	.27
	Current oral contraceptive pill use	–.022	.011	.05
	Current medroxyprogesterone injection use	.001	.023	.96
	Current smoking	–.038	.015	.01
	Chlamydia trachomatis infectiond	<.001	.028	.99
	Neisseria gonorrhoeae infectiond	.023	.132	.86
	Trichomonas vaginalis infectiond, e	–	–	–
	Bacterial vaginosisf	–.047	.039	.23
	Candidal infectionf	–.019	.035	.59
	HPV infection by any of 32 typesf	.002	.006	.78
	HPV infection by high-risk typesf, g	–.008	.012	.55
	HPV infection by low-risk typesf, h	.018	.021	.40

HPV = human papillomavirus.

a Rate of cervical epithelial maturation was defined as the change in percent-immaturity between two sequential visits. Each model was adjusted for percent-immaturity at study entry, and the time variable was entered in 4-month increments. b Referent group was Caucasian. c Referent group was nonuse of condoms. d Defined as any infection during the 4–8 months prior to each visit. e Insufficient number of positive infections to generate a parameter estimate. f defined as any infection during the 2-4 months prior to each visit. g High-risk types: 16, 18, 26, 31, 33, 35, 39, 45, 51, 52, 53, 56, 58, 59, 66, 68, 73, 82. h Low-risk types: 2, 6, 11, 40, 42, 54, 55, 57, 67, 69, 70, 72, 83, 84.

Table 4. Multivariate analysis of the associations between behaviors and the rate of cervical epithelial maturation a

	Behavior	Parameter estimate	Standard error	p Value
	Percent-immaturity at baseline	.617	.035	<.01
	Current use of oral contraceptive pill	−.023	.011	.04
	Current smoking	−.039	.015	.01

Akaike information criterion (AIC) = −775.2. Bayesian information criterion (BIC) = −769.3.

a Rate of cervical epithelial maturation was defined as the change in percent-immaturity between two sequential visits. The time variable was entered in 4-month increments.

For subanalyses of the factors associated with reversal, intervals of reversal were selected for comparison to intervals of no change. We selected the 34 reversal intervals during which the percent-immaturity increased at least +5% and excluded the remaining 14 reversal intervals to avoid possible measurement error. Bivariate models indicated that a pregnant woman was more likely to experience reversal than a nonpregnant woman (OR = 9.27, 95% CI = 2.35–36.51, p < .01), whereas an oral contraceptive user was less likely to experience reversal than a non-user (OR = .45, 95% CI = .21–.96, p = .04). These results remained consistent in a multivariate model controlling for smoking and the number of recent sexual partners. Again, pregnancy was associated with an increased odds for reversal (OR = 6.77, 95% CI = 1.56–29.34, p = .01). Oral contraceptive use was associated with a decreased odds for reversal at a marginal significance level (OR = .43, CI = .18–1.01, p = .05).