Incident cervical infections with high- and low-risk human papillomavirus (HPV) infections among mothers in the prospective Finnish Family HPV Study

Human papillomavirus (HPV)-types that infect the female genital tract belong to the alpha-papillomavirus-genus which includes 15 species and 58 HPV-genotypes. According to their clinical behavior, 15 HPV-types are high-risk (HR)-types, 12 are low-risks (LR)-HPV-types, and three probable HR-types [1, 2]. Worldwide, the eight most common HPV-types in cervical cancer are all included either into species7 (HPV18,45) or 9 (HPV16,31,33,35,52,58) [3]. The studies on incident HPV-infections have been evaluating HR- and LR-types collectively or HPV in general [4‐7] or acquisition of incident HPV-infections have been assessed at genotype level [8‐16].

Among young women, incident infections with HR-HPV-types especially with HPV16 [8‐11, 14‐16] seem to be more common than LR-types [9, 17]. The knowledge on type specificity and factors that increase or decrease the risk of incident HPV-infections is important to better understand the dynamics of HPV-infections and to take appropriate measures for their optimal prevention.

The aim of this study was to assess the frequency of type-specific incident HPV-infections in addition to actuarial and crude incidence-times and -rates of the most common LR- and HR-HPV- genotypes in newly delivered mothers, prospectively followed-up for 6 years in the Finnish Family HPV Study. Risk-factors of incident HPV-infection were analyzed with univariate and multivariate Poisson regression for panel data.

This is one of the first studies to assess type-specific incident HPV-infection [8‐16], in contrast to the previous studies, where acquisition of HPV-infection has been analysed collectively for all HPV-types or acquisition of HR- and LR-types has been compared [4‐7]. Similarly, in currently available literature, it is hard to find the times to the first incidence infections at the genotype-level, most reports being focused on duration of incident infections [8, 9, 16].

We previously analysed crude incidence times collectively for 13 HR-HPV-types detected by Hybrid capture 2-assay [6]. It took an average of 16.6 months to develop the first incident HR-HPV event. In the present study, we went further and provide these times for individual HPV-genotypes (and species) both as actuarial and crude format. To calculate the former, included are also baseline HPV-negative women who never develop an incident event, while the latter is calculated only for women who develop incident events. Thus, the actuarial times reflect the real-life situation, while indicating how long it takes in a cohort of baseline HPV-negative women to develop the first incident infections. On the other hand, the crude times indicate the time that these first incident infections take to develop among women who all experience an incident event. These crude (type-specific) incidence times provide a robust measure to compare different HPV-genotypes in their speed to develop the first incident events (Table 2). In this respect, individual genotypes showed marked differences in their crude incidence times, varying within the range of 2 to 90 months, with the mean of 20.3 and median of 13.5 months (data not shown). Generally speaking, LR-HPV-types had shorter crude mean incidence times than HR-HPV-genotypes. However, despite the significant overall difference (p = 0.0001) between the individual genotypes, these results are hampered by multiple comparisons and a small number of cases in each strata. Of interest is the observation that no significant differences were observed between mean incidence times of HPV16, HPV18 and multiple-type infections, each being represented by sufficient number of cases.

As evident in Figure 3, these incident events show a tendency for clustering around the time points of the control visits, i.e., at 2-, 12-, 24-, 36- months and 6 years. This flaw is inherent to the study design, however, and equally affects all studies with similar design, i.e., HPV sampling at irregular (and relatively long) intervals. Because of the fact that the incident events rarely (if ever) occur exactly at the time point of sampling, but somewhere in between the two subsequent samplings, and because the event leads to censoring of the subject (= no longer at risk for incident event), we can speak about interval censoring. Due to the fact that there are both "left-censored and right-censored" (relative to the point) data that occur at random, we feel that this should not cause a major bias, however. Of course, the exact times can only be obtained by very short (1 to 3 months) sampling times, which in the case of HPV (having protracted clinical course) are not feasible in practice.

Also the IR can be calculated as actuarial and crude, as done here. These two IRs are markedly different, because of the significantly different person months at risk. This also makes comparison between individual studies difficult, particularly when the FU-times and cohort size are substantially different. If one compares a setting where 1000 women are followed-up for 10 months (10.000 wmr), and another one with 100 women followed-up for 100 months (10.000 wmr), certainly the IRs in the latter are higher. This is because the median incidence time is around 12-13 months and only a fraction of events can be detected with a short FU-time. In addition, it is essential to know the age-profile of the cohort, because IR for HPV infection is critically age-dependent [7].

In our cohort, actuarial IR was by far the highest for HPV16, 15.6/1000 wmr. This is consistent with other studies, where HPV16 had the highest actuarial IR, albeit the absolute IRs of these cohorts was much lower. A young women's health study-cohort from USA reported an actuarial IR of 5.9/1000 wmr for HPV16 [9]. In the Ludwig-McGill low-income cohort, the actuarial IR for HPV16 was 1.4/1000 wmr [8], which is very similar to another cohort consisting of sexually active women from Hawaii, 1.77/1000 wmr [16]. Due to the dominant role of HPV16, also the actuarial IR of species 9 was the highest 17.9/1000 wmr, which is three times higher than recently published by Goodman and co-workers. (5.4/1000 wmr) [16]. As pointed out above, these differences are explained by different cohort sizes and particularly by much shorter FU-times of the other studies, 10-15 months only [8, 9, 16]. Another factor is the different age-profile; the mean age of our cohort was 25.5 years which is close to 24.2 years in the study by Giuliano and co-workers [9] where as the mean age in the study by Goodman and co-workers was 35 years [16]. The IRs are higher in our cohort and Giuliano cohort [9] than in studies on older women [8, 16], which gives further support to close age-dependence [6, 7].

This is one of the first studies on crude IRs, which reflect the differences in the rates at which individual genotypes accumulate incident events among women who have these events. Although firm conclusions on single genotypes are difficult because of the rarity of many individual genotypes, it is interesting to note that the crude IR of the key LR-HPV-types (HPV6,11), is almost twice as high as that of the two main HR-types (HPV16,18). This indicates that the former develop incident infections at much higher rate, which is also shown by their significantly shorter times to incident events (Table 2). Using rate-ratio (RR) statistics, one cannot detect significant differences between the different HPV-species in their crude IRs (data not shown). Much larger cohort is needed to establish whether such differences exist between individual HR-HPV-genotypes. If confirmed, however, such data would have important implications e.g. in the selection of the most aggressive genotypes with high IRs as targets of the primary prevention.

As to the risk-factors for incident species 7/9 infections, seven variables were significant predictors in univariate analysis. First of these was linked with HPV-serology [25]; mothers who failed to seroconvert to HR-HPV during the FU were at increased risk for incident infections (Table 1). It is well established that serological response to HPV-infections needs several months to become detectable [25, 29]. Thus, women who experience an incident event close to the end of FU have not enough time to become seroconverted during the observation period. There is some evidence that seroconversion may fail with transient and even with some persistent infections [30]. It is commonly believed that high levels of HPV-antibodies detected after seroconversion will protect against a new infection. Thus, seroconversion detected during the early months of FU could implicate an incident event that took place before the baseline visit, and these high antibody levels, except for being incompatible with new incident events, might also neutralise the virus and make these women baseline HPV DNA-negative.

Initiation of smoking beyond 13 years of age was associated with an increased risk of incident HPV-infection, but lost its significance in multivariate model. Some studies have found no association between smoking and incident infection [10, 16], whereas others indicating current smoking to increase the risk [5, 11, 13, 15]. Our recent analysis disclosed that initiation of smoking at < 13 years of age increased the risk of type-specific persistence [31]. Because incident infections precede viral persistence, the present data are consistent with this observation, implicating that women who start smoking later have had less time to develop persistent infection, and consequently, are at the phase when incident infections accumulate.

Women who started their sexual activity before 13 years of age were found to have a lower risk for incident infections during the FU. Also this association is in line with our recent analysis disclosing early age of sexual debut as a risk factor for HR-HPV persistence [31]. It is likely that women with later onset are at increased risk for incident infections, just because those with longer sexual exposure have already experienced infection, and either cleared or developed a persistent infection by the time they reach their 20s (mean age of this study cohort being 25.5 years).

The same analogy applies to life-time number of sexual partners (also a proxy to HPV exposure), shown in this study to be inversely related to the risk of incident infections, i.e., lower partner number was associated with an increased risk. It seems likely that those with more partners have already experienced such an event before enrolled in the study, i.e., are baseline HPV+ and by definition, no longer at risk for incident infection by the same HPV genotype.

The initiation of OC usage before the age of 20 increased the risk for incident infection and was a significant predictor also in the multivariate model. Previous studies report conflicting results; current use of OCs has been shown to increase [11] or decrease [4] the risk of incident infection. In addition, it has been reported that the current users of OCs were not in increased risk and that the past users of OCs seemed to be protected, as shown by a decrease of incident infections with the years of OC use [16]. Contradictory to the previous finding, a recent study stated that the risk for incident infection increased with increasing years of OC use [13]. It is obvious that further studies are needed to evaluate the association of OC with incident infections.

Another significant protective predictor was the second pregnancy during the FU (Table 1). No previous data are available from a similar setting where newly delivered mothers were prospectively followed-up. We have recently shown that women committed to the second child did not share many of the known life-style behavioural risk factors of HPV-infection [32], and this could be the likely explanation for this significant (IRR = 0.32,95%CI0.17-0.61) protective effect of a new pregnancy against incident species 7/9 HPV-infections in the present analysis. Some recent data suggest that parity was protective especially against LR-HPV-types [10]. In our cohort, increasing parity did not show any such effect. However, in a large screening study (the NIS Cohort), ever being pregnant was an independent predictor of incident infection [6].

The main limitations of this current study are firstly the small cohort size of the women; the total number of incidence infections was 203. To gain more information on genotype specific incident HPV-infections, times and rates, and to increase the statistical power, a much larger cohort is needed. Secondly the women were all pregnant at the time of enrolling to this study, as it is the special strength of this study it also may be considered as a limitation because the detailed mechanism how pregnancy and HPV infections interacts are not that well understood.

Data obtained from this study indicate that the probability of mothers, who test HPV-negative just before delivery, to contract an incident infection by species 7/9-genotypes during the subsequent follow-up is less likely among those who report more than 2 life-time sexual partners at baseline, start using of OCs after age 20, and become pregnant for the second time during the follow-up. To the best of our knowledge, this is the first study where actuarial and crude times to the incident events as well as actuarial and crude IRs are evaluated at the genotype- and HPV-species level. Obtaining genotype-specific data during long-term follow-up is needed for better understanding of the natural history of genotype specific HPV-infections, e.g. the times and rates at which each genotype accumulates incident events. It sounds feasible to reason that in the future, those genotypes with shortest times to incident infections (i.e., the highest IRs), could be the most appropriate targets for early prevention, e.g. by genotype-specific multivalent HPV vaccines.