Population-Based Age-Period-Cohort Analysis of Declining Human Papillomavirus Prevalence

Abstract

Background

Most countries in the world have launched human papillomavirus (HPV) vaccination programs, and declining HPV prevalences are reported. We aimed to disentangle the influences of calendar time, birth cohort, and age by analyzing HPV prevalences in the population-based cervical screening program using age-period-cohort modeling.

Methods

All 813 882 primary HPV-based cervical screening tests from women aged 23–64 years between 2014 and 2023 in the capital region of Sweden were identified in the Swedish National Cervical Screening Registry. The odds ratio (OR) of HPV-16/18 infection was estimated comparing birth cohorts to the unvaccinated 1984-born using an age-period-cohort model. The impact of changing HPV prevalences on the number needed to screen (NNS) to detect and prevent 1 cervical cancer case was calculated.

Results

HPV vaccination coverage was 82%–83% among women born in 1999–2000. Before 2019, the HPV-16/18 prevalence was highest among the youngest women. During 2020–2023 the prevalence consistently decreased among the birth cohorts offered organized school-based vaccination. There was a 98% decline in HPV-16 prevalence (OR, 0.02 [95% confidence interval {CI}, .01–.04]) and a 99% decline in HPV-18 prevalence (OR, 0.01 [95% CI, .00–.04]) among the 2000-born compared to the 1984-born. The declining HPV-16/18 prevalences resulted in major increases in the NNS to detect and to prevent 1 case of cervical cancer.

Conclusions

The declines of HPV-16/18 were considerably larger than the vaccination coverage, suggesting herd immunity. The changing epidemiology of HPV types impacts screening needs, necessitating updated screening programs.

Infection with human papillomavirus (HPV) is a necessary cause of cervical cancer [1]. HPV has historically been extremely common, with most women acquiring HPV infections shortly after sexual debut [2]. In 2006–2007, 2 highly efficacious HPV vaccines (with vaccine efficacies >90% against new HPV-16/18 infections) were licensed [3]. Both provide some cross-protective efficacy against HPV-31 and one of them also against HPV-33 and -45 [4]. HPV vaccination at the population level provides herd protection (indirect effects) to the entire population [5]. The degree of herd protection depends on the vaccination coverage, the vaccine efficacy, and the basic reproduction number (R₀) of the HPV type. The higher the R₀, the higher the vaccination coverage required to reach the critical vaccination coverage threshold needed to eliminate the transmission of the HPV type [6]. Mathematical modeling has consistently found that the vaccination coverage needed for elimination is higher for HPV-16 than HPV-18. One modeling study found the critical vaccination coverage to be 95% and 82% for HPV-16 and HPV-18, respectively, when only females are vaccinated [6].

In Sweden, subsidized “on demand” HPV vaccination commenced in 2007 targeting 13- to 17-year-old girls. However, it achieved low coverage and was unequally distributed according to socioeconomic class [7]. In 2012, organized school-based HPV vaccination began using the quadrivalent vaccine for 11- to 13-year-old girls (born in 1999 and younger), resulting in high and socioeconomically even uptake [7, 8], with additional catch-up vaccination for girls born from 1993 to 1998 (girls aged up to 19 years). In 2020, the HPV vaccination program became gender-neutral, including HPV vaccination of boys. A nationwide trial of concomitant cervical screening and HPV vaccination using the nonavalent vaccine among women born between 1994 and 1999 commenced in 2021 (Supplementary Figure 1) [9].

Prior to HPV vaccination, HPV-16 and -18 were common with the prevalence peaking in the early 20s and declining with age [10, 11]. After HPV vaccination, there was a decrease in HPV-16/18 prevalence among the vaccinated birth cohorts (in comparison to the prevaccination prevalence) and an early decline in condyloma incidence among women aged <25 years [12].

Sweden has population-based organized HPV screening with HPV genotyping targeting women aged 23–64 years, enabling real-time monitoring of type-specific HPV prevalence. To quantify whether the HPV prevalences were changing as expected, we compared the HPV vaccination coverage by birth cohort over time in Sweden with the population-based HPV prevalences by HPV type from 2014 until 2023 among the cervical screening population.

METHODS

HPV vaccination coverage among female residents in Sweden was obtained from the Swedish National Vaccination Register (at the Public Health Agency of Sweden) [13]. Since 1 January 2013, it has been mandatory for all healthcare providers to report vaccinations to the national vaccination register (during 2006–2012 there was voluntary reporting of HPV vaccinations).

The women attending primary HPV-based cervical screening from 1 January 2014 until 31 December 2020 in the Stockholm region were identified in the Swedish National Cervical Screening Registry (NKCx) [14]. Primary HPV-based screening began in 2012, first with a randomized healthcare policy trial of women aged 56–60 years [15], expanded to women aged 30–64 years in August 2014 [16]. In 2017, primary HPV testing was implemented for all women aged ≥30 years, and in autumn 2019, HPV-based cervical screening was extended to all women aged 23–64 years. In 2020, the cervical screening program switched to self-sampling sent directly to the women's homes as the primary screening method for women aged 26–64 years (Supplementary Figure 1) [17]. Women aged 23–25 years were still invited to have a screening sample taken by a midwife. To supplement the study, a random selection of screening samples from 23- to 29-year-old women from 2014 was retrieved from the biobank of the screening program [10] and HPV tested by the same method used in the program.

Laboratory Analyses

The cervical screening samples collected from 2012 onward were collected using liquid-based cytology (ThinPrep) and HPV genotyped using Cobas 4800 (detecting HPV-16, HPV-18, and pooled HPV-31/33/35/39/45/51/52/56/58/59/66/68). The random sample of biobanked cervical screening samples from 2014 were tested using the Cobas 4800 [11]. On 1 April 2022, HPV testing switched from partial HPV genotyping to extended HPV genotyping using BD Onclarity (detecting type-specific HPV-16/18/31/45/51/52 and pooled HPV-33/58, HPV-35/39/68, and HPV-56/59/66).

Data Collection

All primary screening samples in the Stockholm region from 2014 to 2020 were identified in the NKCx, and data from 2021 to 2023 were identified in the centralized screening laboratory information database. For the year 2021, we merged data from the 2 data sources using their unique personal identification number.

Samples with an invalid test result or taken outside the recommended ages for HPV-based screening were excluded from the analysis. Only samples from the organized HPV-based screening program were included (opportunistic screening samples were not included). Only primary screening samples were included (follow-up samples precipitated by the index sample were excluded).

Statistical Analyses

The HPV vaccination coverage (of having received at least 1 vaccine dose) was extracted by birth cohort and calendar year. Likewise, the prevalence of HPV-16 and -18 was estimated among the women attending primary HPV-based screening stratified by birth cohort and year from 2014 until 2023. The accompanying 95% confidence intervals (CIs) to the prevalence estimates were calculated using the Agresti and Coull method [18]. HPV-16/18/31/45/51/52 prevalence and pooled HPV-33/58, HPV-35/39/68, and HPV-56/59/66 prevalences in 2023 were estimated stratified by birth cohort grouped by eligibility for school-based HPV vaccination (the 1999–2000-born), catch-up HPV vaccination (the 1993–1998-born), and those ineligible for any organized HPV vaccination (the 1952–1992-born).

To investigate trends by age, period, and cohort, the prevalence of HPV-16/18 was plotted by (1) age stratified by year, (2) age stratified by birth cohort, (3) calendar year stratified by age, and (4) birth cohort stratified by age, grouped into 2-year intervals.

To evaluate the impact of the changing HPV prevalence on the cervical screening program, we estimated the number of women needed to screen (NNS) in the target population “to detect” and “to prevent” 1 case of invasive cervical cancer caused by a specific HPV type (HPV-16/18/31/45/51/52 and pooled HPV-33/58, HPV-35/39/68, and HPV-56/59/66), stratified by the prevaccination, catch-up HPV vaccination, and school-based HPV vaccination birth cohorts according to the methodology of Wang et al [11]. The 95% CIs were estimated using bootstrap resampling of 2850 cervical cancer cases. To calculate the above impact numbers, we utilized data from the Swedish population registry to calculate the population size, data from the Swedish National Audit of cervical cancer cases in 2002–2011, data from the NKCx to estimate the incidence rate of cervical cancer by a given HPV type among screened and unscreened women, and data from the current study on type-specific HPV prevalence over time according to birth cohort (see more details in the Supplementary Material).

We investigated the effect of birth cohort on HPV prevalence using a binomial age-period-cohort model fitted with natural splines to assess the odds of HPV-16 or -18 positivity by birth cohort relative to a reference birth cohort (women born in 1984, among whom the HPV vaccination coverage was negligible), with the period effect being constrained to zero on average [19]. For the age-period-cohort modeling, we recalculated the HPV-16/18 prevalence stratified by age, calendar year, and birth cohort into 1-year wide triangles. The age, period, and year used was the mean of the triangle. Thus, for a lower triangle the age was offset by 1/3 of the lower age in the age interval (ie, for a lower triangle with age of 23 years, the mean age would be equal to 23 + [1/3]), the calendar year was offset by 2/3 of the 1-year-long interval, and the birth cohort by 1/3 of the 1-year-long birth cohort interval. All statistical analyses were performed using R statistical software (version 4.4.0).

RESULTS

In total, 730 402 HPV-based screening samples were identified in the NKCx (complete up to 31 December 2021) and 456 985 samples were identified from the central screening laboratory database up to 29 December 2023. Among those, 850 317 were primary HPV-based screening samples. We excluded 705 samples due to an invalid person ID, 3037 samples taken prior to 2014, 3228 invalid samples, 7 duplicate samples, and 29 458 samples from women outside the HPV-based screening ages. Finally, there were 813 882 samples from women 23–64 years old attending primary HPV-based cervical cancer screening in the Stockholm region from 2014 to 2023 (Figure 1).

The HPV vaccination coverage (at least 1 dose) was highest among women eligible for school-based vaccination (1999-born and onward), reaching 81.5% and 82.7% among the 1999- and 2000-born, respectively, by 2016. HPV vaccination coverage among the birth cohorts eligible for catch-up HPV vaccination was low from 2007 until 2011, but steadily increased until 2016, ranging from 55.1% to 62.9% among the 1993–1998-born. Among the youngest birth cohort (2010-born), coverage reached 91.7% among females and 88% among males (Table 1, Figure 2).

From 2014 to 2018, the HPV-16 prevalence among women attending primary HPV screening was highest among the youngest birth cohorts, gradually decreasing among the oldest cohorts (Table 2). HPV-16 prevalence among the random selection of women aged 23–29 years attending screening in 2014 was 6.11% (95% CI, 4.43%–8.37%; 36/589 women) (Figure 2). In 2019, the peak HPV-16 prevalence by age shifted to those born in 1986 or 1987 (P = 4.44% [95% CI, 3.55%–5.54%] and P = 4.68% [95% CI, 3.80%–5.77%], respectively) (Figure 2). By 2023 the HPV-16 prevalence was lowest among the youngest birth cohorts. The overall prevalence was 2.59% (95% CI, 2.48%–2.71%) among the 1958–1992-born ineligible for organized HPV vaccination, 2.25% (95% CI, 2.05%–2.47%) among the 1993–1998-born (the catch-up HPV vaccination cohorts), and 0.65% (95% CI, .50%–.85%) among the 1999–2000-born (Figure 2, Table 1, Supplementary Tables 1 and 4). Among the youngest birth cohorts entering screening at 23 years old, the HPV-16 prevalence decreased further from 7.53% (95% CI, 4.13%–13.1%; 11/146 women) in 2014 to 1.81% (95% CI, 1.48%–2.22%; 93/5130) in 2021 among the 1998-born, 0.67% (95% CI, .41%–1.08%; 17/2530) in 2022 among the 1999-born, and 0.54% (95% CI, .34%–.85%; 19/3511) in 2023 among the 2000-born. Among the women born in 1988 to 1952 there was a 1.5-fold increase in the HPV-16 prevalence in 2023 compared to 2022.

The HPV-18 prevalence was generally lower than the HPV-16 prevalence (Table 2, Supplementary Tables 1 and 2). From 2014 until 2020, the HPV-18 prevalence was highest among women aged 30–35 years (the youngest women who were eligible for HPV-based screening at the time). Among the random sample of women aged 23–29 years, the HPV-18 prevalence was high (2.38% [95% CI, 1.38%–3.99%]; 14/589 women) (Figure 2). In 2023, the HPV-18 prevalence was highest among the older birth cohorts at 0.69% (95% CI, .63%–.75%) among the 1958–1992-born who were ineligible for organized HPV vaccination, 0.57% among the 1993–1998-born who were eligible for the catch-up HPV vaccination, and 0.15% among the 1999–2000-born who were eligible for school-based HPV vaccination (Supplementary Table 3). Similarly, among the youngest birth cohorts entering the screening program, the HPV-18 prevalence decreased dramatically from 2.74% (95% CI, .83%–7.07%) in 2014 to 0.66% (95% CI, .47%–.93%) in 2021, 0.12% (95% CI, .02%–.37%) in 2022, and 0.09% (95% CI, .02%–.26%) in 2023, with only 3 of 3511 women being HPV-18 positive (Table 2, Figure 2).

The prevalence of all the non-vaccine-targeted HPV types measured (types 31, 45, 51, 52, 33/58, 35/39/68, and 56/59/66) in 2023 was highest among the youngest age groups except for HPV-31 (Table 3). The HPV-31 prevalence decreased from the 1997-born to the 2000-born, from 3.10% (95% CI, 2.48%–3.86%) among the 1997-born, to 2.04% (95% CI, 1.55%–2.66%) among the 1998-born, to 1.64% (95% CI, 1.32%–2.04%) among the 1999-born, and finally to 1.31% (95% CI, .98%–1.75%) among the 2000-born (P value for trend = .0424) (Table 3).

The number of women who need to be screened to detect 1 case of HPV-16–associated invasive cervical cancer (NNS_detect) among the prevaccination HPV vaccination birth cohorts was 24 518 (95% CI, 21 594–28 004), but increased 9.36-fold among the school-based HPV vaccination birth cohort to 229 377 (95% CI, 202 136–262 182). Similarly, the NNS_detect for HPV-18–associated invasive cervical cancer was 49 036 (95% CI, 39 801–63 603) among the prevaccination birth cohorts but increased 15.8-fold among the school-based HPV vaccination birth cohorts in comparison to the NNS_detect among the prevaccination birth cohorts (773 085 [95% CI, 631 410–1 005 942]) (Table 4).

Likewise, the number of women needed to be screened to prevent 1 case of HPV-16–associated invasive cervical cancer (NNS_prevent) in the next 10 years was 4747 (95% CI, 3938–5967), but increased 2.70- and 9.35-fold among the catch-up and school-based HPV vaccination birth cohorts, respectively, compared to the prevaccination scenario, to 12 828 (95% CI, 10 653–16 191) among the 1993–1998-born and to 44 406 (95% CI, 36 874–56 046) among the 1999–2000-born (Table 4). In the prevaccination birth cohort scenario, the NNS_prevent 1 case of HPV-18–associated cervical cancer was 50 908, already 10.7 times higher than the NNS_prevent for HPV-16. Among the catch-up HPV vaccination and school-based HPV vaccination birth cohorts, the HPV-18–specific NNS_prevent increased 4.15- and 15.8-fold to 211 211 (95% CI, 83 293–∞) and 802 603 (95% CI, 316 512–∞) women, respectively (Table 4).

When plotting the HPV-16/18 prevalence by age, period, and cohort, the prevalence mostly changed according to age with the prevalence being highest among younger women, except among the youngest birth cohorts (the 1991–2000-born), which deviated from the trend (Supplementary Figures 2 and 3). When comparing the adjusted odds of HPV-16 infection among each birth cohort to the odds among women born in 1984 (the reference birth cohort with negligible vaccination coverage), the odds of HPV-16 and HPV-18 infection already began to decrease among the women who were eligible for opportunistic “on demand” HPV vaccination, continuing to consistently decrease with each consecutive birth cohort until among the 2000-born, when the odds ratio (OR) of HPV-16 and -18 infection was 0.02 (95% CI, .01–.04) and 0.01 (95% CI, .00–.04), respectively, in comparison to the HPV-unvaccinated reference birth cohort (Figure 3, Supplementary Figures 4 and 5, Supplementary Table 5).

DISCUSSION

Many countries have already reported dramatic declines in the HPV-16/18 prevalence postvaccination. However, it is difficult to disentangle effects of age, birth cohort, and calendar time. In this study we used an age-period-cohort model to estimate the overall decline in the HPV prevalence. Among the youngest birth cohort attending the screening program who were eligible for organized school-based HPV vaccination, we found the HPV-16 prevalence to have dropped by 98%, while HPV-18 prevalence dropped by 99%.

Despite the vaccination coverage being only 83% among the 2000-born women (with 2000-born men ineligible for school-based HPV vaccination), there were just 3 women with HPV-18 infections out of 3511 women. Given that sexual activity tends to be assortative by age and the average direction of HPV transmission chains move from the older to the younger birth cohorts over time, this finding among the 2000-born suggests that Sweden is rapidly approaching the scenario where HPV-18 is nearing stochastic extinction among the youngest HPV-vaccinated birth cohorts [6, 9, 20]. In comparison, 19 of the 3511 women had an HPV-16 infection, in line with the findings of mathematical modeling that HPV-18 is easier to eliminate than HPV-16 [21].

Among the women eligible for organized school-based HPV vaccination at the age of 11–12 years, the vaccination coverage was consistently high, ranging from 82% to 92%. Prior to 2019 we observed that the HPV-16 and -18 prevalence was highest among the youngest women in the screening program who typically have the highest number of new sexual contacts, in line with earlier studies reporting the age-specific prevalence in the prevaccination era [10]. When the HPV-vaccinated birth cohorts began entering the screening population, the prevalence among the youngest began to decrease, becoming more apparent when the cohorts eligible for catch-up HPV vaccination (the 1993–1998-born) entered the screening program and then consistently decreasing with each consecutive cohort entering.

Most HPV-vaccinated women in Sweden received the quadrivalent vaccine, which has a reported cross-protective efficacy against HPV-31 of 46% [4]. The critical vaccination coverage is lower for HPV-31 than for HPV-16 or -18; however, with a cross-protective efficacy of 46% and only girls being HPV vaccinated, modeling results suggest that HPV-31 would be impossible to eliminate [6]. In line with this, a reduction in HPV-31 prevalence was observed among women who received school-based HPV vaccination, in comparison to those who received catch-up or no HPV vaccination (with 45 cases of the 3511 women from the youngest birth cohort). However, the other low oncogenic HPV types, which are not targeted by the quadrivalent vaccine, showed the same characteristic age-specific prevalence trend (decreasing with age) as was seen for HPV-16 and -18 in 2014 [11]. No decrease in the prevalence of nonavalent vaccine–targeted HPV-52 or HPV-33/58 has been observed among the 1994–1999-born women eligible for the elimination trial.

A second peak in HPV-16 prevalence appeared in 2023 among older women between the ages of 50 and 68 years, which had not been present in previous years [10, 11]. This increase coincides with the change of HPV genotyping platform from Cobas to BD Onclarity, which has a higher sensitivity and targets another region of the HPV genome. The timing suggests that the emergence of a second peak in HPV-16 may reflect changes in the sample testing methodology rather than a real change in the prevalence. A second peak in the age-specific prevalence of HPV-16 has been reported in several (but not all) countries and may be due to new acquisition of infection and/or the detection of an old HPV infection previously under the limit of detection [22, 23]. We cannot explain with certainty why the second peak was seen only for HPV-16 and only during 2023. The consistency of this finding and how it may impact screening needs further investigation.

Although HPV screening has been recommended by the World Health Organization since 2014, implementation has been slow. Surprisingly, most of the debate about HPV screening has not considered the fact that different oncogenic HPV types have very different cervical cancer risks (varying 40-fold, from an OR of 48 for HPV-16 to an OR of 1.2 for HPV-51) [24]. Quantifying how type-specific HPV prevalences are changing could thus be useful to inform screening strategies in vaccinated populations. The NNS to detect or prevent 1 case of cervical cancer was found to differ drastically by HPV type [11]. We found that among the birth cohorts who were eligible for school-based HPV vaccination, the NNS “to detect” and the NNS “to prevent” 1 case in the next 10 years were both 9-fold and 16-fold higher in comparison to the historical numbers for HPV-16 and -18–associated cervical cancer, respectively. As treatment of screening-positive women also contributes harms (ie, an increased risk of preterm birth among women treated with conization [25]), this suggests that the screening intensity among the birth cohorts having received school-based HPV vaccination may need to be lessened and/or management strategies could be more conservative.

Our findings of dramatically decreased HPV-16 and -18 are in line with the findings of high HPV vaccine effectiveness in countries where vaccination has been implemented for more than a decade with high vaccination coverage [14, 26]. The use of HPV screening data enables timely monitoring of the HPV prevalence among the target population also when the infection is so rare that it is approaching extinction (as per the Dahlem criteria for the successful eradication of an infectious pathogen) [27]. Indeed, the HPV-18 prevalences were so low that a traditional cross-sectional survey would have been underpowered to detect it.

Our study is generalizable to the Swedish female population, as both the screening and school-based vaccination programs have high population coverages with limited regional variation [14]. Organized, population-based screening and vaccination programs are known to achieve greater equity in participation across socioeconomic groups [7, 28]. Gender-neutral HPV vaccination is known to induce stronger herd effects than girls-only vaccination; thus, the HPV-16/18 reduction may have occurred earlier if HPV vaccination would have been gender-neutral at initiation [6]. Our study is limited as it reports only vaccination coverage among women having received at least 1 dose of an HPV vaccine. However, the HPV vaccination coverage does not differ substantially if the coverage is defined as among women having received at least 2 doses; among 2010-born females, the 1-dose HPV vaccination coverage was 92%, whereas the 2-dose coverage was 90% [13].

As a proof-of-principle for the elimination of vaccine-targeted HPV, our study may be generalizable to other high-income countries similar to Sweden. However, for the feasibility of global eradication of HPV-16 and -18, our study falls short in proving the operational feasibility. For proof-of-principle of HPV-16 and -18 eradication globally, it will be necessary to conduct such an elimination initiative in at least 2 countries where the difference between the vaccination coverage threshold and the vaccine coverage is the highest, the health infrastructure is weaker, where it is less geopolitically stable, and where there is less gross domestic product per capita and political support for such an initiative [29].

In conclusion, among women from older unvaccinated birth cohorts, the HPV-16/18 prevalence was affected largely by age, with prevalence highest among younger women aged <30 years. However, among birth cohorts targeted by school-based HPV vaccination in a country with vaccination coverage of >80%, HPV-18 is disappearing to a point where stochastic extinction is likely, while HPV-16 is now rare. Our study suggests that elimination of HPV-18 transmission within a defined population is possible. The dramatic decline in the HPV-16 and -18 prevalence suggests that benefits of cervical screening may be lower in birth cohorts targeted by school-based HPV vaccination.

Supplementary Data

Supplementary materials are available at The Journal of Infectious Diseases online (http://jid.oxfordjournals.org/). Supplementary materials consist of data provided by the author that are published to benefit the reader. The posted materials are not copyedited. The contents of all supplementary data are the sole responsibility of the authors. Questions or messages regarding errors should be addressed to the author.

Notes

Author contributions. P. G., J. W., K. M. E., and J. D. conceptualized the study. S. N. K., K. M. E., and P. G. performed data curation and project administration. P. G. and J. W. performed the formal analysis. J. D. was responsible for funding acquisition. K. M. E. and J. D. supervised the project. P. G. wrote the original manuscript draft. All authors reviewed and edited the manuscript.

Acknowledgments. The authors thank Dr Tiia Lepp from the Public Health Agency of Sweden for providing the vaccination coverage data from the Swedish National Vaccination Register.

Data availability. The data used in this study are available from the Swedish National Cervical Screening Registry upon reasonable request and ethical review board approval (https://nkcx.se/research_e.htm).

Ethics statement. This study was granted ethical approval by an ethical review board in Stockholm from the Swedish Ethical Review Agency (a government agency), with the decision number 2011/1026-31/4. All cervical screening data are stored in the NKCx by default and screening samples (from the Stockholm region) are stored in a biobank; women have the right to opt out or retrieve their own data at any time as per the General Data Protection Regulation.

Funding. This study was funded by the Horizon 2020 Framework Program (HEAP, project number 874662); Avtal om Läkarutbildning och Forskning (ALF, RS2021-0855); and Center for Innovative Medicine (CIMED; grant number 976486) from Region Stockholm.

References

Author notes