OUP user menu

Long-term clinical impact of introducing a human papillomavirus 16/18 AS04 adjuvant cervical cancer vaccine in Spain

Aline Gauthier, Victoria Martín-Escudero, Lee Moore, Nicole Ferko, Silvia de Sanjosé, Isabel Pérez-Escolano, Ferrán Catalá-López, Elena Ferrer, F. Xavier Bosch
DOI: http://dx.doi.org/10.1093/eurpub/ckn064 674-680 First published online: 19 July 2008

Abstract

Background: Human papillomavirus (HPV) epidemiology and screening practices vary considerably between countries and specific analyses are required to estimate the impact of HPV vaccination. This study aimed to predict the clinical benefits of introducing a bivalent HPV16/18 vaccine in Spain, where the cervical cancer (CC) incidence is 10.3 per 100 000. Methods: A Markov model based upon the natural history of HPV and CC was developed to simulate transitions between health states, in the presence of specific screening programmes. Published data were used to reflect the Spanish situation in terms of epidemiological characteristics, screening and treatment practices. Calibration consisted of varying disease progression rates within established ranges until model predictions matched observed epidemiological data. The clinical impact of vaccinating a cohort of 12-year-old girls against HPV was assessed over their lifetime using the calibrated model. Results: Vaccination of all 12-year-old girls would result in a reduction of 75% (from 0.32% to 0.08%) in the prevalence of high-grade precancerous lesions due to oncogenic HPV, and a 79% reduction in both CC cases (from 1745 to 365) and CC deaths (from 417 to 86). Assuming a vaccine coverage of 80%, the number of CC cases and deaths would decrease by 63%. Vaccination could also substantially reduce the number of screening tests and treatments required for cervical dysplasia. Conclusion: Our model was successfully adapted to the Spanish epidemiological environment, screening and treatment practices and predicted a substantial long-term benefit of HPV vaccination despite a low HPV prevalence in Spain.

  • papillomavirus vaccines
  • mass screening
  • Spain
  • uterine cervical neoplasms

Introduction

Human papillomavirus (HPV) commonly infects the genital mucosa of sexually active women leading to cervical neoplasia as a complication of the infection.1 Spain has a relatively low HPV prevalence in the general population,2,3 while the prevalence is intermediate among women attending family planning clinics.4,5 A recent world-wide meta-analysis reported an HPV prevalence of 6.8% for Southern Europe and 3% for Spain.6 Spain has an incidence of invasive cervical cancer7 (ICC) of 10.3 per 100 000 which is comparable to other countries such as Sweden, the Netherlands, Canada or the United States. In these countries, extensive screening programmes exist, while in Spain, screening efforts have largely been based on opportunistic demand.8 The low level of HPV-related disease reported in Spain has been attributed to a high proportion of women reporting lifetime monogamy, late age at first sexual intercourse and an absence of large immigration movements for long periods. However, recent data suggest that important changes may occur with more liberal attitudes towards sexual partnership and decreasing age at sexual debut, particularly in younger cohorts.9 Recent migration movements from high-risk countries for cervical cancer (CC) are also taking place at a significant rate. In migrants from Latin America, the prevalence of high-risk HPV has been reported to be 2- to 3-fold greater than in native Spanish.5

In Spain, HPV 16 is the most common type among women with normal cytology or with cervical lesions, while HPV 16 and 18 are the most common types among women with ICC.2

Globally, HPV 16 and 18 account for 65.6% of squamous cell CCs (HPV 16: 54.6% and HPV 18: 11.0%). HPV 16/18/31 and 45 together account for 73.4% of squamous cell CCs.10 A prophylactic vaccine has recently been approved for the prevention of HPV 16/18 infection, demonstrating high levels of efficacy against HPV 16 and 18, and preliminary evidence of cross-protection against HPV 31 and HPV 45.11 In the extended phase of one trial, vaccine efficacy was maintained against HPV infection and pre-cancerous lesions over 6.4 years and protection against other oncogenic HPV types was demonstrated.11–14

As the vaccine was developed only recently, studies assessing the long-term impact of vaccination are not available at present, and mathematical models are needed to simulate the consequences of vaccination. A research programme has been undertaken to produce a generic model to evaluate the benefit of HPV vaccination in various countries. This article aims to describe the adaptation of the model to the Spanish setting and to report the predicted clinical impact of introducing HPV vaccination in Spain.

Methods

Model overview

A mathematical model was developed in Microsoft Excel® to reflect the natural history of type-specific HPV infection and the progression of cervical lesions to cancer. The model is based on a set of mutually exclusive health states corresponding to HPV infection, cervical intraepithelial neoplasia (CIN) lesions and ICC, between which women transition according to age-specific transition probabilities. Further details on model structure are provided in a paper published by Goldie et al.15

Modules

The model is composed of the following three modules:

The natural history module reflects the natural history of HPV infection and its progression to CC in the absence of intervention. The model uses a Markov process based on transitions in 6-month cycles between the following health states: normal (no HPV infection), HPV infection, CIN 1 to 3, CC (Stages 1 to 4) and death. Health states are stratified by HPV type: 16, 18, 31, 45, 52, other oncogenic and low risk. Transition probabilities are assumed to be dependent upon age and HPV type.

The screening module was developed to simulate disease progression in the presence of screening practices. The model reflects the observed compliance to screening via the percentage of women never screened and age-specific screening coverage rates. The clinical management of detected abnormalities is modelled using a decision tree where probabilities associated with each possible procedure capture the variability in tests and treatment pathways. These include repeat test (HPV, cytology), colposcopy, biopsy, treatment and cytological follow-up.16–19

The vaccination module allows evaluation of the vaccination programme. Vaccine efficacy is accounted for by applying reduced infection rates for each HPV type. The model allows consideration of waning vaccine efficacy and the effect of boosters.

Adaptation of the model to Spanish settings

This article focuses on the use of the model in Spain.

Model inputs specific to Spain

Some of the inputs, such as screening coverage rates, treatment practices and stage-specific cancer survival, could be obtained directly from the literature (table 1). National guidelines were used to assess the recommended frequency of screening.20,21 Compliance to guidelines was used to reflect observed screening.22 An observational study of gynaecology clinics was conducted in Spain to delineate the type and probability of occurrence of treatment patterns and diagnostic testing following an abnormal cytology.23

View this table:
Table 1

Model inputs

Model parameterValueReference
Screening patterns
    Start/stop age of screening25, 6420
    Recommended screening interval (in years)320
    Never screened in lifetime (%)20.3021
    Screened every 3 years (% dependent on age)28.9–81.821
Test characteristics
    Cytology—Sensitivity (specificity)0.41–0.67 (0.966)16,46
    Probability of accurate biopsy CIN diagnosis0.53617
    Colposcopy—Sensitivity (specificity)0.96 (0.48)19
Screening practices
    ASCUS to regular screening (%)0
    ASCUS to triage cytology/colposcopy/HPV test (%)64/33/3
    LSIL to triage cytology/colposcopy/HPV test (%)44/48/8
    Greater than or equal to HSIL to colposcopy100
    Negative triage to regular screening (%)100
    Positive triage to colposcopy (%)10022
    Negative diagnosis to routine/increased screening/Dx excision (%)78/17/5
    CIN 1/CIN 2 diagnosis to routine screening (%)7/5
    CIN 1 diagnosis to increased screening/treatment/regular screening (%)28/65/7
    CIN 2 diagnosis to treatment/regular screening (%)95/5
    CIN 3 diagnosis to treatment/regular screening (%)95/5

Calibration

Other model parameters, particularly the probabilities of disease progression, were more challenging to estimate since there is a dearth of published literature on this topic. Thus, a calibration process was carried out. First, a comprehensive review of the literature was completed to determine plausible transition probability ranges. Transition probabilities were varied within the resulting range, until the model-predicted outcomes for unvaccinated girls accurately reflected observed epidemiological data.

The following calibration endpoints and outcomes were used:

  1. Age-specific HPV prevalence: Data were obtained from a population-based study (i.e. mainly women with normal cytology).2

  2. HPV type distribution: HPV-type distribution was not well documented. As there is not much evidence of variation in HPV-type distribution among European countries, lower and upper limits were based on studies conducted in Spain, Italy and the UK. For each state (normal cytology, low- and high-grade lesions and ICC), the proportion of women with each HPV type was taken as the mid-point of the collected values.24–27

  3. CIN prevalence: Published data related to CIN prevalence by age was scarce. This endpoint was estimated from a sample of more than 33 000 cytologies routinely collected from family planning clinics in 1999.28 The prevalence estimates were consistent with figures reported in the bulletin of the Asociación Española de Patología Cervical y Colposcopia.29

  4. Cancer incidence and mortality: Age-specific rates (per 100 000) from GLOBOCAN (2002) were used30; adjusting to the variation observed in more detailed age-specific rates from the International Agency for Research on Cancer (IARC) (1991–97).31,32 Age-specific rates for mortality were gathered from the Instituto Nacional de Estadística (INE) (2003).33

Goodness of fit was evaluated using the mean-error-term, by calculating the percentage deviations between the predicted and observed values, weighted by the age-specific rates. Calibrating the model this way provided a robust foundation for evaluation of the long-term effects of vaccination in the current country-specific screening environment.

Analysis

The objective of this analysis was to estimate the public-health benefits of adopting HPV vaccination in Spain, in terms of HPV and CIN prevalence, CC incidence and mortality, and frequency of screening tests and treatments required for cervical dysplasia.

Base case assumptions

Introduction of the vaccine is likely to influence screening practices over the long term. However, since the extent to which screening practices will change is difficult to predict, this study was conducted in the context of current screening practices. The calibrated model was used to track the clinical events over a lifetime for two comparable cohorts of females that were either vaccinated against HPV or not vaccinated. Rates of screening events were also estimated. The primary objective of this analysis was to assess the impact of vaccinating a cohort of girls, assuming that current screening practices are maintained. As the ultimate objective of a vaccination programme is to vaccinate all girls, a vaccine coverage of 100% was selected for the base case scenario. Moreover, the choice of vaccine coverage facilitates the comparison with published results for other countries.15,34 Vaccine efficacy was assumed to be 95% against HPV 16/18, 90% against HPV 45 and 50% against HPV 31.35 The study conducted by Harper et al.11 showed absence of waning up to 4.5 years after vaccination and this was assumed to remain true over a lifetime.

It was assumed that 12-year-old girls would be vaccinated, according to the recommendations made by the Spanish Health Authorities.36 This would ensure that the vast majority of girls are vaccinated before they commence sexual activity37 and, furthermore, such a strategy could easily be implemented as part of the vaccination calendar for children. A cohort of 209 116 girls was analysed, representing an age 12 cohort in Spain.38

Sensitivity analysis assumptions

Different types of parameter uncertainty exist, including variability in long-term vaccine efficacy, decision policy surrounding age at vaccination, and clinical data uncertainty. Alternative assumptions were explored in the sensitivity analysis. Vaccine efficacy for HPV types 16 and 18 was varied from 90% to 100%, in accordance with the 95% confidence interval obtained from clinical trial results at 5.5 years for the bivalent HPV vaccine.12,39 We also assessed the effects of excluding efficacy against non-16/18 oncogenic types, and of waning protection against HPV 31 and HPV 45 (assuming a linear waning pattern to 0% of initial efficacy after 10 years). This latter scenario was tested under two assumptions concerning whether or not there was a repeat vaccination (booster shot) after 10 years. The impact of vaccinating older and younger girls was also tested (10 and 14 years).

Although the optimal goal of vaccination programmes is to vaccinate the entire population of interest, full compliance may be unlikely. The vaccine coverage was varied to 80%, which is the rate that has been achieved for the hepatitis B vaccine, administered in school for adolescents.40 Finally, because of the uncertainty surrounding the estimates of CIN1 prevalence in Spain (perceived as relatively low by clinical experts), a sensitivity analysis using the probability of transitioning from HPV to CIN 1 from the UK model34 was performed. This article reports the potential impact of introducing the vaccine over the cohort lifetime in terms of CC cases, CC deaths, pre-cancerous lesions and HPV prevalence related to oncogenic HPV (16/18/31/45/52 and other high-risk types).

Results

Model calibration

Following calibration, the model predictions matched Spanish observed epidemiology data (figure 1). The percentage deviations were 3.9, 7.1 and 17.0% for HPV prevalence, cancer incidence and mortality, respectively. The percentage deviation for overall CIN prevalence was 2.3%. Model-predicted CIN prevalence (CIN 1 = 0.67%, CIN 2 and 3 = 0.28%) was very close to observed data (CIN 1 = 0.68%, CIN 2 and 3 = 0.29%). HPV-type distribution was well simulated within normal cytology, cervical lesions and cancer, where values remained within the range observed from published studies.24–27 Using 2002 demographic data, the model predicted 2024 CC cases due to any HPV type, compared with 2103 observed cases7 (3.76% difference). CC mortality also showed strong correspondence with observed data, with the model predicting a crude average of 2.83 compared with a reported rate of 2.82 per 100 000.33

Public health impact of vaccination

Estimated age-specific clinical results are presented in figure 2 for the non-vaccinated and vaccinated cohorts. Over the lifetime of the 12-year-old cohort, the model predicted the occurrence of ∼1745 CC cases and 417 cancer deaths without vaccination. With vaccination, the forecast would be as low as 365 cancer cases (79.1% reduction), and 86 cancer deaths (79.4% reduction).

Vaccination also predicted large reductions in the prevalence of high-grade lesions (CIN 2 and CIN 3) across all ages, with an estimated 95% reduction in the prevalence of lesions associated with HPV 16 and 18 and a 75% reduction in the prevalence of lesions associated with oncogenic HPV types (from 0.32% to 0.08%). Oncogenic HPV prevalence was estimated to fall from 2.30% to 1.26% (45% reduction). For all HPV types the projected reduction in HPV prevalence was 35% (from 2.96% to 1.92%). CIN 1 prevalence was estimated to fall from 0.57% to 0.31% (46% reduction) when considering oncogenic HPV types.

Additional benefit was predicted in terms of reductions in the incidence of screening tests and treatments for cervical dysplasia. Approximately 13 000 fewer abnormal cytology tests were predicted (51% reduction), while the frequency of biopsy, colposcopy and treatment of CIN lesions decreased by 53, 54 and 54%, respectively.

Sensitivity analyses

Table 2 details the results of the sensitivity analyses for the more severe cervical outcomes. Model predictions consistently suggested high efficacy of vaccination in terms of reductions in the number of CC deaths, although estimates varied from 75.3% (for 90% efficacy) to 83.7% (for 100% efficacy).

View this table:
Table 2

Base case and sensitivity analysis results

ScenarioCIN 2 and 3 Prevalence (%) (reduction)Cervical CC n(%) (reduction)Cervical cancer deaths n(%) (reduction)
Scenario 0: no vaccine0.321745417
Scenario 1: base case0.08 (75.0)365 (79.1)86 (79.4)
Scenario 2: low efficacy (HPV 16/18)0.09 (71.9)436 (75.0)103 (75.3)
Scenario 3: high efficacy (HPV 16/18)0.07 (78.1)295 (83.1)68 (83.7)
Scenario 4: no cross protection0.10 (68.8)430 (75.4)100 (76.0)
Scenario 5: vaccine waning (HPV 31, 45 types)0.09 (71.9)427 (75.5)99 (76.3)
Scenario 6: vaccine waning (HPV 31, 45 types) and booster at 10 years0.09 (71.9)422 (75.8)97 (76.7)
Scenario 7: vaccination coverage (80%)0.13 (59.4)641 (63.3)152 (63.5)
Scenario 8: lower age at vaccination (10 years)0.08 (75.0)365 (79.1)86 (79.4)
Scenario 9: higher age at vaccination (14 years)0.10 (68.8)386 (77.9)94 (77.5)
Scenario 10: HPV to CIN transition probabilities from UK model0.07 (78.1)282 (83.8)66 (84.2)

A vaccine without cross-protection would result in a smaller reduction in CC incidence (75.4%). Vaccination of 14-year-old girls was slightly less beneficial in terms of reduction in CC incidence (77.9%), while no extra benefit was predicted by the model through vaccination of 10-year-old girls.

A vaccine coverage rate of 80% offered a reduction in the number of CC deaths of 63.5%. Application of HPV to CIN 1 transition probabilities specific to the UK, resulted in greater reductions in CIN prevalence, cancer incidence and death (78.1, 83.8 and 84.2% reduction, respectively), due to the higher rate of transition from HPV to CIN 1 in the UK. Application of UK transition probabilities resulted in a higher estimated impact of the vaccine; therefore the conservative approach is to use the Spanish transition probabilities.

Discussion

Clinical trials have established a high efficacy of an HPV 16/18 AS04 adjuvant prophylactic CC vaccine against HPV infection and pre-cancerous lesions at 4.5 years.11,35 To evaluate the long-term impact of HPV vaccination, a detailed model reflecting the natural history of HPV infection and CC was developed. Given that CC epidemiology and screening policies vary across countries, it was important to adapt the model to the country of interest. The present study investigated the clinical impact of introducing the vaccine in Spain alongside current screening practices.

The calibration exercise resulted in a model that had a close fit to the HPV and CIN prevalence, ICC incidence and mortality observed in Spain. Calibration of the model provided a robust foundation for evaluation of the long-term impact of vaccination in the current screening setting. Perhaps the most striking prediction made using the model was that the vaccination strategy would provide a 75.0% reduction in the prevalence of high-grade cervical lesions due to oncogenic HPV, and a 79.4% reduction in CC deaths.

Results were sensitive to changes in assumptions for certain parameters. Vaccination strategies in terms of age group and vaccine coverage were investigated. Vaccination programmes targeting 10- or 12-year-old girls appeared to represent more beneficial strategies than those targeting 14-year-old girls, while the projected benefit was lower when assuming 80% vaccination coverage. Preliminary evidence of cross protection of the HPV 16/18 AS04 adjuvant CC vaccine against HPV 31 and HPV 45 has been demonstrated recently,35 and the clinical consequences were assessed by providing results in the presence and absence of cross-protection, illustrating the additional benefit of this further protection.

While HPV prevalence rates in Spain (3%) are lower than those observed world-wide and in Europe (10.4 and 8.1%, respectively6), the percentage reduction in CC incidence implied by the vaccine strategy is still substantial. The Spanish results are comparable to those obtained in similar studies when differences in screening practices and epidemiology are taken account of. In similar model adaptations for the United Kingdom,34 Italy,41 Germany42 and the United States,43 the vaccine strategy was estimated to result in a reduction in the number of CC cases of between 78 and 83%.

Spanish health authorities recommend vaccinating cohorts of young girls (aged between 11 and 14) and these cohorts’ ages have been explored in our analysis. Vaccination policies that include a wider population (older girls and boys) have been considered within the context of Italian and Danish HTAs.44,45 The Danish national board of health concluded that it was more advantageous to vaccinate a wider population of girls (e.g. 13–19 years) than to include boys in the vaccination programme. The Italian HTA unit concluded that although vaccination of older women (aged 21–23) was more beneficial in the short term, vaccination of younger girls (aged 12–14) resulted in better clinical results over the cohort lifetime, further justifying the focus of this study on these younger girls.

This model has two great advantages. First, screening and treatment practices observed in Spain are well reflected in the model. Second, the process of calibration to numerous epidemiological endpoints ensured that the model was capable of reproducing the natural progression of disease in Spain. With these two factors taken into account, the model provided a realistic framework to simulate the introduction of the HPV vaccine in the current Spanish settings.

There are some limitations to this analysis. First, calibration is conducted using cross-sectional data as it cannot take into account future changes to CC risk factors. This approach has been used in previous CC modelling efforts.15,46,47

Ideally, longitudinal data would be used since a single cohort is being modelled over time. However such data do not exist, and if they did, they still might not accurately represent a future cohort, given the evolution of risk factors over time. Second, long-term vaccine efficacy is uncertain and will only be established over the next several decades. We have addressed this uncertainty by conducting sensitivity analyses. Third, the natural history of multiple HPV infections was not explicitly modelled. Future modelling efforts that consider multiple infections will provide a better understanding of the impact of HPV 16/18 vaccination on type-specific HPV prevalence over time. Fourth, the model does not investigate the potential impact of HPV vaccination on pre-cancerous lesions, and neoplasms in other sites, such as the vulva, vagina, anus and penis. A recent study attributed HPV 16 and/or 18 to be the cause of cancer in these sites in 63% to 92% of cases.48 By disregarding these HPV-related diseases, we may have potentially underestimated the clinical benefits of HPV 16/18 vaccination. Fifth, the predicted impact of vaccination may be underestimated since our modelling does not account for transmission and epidemic dynamics. Through the effects of herd immunity, one would expect a reduction in HPV prevalence among males if girls were vaccinated. Thus, a secondary beneficial effect of vaccination, which is not taken into account in our analysis, would be to slow down the growth of the HPV epidemic. However, Edmunds et al.49 have shown that the difference between static and dynamic models was relatively low for high levels of vaccine coverage. Last, studies have suggested that CC incidence is increasing for successive birth cohorts in Spain,50 which is not reflected in our model baseline assumptions. This would imply a greater potential for vaccination to prevent CCs.

In the current screening setting in Spain, the adjuvant prophylactic CC vaccine is highly effective against HPV-infection and pre-cancerous lesions associated with HPV 16/18 and other oncogenic types. The vaccine therefore has potential for reducing the number of cervical screening tests required and the need for dysplasia treatment. This would be welcome in Spain where the number of cytological tests performed has increased steadily since 1993, due to opportunistic screening.51 Our results demonstrate that over 50% of abnormal cytology tests, diagnostic tests and CIN treatments could be averted with a type-specific vaccine. Reducing the number of screening tests performed is important not only from an economic perspective, but also to reduce the distress women face in association with positive tests.52 There is the possibility that cervical screening policy will be altered over the coming decade with the introduction of a vaccine,53 and studies have begun to hypothetically assess the optimal combination of vaccination plus screening: results support maximizing cost-effectiveness in a policy combining HPV vaccination, screening initiation at a later age and screening tests conducted less frequently.43 Further analyses assessing how the characteristics of the screening programme (age range, frequency and compliance) affect the clinical outcomes in Spain would be needed.

HPV vaccination has been recommended by health authorities in many European countries, including France, where the mortality from CC is lower than in Spain, and a more stringent screening programme is in place (covering larger age groups and recommending more frequent screening tests).

Mathematical models provide a means of exploring the long-term impact of vaccination upon clinical outcomes. Calibration of the model to the Spanish setting was performed to obtain robust estimates. This analysis suggests that the public-health benefits of HPV 16/18 vaccination within the context of the existing screening programme may be substantial in Spain, as the model demonstrated large reductions not only in CC incidence and mortality, but also in the prevalence of pre-cancerous lesions and associated diagnostic tests and treatments.

Acknowledgements

This study was carried out independently by i3 Innovus under a grant from GlaxoSmithKline Biologicals. The authors are indebted to the Pathology Laboratory of Hospital Creu Roja de L’Hospitalet for providing the cytology information used in this analysis. All authors approved the manuscript prior to submission.

Conflicts of interest: S.S. has received travel funds from GlaxoSmithKline, Merck, Sanofi Pasteur MSD and Digene, and has independently assessed the data provided in this manuscript. E.F. has received an unrestricted grant from GSK. F.X.B. served on steering committees for Merck and Sanofi Pasteur MSD, is an external adviser for GlaxoSmithKline, and has received travel funds or honoraria from GlaxoSmithKline, Merck, Sanofi Pasteur MSD and Digene. His research unit is involved in vaccine trials organized by GlaxoSmithKline, Merck and Sanofi Pasteur MSD.

Key points

  • Several studies predicting the clinical impact of introducing an HPV vaccine in different countries have been published.

  • No study reported the impact of introducing the vaccine in Spain, a country in which HPV prevalence is low.

  • The present study consisted of adapting a generic model to Spain (HPV related epidemiology, screening policies and clinical practice) to estimate the impact of introducing an HPV vaccine.

  • The model was calibrated to ensure that observed Spanish epidemiological data were reproduced in the absence of vaccination.

  • The model predicted a 79% reduction in both CC cases and deaths, and a substantial reduction in the number of screening tests and treatments required for cervical dysplasia.

References

View Abstract