Background. When the bivalent and the quadrivalent HPV vaccines were marketed they were presented as having comparable efficacy against cervical cancer. Differences between the vaccines are HPV types included and formulation of the adjuvant. Method. A systematic review was conducted to assess the efficacy of the two vaccines against cervical cancer. Outcomes considered were CIN2+, CIN3+, and AIS. Results. Nine reports (38,419 women) were included. At enrolment mean age of women was 20 years, 90% had negative cytology, and 80% were seronegative and/or DNA negative for HPV 16 or 18 (naïve women). In the TVC-naïve, VE against CIN2+ was 58% (95% CI: 35, 72); heterogeneity was detected, VE being 65% (95% CI: 54, 74) for the bivalent and 43% (95% CI: 23, 57) for the quadrivalent. VE against CIN3+ was 78% (95% CI: <0, 97); heterogeneity was substantial, VE being 93% (95% CI: 77, 98) for the bivalent and 43% (95% CI: 12, 63) for the quadrivalent. VE in the TVC was much lower. No sufficient data were available on AIS. Conclusions. In naïve girls bivalent vaccine shows higher efficacy, even if the number of events detected is low. In women already infected the benefit of the vaccination seems negligible.

1. Introduction

In the seventies, Dr. Harald Zur Hausen firstly postulated the link between human papillomaviruses (HPVs) and cervical cancer: studies to develop an anticancer vaccine followed. Approximately 70% of cervical cancers worldwide are associated with two high-risk HPV types (16/18) [1, 2] and almost 90% of genital warts are associated with two low-risk HPV types (6/11). Each year around 500,000 women develop invasive cervical cancer worldwide, with 83% of new cases and 85% of deaths occurring in developing countries [3, 4]. Risk factors associated with HPV infection are younger age at first coitus, higher number of sexual partners, smoking cigarettes, and history of herpes simplex virus infection [5, 6]. Some 75% of sexually active women develop a HPV infection [7], more frequently soon after their sexual debut: the majority of these infections (between 70% and 90%) spontaneously clear [810]. A minority progress from acute infection to cervical cancer, a process taking decades and going through precancerous lesions named cervical intraepithelial neoplasia (CIN) of increasing severity, from CIN1 to CIN3; spontaneous regression of the lesions is possible at any point [4, 8]. Incidence of genital warts is less precisely known, due to lack of data on the general population, but it is estimated to be around 1% [8].

Based on the results of five randomized controlled trials (RCTs) involving 40,000 women [1117] two HPV vaccines entered the market. The quadrivalent vaccine against HPV6/11/16/18 was approved by the FDA and the EMA in 2006 [18, 19], whereas the bivalent vaccine against HPV16/18 was first approved in Europe in 2007 [20] and then in the USA in 2009 [21]. Soon after, several western countries such as the USA, Australia, and five European states [22] started national immunization campaigns targeting adolescent girls. The number of countries adopting the vaccines has since increased: in April 2014, 23 out of 29 European countries were reported to have implemented it [23], budgetary constraints being one relevant obstacle for the remaining countries. The primary target of HPV vaccination is adolescent girls aged 11 to 13 years, with some minor differences in national recommendations: in the USA routine vaccination is recommended at age 11 or 12 years with quadrivalent or bivalent vaccine for females and with quadrivalent vaccine for males in a 3-dose schedule during a 6-month interval [24]; in the UK HPV vaccine is recommended for girls under 15 years of age and consists in two injections spaced at least six and not more than 24 months apart. The vaccine is also recommended for men who have sex with men [25]. No prior assessment with Pap testing or screening for existing HPV infection is required.

Both vaccines contain human papillomavirus L1 self-assembling virus-like particles and are not infectious. Differences between the two vaccines are the number of HPV types included and formulation of the adjuvant (Table 1) possibly leading to different vaccine efficacy (VE) [3638]. Head-to-head comparisons between the two vaccines are still exclusively based on immunogenicity [38, 39], although an immune correlate of protection has not yet been established [40].

Three meta-analyses, published when vaccines were marketed, showed comparably high efficacy of the two vaccines against precancerous lesions associated with HPV16/18 [30, 41, 42]. The meta-analyses subsequently published [43, 44] have confirmed the high efficacy of the vaccines against lesions associated with HPV16/18, but they did not provide information about VE against any cervical lesions irrespective of HPV type, nor did they discuss possible differences in terms of efficacy of the two vaccines. Thus, our systematic review that includes studies with longer follow-up aims to assess differences between the two vaccines from a public health perspective, considering all cervical lesions.

2. Methods

The systematic review was developed based on a prespecified protocol (protocol number FARM8N2ZFL) funded by the Italian Medicines Agency (AIFA) within a program of independent research on drugs [45]. The PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) statement guided the content and reporting of the review [46]. Published and unpublished RCTs comparing any of the two HPV vaccines versus placebo or any other control were considered for inclusion. We exclusively considered studies involving women, irrespective of age at enrolment.

As for protocol, primary outcome measures were cervical lesions (i.e., cervical cancer, CIN2, CIN3, and AIS) associated with any HPV type and cervical lesions exclusively associated with HPV16/18 occurring in three study populations: according to protocol population (ATP), the general population of vaccinated women (total vaccine cohort (TVC)) approximating all women regardless of status of HPV infection at vaccination, and a selected population of women seronegative for HPV16/18 and HPV DNA negative for 14 oncogenic HPV types, approximating the group of young adolescent girls targeted in the national immunization campaigns (TVC-naïve). In this paper only data related to cervical lesions associated with any HPV type occurring in the TVC and in the TVC-naïve population are reported, as the other data are not relevant from a public health perspective.

2.1. Literature Search

Trial identification: we searched the Cochrane Library (to Issue 3, 2014), MEDLINE (to March 2014), and EMBASE (to March 2014) using keywords and MeSH terms as reported in Annex 1 online, in combination with a highly sensitive filter for identifying RCT [47]. There were no language or time restrictions. Reference lists of relevant papers were also examined to identify additional studies. For unpublished RCTs, we searched the Internet for prepublication study presentations at conferences or meetings. Moreover experts in the field and vaccine manufacturers were contacted for further information (unpublished studies and single patient data). Clinical trial registers were searched for ongoing studies. Two review authors independently screened abstracts of potential studies and retrieved full articles for those deemed eligible.

2.2. Study Selection

Two reviewers carried out independent assessment of citations retrieved. When more than one publication reported the same trial, the one with a longer follow-up was selected. Reasons for exclusion were recorded. Quality of trials was assessed using the criteria outlined in the Cochrane Handbook [47] and included the assessment of (i) generation of the randomization sequence, (ii) quality of the allocation concealment, (iii) completeness of follow-up, and (iv) blinding of the outcome assessment. Based on quality assessment the risk of bias of the included trials was defined as low, high, or unclear. Differences in opinion were resolved through discussion involving a third author if needed.

2.3. Statistical Analysis

Two review authors extracted the data independently using a data extraction form. Statistical analyses were carried out using the STATA software version 11. For time to event data the hazard ratio (HR) was used as a measure of association. The results were summarized by using the inverse of variance method and the random effects model. Point estimates as well as their 95% confidence intervals (95% CI) were calculated and represented by the forest plot. Vaccine efficacy (VE) was calculated as , where RU is the rate of disease in the unvaccinated and RV is the rate in the vaccinated [48]. The equation can be rewritten to use the HR in the following way: , where HRv/u is the HR of the vaccinated versus the unvaccinated. Heterogeneity among studies was assessed using the statistics [47].

2.4. Subgroup Analyses

The following prespecified subgroup analyses were planned:(i)Geographical areas (Europe, Africa, Asia, North America, and South and Central America): the literature suggests that prevalence and circulation of HPV high risk types varies according to geographical areas [4951].(ii)Vaccine type (bivalent, quadrivalent): data on VE can be influenced by type of vaccine used as differences between the two formulations could be not negligible.

3. Results

3.1. Selection of Studies

Study identification and selection process is outlined in Figure 1. Of the 726 records initially identified, 3 were duplicates and 670 were excluded based on title and abstract assessment. The most common reasons for exclusion were the following: reports were not RCTs, did not assess the outcome of interest, were not related to oncogenic HPV or to vaccine administration, or were studies reporting exclusively laboratory or immunogenicity data. We assessed the full text of 53 articles and excluded 44 (reasons reported in Figure 1). Thus nine reports [2629, 3135], corresponding to five registered protocols [5256], were included in the systematic review: in three trials (20,797 women) the bivalent vaccine was used (PATRICIA trial being the larger study) [2729] and in two trials (17,622 women) the quadrivalent vaccine was assessed (FUTURE I and FUTURE II trials) [26]. Five publications reporting subset of data of the above-mentioned trials relating to specific geographical areas were also included [3135].

The data of the phase III trials included in our meta-analysis have a mean follow-up of 4 years for the bivalent trial [27] and 3.6 years for the quadrivalent trial [26], longer than that of the systematic reviews with meta-analysis published up to now.

3.2. Quality of Included Trials and Characteristics of Trial Participants

Characteristics of the trials are summarized in Table 2.

Studies were all double blind RCTs, with an adequate sample size. They were all manufacturer-sponsored. Eligible participants were healthy, not pregnant women aged between 15 and 26 years, with 6 or less lifetime sexual partners and no history of abnormal Pap smear at enrolment. Almost 90% of women had normal cytology at study entry. Risk of bias was low in 4 out of 5 trials: generation of the randomization sequence, quality of allocation concealment, completeness of follow-up, and blinding of outcome assessment were adequate. Risk of bias was unclear (generation of the randomization sequence and allocation concealment were not described) in one smaller trial conducted in Japan [29]; as the Japanese trial provided only data on CIN2+ lesions exclusively associated with HPV16/18, its results are not reported here. Characteristics of the women enrolled in the trials used in this meta-analysis are reported in Table 3.

3.3. Pooling of the Data

Pooling of the data was possible for CIN2+ [2628], CIN3+ [26, 27], and AIS [26, 28]. Data for each outcome are presented below for the TVC and the TVC-naïve cohorts.

CIN2+. The pooled HRs for CIN2+ lesions associated with any type of HPV in the TVC and in the TVC-naïve are reported in Figure 2. Corresponding values of VE were 26% (95% CI: 11, 39) in the TVC [26, 27] and 58% (95% CI: 35, 72) in the TVC-naïve [2628]. Results suggested substantial heterogeneity among bivalent and quadrivalent vaccines: test was 68.7% and 66.4% in the TVC and in the TVC-naïve, respectively.

CIN3+. The pooled HRs for CIN3+ lesions associated with any type of HPV in the TVC and in the TVC-naïve are reported in Figure 3. Corresponding values of VE were 32% (95% CI: <0, 56) in the TVC [26, 27] and 78% (95% CI: <0, 97) in the TVC-naïve [26, 27]. Results suggested substantial heterogeneity among bivalent and quadrivalent vaccines: test was 86.3% and 90.7% in the TVC and in the TVC-naïve, respectively.

AIS. The pooled HR for lesions associated with any type of HPV was 0.31 (95% CI: 0.14–0.70) in the TVC [26, 27], corresponding values of VE being 69% (95% CI: 30, 86).

AIS cases in the TVC-naïve cohort were zero in the vaccine group and ten in the placebo; thus only an approximate estimate of the efficacy was possible. The pooled HR for lesions associated with any type of HPV was 0.01 (95% CI: 0.01–0.22) resulting in a VE of 99% (95% CI: 78, 99) [26, 27]. For all the comparisons suggested low heterogeneity.

3.4. Analysis by Geographical Area

Although formally required by the scientific advisory unit of ECDC in Stockholm, the two manufacturers did not provide single patient data (the authors received only partial data unsuitable for the analysis from GSK and no answer at all from Sanofi Pasteur MSD).

Five published papers [3135] reported data according to geographical areas. Pooling of the data was not appropriate since geographical areas definition differed (Table 4).

4. Discussion

Since their introduction into the market, the effectiveness of the two vaccines against cervical cancer, based on first published data [1117], has been subject of debate. Enthusiastic positions assumed that if vaccines are immunogenic and prevent infections associated with HPV16/18 they also prevent cervical cancer and therefore should be widely used [5759]. Uncertainty was related to the following issues: correlation between immune response and clinical outcomes, need for a booster dose, replacement with other oncogenic strains, and possible reduction of Pap-test screening among the vaccinated [6065].

Our systematic review highlights that, for precancerous lesions (CIN), the only available proxy of cervical cancer and heterogeneity among pooled studies is substantial (Figures 2 and 3): the bivalent vaccine shows higher efficacy against precancerous lesions. We focus our comments on the TVC-naïve cohort, as VE in the TVC is confirmed to be much lower and HPV vaccination is not universally offered to women already sexually active and data on AIS are too sparse to make sensible comments. For CIN2+ lesions estimates of efficacy of the two vaccines in the TVC-naïve cohort differ but the wide limits of the confidence intervals partially overlap (VE 65%; 95% CI: 54, 74 for the bivalent and VE 43%; 95% CI: 23, 57 for the quadrivalent), whereas for CIN3+ lesions estimates of efficacy largely differ and the limits of the confidence intervals do not overlap (VE 93%; 95% CI: 77, 98 for the bivalent and VE 43%; 95% CI: 12, 63 for the quadrivalent). However, due to limited number of patients with lesions detected, leading to wide confidence intervals of our estimates of effect, our conclusions should be interpreted with caution. The heterogeneity observed might be due to higher efficacy of the bivalent vaccine against cervical cancer possibly related to the specific adjuvant used (ASO4 adjuvant system), as suggested in two recent immune response head-to-head studies that consistently showed a higher neutralizing antibody production [38, 39] and a higher CD4+ T cell response [38] in bivalent than in quadrivalent vaccine recipients. Heterogeneity can also be due to baseline differences between the populations enrolled in the two trials, although such differences were not reported in the two trials (Table 3). However, misclassification of naïve women cannot be ruled out since the two manufacturers used different laboratory tests to measure immune response and to identify naïve girls. In fact cLIA test was used in the FUTURE trials whereas ELISA test that has a higher sensitivity than cLIA [66] was used in the PATRICIA trial. Moreover, data are often differently and poorly reported in the published trials [2629, 3135]; thus our ability to make meaningful comparisons and further analysis (e.g., assess the possible effect modification by smoking status or age) is hindered. We asked the manufacturers to provide individual patient data, but we did not receive a positive answer.

Another relevant point is that the length of the follow-up in the trials assessed seems insufficient to detect information relevant to the public and to policy-makers: as time interval from HPV infection to cervical cancer development is approximately 20 years, all information gathered in a much shorter period of time is not conclusive. Nevertheless the PATRICIA trial has a planned follow-up of 4 years [27] and longer follow-up data on bivalent vaccine are only available for 436 Brazilian women enrolled in a previous phase II trial [67]. We will have more information in 2020, when the results of the Finnish study that extended the follow-up for Finnish girls enrolled in the PATRICIA study will be published [68] and when the extension studies for the FUTURE II trial assessing the quadrivalent vaccine will also be available. In the meantime an open debate in this respect is urgently needed: national health agencies should set up a surveillance system to provide data on actual vaccines efficacy in the field and to allow international comparisons. At the moment, this comparison is not possible and we still do not know how to choose between the two vaccines [69]. Contrary to what is previously reported in other meta-analyses [30, 4144], our systematic review suggests that the quadrivalent and the bivalent vaccines differ in terms of efficacy. This could be attributable to the different adjuvants contained in the two vaccines. Such difference in efficacy has not been widely recognized by national health agencies. For example, in Italy HPV vaccines are chosen and purchased through tendering schemes organized by regional health authorities that are based on the lowest price [70].

Apart from cervical cancer prevention, quadrivalent vaccine is known to effectively prevent genital warts [71, 72], whereas the bivalent vaccine can only marginally impact on these benign but distressing lesions [73]. The UK and a few Italian regions have recently substituted the bivalent with the quadrivalent vaccine, assuming comparable efficacy of the two vaccines against cervical cancer and giving an additional value to the activity against genital warts of the quadrivalent vaccine [10, 74]. Unfortunately, it is not possible to anticipate the consequences of the UK’s and Italian choice: if different vaccine efficacy of the bivalent and the quadrivalent vaccine is going to be confirmed, lack of equivalence in terms of cervical cancer prevention could become an issue. The availability in the coming years of new broader spectrum HPV prophylactic vaccines could provide more insight into the current debate [75, 76].

5. Conclusions

In conclusion, we acknowledge that this systematic review has some limitations due to a low number of women with events and a high heterogeneity among trials that suggest caution in the interpretation of results. However, our conclusions are consistent with those from recent immunogenicity head-to-head studies [38, 39] and that provides strength to our interpretation. Our systematic review suggests that after nine years since HPV vaccines were introduced, their estimates of efficacy seem to diverge over time. The decision to consider the two vaccines similar in terms of cervical cancer prevention seems challenged by our longer term follow-up analyses. This might have implications for policy and pragmatic choices and deserves an open and comprehensive discussion. Moreover, international comparisons of the actual effectiveness of the two vaccines used in the field can add valuable information.

Finally, regulatory agencies should encourage the pharmaceutical companies to provide data across trials to assess all relevant outcomes in a comparable way, to reduce uncertainty, and to support health policy-makers that have to choose between alternative options [77, 78].

Conflict of Interests

The authors declare that no competing interests exist.


The authors are grateful to Maurizio Bonati, Head of Department of Public Health, Mario Negri Institute, Milano, Italy; Vittorio Demicheli, Head of Department of Epidemiology SEREMI-Cochrane Vaccines Field, ASL Alessandria, Italy; and Maria Luisa Moro, Head of Regional Health and Social Agency, Bologna, Italy, for providing useful feedback and comments on the final version of the paper. They also want to thank Maria Chiara Bassi for developing and running the search strategy and Christine Kieran for revising and editing the English text.

Supplementary Materials

The supplementary material shows the search strategy that was adapted for each database searched.

  1. Supplementary Material