Human papillomavirus type distribution in 30,848 invasive cervical cancers worldwide: Variation by geographical region, histological type and year of publication

Pooled data on human papillomavirus (HPV) type distribution in invasive cervical cancer (ICC) can help to predict the potential impact of HPV type‐specific vaccines and screening tests, and to understand the carcinogenicity of HPV types. We performed a meta‐analysis of HPV type‐specific prevalence data published from 1990 to 2010, including a total of 243 studies and 30,848 ICC. The proportion of ICC associated with HPV16 and/or 18 (HPV16/18) was between 70 and 76% in all world regions except Asia. In Western/Central Asia, 82% of ICC was HPV16/18‐associated compared to only 68% in Eastern Asia. The 12 most common HPV types identified, in order of decreasing prevalence, were HPV16 (57%), 18 (16%), 58, 33, 45, 31, 52, 35, 59, 39, 51 and 56. The prevalence of other types, phylogenetically related to those above, ranged from <0.1% for HPV85 to 0.6% for HPV68. Overall HPV prevalence increased significantly from 85.9% in studies published from 1990 to 1999 to 92.9% in studies published from 2006 to 2010. Prevalence increases were large for multiple infections (from 4.0 to 15.7%) and for HPV16 (from 51.8 to 60.0%, including HPV16 alone or in multiple infections). Smaller but significant increases in prevalence were also seen for HPV39, 53 and 58. A large amount of recently published data has improved the understanding of the contribution of a broad range of HPV types to ICC in different world regions. However, estimating the fraction of ICC attributable to different types is increasingly complicated by the detection of multiple HPV infections in ICC.

Since the establishment of human papillomavirus (HPV) as the central cause of invasive cervical cancer (ICC), 1 data on HPV type distribution in ICC have proven useful to predict the potential impact of HPV16 and 18 vaccines, as well as to determine priorities for inclusion of carcinogenic HPV types in future HPV vaccines and HPV-based screening tests. 2,3 A previous meta-analysis including 14,595 cases confirmed that a majority of ICC in all world regions is HPV16 and/or 18 (HPV16/18)-related. 3 These two types, together with the less frequently detected HPV31, 33, 35, 39, 45, 51, 52, 56, 58 and 59, were recently classified as carcinogenic to humans, 3 and they cluster together in a few species (alpha-species 5, 6, 7, 9 and 11) that are phylogenetically related (i.e., they belong to the same evolutionary branch or clade of the mucosotropic HPV alpha-genus). 4,5 This clade has therefore been referred to as the high-risk clade. 5 Many of the rarest types of the high-risk clade have not been studied systematically. However, the number of studies available and the sensitivity of polymerase chain reaction (PCR)-based HPV testing protocols have increased over time, as has the ability to detect a broad range of HPV types and multiple infections.
The aims of our study were thus to update previous metaanalyses with the large amount of new data on HPV typespecific prevalence in ICC published in 2006-2010 and to evaluate whether HPV type-specific prevalence has changed during 20 years of evolving HPV DNA testing protocols.

Material and Methods
The detailed methods used for this meta-analysis of HPV type-specific prevalence in ICC have been reported previously. 2,3 In brief, Medline was used to search for citations published from January 1990 to February 2010 using combinations of the following MeSH terms: ''cervical cancer'', ''cervical intraepithelial neoplasia'', ''HPV'', ''human'', ''female'' and ''polymerase chain reaction''. Additional relevant references cited in retrieved articles were also evaluated. Included studies had to meet the following criteria: (i) use of PCRbased technology to detect HPV DNA and (ii) reporting of type-specific prevalence for at least one HPV type. For the purpose of our study, ICC refers to squamous cell carcinoma (SCC), adeno/adenosquamous carcinoma (ADC) and cervical cancer of unspecified histology, but does not include carcinoma in situ. For each included study, the following key information was extracted: country of sample, sample size, distribution of cancer cases by histological type (SCC, ADC or unspecified), HPV DNA source (fresh/fixed biopsies or exfoliated cervical cells), PCR primers used to detect HPV-positive samples, and type-specific and overall prevalence of HPV DNA. Studies were classified into seven broad geographical regions (Africa, Eastern Asia, Western/Central Asia, Europe, North America, South/Central America and Oceania). Crude type-specific prevalence is presented for all HPV types in the high-risk clade (i.e., HPV16,18,26,30,31,33,34,35,39,45,51,52,53,56,58,59,66,67,68,69,70,73, 82 and 85) 5 other than HPV97, which has been too recently discovered to be included in our study. HPV6 and 11 from alpha-species 10 (the most well-studied non-carcinogenic types) were also included for comparison purposes. Each HPV type was evaluated independently of all others. Type-specific prevalence was estimated only among those studies testing for the HPV type in question, and thus denominators of prevalence estimates vary by type. Unless otherwise specified, therefore, type-specific prevalence includes that in either single or multiple HPV infections. However, the prevalence of single and multiple infections could be evaluated separately in a subset of studies from which such data were available. Owing to the predominance of SCC worldwide, ICC of unspecified histology was combined with SCC for comparison of HPV typespecific prevalence by SCC and ADC.

Statistical analysis
HPV prevalence in SCC and ADC was compared by logistic regression, adjusted, when appropriate, for region (Africa, Eastern Asia, Western/Central Asia, Europe, North America, South/Central America, Oceania), year of publication (1990-1999, 2000-2005, 2006-2010), specimen type (fresh/fixed biopsies, exfoliated cervical cells) and PCR primer used (GP5þ/6þ, MY09/11, PGMY09/11, SPF10, combination of the above, other). The statistical significance of trends in HPV prevalence by year of publication was assessed by considering the categorical variable as a continuous variable in the logistic model. Because of multiple statistical testing, only p values equal to or smaller than 0.01 were considered significant. Trends in single versus multiple infections with any HPV type, HPV16 and HPV18 were assessed in a subset of studies in which all 12 carcinogenic HPV types were tested for, and information on whether an individual type was found in single or multiple infections could be retrieved.

Results
Two hundred forty-three studies met our inclusion criteria, including a total of 30,848 ICC (Table 1). This constituted a gain of 16,253 ICC over a previous meta-analysis. 3 The proportional gain was greatest for Asia, particularly for Western/ Central Asia (from 386 to 2,051 ICC) where five countries were newly represented. Thirty-eight percent of all ICC came from Eastern Asia, 29% from Europe, 10% from South/Central America, 8% from North America, 7% from Western/ Central Asia, 7% from Africa and 2% from Oceania ( Table  1). The proportion of cases diagnosed as SCC, ADC and unspecified histology varied by region, with the proportion of ICC classified as ADC ranging from 3.8% in Western/Central Asia up to 23.7% in North America ( Table 1).
The prevalence of 26 selected HPV types are shown overall and by histological type in Table 2. Overall, prevalence of any HPV type, HPV16 and HPV18 were 89.9%, 56.6% and 16.0%, respectively. Among the studies reporting the relevant information, overall prevalence of single and multiple infections were 79.0% and 11.2%, respectively. The 12 most common HPV types identified were, in order of decreasing prevalence, HPV16, 18, 58, 33, 45, 31, 52, 35, 59, 39, 51 and 56. Prevalence of other types belonging to the high-risk clade ranged from less than 0.1% for HPV85 to 0.6% for HPV68. HPV6 and 11 were detected in 0.4% and 0.5% of ICC, respectively.
Overall HPV positivity was significantly higher among 26,667 cases of SCC/unspecified histology (90.9%) than in 3,525 ADC (82.0%; Table 2). HPV18 was over-represented in ADC (36.8%) compared to SCC (13.2%). Conversely, HPV16 was significantly under-represented in ADC (36.3%) compared to SCC (59.3%), as were many other carcinogenic types belonging to the same species as HPV16 (i.e., HPV31, 33, 52 and 58). Two very rarely detected types (HPV6 and 69) were significantly more frequent in SCC and ADC, respectively. The prevalence of multiple infections did not differ significantly by histological type.
Overall HPV prevalence ranged from 88.3% in studies from Oceania to 94.2% in those from Africa (Fig. 1). The ten most common HPV types in ICC are shown by region in Figure 1. HPV16 was the most common type in all regions (ranging from 52% in Africa to 67% in Western/Central Asia), and HPV18 the second most common type (ranging from 14% in Eastern Asia to 20% in Oceania). The third to eighth most common HPV types were almost always HPV31, 33, 35, 45, 52 and 58, although their relative importance slightly differed by region. Of notice, HPV58 was the third most common type in Eastern Asia. In addition, HPV51 was in the seventh place in Africa, HPV59 in the eighth place in Eastern Asia, HPV39 in the eighth place in North America and HPV73, 39 and 53 in the fourth, fifth and eighth place, respectively (although sample sizes were small), in Oceania.
Overall and type-specific HPV prevalence in ICC and ADC are shown in Table 3 stratified by year of publication. The prevalence of any HPV type in ICC increased significantly from 85.9% in studies published in 1990-1999 to 92.9% in studies published in 2006-2010. Over the same time period, the prevalence of multiple infections increased from 4.0 to 15.7%, whereas the prevalence of single infections slightly decreased from 82.1 to 76.8%. The prevalence of HPV16, including both single and multiple infections, increased significantly from 51.8% in 1990-1999 to 60.0% in 2006-2010. After adjustment for histological type, region, specimen type and PCR primer, statistically significant increases in prevalence were also seen for HPV58, 39 and 53.
The absolute increase in overall HPV prevalence in ADC (from 77.8 to 86.0%) was similar to that for all ICC, as was the significant increase in multiple infections (from 3.9 to 15.9%). The prevalence of HPV16 increased (from 28.8% in 1990-1999 to 41.6% in 2006-2010), whereas HPV18 prevalence did not significantly change.
The distribution of ICC by year of publication is shown across strata of region, specimen type and PCR primer in Table 4, as well as the corresponding trends in HPV16 prevalence. With respect to distribution by region, there was an increase in the proportion of cases from Asia over time. The increase in HPV16 prevalence was observed in most regions and it was statistically significant in Eastern Asia. HPV16 prevalence was consistently higher among ICC tested from fresh/fixed biopsies than exfoliated cervical cells, but the increase in HPV16 prevalence over time was consistent in both types of specimen. The two most frequently used PCR primer sets over the entire study period were GP5þ/6þ and MY09/11. The proportion of cases tested with MY09/11 decreased over time, with a parallel increase in those tested by PGMY09/11, SPF10 or a combination of the four primers listed in Table 4. A significant upward trend in HPV16 prevalence over time was found only for MY09/11 (Table 4).
A subset of 66 studies, including 12,106 HPV-positive ICC and 1,156 ADC, tested for all 12 carcinogenic types and had the information to distinguish whether an individual HPV type was found in single or multiple infections (Table  5). Overall estimates of HPV prevalence were slightly higher in this subset of studies than the respective figures among the overall meta-analysis, but prevalence of HPV16 and HPV18 were similar. The increases in the prevalence of any HPV type (from 91.4 to 94.0%), multiple HPV infections (from 3.0 to 16.4%) and HPV16 (from 51.6 to 59.1%) by the year of publication were consistent with the overall meta-analysis. In ICC, the prevalence of HPV16 (49.7% in 1990-1999 and 49.0% in 2006-2010) and HPV18 (14.6 and 10.4%) in single infections stayed constant, while the corresponding prevalence in multiple infections increased substantially (from 2.0 to 10.1%, and from 1.6 to 5.9%, respectively). These shifts from single to multiple infections for any HPV-, HPV16and HPV18-positive ICC were all significant. In ADC, similar shifts from single to multiple infections were seen by year of publication and trends over time were significant for any HPV type (Table 5).

Discussion
By updating with the large amount of data published after 2006, our study includes twice as many cases as previous meta-analyses of HPV types in ICC. 2,3 The accrual of new information has been especially large for studies from Asia, and for the rarest types in the high-risk clade. 4 Furthermore, for the first time, our study allowed a description of time trends in the prevalence of HPV types and multiple infections in ICC over 20 years of publications on the topic.
Overall, approximately 73% of ICC were associated with either HPV16 (57%) or 18 (16%), which, as shown previously, 3 were by far the most common types in ICC in all regions studied. As new data accrue, the proportion of cases associated with HPV16/18 appears increasingly similar across all regions (70-76%), with the only partial exception of Asia. In Western/Central Asia, 82% of ICC was HPV16/18-associated, compared to only 68% in Eastern Asia. These figures suggest, therefore, a difference in HPV type distribution in  ICC across the Asian continent, which is estimated to bear more than half of the world's cervical cancer burden. 6 The most common types in ICC after HPV16 and HPV18 were confirmed to be consistent in all world regions and over time, namely a combination of HPV31, 33, 35, 45, 52 and 58, with few exceptions. Our present findings are in agreement with those from our previous meta-analysis, 3 as well as with those from the uniformly tested samples of the IARC

Epidemiology
Multicentric Cervical Cancer Study of 3,085 ICC, 7 and ad interim data from a large multi-centre study. 8 The most important regional differences in the ranking of HPV types was represented by HPV58, which is more frequent in Eastern Asia compared to elsewhere. For this reason, the total prevalence of HPV58 in published studies (4.7%) was inflated by the large contribution of cases from Eastern Asia. HPV31, 33, 35, 45, 52 and 58 clearly remain the priority candidates for additions in future generations of prophylactic vaccines. The remaining four HPV types recently classified as carcinogenic to humans, 5,9 namely HPV39, 59, 51 and 56, were the next most common HPV types in ICC, each found in about 1% of cases.
HPV68, which was classified as probably carcinogenic to humans (Group 2A), was detected in 0.6% of ICC.
In a recent working group at IARC, other types in the high-risk clade (HPV26, 30, 34, 53, 66, 67, 69, 70, 73, 82 and 85) were all classified for cautionary reasons as possibly carcinogenic to humans (Group 2B) even in the absence of adequate epidemiological data on their carcinogenic potential. 5,9 Our study revealed that most of these types were identified in less than 0.5% of ICC and less often than HPV6 and/or HPV11, which are not considered to be causally related to ICC. Furthermore, some of these types, such as HPV53 and HPV66, are relatively common in women with normal cytology 9 and in women with low-grade cervical lesions, 10 Hence, including these types in screening assays would decrease the specificity and positive predictive value of the assays with virtually no gain in sensitivity and negative predictive value towards ICC and severe precancerous lesions. 11 Ninety percent of all ICC in our present study were positive for HPV DNA, and this percentage has increased over time. Given that HPV is accepted as a necessary cause of ICC, being found in more than 99% of ICC tested under the best conditions, 12 this increase is expected to be related to general improvements in HPV DNA testing protocols and not to any real changes in the contribution of HPV to ICC aetiology. It is also worth noting that year of publication is not always representative of the year of cancer diagnosis and sample collection, which may have been much earlier. The statistical significance of these increases remained after adjustment for specimen type and PCR primers used, suggesting that other improvements, not captured in our present meta-analysis approach, were at least partly responsible for the increases in HPV detection in ICC (e.g., the general quality of biological specimen and/or reagents used, including the use of different DNA polymerases).
The progressive reduction in the proportion of HPV-negative ICC seen in our present meta-analysis has been accom-panied by a substantial increase in the detection of HPV16, as well as smaller increases in some other rarer types. This would be expected if the type distribution in the falsely HPV-negative ICC in earlier published studies was similar to that of HPV-positive cases. Indeed, this was found to be the case when the HPV-negative cases were retested under goldstandard conditions by Walboomers et al. 12 Although the increase in HPV16 prevalence was consistently observed across many strata of region, specimen type, histological type and PCR primers, the phenomenon was particularly strong in ADC. Thus, the contribution of HPV16 and HPV18 to ADC now seems more similar than in the earliest studies that showed a predominance of HPV18. 2,3 This increase in the proportion of HPV16-positive ADC may be in part related either to increases in the joint detection of ADC and squamous intraepithelial lesions (in large part HPV16-positive) in screening programmes or to an increase in the fraction contributed by adenosquamous carcinoma, which appears to be related to HPV16 more often than adenocarcinoma. 13 It is interesting that differences in prevalence by histology are not limited to HPV16 and HPV18, but extend to other HPV types that are phylogenetically related to these two types. 14 Our study, however, also sends a warning: the attribution of ICC to different HPV types is increasingly complicated by the growing prevalence of multiple HPV types, which reached 15.7% in the most recently published studies. This increase is expected to be due, in large part, to the same factors that have driven the overall increase in HPV prevalence (i.e., increased sensitivity). In addition, the increase in the types included in genotyping protocols of HPV-positive cases has certainly affected the detection of multiple infections. A full assessment of multiple HPV infections is not possible in the entirety of our present meta-analysis because of the absence of individual-level data on the type-specific breakdown of multiple infections. Nevertheless, the extent to which multiple infections affect the attribution of ICC to HPV16 and HPV18 could be evaluated in a subset of studies including more than 12,000 ICC. This subset of studies showed that the prevalence of HPV16 in multiple infections has increased more than it has in single infections. HPV16 is expected to be causally related to ICC in a large proportion of these multiple infections, especially as an unknown number of multiple infections may involve non-carcinogenic types, but the estimate of the fraction truly attributable to HPV16 is uncertain. It can vary from the most conservative estimate of 49% (i.e., HPV prevalence in single infections only) up to 63% (the prevalence of HPV16 among HPV-positive ICC only, irrespective of whether HPV16 was found in single or multiple infections). 15 The increasing prevalence of multiple infections also complicates the estimation of the proportion of cancers that can be attributable to groups of HPV types. For example, we estimated the fraction of ICC attributable to HPV16/18 from our entire meta-analysis to be 73% by simple addition of the type-specific prevalence for HPV16 to that for HPV18. However, the number of HPV16/18 multiple infections cannot be extracted from the majority of published data, and HPV16 and HPV18 are thus sometimes counted twice. The problem of overlapping multiple infections obviously becomes more severe as more types are included in the measure of the attributable fraction. For example, based upon studies published in 1990-1999, where multiple infections accounted for only 4%, the proportion of ICC attributable to the eight most common HPV types could be conservatively estimated at 86%. However, 14.1% of ICC remained HPV-negative in these early studies, suggesting that the fraction attributable to these eight types might actually be much higher. In more recently published studies, the HPV-negative fraction has halved (7.1%), but the newer estimation of the fraction of ICC attributable to the combination of these types (101.0%) is certainly inflated by the multiple counting of HPV types in multiple infections.
In conclusion, the rise of the fraction of multiple infections in ICC, which is understood to arise from a single cell, poses serious interpretation problems. Unless functional involvement of HPV types in multiple infections is demonstrated, the attribution of the various HPV types present in multiple infections remains uncertain. However, demonstrating which HPV type is functionally relevant to a tumour, by using techniques such as microdissection and/or detection of E6 or E7 transcripts, remains difficult. 16 It is therefore increasingly challenging to distinguish between clinically relevant and irrelevant HPV infections and thereby to estimate the benefits of tests and vaccines including more or less broad ranges of HPV types. 17