Prevalence of Human Papillomavirus in Self-Taken Samples from Screening Nonattenders

ABSTRACT The Copenhagen Self-Sampling Initiative (CSi) has shown how human papillomavirus (HPV)-based self-sampling can be used to increase screening participation among 23,632 nonattenders in the Capital Region of Denmark. In this study, we describe HPV prevalence and genotype frequency in 4,824 self-samples as determined by three HPV assays (the CLART, Onclarity, and Hybrid Capture 2 [HC2] assays) and compare the results with those for physician-taken follow-up samples. The HPV self-sample findings were also compared to the findings for a reference population of 3,347 routinely screened women from the Horizon study, which had been undertaken in the same screening laboratory. Nonattenders had an HPV prevalence of 11.3% as determined by the CLART assay, which was lower than that for women from the Horizon study (18.5%). One-third of the CSi women who tested HPV positive by self-sampling tested HPV negative on the physician-taken follow-up sample. The CLART and Onclarity assays agreed on 64% (95% confidence interval [CI], 60 to 68%) of the HPV-positive self-taken samples. When the HC2 assay results were added into a three-way comparison, the level of agreement decreased to 27% (95% CI, 24 to 29%). Our findings suggest that further validation of HPV assays on self-taken samples is needed for optimal HPV detection and correct clinical management of HPV-positive women.

H uman papillomavirus (HPV)-based self-taken samples can be an alternative to cervical cytology screening for women who do not attend screening (nonattenders) (1,2). A meta-analysis showed that with specific combinations of collection devices and HPV assays, the accuracy of detection of high-grade cervical intraepithelial neoplasia (ՆCIN2) was similar to that with physician-based sampling (3). Most selfsampling studies to date have used the Hybrid Capture 2 (HC2) assay for HPV testing (2,(4)(5)(6)(7)(8)(9)(10)(11), but currently, no single HPV assay is recognized as the gold standard. On routine screening with physician-taken samples, assays are considered validated if they show noninferiority to the HC2 assay or GP5ϩ/6ϩ PCR, the two assays that were the basis for four European randomized trials (12). However, no guideline exists on the criteria for the use of HPV assays with self-taken samples.
The pilot implementation of "The Copenhagen Self-Sampling Initiative" (CSi) was initiated to investigate whether self-sampling by nonattenders was a viable option for increasing participation in screening. Another aim was to compare three HPV assays on self-taken samples. By use of an opt-in strategy, 20% of 23,632 invited nonattenders returned self-taken samples (13). All samples were tested in a split-sample protocol with the CLART HPV2 (Genomica, Madrid, Spain) and BD Onclarity HPV (BD, Sparks, MD, USA) tests. Moreover, a limited subset of 1,008 samples was also tested using the HC2 assay (Qiagen, Hilden, Germany), which was the routine assay in the laboratory at the time the pilot started. Women with a positive result on any of the HPV assays were recommended to have a physician take a cytology sample, to be cotested for cytology and HPV, as follow-up.
The frequencies of different HPV genotypes among self-sampling nonattenders are infrequently reported in the literature, mainly because the majority of the studies used the HC2 assay, which does not allow one to distinguish between the HPV genotypes detected. In this study, we not only employed assays with extended or full genotyping but also compared the findings to those of the Danish Horizon study, a previous study using physician-taken samples from routinely screened women, where the CLART assay was also used in that study.
One of the major outcomes of the Horizon study was that the frequencies of ՆCIN2 detection were fairly similar for the assays compared, but the assays detected different HPV infections in the same samples (14). On samples from approximately 3,000 women undergoing primary screening, only 27% of all infections detected by four commercially available assays were confirmed by all assays, and when the assays were compared pairwise, only half of all HPV infections were confirmed on both assays. No similar HPV assay performance issues on self-taken samples have been described earlier.
In this study, we compared the detection of HPV infections by three assays (CLART, Onclarity, and HC2) and the subsequent cytology results for women with detected infections. The findings were compared to those for routine screening samples from the Horizon study (15,16).

RESULTS
High-risk HPV (HR-HPV) infections were detected in 769 (15.9%) self-taken samples, based on a positive test result on any assay (where testing by the HC2 assay was undertaken only on the first 1,008 samples). The Onclarity assay showed the highest prevalence (14.1%), followed by the CLART and HC2 assays (11.3% and 7.2%, respectively) ( Table 1). The HPV prevalence detected in self-taken samples appeared to be lower than that for physician-taken samples in the Horizon study (18.5%) when tested by the CLART assay. This was also the case with the age-specific prevalences. The prevalence of HPV appeared not to be related to the women's screening history. Of the 769 women who tested positive for HPV, 641 went for follow-up (follow-up compliance rates were 83% for all ages, 90% at the ages of 27 to 29 years, 82% at 30 to 39 years, 84% at 40 to 49 years, 77% at 50 to 59 years, and 87% at 60 to 65 years [data not tabulated]). Almost one-third of the women with physician-taken follow-up samples after a positive self-taken sample had a diagnosis of atypical squamous cells of undetermined significance or higher-grade lesions (ՆASCUS), a proportion 2-fold higher than that for women in the Horizon study with a positive HPV test at baseline.
HPV16 was the genotype most frequently observed both for routinely screened women and in self-sampling, constituting 25.4% and 24.3%, respectively, of the HR-HPV types (data not tabulated), and with a detected prevalence of 2.9% in self-sampling. The second most frequent HR-HPV genotype in self-sampling was HPV51 (2.0%), followed by HPV31 (1.6%) ( Table 2). In the Horizon study, the most frequent genotypes were HPV16 (4.5%), HPV31 (2.8%), and HPV52 (3.2%). Infections with more than one HPV genotype were observed in approximately 56% of all HPV-positive samples in the Horizon study, whereas for self-sampling, the prevalence of multiple infections was below 48% (data not tabulated). The numbers and proportions of women with highand low-risk infections are shown in Table 3.
Of 736 HR-HPV infections detected by the Onclarity or CLART assay in the CSi study, a total of 471 infections (64% [95% confidence interval {CI}, 60 to 68%]) were detected by both assays (Table 4; Fig. 1A). Women for whom infections were detected by both assays were more likely to have a diagnosis of ՆASCUS.
The relatively low agreement was not substantially increased when genotype HPV66 was included on the CLART assay so that it would match the genotypes targeted by the Onclarity assay. The levels of agreement were highest for HPV16, HPV31, and HPV52 (all ϳ75%), whereas the lowest level of agreement was observed for HPV45 (40%) ( Table 4). When the HC2 assay was added into the comparison, with 964 valid test results, the  level of positive agreement was 27% (Fig. 1B). In pairwise comparisons, the level of agreement between the CLART and HC2 assays was 38%, and that between the Onclarity and HC2 assays was 33%. The detected prevalence of HPV infections in this subset was similar to that in the entire study (14.8% by the Onclarity assay, 10.0% by the CLART assay, 7.4% by the HC2 assay, and 17.2% by any of the three assays).
Almost two-thirds of women with HPV-positive test results were also found HPV positive in the follow-up sample (69.1% for the CLART assay and 63.7% for the Onclarity assay) ( Table 5). Women who came early for follow-up had a slightly lower HPV prevalence on the follow-up sample than women who came later, although the difference was not statistically significant (P ϭ 0.67) (data not reported).
The use of the cycle threshold (C T ) values of the internal controls included in the Onclarity assay, as well as the C T values for positive samples, allowed a quantitative evaluation of sample quality. Mean C T values on self-taken samples were approximately 2 cycles lower than those on physician-taken samples for all three human beta-globin (HBB) control wells (mean differences, Ϫ2.00 with HBB1, Ϫ1.99 with HBB2, and Ϫ1.99 with HBB3), indicating a larger concentration by volume of human material in self-taken samples than in physician-taken samples ( Table 6). Discrepant sample sets with HPVpositive self-taken samples but HPV-negative physician-taken samples showed higher genotype-specific C T values on the self-taken sample than did the concordant sample sets (by 2 to 5 cycles) ( Table 7). Concordant sample sets tended to have similar  genotype-specific signal strengths. Finally, C T values tended to be lower in self-taken samples found positive by all three HPV assays than in self-taken samples that were positive only by the Onclarity assay (difference in median signal strengths, about 8 cycles) ( Table 8).

DISCUSSION
General findings. Among self-sampling attenders, the prevalence of HR-HPV was between 11% (by the CLART assay) and 14% (by the Onclarity assay), and about one-third of HPV-positive women had abnormal cytology upon follow-up. Agreement in the detection of HR-HPV between the CLART and Onclarity assays was observed in approximately two-thirds of cases overall, and in three-fourths of cases when HPV16, HPV31, or HPV52 infections were involved. About one-third of the women whose  self-taken samples tested HPV positive were found to be HPV negative on the physician-taken follow-up sample, and this was particularly the case when the signal strength in the self-taken sample was close to the manufacturer-set cutoff, indicating small amounts of viral input material. The detected HR-HPV prevalences, overall and genotype specific, were lower than those in a routinely screened population from the same uptake area, although among the infected women, the proportion with abnormal cytology was higher.
Strength and limitations. Both the CSi and the Horizon study were populationbased studies undertaken in a routine screening laboratory. Our laboratory is the sole provider of cervical screening for this part of Denmark; it has a long-standing cervical screening program with a ϳ75% coverage rate (17). Hence, our findings are representative for the general population in a real-world setting, even though the laboratory's catchment area during the Horizon study included only the urban Copenhagen and Frederiksberg municipalities, whereas during the CSi study, the catchment area had been expanded to include the periurban areas of the entire Capital Region (including the former Copenhagen, Frederiksborg, and Bornholm counties besides the two municipalities). Moreover, the same senior laboratory staff participated in the two studies, ensuring the same standards for both studies. Finally, since our laboratory is the principal site for the validation of the Onclarity assay on SurePath samples (18), we have several years of experience in operating the Onclarity assay and instrumentation.
A limitation of the comparisons between the HC2, Onclarity, and CLART assays in the CSi may be that BD's Cervical Brush Diluent (CBD) medium, which was used to  HBB2, and HBB3). For eight women, the C T value was reported as zero for either the selftaken sample or the follow-up sample with HBB1, but valid C T values were returned with HBB2 and HBB3, leaving 496 pairs of valid C T values for the analysis. Four samples with HBB2 and six samples with HBB3 failed to yield valid results. c Calculated by subtracting the C T for the physician-taken sample from the C T for the self-taken sample. resuspend the self-sampling brush heads prior to analysis, has not been used for self-sampling previously, although it has been used extensively on physician-taken screening samples (18). The CLART assay is a well-described assay with good performance compared to the HC2 assay on both SurePath (19) and ThinPrep (20) samples and has been used for self-sampling in our laboratory previously (21).
Clinical implications and comparison with other studies. This is one of the first studies comparing different HPV assays in terms of their ability to detect HR-HPV in self-taken samples. Besides, this is an unsupervised self-sampling setting, which is the most relevant setting for a routine rollout of this service to screening nonattenders.
In line with previously published experience on physician-taken samples (14,22), we found the positive agreement between the Onclarity and CLART assays to be just above 60%. This is comparable to the detection of HPV in routinely screened women, where pairwise agreement in a screening population reached 58% for the CLART and cobas assays (14). The agreement with the HC2 assay in this self-sampling population was below 40%, and the detected HPV prevalence was also lower on the HC2 assay than on the other two assays. Whereas the majority of European self-sampling studies in well-screened populations used only the HC2 assay for the detection of HPV infections (2, 4-8, 10), Enerly and colleagues (21) used the HC2 and CLART assays to investigate HPV prevalence among 169 women who returned a self-taken sample for HPV detection. In this small study, with the HPV analyses performed in our laboratory, the positive agreement for the assays was still only 50%. The disagreement is mainly technology driven (14), but our results underline the need for validation criteria for HPV testing on self-taken samples to facilitate a minimum of quality assurance.
For screening purposes, the relevant endpoint is histological confirmation of precancerous high-grade cervical lesions especially; this is beyond the scope of the current analysis and will be reported in detail separately.
One-third of the women found HPV positive on the self-taken sample were found HPV negative on the follow-up physician-taken sample. This discrepancy is highly interesting and relevant to the future implementation of self-sampling. One explanation for this observation could be spontaneous HPV clearance, yet the same decrease in HPV detection in the physician-taken sample was also seen in women who came for follow-up within 1 month after returning the self-taken sample. Alternatively, some women may have inadvertently sampled the vaginal canal, whereas the physicians are taught to visualize the cervix prior to taking the screening sample. Finally, self-taken samples were preprocessed in 3 ml CBD, whereas the physician-taken samples were suspended in 10 ml SurePath. The higher concentration of biological material per volume unit in a processed self-taken sample may lead to higher signal strengths (i.e., lower C T values) for real-time PCR and a higher proportion of samples above the cutoff for HPV positivity. The C T values on positive self-taken samples of women whose physician-taken samples were also HPV positive tended to be lower than those for women whose physician-taken samples were HPV negative. Finally, stronger signals a C T value on the genotype-specific well with the strongest signal. b For 964 samples, all three assays (Onclarity, CLART, and HC2) gave valid results. In total, 166 samples (17.2%) were HPV positive on at least one of the assays (Fig. 1). Of these, 143 (86.1%) were positive by Onclarity, with C T values available for reporting.
were found when all three assays confirmed the HPV-positive test result than when only one assay detected the infection. In line with our research on routinely screened women from the Horizon study, this could also suggest that women whose HPV infections had higher signal strengths may have been more likely to harbor high-grade CIN (16). Nevertheless, this finding remains to be confirmed for women undergoing self-sampling. Conclusion. Cervical cancer screening of nonattenders who participated in selfsampling yielded an overall HPV prevalence of 16%, detected by any of the three HPV assays used, which was somewhat lower than the prevalence detected among women attending routine screening in the same screening laboratory. Self-sampling women, however, had a higher proportion of cytological abnormalities upon follow-up. As in routine screening, self-taken samples showed substantial discordance between assays detecting HPV infections.

MATERIALS AND METHODS
Study design. In Denmark, women are invited for screening every 3 years (at the ages of 23 to 49 years) or every 5 years (at the ages of 50 to 65 years). The design of the CSi study has been explained elsewhere (13). Briefly, 23,632 nonattenders (defined as having had no screening for Ն4 years [at ages 27 to 49] or Ն6 years [at ages 50 to 65]) from the Capital Region were invited to order a self-sampling brush (Evalyn; Rovers, Oss, the Netherlands). In total, 4,824 (20%) women returned the self-taken sample. Each brush was preprocessed and was aliquoted into four split samples (Fig. 2). For clinical management Self-taken samples were aliquoted into four split samples. The first three aliquots were for HPV testing, and the fourth (not shown) served as backup material and constituted a biobank. (c) All women with detected HPV infections were recommended to see their GP for a cytology sample. Compliance to follow-up was 83.3% (641/769). (d) By mistake, 35 of the 641 (5.3%) follow-up samples were not tested for HPV, but they all underwent cytology evaluation.
within the study, a woman was considered HPV positive and recommended to undergo physicianadministered sampling if any of the three assays detected high-risk HPV (HR-HPV). Women were notified of their results by letter, and the result was also sent to the general practitioner (GP) if consent was given by the woman. The subsequent cytology follow-up samples were cotested for HPV and used for triage to colposcopy. Women who were normal by cytology, regardless of the HPV test result, were recommended for cytology and HPV cotesting after 12 months. Women who had ϾASCUS, regardless of the HPV test result, were referred directly to colposcopy, which was also the case for women who had ASCUS and were HPV positive. Women who were HPV negative but had ASCUS were recommended to have a cytology retest after 6 months. The cytotechnician was not blinded to the woman's HPV status. Women with no HPV infections detected in their self-taken sample were recommended to participate in the next round of routine screening.
Preprocessing of self-sampling brushes. All returned brushes were preprocessed and were analyzed for HR-HPV. Brush heads were removed from the holders and were placed in empty 5-ml Eppendorf tubes (see Text S1 in the supplemental material). Three milliliters of BD's Cervical Brush Diluent (CBD) was added. Brush heads were removed after 15 min, and samples were vortexed to achieve homogeneous mixtures. Each sample was aliquoted into four empty tubes using the following schedule: 450 l was transferred to an Eppendorf tube for DNA extraction and CLART analysis, 1,000 l to a tube for Onclarity testing, and 500 l to a tube for HC2 analysis (the first 1,008 brushes), while all residual material was stored at Ϫ20°C for backup.
HPV DNA testing. For the CLART assay, two 200-l aliquots were transferred directly to a MagNA Pure 96 system (Roche Diagnostics, Mannheim, Germany) for DNA purification. The two 100-l purified DNA outputs were then pooled. The assay provided full genotyping of 35 HPV types, without quantitative measurement of signal strengths. For clinical management, HR-HPV types were defined as HPV16, For the Onclarity assay, the 1,000-l aliquot was prewarmed for cell lysis. The sample was loaded into the Viper Lt real-time PCR system (BD, Sparks, MD, USA). The Onclarity assay detects six HR-HPV genotypes individually (HPV16, -18, -31, -45, -51, and -52), whereas another seven HR-HPV genotypes and one low-risk HPV genotype are detected in three combinations (HPV33/58, HPV35/39/68, and HPV56/ 59/66). The standard cycle threshold (C T ) cutoff for HPV positivity is 34.2.
For the HC2 assay, DNA denaturation was performed manually on the 500-l CBD medium aliquot before testing. HPV testing was performed on the automated Rapid Capture system (Qiagen, Gaithersburg, MD, USA), and the results were read automatically on a luminometer. The threshold for positive test results was 1 relative light unit (RLU)/cutoff (CO). Due to budgetary restrictions, HPV testing by the HC2 assay was completed only on the first 1,008 consecutive self-samples.
Data sources. Women's screening histories since January 2000 were retrieved from the nationwide Pathology Database (Patobank), which has been virtually complete for the entire country since the mid-2000s.
The invitations for self-sampling were sent between May 2014 and April 2015. For the current analysis, we retrieved the screening activities until November 2015, which allowed for delays in returning the brush (13), the laboratory testing and reporting, and the cytology follow-up.
Women were categorized as "intermittently screened" if they had a cytology sample within the last 10 years before the self-sampling invitation and as "long-term unscreened" if they had no sample for Ն10 years. The analyses using this categorization were restricted to women Ն34 years old, i.e., those who had been eligible for screening for Ն10 years.
Statistical analyses. Due to hardware failure and technical issues with the Viper Lt system, some of the samples were stored at Ϫ20°C before testing with the Onclarity assay. No significant differences were seen when HPV prevalences determined for frozen and fresh samples were compared, including a comparison of the respective C T values and adjustment for age (data not shown). Hence, we included all test results in the analyses.
As in the Horizon study (14), positive agreement between assays was calculated as the proportion of samples testing positive on any assay that were confirmed by all assays compared.
For the comparison of positive HPV test results in self-taken samples and physician-taken follow-up samples, we calculated medians and interquartile ranges (IQR) for the signal strengths (C T values) of the Onclarity assay. The C T medians were stratified according to the median follow-up time (27 days) for women who went for follow-up.
The 95% confidence intervals (CIs) for proportions and for prevalence agreement were calculated on the assumption that they were binomially distributed, whereas the 95% CIs for mean differences in C T values were calculated by taking into account that the observed differences were normally distributed.
Stata SE 13.1 and Microsoft Excel 2010 were used for data analyses. Ethical approval. The CSi was a time-limited pilot implementation initiative mandated by the Danish Health Authority. The Horizon study was undertaken as a quality development study in accordance with