Human papillomavirus testing on self‐collected samples to detect high‐grade cervical lesions in rural Bhutan: The REACH‐Bhutan study

Abstract Background “REACH‐Bhutan” aimed to evaluate the feasibility and clinical performance of a community‐based screening program for cervical cancer in rural Bhutan using self‐collected samples for high‐risk human papillomavirus (HR‐HPV) testing. Methods In April/May 2016, 2590 women aged 30–60 years were screened across rural Bhutan by providing a self‐collected sample for careHPV testing. All careHPV‐positive women, plus a random sample of careHPV‐negative women, were recalled for colposcopy and biopsy. Self‐samples also underwent GP5+/6+ polymerase chain reaction (PCR)‐based HR‐HPV DNA detection and genotyping. Cross‐sectional screening indices were estimated against histological high‐grade squamous intraepithelial lesions or worse (hHSIL+), including imputation of hHSIL+ in women without colposcopy. Results HR‐HPV positivity was 10.2% by careHPV and 14.8% by GP5+/6+ PCR. Twenty‐two cases of hHSIL+ were histologically diagnosed, including one invasive cancer; an additional 7 hHSIL+ were imputed in women without colposcopy. HR‐HPV testing by GP5+/6+ showed higher sensitivity for hHSIL+ (89.7%, 95% CI 72.6–97.8) than careHPV (75.9%, 95% CI 56.5–89.7). Negative predictive value was also slightly higher for GP5+/6+ (99.9%, 95% CI 99.6–100) than careHPV (99.7%, 95% CI 99.4–99.9). Specificity, however, was lower for GP5+/6+ (86.1%, 95% CI 84.6–87.4) than careHPV (90.6%, 95% CI 89.4–91.7), as was positive predictive value (6.9%, 95% CI 4.5–9.9 vs. 8.5%, 95% CI 5.4–12.6). Of 377 HR‐HPV‐positive women by GP5+/6+, 173 (45.9%) were careHPV‐positive, including 54.7% HPV16‐positive and 30.2% HPV18‐positive women. Conclusions The final REACH‐Bhutan results show that screening for cervical cancer with self‐collection of samples and HR‐HPV testing, in addition to our previous report of achieving high participation, can also perform well to detect women with hHSIL+.


| INTRODUCTION
Cervical cancer is the most frequent cancer among women in Bhutan, 1 a country that is actively committed to cervical cancer prevention.In 2000, the Ministry of Health (MoH) of Bhutan launched a national cytology-based screening program, 2 followed by a national program of vaccination against human papillomavirus (HPV), reaching >90% coverage in girls aged 12-18 years in 2010. 3Cytology-based screening coverage has been less widespread and unevenly implemented across the country, 4 being particularly low in rural areas 5 where the majority of the Bhutanese live.Follow-up and treatment of screen-positive women also presents a challenge.
High-risk HPV (HR-HPV)-based screening has the advantage of extended screening intervals, self-sampling, and test automation, [6][7][8] providing an excellent opportunity to improve the coverage and cost-effectiveness of cervical screening.Thus, in 2016, the Bhutan MoH and the International Agency for Research on Cancer (IARC) initiated the REACH-Bhutan study, with the aim of assessing the feasibility, clinical performance, and challenges of screening for cervical cancer using the careHPV (Qiagen, Gaithersburg, MD) test on self-collected vaginal samples in women 30-60 years of age from rural Bhutan.The design and implementation of the REACH-Bhutan study, as well as the determinants of screening participation (most notably age and time taken to travel to the health center) and HPV positivity (most notably sexual behavior), have been reported previously, and showed that HPVbased cervical cancer screening using self-collection can achieve high coverage in rural Bhutan. 9Here, we discuss the clinical performance of the REACH-Bhutan screening protocol.The cross-sectional performance of careHPV on self-collected samples was assessed against the gold standard of colposcopy and histologically proven highgrade squamous epithelial lesions or worse (hHSIL+).Ascertainment bias was addressed by random biopsies among a subset of careHPV-negative women and validation with a clinically validated HR-HPV test that has shown good performance on self-collected samples. 10,11 2 |MATERIALS AND METHODS

| Study population and recruitment
The design of the REACH-Bhutan study has been previously described in detail. 9In brief, in April/May 2016, women aged 30-60 years were invited to attend cervical screening in 15 Basic Health Units (BHUs) providing primary health care in rural areas of Bhutan.Locallybased health workers (HWs), accustomed to community mobilization, attended public information sessions in the villages covered by the preselected BHUs, to explain the benefits of screening for cervical cancer.Women who were known to be pregnant or who had undergone hysterectomy were not eligible.At the screening visit, women provided informed consent and completed a short electronic questionnaire.The study was approved by both the Research Ethical Board of the Bhutan Ministry of Health (REBH/PO/15/023) and the IARC Ethics Committee.

| Sample collection, transportation, and laboratory analysis in Bhutan
Each participant provided a self-collected cervicovaginal sample into careHPV universal collection medium (UCM) medium using a careBrush, as detailed previously. 9Specimen vials were transported to Jigme Dorji Wangchuck National Referral Hospital (JDWNRH) in Thimphu where an aliquot was tested on the careHPV platform (Qiagen Corporation, Gaithersburg, MD, USA) according to the manufacturer's instructions.The careHPV test is a signal amplification, rapid batch diagnostic test designed for the detection of the DNA of 13 HR-HPV types (16, 18, 31, 33, 35, 39, 45,  51, 52, 56, 58, 59, and 68) and HPV66. 12careHPV results were given to each BHU to arrange colposcopy for all careHPV-positive women, as well as for a random subset of careHPV-negative women (~5 per BHU; total n = 83).previous report of achieving high participation, can also perform well to detect women with hHSIL+.

K E Y W O R D S
cancer prevention, screening, viral infection, women's cancer

| Cervical disease assessment
Mobile teams with a portable colposcope and cryotherapy equipment visited BHUs (or, exceptionally, closest hospitals), at which time all careHPV-positive women (and the randomly selected subset of careHPV-negative women) were invited to attend a follow-up visit.Colposcopy was used to take biopsies from all suspicious areas in women with abnormal colposcopic findings or, in the absence of a specific suspicious area, randomly from 12 o'clock of the squamocolumnar junction.Treatment of colposcopydetected lesions was performed according to local protocols, primarily using cryotherapy on-site (n = 119) or referral to closest hospitals for loop electrosurgical excision procedure (n = 214).
Histological reading of biopsies was first performed at the Department of Pathology at JDWNRH (T.T.), with a second expert reading (T.M.D.), both blinded to individual level HPV status (even if pathologists knew that most biopsies were obtained from HR-HPV-positive women).Results were reported according to LAST criteria 13 which was adopted as the reference diagnosis.
2.4 | High-risk HPV DNA detection and genotyping by GP5+/6+ PCR Vials containing cellular material in careHPV UCM medium were shipped to the Department of Pathology at Amsterdam UMC location Vrije Universiteit Amsterdam, the Netherlands, where DNA was extracted using magnetic beads on a robotic system.The presence of human DNA in all specimens was confirmed by βglobin PCR analysis as quality control for the extraction procedure and subsequent PCR.HR-HPV positivity was assessed by GP5+/6+−mediated PCR 14 followed by hybridization of PCR products in an enzyme immunoassay (EIA) with an oligoprobe cocktail for detection of 13 HR-HPV types (16,  18, 31, 33, 35, 39, 45, 51, 52, 56, 58, 59, and 68).Genotyping of EIA-positive samples was subsequently conducted by luminex hybridization of GP5+/6+−PCR products as described previously. 15EIA-positive samples that failed to reveal a positive signal in the genotyping assay were designated as HR-HPV type X (HPVX).

| Statistical analyses
Standard screening indices of accuracy, including sensitivity, specificity, positive predictive value, negative predictive value, and their 95% confidence intervals (CI) were calculated for hHSIL+.Corrected indices were calculated after imputation of missing data for women who did not attend colposcopy. 16,17In the corrected model, pseudoobservations were created for women without a valid histology result and weighted by the probability of hHSIL+ among women with the same combination of careHPV and GP5+/6+ results who underwent colposcopy.This approach is valid under the assumption that lack of colposcopy was independent of the underlying hHSIL+ status, given the same combination of careHPV and GP5+/6+ test results ("missing at random").All analyses were performed using STATA version 14.

| RESULTS
Of 2590 women screened with careHPV in the REACH-Bhutan study, 2547 with a valid HR-HPV GP5+/6+ PCR-EIA test result were included in the current analysis (Figure 1).Of the 43 excluded for lack of GP5+/6+ test, six were careHPV-positive: five had histologically benign biopsy results and one did not undergo colposcopy (Figure 1).Adequate histology results were obtained from 331 of 332 women who underwent colposcopy, among whom 22 cases were diagnosed with hHSIL+, including one invasive squamous cell carcinoma.In total, 29 hHSIL+ cases (including seven pseudo-observations among women without colposcopy) were included as outcomes in analyses (Table 1).Only corrected indices are shown, but crude indices, assuming that all women without valid histology had no hHSIL+ (albeit less methodologically valid), can also be calculated using the data presented in Table 1 (e.g., 21 out of 22 confirmed hHSIL+ were careHPV-positive; crude sensitivity for careHPV = 95.5%,95% CI 77.2%-99.9%)(see also Tables S1 and 2).
Type-specific HR-HPV prevalence, according to GP5+/6+ PCR-based genotyping, is described in Table 3.Of the 2547 women tested, the most commonly detected HR-HPV type was HPV16 (6.7%), followed by HPV18

| DISCUSSION
In addition to achieving high acceptability in rural Bhutan as we previously reported, these final REACH-Bhutan results show that HPV-based screening for cervical cancer with self-collection of vaginal samples can achieve high performance in detecting women with hHSIL+.Our report adds to several existing assessments of HPV testing in self-collected samples in underserved populations in lowor middle-resource settings, 6,[18][19][20][21][22] including those based upon careHPV testing. 23,24lthough careHPV was associated with a slightly lower sensitivity than the reference PCR-based HPV test, GP5+/6+ PCR-EIA, the performance of selfcollected careHPV to detect hHSIL+ in our study, with its 76% sensitivity and 91% specificity, is consistent with that estimated by a 2017 meta-analysis of four careHPV evaluations (74% sensitivity and 88% specificity), 23 and with a larger recently published report of self-collected careHPV screening in China (73% sensitivity and 97% specificity). 24revious studies have shown that, similar to other signal amplification methods, for instance Hybrid Capture 2 (HC2, Qiagen), 11 careHPV has slightly lower sensitivity and higher specificity against cervical intraepithelial neoplasia grade 2 or worse (CIN2+) when performed on selfcollected versus clinician-collected samples in the same study, 12,23,25,26 and that differences in sensitivity in between careHPV and HC2 may be bigger for self-collected than clinician-collected samples. 25Nevertheless, careHPV on self-collected samples has been shown to remain more sensitive than either Pap tests 26 or visual acetic acid (VIA) inspection 12,[25][26][27] in the settings in which it has been evaluated head-to-head.Furthermore, the wider advantages of self-collected over clinician-collected sampling in terms of feasibility and acceptability to improve coverage 9,11 are expected to be more important for impact on cervical cancer prevention at a population level than small losses in test sensitivity.
There is a wider literature describing the performance of careHPV testing on clinician-collected cervical specimens, which has also shown greater cross-sectional sensitivity for CIN2+ than either VIA, 12,26,[28][29][30] or Pap tests, 26,31 and resulted in significantly higher CIN2+ yields than VIA or liquid-based cytology in a large randomized control trial (>15,000 women tested for careHPV in rural China 32 ).On the contrary, clinician-collected careHPV has been associated with slightly lower sensitivity to detect CIN2+ versus HC2 27,28,31 or PCR assays, for example Sansure (Sansure Biotech Inc.), 25 INNO-LiPA (Innogenetics N. V.), 35 or GP5+/6+ PCR-EIA. 32n the REACH-Bhutan study on self-collected samples, careHPV was also associated with lower sensitivity than GP5+/6+ PCR-EIA.Indeed, we retested all REACH-Bhutan samples with a reference PCR test for a number of reasons.Firstly, to improve the imputation of CIN2+ among careHPV-negative women (see below).In addition, by performing full HR-HPV genotyping on GP5+/6+ PCR-EIA-positive samples, we were able to investigate care-HPV sensitivity according to HR-HPV genotype.Overall, we found 46% of GP5+/6+ PCR-based HR-HPV-positives to be careHPV-positive from the same sample.This compares to 55% of HR-HPV positives from the only previous comparison of careHPV with a PCR assay according to genotype (albeit using the more analytically sensitive INNO-LiPA assay). 33With respect to the most carcinogenic HPV types relevant for cervical cancer screening, we found that 55% of HPV16-positives and 30% of HPV18positives by GP5+/6+ PCR-based genotyping were also detected by careHPV, which compares with 46% of HPV16 and 49% of HPV18 INNO-LiPA-positive samples, respectively. 33Of 45 samples that were positive for the HR-HPV GP5+/6+ PCR-EIA cocktail probe, but for which no specific HR-HPV genotype could be detected, only two (4%) were careHPV-positive, suggesting that these samples contain infections not easily identified by either assay.The one HPVX with hHSIL was investigated and found to be HPV66-positive which is not considered as a HR-HPV type.Furthermore, careHPV positivity was strongly related to the strength of the semiquantitative GP5+/6+ PCR-EIA optical density signal, a surrogate for HR-HPV viral load.Of note, the distribution of HR-HPV types in the REACH-Bhutan sample of unvaccinated rural women aged 30-60 years, characterized by a strong predominance of HPV16, followed by HPV18 and HPV59, is comparable with that reported in an urban sample of unvaccinated women in the Bhutanese capital, Thimphu. 34HSIL is considered the gold standard diagnosis for cervical precancerous lesions.However, verification bias may occur in histology when biopsies are not taken from all screened women.Here, we tried to overcome this issue by taking biopsies from a proportion of careHPVnegative women, by submitting all biopsies to specialist review according to recommended LAST criteria, 13 and by imputing underlying hHSIL+ in the few women with no histological reading, informed also by GP5+/6+ HPV testing results.Nevertheless, the number of careHPVnegative women that were randomly recalled for colposcopy (n = 83), and the number of observed hHSIL+ with discordant careHPV and GP5+/6+ PCR-EIA results was small, and so the accuracy of imputation of hHSIL+ prevalence to careHPV-negative women without colposcopy is a limitation of our study.
A number of successes in the implementation of this screening initiative should be highlighted.Firstly, we previously reported on the high acceptability and feasibility of self-collected samples in the rural Bhutanese population. 9With this report, we can also now add high completion rates for colposcopy and treatment among HR-HPV-positive women in Bhutan, showing that this can also be achieved among women in remote areas.Only 11 of 264 (4%) careHPV-positive women did not attend colposcopy and undergo a biopsy.This low rate of loss to follow-up is expected to be related to the high trust of Bhutanese population in their public health system, and the close interaction with local health workers.Of course, the loss to follow-up in this well-supported and punctual research study may be less than in a more widespread government campaign in Bhutan, if the follow-up of screen-positive women is not carefully planned and supported.Similarly, our present findings cannot necessarily be expected to be representative of other settings outside Bhutan.Indeed, this number of 4% compares with 25%-30% loss to follow-up in some other experiences of recalling HPV-positive women in low/mediumincome countries. 30,35,36n the contrary, we also encountered certain technical challenges when implementing careHPV testing, as might be expected, it being the first experience in Bhutan.For instance, as previously reported, 9 there continued to be considerable wastage through invalid careHPV runs, even after validation of the initial training course.Indeed, lab implementation problems for careHPV were first noted for REACH-Bhutan, 9 but similar issues have since been mentioned in other reports, 30,[37][38][39] even leading to the active development of a statistical model for quality assurance of careHPV contamination issues. 40][47][48] Overall, the findings of the REACH-Bhutan study highlight the potential for cervical cancer screening programs in low-resource settings based on self-collection of samples for HR-HPV DNA testing.This is an approach that has been strongly endorsed by latest WHO guidelines for screening and treatment of cervical precancer lesions for cervical cancer prevention. 7This recommendation was informed by an IARC evaluation of cervical cancer screening methods, 8 showing that the use of vaginal samples collected by women themselves can achieve a similar sensitivity and specificity for the detection of CIN2+ or CIN3+, at least when PCR-based assays are used.Indeed, in addition to being implemented as the primary program in largely unscreened populations in low-resource settings, self-collection is increasingly being evaluated as a primary modality in organized HPV-based programs in higher income settings. 10,49,50Self-sampling may thus have a future role in the recently initiated policy of the Bhutanese government (https://www.moh.gov.bt/hspd/healt h-flagship) to shift their entire national cytology-based cervical cancer screening program to primary testing for HPV DNA.