Selecting human papillomavirus genotypes to optimize the performance of screening tests among South African women

Abstract Human papillomavirus (HPV) testing is highly sensitive compared to cytology, with the trade‐off of being less specific. We investigated whether select combinations of HPV genotypes, ascertained by Linear Array (LA) and Xpert HPV (GX), can optimize sensitivity/specificity trade‐offs to detect high‐grade cervical intraepithelial neoplasia (CIN2+). In a study in Cape Town, South Africa, 586 women living without and 535 living with HIV, aged 30‐65 years, were recruited. Each woman underwent a pelvic exam to collect cervical samples (tested by LA and GX for 14 high‐risk HPV genotypes) and underwent colposcopy with histological sampling to determine CIN2+. In multivariable logistic regression of LA results, only HPV genotypes 16, 18, 31, 33, 35, 52, 58 were significantly associated with CIN2+ (P < .05). Xpert includes these seven types along with HPV 45 within three of the test's five channels and we defined these eight types as restricted genotyping (ie 16, 18, 31, 33, 35, 45, 52, 58). Full genotyping was defined as all 14 high‐risk types. Sensitivity estimates for full genotyping using LA were similar to that of restricted genotyping: 83.9% (full) vs 79.0% (restricted) in women without HIV and 93.0% (full) vs 88.9% (restricted) in women living with HIV. Specificity estimates improved for restricted vs full genotyping: 87.4% (full) vs 90.8% (restricted) in women without HIV and 63.7% (full) vs 71.4% (restricted) in women living with HIV. To optimize the performance of HPV testing for cervical cancer screening in high‐burden, under‐resourced settings like South Africa, only HPV 16, 18, 31, 33, 35, 45, 52, 58 could be included to define screen‐positive. We recommend the inclusion of HPV45 for its known link to adenocarcinoma.


| INTRODUCTION
HPV primary screening is increasingly being considered as the preferred approach for cervical cancer screening in many countries with established screening programs. 1 HPV primary screening also provides a viable option for high-burden, low-resource settings, especially since affordable, point-of-care assays are now available. [2][3][4] Point-of-care tests are conducted in the clinic setting with non-physician providers and produce results within the same day, alleviating the burden of limited laboratory infrastructure and personnel. 5 Furthermore, HPV DNA testing has been demonstrated to be considerably more sensitive than cytology 6,7 and is more sensitive, specific, and reproducible than the visual inspection with acetic acid (VIA). 8 The low sensitivity of cytology necessitates frequent, repeated screening to be effective, making it susceptible to high attrition and costly follow-up. 2 VIA is a subjective, visual skill that requires intensive quality assurance monitoring and frequent refresher courses to ensure test performance does not fluctuate by provider. 3,5,8 Although advantageous in many regards, HPV DNA testing has the trade-off of having relatively low specificity, 7,[9][10][11] particularly in women living with HIV. 5,12 In the screen-and-treat model, now widely recommended as the most appropriate approach for low-resource settings, the relatively low specificity of HPV DNA testing means that sizable numbers of women may be overtreated for transient infections that will ultimately regress. 13 Scarce resources would be unnecessarily spent on women at low risk for CIN2+ being referred for further colposcopy follow-up or treatment. 14 A potential option for reducing the risk of overtreatment is triaging HPV-positive women with VIA or cytology screening, [15][16][17][18][19][20] which guidelines from WHO and ASCO have outlined. 13,21 However, both of these approaches have limitations and other alternatives would be desirable.
Here we evaluate whether the strategy of limiting the number of high-risk HPV genotypes included in HPV screening assays could improve the clinical performance of HPV testing. Genotyping for specific HPV genotypes could provide risk stratification of HPV-positive women by identifying HPV infections that are at highest risk of persisting and developing into invasive cervical cancer. 20,22 Of the approximately 40 HPV genotypes that affect the anogenital tract, 14 types (16,18,31,33,35,39,45,51,52,56,58,59,66, 68) are considered high-risk for cervical cancer but vary in their degrees of oncogenicity. 23 HPV 16 and 18 are universally accepted as being the most carcinogenic (present in ~70% of invasive cancers). [24][25][26] However, there is less consensus surrounding the relative importance of the other high-risk genotypes. 22 Based on the results from genotyping over 12 000 invasive anogenital cancers, the relative contribution of HPV 16,18,31,33,45, 52, 58 to female anogenital cancer was found to be approximately 90% and, based on this, these are the high-risk genotypes incorporated in the nanovalent HPV vaccine. 27 This suggests that restricted genotyping could be effectively utilized to triage HPV-positive women in the context of screening.
In this study, we evaluated the impact on sensitivity and specificity for CIN2+ of restricting the number of HPV genotypes identified using the Roche Linear Array (LA) method that identifies each of the high-risk HPV genotypes individually. We also compared the individual LA results to the grouped HPV typing inherent in the Xpert HPV (GX) assay.

| Setting and population
We recruited women from two populations in Cape Town, South Africa. One was a community-based clinic (screening population) located in Khayelitsha township which provides general women's health services to the local community. The other was women referred to a colposcopy clinic at Groote Schuur Hospital, a university teaching hospital, for evaluation of abnormal cervical cytology (referral population). Women aged 30-65 years were eligible to participate if they: (a) had a known and documented HIV status, (b) were not previously treated for cervical disease, (c) did not have a hysterectomy, and (d) were not pregnant. The study aimed to (restricted) in women living with HIV. Specificity estimates improved for restricted vs full genotyping: 87.4% (full) vs 90.8% (restricted) in women without HIV and 63.7% (full) vs 71.4% (restricted) in women living with HIV . To optimize the performance of HPV testing for cervical cancer screening in high-burden, under-resourced   settings like South Africa, only HPV 16, 18, 31, 33, 35, 45, 52, 58 could be included to define screen-positive. We recommend the inclusion of HPV45 for its known link to adenocarcinoma.

K E Y W O R D S
HPV genotyping assays, human papillomavirus, sensitivity, South Africa, specificity recruit approximately equal numbers of women living with and without HIV. Seven hundred and fifteen women were recruited at the community-based clinic and 406 from the referral colposcopy clinic (total of 1121 women).

| Study procedures
This study was approved by institutional review boards at Columbia University (AAAO3652) and the University of Cape Town. Following written informed consent, each woman underwent a pelvic examination by a physician and had two liquid-based cytology specimens (ThinPrep Pap Test, Hologic) collected using a cytobrush and plastic spatula for each (Medscand).
After the collection of the cervical samples, a colposcopic examination was performed. Multiple cervical biopsies were obtained of the most abnormal-appearing areas of the cervix and/or an endocervical curettage (ECC) if there was no visible cervical lesion. If clinically indicated based on clinical protocol, a Loop Electrosurgical Excision Procedure (LEEP) was performed. Women from the screening population were told to return in 6 weeks after enrollment to receive their histology results. For those women recruited at the screening clinic, a second colposcopy was undertaken in those with HPV at baseline who had not had cervical intraepithelial neoplasia grade 2 or greater (CIN2+) detected at baseline. A third colposcopy was undertaken if the second colposcopy still failed to identify CIN2+. Random four-quadrant biopsies were undertaken at these repeat colposcopies to reduce risks of missing disease.
Histological samples collected via biopsy, LEEP, and/ or ECC were first evaluated by a pathologist in Cape Town and then blindly reviewed by an expert pathologist in the US. Results were classified using the cervical intraepithelial neoplasia (CIN) classification system: within normal limits (WNL), CIN grade 1 (CIN1), CIN grade 2 (CIN2), CIN grade 3 (CIN3), and cancer. 28 Any discrepancies were resolved by a third pathologist as a tie-breaker to reach a final consensus diagnosis.

| HPV testing
One mL of the first cervical sample was tested at the Khayelitsha screening clinic using the Xpert HPV assay (Cepheid). The remaining 19 mL were stored and tested later in batches at the University of Cape Town using Linear Array (LA) (Roche Diagnostics). The second sample was stored for possible later testing.
Xpert HPV (GX) is a clinic-based, point-of-care HPV assay that tests for 14 high-risk HPV types in five separate channels through real-time PCR. Each single-use Xpert HPV cartridge is prefilled with the necessary reagents, primers, and fluorescent probes to amplify the E6 and E7 regions of HPV DNA automatically on the GeneXpert System (Cepheid). Presence of HPV DNA that meets cycle thresholds is indicated by fluorescence in the following five channels: HPV16; HPV18, 45; HPV31, 33, 35, 52, 58; HPV51, 59; and HPV39, 56, 66, 68.
The Roche Linear Array (LA) is a laboratory-based assay that is not generally used in clinical practice but which enables the identification of 37 high-risk HPV genotypes individually. PGMY09/11 primers are used to amplify the L1 region of the HPV genome using real-time PCR. The resulting amplicons undergo reverse-line blot hybridization to individually detect genotypes. 29 Only the 14 high-risk types that are identified by the Xpert HPV assay were included in this analysis.

| Statistical analysis
Clinical and demographic information was summarized using descriptive statistics. Groups were compared using two-sample t-tests for continuous variables and chi-squared or Fisher's exact tests for categorical variables. Overall and type-specific HPV prevalence was calculated using only data from the general population. Frequency tables were used to determine HPV distributions by age group, HIV status, and cervical disease status. Diagnostic agreement between GX and LA was compared using percent overall agreement and Cohen's kappa statistic. We used multivariable logistic regression to determine which HPV genotypes were significantly associated with CIN2+ and CIN3+ (P < .05). By study design, women with CIN2+ and CIN3+ identified in both the general and the referral populations were included whereas women with <CIN2 or <CIN3 only from the general population were included in the analyses. This design was selected to obtain a sufficiently large number of women with cervical disease with the resources available. 30 After selecting the significant genotypes from the multivariable logistic regression results, we calculated sensitivity and specificity comparing full vs restricted genotyping. Ninety-five percent confidence intervals (CI) were calculated using the binomial estimates of the standard errors around these proportions. All statistical analyses were conducted using SAS 9.4. The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.

| Demographic and clinical characteristics
Seven hundred and fifteen women were recruited at the community-based clinic in Khayelitsha and 406 women from the referral colposcopy clinic. Of the 1121 women, 1118 had valid GX and LA results. Baseline demographic and clinical characteristics of the women recruited at the community-based clinic are summarized by HIV status in Table 1. Our sample had similar proportions of women without HIV (382, 53.50%) and women living with HIV (332, 46.50%) by design. The prevalence of cervical intraepithelial neoplasia was higher for women living with HIV (17.02% CIN2+, 8.51% CIN3+) compared to women without HIV (5.29% CIN2+, 3.17% CIN3+). Women living with HIV were significantly younger (P < .001), had fewer children (P < .001), were less educated (P = .001), and had more prior cytology screenings (P = .006) than women without HIV. Of the women living with HIV, 265 (79.82%) reported currently taking antiretroviral medications. There were no differences observed for tobacco use (P = .954) and employment status (P = .740) by HIV status.

| Overall
For both assays, HPV prevalence for any type was higher regardless of age group for women living with HIV (48.   (Figure 1).

| Type-specific
Using GX in women living with HIV, the channel detecting HPV

| Diagnostic concordance
The overall concordance between GX and LA for any of the 14   and 58 were significantly associated with CIN2+ in multivariable analysis. HPV 59, which was significantly associated with CIN2+ in univariable analysis, was no longer associated with CIN2+ once the above-mentioned HPV types were included in the model. In sum, both GX and LA assays identified the same seven HPV genotypes. The exception was HPV45 which was not significant as an individual type in the LA analysis but was included along with HPV 18 in one channel on GX. For the purpose of further analysis, we included HPV45 (due to its known links to adenocarcinoma) along with the seven consistently identified genotypes, in our calculations of selected typing. Full typing was defined as all five channels for GX and the 14 individual high-risk HPV types for LA. Results from the univariable and multivariable logistic regression analyses for CIN3+ were similar and are included as Table S1.

| Comparison of full vs restricted genotyping
After selecting the genotypes that were significantly associated with CIN2+ from the multivariable logistic regression results, we calculated sensitivity and specificity comparing full vs restricted genotyping. For both assays, there was only a slight decrease in sensitivity for restricted vs full genotyping.  (Table 3).
Sensitivity was slightly higher using GX than LA. In 11 women with CIN2+ who did not have one of the 14 genotypes detected by LA, but who did have them detected by GX, two were positive on the HPV 16 channel, two on the HPV18,45 channel, and nine positives on the HPV 31, 33, 35, 52, 58 channel (2 cases overlapped with the HPV 18, 45 channel). Nine of the 11 were reported as "no isolates" on LA, one was HPV 53 and one was IS39. In four women with CIN2+ who did not have one of the 14 genotypes detected by GX, but who did have them detected by LA, the HPV genotypes detected in LA were: 45; 16 Sensitivity estimates were slightly higher and specificity estimates lower for detection of CIN3+ compared to CIN2+ but the benefits of restricted compared to full genotyping were similar (Table S2).

| Discussion
To optimize the clinical performance of HPV testing in screening programs, our results indicate that only HPV 16,18,45,31,33,35,52, and 58 should be tested for. Although HPV 45 was not significantly associated with our disease endpoint, CIN2+, we include this genotype since it is the third most commonly identified genotype in invasive cervical carcinoma. 25 Lack of significance in our analysis may be attributed to the small number of cases of invasive carcinoma in our sample. In this analysis, we observed that restricted T A B L E 3 Sensitivity to detect cervical intraepithelial neoplasia grade 2 or worse (CIN2+) with selected typing vs full typing using Xpert HPV on GeneXpert or Linear Array genotyping, as compared to full genotyping, resulted in only marginal reductions in sensitivity but sizeable improvements in specificity. Since the inclusion of HPV 51, 59, 39, 56, 66, 68 when testing did not appreciably improve clinical performance, these genotypes could be excluded when HPV testing in high-burden low-resource settings where the capacity to treat or follow-up HPV-positive women is limited. Genotyperestriction was even more effective for women living with HIV compared to women without HIV, with a greater increase in specificity observed moving from full to restricted genotyping. However, the highest specificity achieved with restricted genotyping for women living with HIV (67.77% GX, 71.43% LA) was still considerably lower than that of HIV negative women (89.66% GX, 90.78% LA). While improvements have been made, there is still a need to further explore mechanisms to improve specificity for women living with HIV, especially in settings such as South Africa which had an estimated HIV prevalence of 18.8% in 2017 (4th highest in the world). 2 We examined whether the inclusion of only types 16 and 18 might be a simpler approach to HPV type selection. We found that restricting only to these types led to an unacceptable decline in sensitivity. Positive predictive value in the group with types 16 and 18 was not appreciably better than the PPV in the group of eight types that we selected.
In our analysis, we have additionally shown that pointof-care HPV DNA testing is a robust option for primary cervical cancer screening in limited-resource settings like South Africa. The point-of-care GX used in South Africa has the benefit of rapid same-day results and minimal lab infrastructure/personnel. In this analysis we found it to have equivalent clinical performance characteristics as laboratory-based LA. 31 Concordance between the two assays was excellent (94%), with a kappa of 0.88 in women without HIV and 0.79 in women living with HIV. Furthermore, this analysis confirms the validity of grouping HPV genotypes into the five separate channels offered by GX. The seven individual high-risk HPV genotypes identified as significantly associated with CIN2+ by LA aligned with the same genotypes included in the first three channels of GX. Since the carcinogenic behavior of each HPV type is closely related to its phylogenic category, it is not surprising that these seven HPV types belong to the higher risk species related to cervical cancer: alpha-9 (HPV 16, 31, 33, 35, 52, 58) and alpha-7 (HPV 18). 32 In particular, HPV35 was highly correlated to CIN2+/CIN3+ within our sample, which is consistent with recent studies that have found a strong link to cervical carcinogenesis for women of African ancestry with HPV35. 33 The GX platform is a pragmatic choice for South Africa given its current widespread usage across the country for tuberculosis testing 34 which offers potential leverage to utilize the same instruments for HPV screening. The global placements of the GX system now exceed 23 000 instruments, a large portion of which are in low resource settings (where TB is common), including Africa, Asia, and Latin America. 35 In addition, GX HPV testing has met international performance standards for use in primary cervical cancer screening. 30 Several countries have piloted the GX HPV test including South Africa, 36 Cameroon, 37 UK, 30 Zambia, 38 Saudia Arabia, and the USA. 39,40 Sensitivity rates in detecting CIN2+ ranged from 63.2% to 98.7% and specificity ranged from 42.6% to 90.3%. In our study, restricting HPV genotypes allowed us to achieve specificity rates that exceed estimates reported in most studies and should be considered for improving test performance. We observed only marginally lower specificity with GX compared to LA consistent with slightly higher sensitivity in GX compared to LA. It is possible that cross-reactivity with low-risk types or amplification of low signal when HPV is detected in pooled channels may contribute to these results. 41 There are a few limitations to our study. First, we used disease endpoints (CIN2+/CIN3+) in our analysis but realize that not all of these cervical lesions will progress to cancer. Some researchers criticize the use of CIN2+ due to lack of reproducibility and the higher likelihood of these lesions regressing compared to CIN3+. 7,36 However, most published studies have used CIN2+ as an endpoint to have the international performance standards, which allows us to make meaningful comparisons. 42 Nonetheless, our comparison of CIN2+ and CIN3+ showed similar results and the same conclusions about the clinical performance of GX and LA were drawn from either endpoint. Another limitation was that our sample size was small. Due to limited resources, we were unable to obtain sufficient cases of CIN2+ from the general screening population. Therefore, we needed to sample patients from a colposcopy referral clinic to obtain sufficient numbers of women with cervical disease for our analysis. Therefore, our sensitivity calculations included patients from the screening and referral populations, while specificity was calculated using only the screening population. This enrichment approach is supported in the recommendations for HPV test validation. 30 The landscape of cervical cancer prevention is rapidly changing and on the horizon is primary screening with HPV DNA testing in resource-constrained settings. 2 Restricting the definition of screen-positive to women with HPV genotypes 16, 18, 45, 31, 33, 35, 52, or 58 substantially improves the specificity of screening, while producing only minor reductions in sensitivity compared to inclusion of all 14 HPV genotypes conventionally included as high risk in HPV assays. Using the GX assay, genotype restriction can be easily accommodated given the existing channels. The lack of population benefit of testing for some of the HPV genotypes typically classified as "high-risk" has been previously noted. 43 Comparison of the distribution of individual high-risk genotypes identified using LA did detect other HPV groupings that appreciably improved specificity on the groupings of the GX assay. Furthermore, comparable performance between LA and GX shows that HPV screening can be done at point-of-care with considerably less demand for resources. Point-of-care testing with GX is feasible and restricted genotyping can maintain sensitivity while improving specificity. As such, the next step is to integrate HPV testing into cervical cancer screening programs to improve the health of women.