Main Findings
Our study found a high burden of high-risk HPV infections among women attending family planning services in Western Kenya with a more than 2-fold higher burden among HIV infected women. VIA, the current screening test in Kenya and most sub-Saharan African countries (SSA), only poorly correlated with the HPV test, as nearly 2/3 of VIA positive women had no HPV infection and would thus be overtreated.
We used p16INK4a/Ki-67 dual stain cytology in order to differentiate common, and often spontaneously clearing HPV infections from persistent, transforming infections, and their morphological correlate, a CIN2+ lesions. When comparing VIA with dual stain cytology we found that VIA only detected a small fraction of transforming HPV infections and only few among them were HPV 16 and/or 18 infections, which have the highest potential to develop cancer.
A primary HPV testing strategy should be considered also in resource-poor countries. Provided its cost-effectiveness is established, dual stain cytology could be suitable triage test. The assistance of an electronic data system will greatly facilitate such a multi-contact approach.
Interpretations of results
HPV burden
We found a high overall burden (32.5%) of high-risk HPV infections in our study population, similar to results from another study in our Western Kenya region 33. A number of other studies done in SSA report lower 34 35 36 37 38 39 or similarly high 40 41 HPV burden. All these studies were facility-based with often unknown cervical morbidity and different distribution of risk factors for HPV acquisition, i.e. enrolled populations differed by age distributions and HIV status. A true population HPV prevalence is still lacking for the region. Consistent among all of these studies was the 1.5-2 fold increase of HPV infection among the HIV infected as compared to HIV-neg women- when reported.
HPV genotypes
The five most common genotypes in our study were HPV16, 52, 68, 58, 35. Again, the genotype distribution is influenced by the state of the cervical disease, the geographic region as well as by technical issues including the HPV platform used, assay cut-offs and the selection of hrHPV types included in the analysis. In our study we were able to classify the genotype distribution according to the dual stain cytology findings. A strong association of HPV16, 18 and also HPV31, 58, 68 was found with dual stain cytology positive cases consistent with findings of a meta-analysis that compared normal and HSIL cases in East Africa 42. In the dual stain negative group HPV16, 52, 35, 58, 68 were most common which is in line with a meta-analysis among African women with normal cytology 43 44 .
Surprisingly we found a high burden of HPV68, a genotype that is uncommon in epidemiological surveys worldwide. In our study the high proportion of HPV68 was found both in the dual stain positive and dual- tain negative cases. Two-third of HPV68 was associated with multiple infections and its frequency was more than 2 times higher in HIV infected women. This is consistent with another study in the same region of Kenya where a high HPV68 prevalence among HIV-infected women was found 33. A study conducted in an isolated rural community in Brazil reported HPV68 as the most prevalent genotype, however it was not present in women with cytological abnormalities 45. The epidemiological importance of HPV68 needs to be further evaluated, especially as assay-related variation in HPV68 detection has been reported 46.
VIA
The overall VIA positivity in our study population (7.1%), diagnosed by well trained and experienced VIA nurses was similar to reports from India 47, Cameroon 39 and Tanzania 48 but lower than in a meta-analysis of 15 studies in SSA 49 were the pooled estimate of positivity was 17.4% (95&CI 10.4 to 25.6). A high variation in test positivity is observed among VIA studies conducted world wide 49, such discrepancies in VIA positivity is explained by patient characteristics (age, HIV status, precancer prevalence) and inherent procedure issues (high inter-operator variability, unamenable to quality control).
Dual stain cytology
The dual stain cytology positivity of 1.9% in the HIV uninfected population reflects the expected proportion of CIN2+ lesions in similar unscreened populations 49 50. Also, the strong association of the highly oncogenic HPV16, 18, 31 with dual stain positivity is well in line with the increased risk of (pre-)cancer associated with these HPV types 51.
All but one dual stain positive samples were also HPV positive. Possible explanation for such rarely found case 52 53 54 could be a low viral load or an HPV type not included in the common tests.
HIV/HPV co-infection
The detrimental effect of an HIV /HPV co-infection is well known 55,56. HIV significantly impacted on the outcome of VIA and dual stain cytology screening. The proportion of dual stain positivity was three times higher among the HIV infected population. The VIA positivity among the HIV infected women was twice as high (12.5%) compared to HIV uninfected women (6.7%). Studies of cohorts with high HIV burden in South Africa and Western Kenya reported high VIA positivity rates ranging from 22% to 55% 13 57 58 59.
HIV-infected women in our study carried significantly more single and multiple hrHPV types compared to HIV-negative women. Among the currently available HPV vaccines the nonavalent (9v) vaccine would provide additional protection. Given the substantial number of non-9v hrHPV infections among HIV-infected women found by us and others 60 61, however, careful post-vaccination followup will be important.
Evaluation of screening technologies
When comparing the studied screening techniques, the poor agreement between HPV status and VIA diagnosis is striking. Only 32% (16/50) of VIA positive cases were hrHPV positive or two third of women would be overtreated based on VIA alone which is consistent with a study from Cameroon where half of all VIA/VILI-DC positive women had no associated hrHPV infection 39.
Our complete dataset gave us the opportunity to evaluate the performance of VIA and HPV testing when using dual stain cytology as a surrogate marker of high-grade cervical lesions. The dual stain biomarker p16INK4a/Ki-67 has a well documented high sensitivity and specificity for identifying the presence of (pre-)cancer 54 62. The low sensitivity of VIA in detecting a p16INK4a/Ki-67 positive transforming infections (Table 4) casts serious doubts on the use of VIA as a primary screening test.
As primary HPV testing will likely become the standard of care for cervical cancer screening in both low- and high-resource settings, the search for a suitable triage strategy to avoid overtreatment is crucial 63 64 65 27. The relatively high specificity of VIA suggests that VIA could be used as triage test for HPV positive women 66 67. In our study such approach would have slightly improved the positive predictive value of VIA but still be burdened with a low sensitivity.
Dual stain cytology is currently the most extensively studied molecular triage test 68 69 31 70, that shows superior performance compared to other disease markers 71 72. In centralized laboratories the assay is performed on automated platforms offering high sample throughput. Recent advances in digital imaging and machine learning show promise to also automate the slide evaluation 73. Although dual stain cytology is unsuitable for self-collected samples 74, it is worthwhile to evaluate the test as part of novel screening algorithms also in LMICS 75 76 77.
Limitations
There were some limitations to this study. p16INK4a/Ki-67 is a very strong but not a definitive predictor of malignant degeneration (high-grade dysplasia). The gold standard for the diagnosis of high-grade cervical dysplasia is a cervical biopsy. Unfortunately, we could not perform colposcopies and biopsies for case ascertainment. Also, we used two different HPV DNA tests in our study, a nucleic acid hybridization test 78 and a PCR-based (GP5+/6+) test. Both tests, however, were validated in the Valgent study 78. Finally, the facility-based recruitment, the rural setting and the relatively low sample size preclude generalization of the study findings.