Cervical Cancer Screening and Human Papillomavirus (HPV ... - KCE

34 Cervical cancer screening and HPV KCE reports vol.38
We also pooled the relative sensitivity and specificity of HPV testing compared to
cytology and of the combination of both cytology and HPV testing compared to each
test alone. The evaluation of these relative accuracy measures offers the advantage that
all types of studies involving concomitant testing with complete or incomplete
verification and randomised trials - can be included.
Summary ROC curve (sROC) regression was performed to assess the impact of study
characteristics on the diagnostic odds ratio 167 .
Absolute accuracy
We retrieved 24 cross-sectional studies where women were tested concomitantly with
168-171, 87, 172, 173, 136,
a Pap smear and an HPV assay in the framework of primary screening
174-176, 125, 177, 178, 54, 85, 179, 180, 137, 181-184
. The trials, carried out in three different areas in India
but described in one report were considered as separate studies 137 . In 10 studies,
women were referred for confirmation of disease status only when at least one
screening test was positive. In 8 studies a random sample of screen negatives was
referred allowing adjustment for verification bias, whereas in 6 other studies, all
enrolled subjects were submitted to colposcopy with biopsy if colposcopically
suspicious. In addition, the base-line results of 2 randomized clinical trials comparing
HPV versus cytology based screening were included in the meta-analyses of the relative
sensitivity 185, 186 .
Overall, the sensitivity of HC2 for finding underlying high grade intra-epithelial neoplasia
was 89.3% (95% CI: 85.2-93.4%) but varied over a large range between 50% 137 and
100% 174 (see Table 7). The observed sensitivity of HC2 was extremely low in the three
cross-sectional studies conducted in India: respectively 50, 70 and 80% 137 , and was also
lower than average in other developing countries (81% in Zimbabwe 136 , 83% in Brazil
184 , 88% in South-Africa 170 ). However, the sensitivity for CIN2+ was consistently high in
six studies conducted in Europe and North-America: pooled estimate of 97.9% (95% CI:
95.9-99.9%; p for inter-study heterogeneity = 0.22) 171, 174, 54, 85, 179, 183 .
The pooled specificity of HC2 in excluding high-grade cervical pre-cancer was 87.8%
(95% CI: 85.5-90.0%; range: 81-95%). In North-America and Europe, the pooled
specificity was higher: 91.3% (95%: 89.5-93.1%; range: 85-95%).
168, 169, 172, 175, 176,
In six studies, a PCR system was used for detecting HPV DNA sequences
125
. Its pooled sensitivity for CIN2+ (80.9%; 95%: 70.0-91.7%) was lower, but its pooled
specificity (94.7%; 95%: 92.5-96.9%) was higher compared to the HC2 assay.
Nevertheless, given the use of different primers and detection of amplified sequences,
this conclusion cannot be generalized. For instance: the sensitivity was 95% in a
German study where GP5+/GP6+ primers were used followed by hybridization with a
cocktail of oligonucleotides of 14 high risk HPV types 172 and only 64% in a British study
where the PCR/Sharp assay was used (MY09/MY11 primers, hybridisation with 10 highrisk
types) 169 .
The sensitivity and specificity of the combination of the HC2 assay and cytology,
considering ASCUS as cut-off for positivity, for detecting CIN2+, pooled from the 6
North-American and European studies, was 99.2% (95% CI: 97.4-100%, p=0.95) and
87.3% (84.2-90.4%) respectively. Overall, 14.5% (95% CI: 11.0-18.1%) of screened
women showed a positive result for at least one test.
The accuracy of HPV DNA testing with the purpose of finding CIN2 or CIN3 or
cervical cancer, showed substantial and statistically very significant heterogeneity, even
when separated by type of HPV test system. The simultaneous variation of the
sensitivity and specificity of HPV DNA testing is shown in the sROC curve in Figure 7.
Studies conducted in Europe or North-America, where HC2 was used, are clustered in
the upper right corner of the ROC space. The area under the sROC curve was 96.1%
(95% CI: 94.2-97.5%). The main factor that explained heterogeneity was the
geographical continent. The diagnostic odds ratio (DOR) did not vary significantly by
completeness of gold standard verification, indicating that verification bias was limited.