Human papilloma virus genotyping for the cross-sectional and longitudinal probability of developing cervical intraepithelial neoplasia grade 2 or more.

Human papilloma virus (HPV) testing is more sensitive but less specific than cytology. We evaluated stand-alone genotyping as a possible triage method. During a multicentre randomised controlled trial comparing HPV testing to conventional cytology, HPV-positive women were referred to colposcopy and followed up if no high-grade lesion was detected. HPV-positive samples were genotyped by GP5+/GP6+ primed polymerase chain reaction followed by reverse line blot. Genotypes were hierarchically ordered by positive predictive value (PPV) for CIN grade 2 or more (CIN2+), and grouped by cluster analysis into three groups (A, B and C in decreasing order). Receiver operating characteristic curves were computed. Among 2,255 HPV-positive women with genotyping, 239 CIN2+ (including 113 CIN3+) were detected at baseline or during a 3-year follow-up. HPV33 had the highest PPV with CIN2+ and CIN3+ as the endpoint and when considering lesions detected at baseline or also during follow-up. HPV16 and HPV35 were the second and third, respectively. Cross-sectional sensitivity for CIN2+ at baseline was 67.3% (95% CI 59.7-74.2), 91.8% (95% CI 86.6-95.5) and 94.7% (95% CI 90.2-97.6), respectively, when considering as "positive" any of the HPV types in group A (33, 16 and 35), A or B (31, 52, 18, 59 and 58) and A or B or C (39, 51, 56, 45 and 68). The corresponding cross-sectional PPVs for CIN2+ were 15.8% 95% (CI 13.2-18.7), 12.0% (95% CI 10.3-13.9) and 9.6% (95% CI 8.2-11.1), respectively. HPV33, 16 and 35 confer a high probability of CIN2+ but this rapidly decreases when adding other genotypes.

RCT Entities:   Population Interventions Outcomes

confidence interval (95%)

cervical intraepithelial neoplasia

human papilloma virus

liquid‐based cytology

negative predictive value

New Technologies for Cervical Cancer

polymerase chain reaction

positive predictive value

randomized controlled trial

receiver operating characteristic

Introduction

Cervical cancer screening based on HPV testing allows earlier diagnosis of persistent high‐grade cervical intraepithelial neoplasia (CIN) than cytology‐based screening1 and is more effective in preventing invasive cervical cancer.2 However, as the specificity of HPV testing for high‐grade CIN is low,1 a triage for selecting which HPV‐positive women are referred to colposcopy is needed. All randomized controlled trials (RCTs) except one (NTCC, New Technologies for Cervical Cancer) referred to colposcopy only HPV‐positive women with abnormal cytology or persistent HPV infection. This approach, compared to stand alone cytology, increased efficacy without increasing the biopsy rate.2 Nonetheless, short‐term test repeats are needed, which is disturbing for women3 and entail appreciable loss to follow‐up4 so investigating other triage methods is of interest.

It has been shown that different oncogenic HPV genotypes5 have different cross‐sectional6 and longitudinal7 strengths of association with high‐grade CIN and invasive cervical cancer. In particular, HPV16 has been shown to be strongly associated with both high‐grade CIN and invasive cervical cancer.8, 9, 10 HPV31, 33, 35, 52 and 58 are particularly associated to high‐grade lesions.8 HPV18 and 45 are not clearly associated with CIN38 but are the second and third most common types in invasive cervical carcinoma and are specifically associated with adenocarcinoma and glandular intraepithelial lesion.11

We used the biobank stored during the NTCC trial to study the cross‐sectional and longitudinal association of HPV genotypes with high‐grade CIN. Previously, partial genotyping for HPV16 and 18 has been studied as a triage method in association with cytology.6, 10 This strategy has already been included in the American multi‐societal and United States Preventive Services Task Force (USPSTF) guidelines.12, 13 Here, we report the accuracy of stand‐alone genotyping as a triage test.

Materials and Methods

NTCC is an RCT, with two pre‐planned recruitment phases, which was conducted in nine population‐based cervical screening programs in Italy. Women aged 25–60 years who were not pregnant, had never underwent hysterectomy, had not been treated for CIN in the last 5 years and who were attending for a new routine cervical screening episode were randomly assigned, between February 2002 and December 2004, to conventional cytology (classified according to the Bethesda 1991 system and managed according to the standard protocol of each centre) or to HPV‐based screening, either in combination with liquid‐based cytology (phase 1)14, 15 or alone (phase 2).16 Details on randomization and masking have been reported previously.14, 15, 16, 17

Cervical specimens were collected using a plastic Ayre's spatula and a cytobrush, and eluted into the PreservCyt buffer (Hologic/Gen‐Probe, San Diego, CA (phase 1) or in the standard transport medium (STM, Qiagen, Hilden, Germany) (phase 2). The presence of HPV in cervical specimens was evaluated by hybrid capture 2 (HC2, Qiagen) using probes designed to detect 13 high‐risk HPV types (16, 18, 31, 33, 35, 39, 45, 51, 52, 56, 58, 59 and 68).

NTCC is registered as an International Standard RCT, number ISRCTN81678807. We obtained central and local research ethics approvals.

Colposcopy and post‐colposcopy follow‐up

During phase 1, all women in the HPV arm were referred to colposcopy if cytology was atypical squamous cells of undetermined significance (ASCUS) or more severe. In addition, HPV‐positive women with cytology less than ASCUS were directly referred to colposcopy if aged 35–60 years,14 while those aged 25–34 years were referred to colposcopy only if after 1 year HC2 remained positive or cytology was ASCUS or greater.15 During phase 2, women in the HPV arm were referred for colposcopy if the HPV test was positive.16 Therefore, in the experimental arm, management at recruitment varied by phase and age.

As a rule, women with CIN2+ were treated and those HPV arm, HPV‐positive women were actively recalled after colposcopy for annual repeats of HC2 and liquid‐based cytology (LBC), as long as HPV remained positive. They were referred to colposcopy if LBC was ASCUS or more severe.17

Genotyping

Storage of residual cell samples after a positive HC2 test was set up during both study phases in Florence, Padua, Trento and Turin, while in Bologna, Imola, Ravenna and Viterbo, it was only performed during the second phase and in Verona, it was never performed. Residual cells from samples collected in PreservCyt were stored at −80°C following double PBS washing to remove methanol; aliquots (400 μl) from those collected in STM were taken before HC2 testing and directly stored at −80°C in their own solution.

The first HC2‐positive sample of each woman was genotyped. Genotyping was performed, blindly to histology results, by GP5+/GP6+ primed polymerase chain reaction (PCR), followed by reverse line blot.18, 19 Although GP5+/GP6+ PCR could in principle detect >40 HPV types, in this analysis, HPV genotyping was restricted to the 13 high‐risk HPV types targeted by HC2. In addition, although other methods were applied in samples where no HC2‐targeted HPV types were identified,20, 21 their results are not considered in the present analysis. Typing results were not used for women's management.

Endpoint assessment

Endpoints were histologically confirmed CIN2, CIN3 or invasive cervical cancer.

We recorded test results and histological findings from the computerized registration systems of participating screening centres. At the end of the recruitment phase, for women who had a histology originally diagnosed as CIN1 or more severe by the local pathology laboratory, all histological specimens taken within 1 year of referral to colposcopy were reviewed. For each woman, all histological slides, both from colposcopy‐guided biopsies and treatments, were provided together, and the most severe diagnosis was used. Each case was randomly assigned to one or two pathologists randomly selected from a group of nine who were unaware of the original diagnosis, randomization or genotyping. Only morphological criteria were used. If a pathologist did not agree with the original diagnosis regarding the presence of CIN2+, then the information was discussed by at least two further pathologists. This consensus diagnosis was then used in all the published analyses of NTCC data. The same exercise was repeated at the time of the second screening round. The results of such reviews were previously reported.14, 16, 17, 22, 23 Adenocarcinoma in situ was considered with CIN3. In addition, to obtain histological diagnoses performed outside the trial, after the end of the second round, we linked the database of recruited women to those of the cancer registries (covering all centres except Viterbo) and of the Pathology Units present in the catchment areas of NTCC.17 Considering the women included in this analysis, 44 of 224 (19.6%) with histology locally diagnosed as CIN2+ at baseline or follow‐up were downgraded to CIN1 or no CIN while 19 of 406 (4.7%) with histology locally diagnosed as CIN1 were upgraded to CIN2 or more severe. For 41 women with CIN2+ diagnosis, slides for review could not be retrieved and the local diagnosis was kept. Five other were added from cancer registries.

For this analysis, lesions detected at baseline were defined as those detected within 1 year from referral to colposcopy. Lesions detected thereafter were defined as detected during follow‐up.

Patients

This analysis was restricted to women in the experimental arm who received HPV‐based screening (n = 47,369). To have colposcopic verification among almost all HPV‐positive women, those aged 25–34 years recruited during phase 1 were excluded from all analyses (n = 6002), and a further 7,877 women from centres where samples for genotyping were not stored were excluded leaving 33,490 women. Among them, 33,249 women had a valid HC2 result and 2,498 were found to be HPV positive. A further 100 HC2‐positive women were excluded from our analysis because their corresponding samples for genotyping were not available, and 143 women were excluded because they did not have colposcopy. The final sample for analysis was 2,255 HPV‐positive women (Supporting Information Fig. S1). Of these, 1,685 had at least one HPV genotype targeted by HC2, while in the remaining either low‐risk HPV types (n = 302) or no HPV types were detected by PCR (n = 268).

Of the 2,084 women who had no high‐grade CIN detected at enrolment, 1,900 (91%) received further tests as part of the post colposcopy follow‐up. Median duration of their follow‐up was 1145 days (IQR 796–1473). Of the 1,900 women, 1,636 (86%) were actively followed up until either detection of high‐grade CIN, negative for a HPV test or for at least 3 years, while 264 (14%) had a shorter follow‐up without disease or resolution of HPV infection (not presented for retesting).

Statistical methods

For each individual high‐risk HPV genotype, prevalence and positive predictive values (PPV) for CIN2+ and CIN3+ were calculated. Hierarchical ranking of HPV types was obtained by iteratively determining the subsequent genotype with the highest PPV excluding cases with multiple infections with higher ranking types. Receiver operating characteristic (ROC) curves were plotted showing cumulative sensitivity and specificity accounting for the genotype hierarchy. A k‐means cluster analysis was used to group the hierarchical ordered individual genotypes into three (highest risk A to lowest risk C) groups based on the hierarchical PPV values. Clusters were determined by iteratively calculating the sum of squares for each cluster with the aim of minimizing the within cluster variation. Sensitivity, specificity and absolute risk of disease for women positive (PPV) and negative (1 minus negative predictive value [NPV]) for certain HPV types were calculated. Relative risks for CIN2+ and CIN3+, if positive for different genotypes, were also calculated. Diagnostic accuracy was computed for individual genotypes in cluster A (very highly predictive), then cumulatively for cluster B (highly predictive) and C (intermediately predictive). Associated 95% confidence intervals (95% CI) were computed based on the exact binomial distribution. Data were also computed for the HPV16 and 18 partial genotyping, since this is proposed by some commercially available tests.

Pretest–posttest probability (PPP) plots starting from the HPV test were constructed. For groups of genotypes, the post‐positive test and post‐negative test probabilities were the corresponding PPV and 1‐NPV, respectively. The post‐negative‐HPV probability was computed including all the lesions detected up to the following screening round within NTCC (up to 3 years).

All analysis was conducted in Stata 13.1 (StataCorp, College Station, Texas).

Results

Overall, 2,255 HPV‐positive women with genotyping for the 13 types targeted by HC2 were included in the analysis. At enrolment, 171 (7.6%) women had CIN2+ histology (including 80 CIN3+), and among them, 9 CIN2+ (3 CIN3+) cases had no high‐risk HPV genotypes detected. Furthermore, 68 CIN2+ were detected during the 3‐year follow‐up (including 33 CIN3+). Thus, 239 CIN2+ (113 CIN3+) cases were detected overall, of which, 13 CIN2+ (4 CIN3+) cases (5.44%) had no high‐risk HPV types detected.

HPV16 was the most common genotype identified in this HC2‐positive population, followed by HPV31 and 18, with 28.0%, 12.9% and 9.0% prevalences, respectively (Table 1). The number of women positive for each HPV genotype and of those with CIN2+ and CIN3+ detected at baseline and during follow‐up, after adjusting for the type hierarchy, are reported in Table 2. For each endpoint and time period, genotypes are ordered by PPV and clustered. HPV33 was the genotype with the highest PPV, both when considering the cross‐sectional and longitudinal analysis, and for both CIN2+ (22.4% and 25.4%, respectively) and CIN3+ (11.9% and 11.9%, respectively). HPV16 was always second and HPV35 third in the hierarchy, except for outcome CIN2+ in the longitudinal analysis, where the order was reversed. When excluding cases with disease detected at baseline, HPV35 was very predictive of CIN2+ during follow‐up. When grouping genotypes by cluster analysis (Table 2), with CIN2+ at baseline as the endpoint, HPV types 33, 16 and 35 formed cluster A, the most predictive of high‐grade lesions. The cumulative prevalence of cluster A was 32.2% of all HC2 positive women. HPV18, 31, 52, 58 and 59 were included in cluster B, while HPV 39, 45, 51 and 56 were included in cluster C. Types 18, 31, 58 and 59 were included in cluster B for all endpoints and type 52 for all except CIN3+ (baseline or follow‐up). Conversely, HPV51 and 45 were always and HPV68 three out of four times (excluding CIN3+ at baseline) included in cluster C. Considering the original histological diagnoses provided locally had limited effect on the order of genotypes by PPV. In particular, the first 3 genotypes remained unchanged independently of the endpoint grade and of cross‐sectional or longitudinal analysis. PPV slightly decreased as expected with a more misclassified endpoint.

Table 1

Number of women with high‐risk genotype, alone or with other types, and prevalence among study women

Genotype	Number of women with genotype, alone or with other types	Prevalence in HPV+ women (%)
HPV 16	632	28.03
HPV 31	291	12.90
HPV 18	203	9.00
HPV 56	181	8.03
HPV 45	129	5.72
HPV 58	119	5.28
HPV 51	117	5.19
HPV 39	94	4.17
HPV 52	90	3.99
HPV 59	72	3.19
HPV 33	67	2.97
HPV 35	50	2.22
HPV 68	45	2.00

In this table, women appear many times in the case of multiple infections.

Table 2

Number of women and of those among them having or developing CIN2+ or CIN3+ who are positive for each new genotype in a hierarchical analysis.

Cluster	Genotype	Women1	CIN2+	PPV (%)	Cumulative prevalence in HPV+ (%)	Cumulative PPV (%)	Cluster	Genotype	Women1	CIN3+	PPV (%)	Cumulative prevalence in HPV+ (%)	Cumulative PPV (%)
A. Endpoint CIN2+ at baseline							B. Endpoint CIN3+ at baseline
A	HPV 33	67	15	22.39	2.97	22.39	A	HPV 33	67	8	11.94	2.97	11.94
	HPV 16	616	93	15.10	30.29	15.81		HPV 16	616	52	8.44	30.29	8.78
	HPV 35	44	7	15.91	32.24	15.82		HPV 35	44	3	6.82	32.22	8.87
B	HPV 31	238	20	8.40	42.79	13.99	B	HPV 58	100	4	4.00	36.67	8.10
	HPV 52	71	5	7.04	45.94	13.51		HPV 59	58	2	3.45	39.25	7.80
	HPV 18	139	9	6.47	52.11	12.68		HPV 68	38	1	2.63	40.93	7.58
	HPV 59	45	3	6.67	54.10	12.46		HPV 31	227	4	1.76	51.00	6.43
	HPV 58	88	5	5.68	58.00	12.00		HPV 18	132	2	1.52	56.85	5.93
C	HPV 39	58	2	3.45	60.58	11.64		HPV 52	59	1	1.69	59.47	5.74
	HPV 51	80	2	2.50	64.12	11.13	C	HPV 39	57	0	0.00	62.00	5.51
	HPV 56	125	1	0.80	69.67	10.31		HPV 45	93	0	0.00	66.12	5.16
	HPV 45	87	0	0.00	73.53	9.77		HPV 51	75	0	0.00	69.45	4.92
	HPV 68	27	0	0.00	74.72	9.61		HPV 56	119	0	0.00	74.72	4.57
C. Endpoint CIN2+ at baseline or follow‐up							D. Endpoint CIN3+ at baseline or follow‐up
A	HPV 33	67	17	25.37	2.97	25.37	A	HPV 33	67	8	11.94	2.97	11.94
	HPV 35	50	12	24.00	5.19	24.79		HPV 16	616	67	10.88	30.29	10.98
	HPV 16	610	121	19.84	32.24	20.63		HPV 35	44	5	11.36	32.24	11.00
B	HPV 31	238	28	11.76	42.79	18.45	B	HPV 39	74	5	6.76	35.52	10.61
	HPV 52	71	7	9.86	45.94	17.86		HPV 58	99	5	5.05	39.91	10.00
	HPV 39	65	6	9.23	48.82	17.35		HPV 59	55	2	3.64	42.35	9.63
	HPV 58	96	8	8.33	53.08	16.62		HPV 18	137	5	3.65	48.43	8.88
	HPV 18	125	10	8.00	58.63	15.81		HPV 56	135	4	2.96	54.41	8.23
	HPV 59	44	3	6.82	60.58	15.52		HPV 31	208	6	2.88	63.64	7.46
	HPV 56	127	8	6.30	66.21	14.74	C	HPV 52	58	1	1.72	66.21	7.23
C	HPV 51	78	3	3.85	69.67	14.19		HPV 45	90	1	1.11	70.20	6.89
	HPV 45	87	3	3.45	73.53	13.63		HPV 51	75	0	0.00	73.53	6.57
	HPV 68	27	0	0.00	74.72	13.41		HPV 68	27	0	0.00	74.72	6.47

PPV for each new genotype and cumulatively. Cumulative prevalence in HPV+ women. Genotypes are ordered by PPV for each endpoint.

1The number of women, CIN2+ and CIN3+ is computed considering only those not included in upper lines because they had multiple infections with previously considered genotypes. Each woman appears only once for each endpoint.

When considering as “positive” any of the types in group A, cross‐sectional sensitivity for CIN2+ at baseline was 67.3% (95% CI 59.7–74.2) (Table 3). This increased to 91.8% (95% CI 86.6–95.5) and 94.7% (95% CI 90.2–97.6) when adding cluster B (31, 52, 18, 59 and 58) and C (39, 51, 56, 45 and 68). The cumulative sensitivity was very similar for both endpoints, and for lesions detected at baseline and follow‐up. The risk of being CIN2+ was 3.7% (95% CI 2.8–4.7)) for women negative to the genotypes in cluster A (1‐NPV), and 1.5% (95% CI 0.8–2.5) for those negative to the genotypes in clusters A and B. The risk was similar for women who had group C types and for those who were HC2 positive but had no oncogenic type detected.

Table 3

Accuracy of individual genotypes and clusters for HC2 positive womena

Endpoint	HPV types	n/N b	Sensitivity	Specificity	PPV	1‐NPV	Relative riskc
CIN2+ (N = 171)	HPV 33	15/67	8.8 (5.0, 14.1)	97.5 (96.7, 98.1)	22.4 (13.1, 34.2)	7.1 (6.1, 8.3)	3.1 (1.9, 5.0)
	HPV 16	95/632	55.6 (47.8, 63.1)	74.2 (72.3, 76.1)	15.0 (12.3, 18.1)	4.7 (3.7, 5.8)	3.2 (2.4, 4.3)
	HPV 35	7/50	4.1 (1.7, 8.3)	97.9 (97.2, 98.5)	14.0 (5.8, 26.7)	7.4 (6.4, 8.6)	1.9 (0.9, 3.8)
	Cluster A	115/727	67.3 (59.7, 74.2)	70.6 (68.6, 72.6)	15.8 (13.2, 18.7)	3.7 (2.8, 4.7)	4.3 (3.2, 5.8)
	Cluster A or B	157/1308	91.8 (86.6, 95.5)	44.8 (42.6, 46.9)	12.0 (10.3, 13.9)	1.5 (0.8, 2.5)	8.1 (4.7, 13.9)
	Cluster A or B or C	162/1685	94.7 (90.2, 97.6)	26.9 (25.0, 28.9)	9.6 (8.2, 11.1)	1.6 (0.7, 3.0)	6.1 (3.1, 11.9)
	HPV 18	14/203	8.2 (4.5, 13.4)	90.9 (89.6, 92.1)	6.9 (3.8, 11.3)	7.7 (6.5, 8.9)	0.9 (0.5, 1.5)
	HPV 16/18	106/795	62.0 (54.3, 69.3)	66.9 (64.9, 69.0)	13.3 (11.0, 15.9)	4.5 (3.5, 5.6)	3.0 (2.2, 4.0)
CIN3+ (N = 80)	HPV 33	8/67	10.0 (4.4, 18.8)	97.3 (96.5, 97.9)	11.9 (5.3, 22.2)	3.3 (2.6, 4.1)	3.6 (1.8, 7.2)
	HPV 16	53/632	66.3 (54.8, 76.4)	73.4 (71.5, 75.2)	8.4 (6.3, 10.8)	1.7 (1.1, 2.4)	5.0 (3.2, 7.9)
	HPV 35	3/50	3.8 (0.8, 10.6)	97.8 (97.1, 98.4)	6.0 (1.3, 16.5)	3.5 (2.8, 4.3)	1.7 (0.6, 5.2)
	Cluster A	63/727	78.8 (68.2, 87.1)	69.5 (67.5, 71.4)	8.7 (6.7, 11.0)	1.1 (0.6, 1.8)	7.8 (4.6, 13.2)
	Cluster A or B	77/1341	96.3 (89.4, 99.2)	41.9 (39.8, 44.0)	5.7 (4.6, 7.1)	0.3 (0.1, 1.0)	17.5 (5.5, 55.3)
	Cluster A or B or C	77/1685	96.3 (89.4, 99.2)	26.1 (24.2, 28.0)	4.6 (3.6, 5.7)	0.5 (0.1, 1.5)	8.7 (2.8, 27.5)
	HPV 18	4/203	5.0 (1.4, 12.3)	90.9 (89.6, 92.0)	2.0 (0.5, 5.0)	3.7 (2.9, 4.6)	0.5 (0.2, 1.4)
	HPV 16/18	57/795	71.3 (60.0, 80.0)	66.1 (64.0, 68.1)	7.2 (5.5, 9.2)	1.6 (1.0, 2.4)	4.6 (2.8, 7.3)

Lesions detected at baseline within 1 year of enrolment.

n is the number of CIN2+ or CIN3+ cases; N is the number of women positive for the HPV type.

Risk of carrying CIN2+ (or CIN3+) among HC2 positive women who were positive for the relevant genotypes divided for the same risk in HC2 positive women who were negative for that genotypes.

Among women positive for types included in clusters A, A or B and A or B or C, the probability of CIN2+ (PPV) was 15.8% (95% CI 13.2–18.7), 12.0% (95% CI 10.3–13.9) and 9.6% (95% CI 8.2–11.1), respectively, and for CIN3+, it was 8.7% (95% CI 6.7–11.0), 5.7% (95% CI 4.6–7.1) and 4.6% (95% CI 3.6–5.7), respectively.

When including lesions detected during follow‐up (Table 4), the risk of CIN2+ if negative for HPV types in cluster A was 5.8% (95% CI 4.7–7.1) and decreased to 2.3% (95% CI 1.2–3.9) for types included in all three clusters. The PPV of CIN2+ was 20.6% (95% CI 17.7–23.8) if positive for HPV types in cluster A, and 14.7% (95% CI 13.0–16.6) and 13.4% (95% CI 11.8–15.1) cumulatively adding cluster B and C respectively.

Table 4

Accuracy of individual genotypes and clusters for HC2 positive womena

Endpoint	HPV types	n/N b	Sensitivity	Specificity	PPV	1‐NPV	Relative riskc
CIN2+ (N = 239)	HPV 33	17/67	7.1 (4.2, 11.1)	97.5 (96.7, 98.2)	25.4 (15.5, 37.5)	10.1 (8.9, 11.5)	2.5 (1.6, 3.8)
	HPV 35	12/50	5.0 (2.6, 8.6)	98.1 (97.4, 98.7)	24.0 (13.1, 38.2)	10.3 (9.1, 11.6)	2.3 (1.4, 3.8)
	HPV 16	124/632	51.9 (45.3, 58.4)	74.8 (72.8, 76.7)	19.6 (16.6, 22.9)	7.1 (5.9, 8.4)	2.8 (2.2, 3.5)
	Cluster A	150/727	62.8 (56.3, 68.9)	71.4 (69.4, 73.3)	20.6 (17.7, 23.8)	5.8 (4.7, 7.1)	3.5 (2.7, 4.5)
	Cluster A or B	220/1493	92.1 (87.9, 95.1)	36.9 (34.7, 39.0)	14.7 (13.0, 16.6)	2.5 (1.5, 3.9)	5.9 (3.7, 9.4)
	Cluster A or B or C	226/1685	94.6 (90.9, 97.1)	27.6 (25.7, 29.6)	13.4 (11.8, 15.1)	2.3 (1.2, 3.9)	5.9 (3.4, 10.2)
	HPV 18	19/203	7.9 (4.9, 12.1)	90.9 (89.5, 92.1)	9.4 (5.7, 14.2)	10.7 (9.4, 12.1)	0.9 (0.6, 1.4)
	HPV 16/18	139/795	58.2 (51.6, 64.5)	67.5 (65.4, 69.5)	17.5 (14.9, 20.3)	6.8 (5.6, 8.3)	2.6 (2.0, 3.3)
CIN3+ (N = 113)	HPV 33	8/67	7.1 (3.1, 13.5)	97.2 (96.5, 97.9)	11.9 (5.3, 22.2)	4.8 (3.9, 5.8)	2.5 (1.3, 4.9)
	HPV 16	68/632	60.2 (50.5, 69.3)	73.7 (71.7, 75.5)	10.8 (8.5, 13.4)	2.8 (2.0, 3.7)	3.8 (2.6, 5.5)
	HPV 35	5/50	4.4 (1.5, 10.0)	97.9 (97.2, 98.5)	10.0 (3.3, 21.8)	4.9 (4.0, 5.9)	2.0 (0.9, 4.7)
	Cluster A	80/727	70.8 (61.5, 79.0)	69.8 (67.8, 71.7)	11.0 (8.8, 13.5)	2.2 (1.5, 3.0)	5.0 (3.4, 7.4)
	Cluster A or B	107/1435	94.7 (88.8, 98.0)	38.0 (35.9, 40.1)	7.5 (6.2, 8.9)	0.7 (0.3, 1.6)	10.7 (4.7, 24.2)
	Cluster A or B or C	109/1685	96.5 (91.2, 99.0)	26.4 (24.6, 28.3)	6.5 (5.3, 7.8)	0.7 (0.2, 1.8)	9.2 (3.4, 24.8)
	HPV 18	7/203	6.2 (2.5, 12.3)	90.8 (89.5, 92.0)	3.4 (1.4, 7.0)	5.2 (4.2, 6.2)	0.7 (0.3, 1.4)
	HPV 16/18	75/795	66.4 (56.9, 75.0)	66.4 (64.3, 68.4)	9.4 (7.5, 11.7)	2.6 (1.8, 3.6)	3.6 (2.5, 5.3)

Lesions detected at baseline or during follow‐up analysis Within 3 years of recruitment.

n is the number of CIN2+ or CIN3+ cases; N is the number of cases positive for the HPV type.

Risk of carrying or developing CIN2+ (or CIN3+) among HC2 positive women who were positive for the relevant genotypes divided for the same risk in HC2 positive women who were negative for that genotypes.

The corresponding ROC curves for cumulative sensitivity and specificity after adjusting for type hierarchy are shown in Figure 1. The areas under the ROC curves were 78.0% for CIN3+ and 74.1% for CIN2+ when considering lesions detected at baseline (panel A), and slightly decreased to 74.8% for CIN3+ and 72.3% for CIN2+ when considering all lesions detected at baseline or follow‐up (panel B).

Figure 1

Hierarchical ROC Curves for genotyping among HPV‐positive women. Hierarchical ranking of HPV types was obtained by iteratively determining the subsequent genotype with the highest PPV excluding cases with multiple infections with higher ranking types. ROC curves show cumulative sensitivity and specificity accounting for the genotype hierarchy. [Color figure can be viewed at http://wileyonlinelibrary.com]

The relative risk of carrying a CIN3+ reached a maximum (17.5%) for women positive to the genotypes included in clusters A and B versus those negatives to the same types and then decreased to 8.7% when comparing the women positives to the genotypes included in clusters A, B and C to those negative. The same pattern was present for carrying CIN2+ and for longitudinal data. In any case, the single genotype associated to the highest relative risk (RR) was always HPV16.

The PPP plots, relative to positivity for HPV16, HPV16 or 18, cluster A, and cluster A + B are reported in Figure 2. A post‐test risk higher than 10% was obtained by positivity for HPV16 or cluster A, while the adding of other types slightly lowered this level.

Figure 2

Pre‐test and post‐test probability plots for screening by HC2 and triage by selected groups of genotypes. The plotted probabilities are of carrying CIN2+. For groups of genotypes, the post‐positive‐test and post‐negative‐test probabilities were the corresponding PPV and 1‐NPV, respectively. The post‐negative‐HPV probability was computed including all the lesions detected up to the following screening round within NTCC (up to 3 years).

Discussion

We have analyzed the cross‐sectional and longitudinal association of high‐risk HPV genotypes with CIN2+ and CIN3+ among women who tested positive for HC2 in a large cohort of Italian women enrolled in the NTCC trial. The hierarchical analysis for cross‐sectional and longitudinal risk of developing high‐grade lesions allowed grouping high‐risk HPV types into three clusters. Although ranking differences were observed in relation to different endpoints (lesion grade and time of detection), results were highly consistent. Cluster A, characterized by the highest PPVs, included, besides HPV16 (an expected observation), HPV33 and HPV35, while HPV18 was ranked lower in cluster B and HPV45 in cluster C.

Both HPV33 and 35 had a low prevalence in the study population. HPV33 ranked first based on PPV thus showing strong association with present and future CIN2+ and CIN3+ risk. This result is consistent with other recent reports which found a strong association between HPV33 and high‐grade CIN despite a much weaker association with invasive cancer.8, 24, 25, 26, 27 Conversely, the importance of HPV35 based on PPV was more unexpected even though a previous analysis within the future study28 reported a high probability of transition from HPV35 incident infection to CIN2 and CIN3 lesions after 3 years of follow‐up. In other studies, the association between HPV35 and high‐grade CIN varied remarkably between different geographical areas.8, 29 In a recent analysis on viral load and CIN risk,30 HPV35 was one of the four non HPV16/18 types (with HPV31, 52 and 58) whose viral load at baseline was significantly associated with a subsequent increase in CIN2/3 risk.

On the other hand, HPV18 and 45 showed a low PPV for CIN2/3+ despite being responsible for a considerable proportion of invasive cancers.31, 32 This discrepancy, already observed in many studies having high‐grade CIN as the endpoint,6, 10, 33, 34 could result from a different natural history of cell transformation induced by HPV18 and 45. In fact, these genotypes are over‐represented in adenocarcinomas and early onset cancers35, 36, 37 and are rarely prevented through cytological screening.2, 38, 39 It is, however, not clear if for these genotypes the co‐existing high risk of developing invasive cancer and low risk of having a pre‐cancerous lesion detected depends on a high rate of progression of such lesions to invasion or on the difficulty in detecting them by colposcopy. In any case, these two possible mechanisms result in a too short detectable sojourn time of pre‐invasive lesions. This makes cytological screening relatively ineffective in preventing invasive cancers caused by these genotypes.40

The risk ratios between HPV16 and the remaining hrHPV types observed in cohort studies evaluating the long‐term CIN3+ risk were 4.5 at 12 years in Kjær et al.10 and 5.8 at 10 years in Khan et al.9 Our corresponding RRs comparing hierarchically HPV16 to the remaining HC2 positive women were quite similar (5.0 [95%CI 3.2–7.9] and 3.8 [95%CI 2.6–5.5]). Some characteristics of our study, such as higher women's age at enrolment (median 38 years vs. 34 and 28 years) and shorter length of follow‐up can partly account for these differences. CIN3 and invasive cervical cancers associated with HPV16 are diagnosed at an earlier age than the corresponding lesions associated with other HPV types.11

Concerning the use as a triage test, the sensitivity of HPV16 does not seem sufficient (66.3% for CIN3+ cross‐sectional) to be used alone as a triage test in high‐income countries. Adding HPV33 and 35 could be reasonable but it must be taken into account that high‐grade CIN caused by these types could have a lower progression rate to invasion. In cluster A, the cross‐sectional sensitivity for CIN3+ reached 78.8% and the risk of CIN3+ in HC2 positive women without these types was 1.1%. By comparison, the cross‐sectional sensitivity for CIN3+ was 88.1% with ASC‐US+ cytology with knowledge of HPV infection (corresponding 1‐NPV: 0.6%)41 and 91% with p16.42 Adding other genotypes would increase sensitivity but at the cost of a strong loss in specificity. HPV18 and 45 represent a possible exception if the corresponding CIN had high probability of progression to invasion, which is unknown. The sensitivity and specificity of partial HPV16/18 genotyping, available for some clinically validated screening assays, was lower than that of HPV16 genotyping alone.

The areas under the ROC curves for cumulative sensitivity and specificity gave values around and above 75%, indicating good discriminating capacity as a triage test. Nonetheless, although this evaluation allows to compare two tests with very high accuracy, other factors must be taken into account, and it is considered a theoretical measure. On the other hand, the PPP plots, constructed to estimate the clinical use of different strategies of HPV genotyping, highlight that positivity for HPV16 or cluster A types (i.e. 33, 16 and 35) confer a risk >10%, while negativity for cluster A + B types confer a risk <1%. Indeed, while immediate colposcopy is generally warranted for a risk >10%, different and country‐specific decision thresholds do exist for risk levels in the 1–10% range.43

Combining genotyping with other triage methods (e.g., cytology) seems more promising than increasing the number of “positive” genotypes44 but this will be examined elsewhere.

This study has strengths and limitations. Our results are derived from a large randomized trial nested in an organized population‐based screening program, where over 70% of eligible women were enrolled, suggesting good applicability to routine activity. All HPV‐positive women were referred to colposcopy, and both participation to and completeness of follow‐up were high, so minimizing verification bias. Genotyping was blind to cytology and histology and its completeness within the areas/phases in which it was done was high and plausibly unselected. On the other hand, despite the large size, the precision of estimates for some less frequent HPV types is low and random variation could explain the differences between our observations and those obtained in other studies.8, 44 Previous screening intensity can account for differences in PPV compared to other studies24 and local variations in type distribution can result in differences in sensitivity and specificity. Indeed, type distribution changes between continents45 but also between areas in the same country.21 Differences with other studies can also result from different analytical sensitivity and specificity of different genotyping assays for different genotypes. For this reason, our result may not reflect exactly the PPV of typing, if we start from a different HPV DNA primary test (including starting directly with GP5+/GP6+ primed PCR). In fact some 10.7% of HC2 positive women were negative for GP5+/GP6+ primed PCR and plausibly another proportion negative to HC2 would have been positive to GP5+/GP6+. Nonetheless, cross‐sectional PPV and sensitivity for CIN3+ of HPV16 (8.4% and 66.3%, respectively) was consistent with what observed in the Addressing the Need for Advanced HPV Diagnostics (ATHENA) study34 (18.3% and 50.4%, respectively) and with a recent Italian study (18.8% and 55.6%, respectively),46 both using the COBAS 4800 HPV test (Roche) with partial genotyping.

In conclusion, we studied the cross‐sectional and short‐term longitudinal association of high‐risk HPV genotypes with high grade CIN on a large unselected population and found HPV 33, 16 and 35 to be the most strongly associated. Nevertheless, in this study the gradient was not sufficient to use genotyping alone as a method to triage HPV‐positive women, and combination with other tests is needed, taking also into account that high‐grade CIN caused by different genotypes could have different probabilities of progression to invasion.

Supporting Information

Click here for additional data file.