The role of human papillomavirus testing in the management of women with low-grade abnormalities: multicentre randomised controlled trial

Transcription

1 DOI: /j x Epidemiology The role of human papillomavirus testing in the management of women with low-grade abnormalities: multicentre randomised controlled trial S Cotton, a L Sharp, b J Little, c M Cruickshank, d R Seth, e L Smart, f I Duncan, g K Harrild, a K Neal, h N Waugh, a The TOMBOLA (Trial Of Management of Borderline and Other Low-grade Abnormal smears) Group* a Department of Public Health, University of Aberdeen, Aberdeen, UK b National Cancer Registry Ireland, Cork, Ireland c Department of Epidemiology and Community Medicine, University of Ottawa, Ottawa, ON, Canada d Department of Obstetrics and Gynaecology, University of Aberdeen, Aberdeen, UK e Division of Pathology, Queen s Medical Centre, Nottingham, UK f Department of Pathology, Medical School Building, Aberdeen, UK g Department of Maternal and Child Health Sciences, University of Dundee, Ninewells Hospital, Dundee, UK h Epidemiology and Public Health, Community Health Sciences, Queen s Medical Centre, Nottingham, UK Correspondence: Prof. J Little, Canada Research Chair in Human Genome Epidemiology, Department of Epidemiology and Community Medicine, University of Ottawa, 451 Smyth Road, Ottawa, ON, K1H 8M5, Canada. jlittle@uottawa.ca Accepted 20 December Published Online 18 March Objective To assess the value of a single human papillomavirus (HPV) test in making decisions on management of women with cervical cytology showing borderline nuclear abnormality (BNA) or mild dyskaryosis. In particular, to determine whether information on high-risk (hr) HPV status would be valuable in the choice between (1) cytological surveillance versus immediate referral to colposcopy, and (2) at colposcopy, between biopsy and recall versus immediate large loop excision of the transformation zone (LLETZ). Design Multicentre individually randomised controlled trial, nested within the NHS Cervical Screening Programmes, investigating the value of HPV testing by testing for interactions between HPV status and (1) cytological surveillance versus colposcopy, and (2) biopsy and recall versus immediate LLETZ. Setting Grampian, Tayside and Nottingham. Population Women (n = 4439), aged years, with a cytology test showing borderline nuclear abnormalities or mild dyskaryosis during October 1999 to October Methods High-risk HPV status was determined at recruitment using the polymerase chain reaction assay with the GP5+/6+ general primer system. The results of this HPV testing were not disclosed to either the participating women or to those involved in their management. Women were randomised to either (1) 6-monthly cytological screening in primary care or (2) referral for *Group members are listed at the end of the report. Name of trial, registration: TOMBOLA (Trial of Management of Borderline and Other Low-grade Abnormal smears), ISRCTN colposcopy. Human papillomavirus status was used to stratify both randomisations. All women were followed for 3 years, concluding with an invitation to an exit appointment at which colposcopic examination was undertaken. In addition, in women who were randomised to initial colposcopy and underwent colposcopy, the association between hrhpv status and presence of cervical intraepithelial neoplasia (CIN) grade 2 or more severe disease (henceforth CIN2 or worse) was examined. Main outcome measures Sensitivity, specificity, positive and negative predictive values of the HPV test for predicting CIN2 or worse and the implications for the choice of management between cytological surveillance and immediate referral for colposcopy. Results There were no significant interactions between management and HPV status. Hence, in women with mild dyskaryosis or BNA who are HPV positive, there is no advantage of (1) immediate colposcopy over cytological surveillance (P = 0.76) or (2) immediate LLETZ over biopsy and recall (P = 0.27). The sensitivity of HPV testing for detection of CIN2 or worse was 75.2% (95% CI %) among women with mild dyskaryosis and 69.9% (95% CI %) among those with BNA. Specificity was higher in those with BNA (71.3%; 95% CI %) than in those with mild dyskaryosis (46.9%; 95% CI %). Sensitivity decreased with increasing age whereas specificity increased. The negative predictive value was high, particularly among women with BNA (94.5%; 95% CI %). Across all ages, 22% of women who had CIN2 or worse were HPV negative. Conversely, 40% of those who were HPV positive did not have CIN. HPV was a much more reliable predictor in women aged over 40 years. ª 2010 The Authors Journal compilation ª RCOG 2010 BJOG An International Journal of Obstetrics and Gynaecology 645

2 Cotton et al. Conclusions We conclude that in younger women with low-grade cytological abnormalities, a single HPV test would not be useful in determining who should be referred for colposcopy or the most effective management at colposcopy. In women over 40, a negative HPV test could be used to rule out further investigation. Keywords Borderline nuclear abnormality, cervical intraepithelial neoplasia, cervical screening, cytological surveillance, colposcopy, follow-up, human papillomavirus, low-grade abnormalities, management, mass screening, mild dyskaryosis, randomised controlled trial. Please cite this paper as: Cotton S, Sharp L, Little J, Cruickshank M, Seth R, Smart L, Duncan I, Harrild K, Neal K, Waugh N, The TOMBOLA (Trial Of Management of Borderline and Other Low-grade Abnormal smears) Group. The role of human papillomavirus testing in the management of women with low-grade abnormalities: multicentre randomised controlled trial. BJOG 2010;117: Introduction In the UK, the NHS Cervical Screening Programmes (CSPs) have a high degree of sensitivity for the detection of cervical changes which may be precancerous, but this results each year in over a quarter of a million cytological tests showing low-grade abnormality (i.e. borderline nuclear abnormalities [BNA] and mild dyskaryosis). 1,2 The vast majority of these women are at a very low risk of cancer, but experience the anxiety and inconvenience of follow-up, which may result in colposcopy, biopsy and possibly treatment, with attendant adverse effects. There are substantial cost implications for the women concerned, and for the health services. 3 There has been controversy about the most effective and efficient management of the large numbers of women who have minor cytological abnormalities. 4 The main methods proposed are cytological surveillance or referral for colposcopy. In view of the role of certain types of human papillomavirus (HPV) in the aetiology of cervical cancer and precancer, 5 7 it has been suggested that an HPV test might be used to decide which women should be referred for colposcopy and which could safely be returned to routine recall However, HPV infections are common, and most are transient, with a median duration of 6 14 months Hence there is a need for rigorous evaluation of this potential application of HPV testing. The 15 types of HPV that have been associated with cervical neoplasia are known as high-risk (hr) HPV. 15 In a pilot implementation of testing for hrhpv in the triage of women with low-grade abnormalities in England, women who were hrhpv positive were referred for colposcopy. There was a 74% reduction in the rate of repeat cytology tests, but a doubling in referrals to colposcopy, compared with the period before the pilot. 16 In a cost-effectiveness analysis, it was noted that there was considerable uncertainty in both the incremental costs and incremental gains that would result from the use of HPV testing in triage, and a lack of data on the quality of life implications and societal costs. 17 A key limitation of the assessment of the pilot was that only women who were hrhpv positive were referred for colposcopy, so that only the positive predictive value (PPV) of an HPV test for cervical intraepithelial neoplasia grade 2 (CIN2) or worse (20.2%) could be determined, and not the sensitivity, specificity or negative predictive value (NPV). Despite this, the NHS CSP in England introduced HPV triage of BNA and mild dyskaryosis in six cytology laboratories in There have been a number of cross-sectional comparisons of the possible value of HPV testing and cytology in the detection of CIN2 or worse, most of which have been included in meta-analyses. 19,20 These meta-analyses suggest that in women with BNA or cytology indicating atypical squamous cells of uncertain significance (ASCUS), HPV testing is more sensitive in the detection of CIN2 or worse than repeat cytology, and has similar specificity. In women with mild dyskaryosis or cytology indicating low-grade squamous intraepithelial lesions (LSIL), the meta-analyses suggest that HPV testing is not more sensitive than a repeat cytology test, and has lower specificity. The results of these meta-analyses were predominantly driven by the findings of cross-sectional analysis of data from a large randomised controlled trial (RCT) in the USA, known as the ASCUS and LSIL Triage Study (ALTS). 21 Cross-sectional studies cannot address the question as to whether there is any medium-term or long-term effect of HPV testing as a tool for triage. Few trials have addressed this question. In the ALTS trial, HPV testing (in fact a combination of HPV testing and cytology) was not effective in triage of women with LSIL. 21,22 For women with ASCUS, the ALTS trial suggested that the sensitivities of HPV testing and repeat cytology in the detection of CIN3 or worse were similar but that HPV testing (in combination with cytology) would result in fewer referrals to colposcopy, 23 and was considered to be an economically viable strategy. 24 The generalisability of the findings of ALTS is uncertain, particularly to populations in which the prevalence of infection with HPV in women with mild dyskaryosis or LSIL is lower, 4 as reported in the UK In a Canadian trial in women with ASCUS or LSIL, addition of HPV testing to repeat cytology resulted in higher sensitivity for detection of CIN2 or worse, but with a substantial loss of specificity. 28 These findings may be highly context dependent, influenced by the relation between cytology interpretations and HPV test result, 29 different grading systems for cytological abnormalities, the character- 646 ª 2010 The Authors Journal compilation ª RCOG 2010 BJOG An International Journal of Obstetrics and Gynaecology

3 HPV in the triage of low-grade cytological abnormalities istics of the women participating in screening, and the prevalence of HPV in screened women. 4 Discrepancies between cross-sectional results regarding HPV obtained in the UK 30 and those from US studies 31 indicate the importance of obtaining good data from an RCT performed in the UK context. Two studies, one in Canada 32 and one in the UK, 33 randomised women with low-grade or negative cytology to management informed by an HPV test result or to management without HPV testing. However, the different lengths of follow-up between trial arms made it difficult to determine whether any differences in the outcome between the arms merely reflects the noncontemporaneous comparison, and the studies lacked statistical power in relation to women with low-grade abnormalities. Hence, there has been a need for an evaluation of HPV testing in the management of women with low-grade abnormalities by means of an RCT with contemporaneous comparison of the trial arms in the setting of the UK NHS CSPs. Another debate in the NHS CSPs concerns the most effective management for women with a visible abnormality on colposcopy, and specifically whether HPV testing might be useful in deciding whether to carry out an immediate large loop excision of the transformation zone (LLETZ) or to perform biopsy with selective recall for treatment. 4 Excisional treatment, including LLETZ, appears to eliminate most HPV infection Therefore, the presence of HPV infection might influence the comparative effectiveness of these alternative treatment strategies. This has not previously been investigated in an RCT. The Trial Of Management of Borderline and Other Lowgrade Abnormal smears (TOMBOLA), a pragmatic RCT embedded in the NHS CSPs, has shown that a policy of referral for colposcopy after low-grade cervical abnormalities confers no clear benefit compared with cytological surveillance and causes more adverse effects. 38 Within the colposcopy arm of this trial, a further randomisation to immediate LLETZ or biopsy with selective recall for treatment was carried out; this showed no advantage of one treatment policy over the other. 39 Here we use data from the TOMBOLA trial to assess first whether an hrhpv test is useful in deciding between cytological surveillance and colposcopy. Second, among women referred for colposcopy, we consider (1) the value of a single hrhpv test in determining what treatment to use (LLETZ or biopsy with selective recall) and (2) the sensitivity, specificity, PPV and NPV in the detection of CIN2 or worse. Methods Full details of the TOMBOLA trial design, a pragmatic RCT in which women were followed up for 3 years, are given elsewhere Briefly, women resident in Grampian, Tayside and Nottingham who had low-grade abnormal cytology (mild dyskaryosis or BNA) detected in the NHS CSPs in the period October 1999 to October 2002, were aged years, and who had no previous cervical treatment were eligible. Women were invited to attend a recruitment clinic, and after informed consent, they were asked to complete questionnaires on sociodemographic and other factors, and an endocervical sample for further testing was sought. Women were intentionally not informed that the further testing was for HPV testing, although they were told this if they asked. The results of HPV testing were not disclosed to either the participating women or to those involved in their management. Human papillomavirus status was used to stratify both randomisations, in addition to age-group, index cytology result and recruitment centre. The first randomisation, to cytological surveillance or immediate colposcopy, was undertaken after the endocervical sample was analysed (1 4 weeks after recruitment). Women assigned to undergo immediate colposcopy who attended were invited to participate in a second randomisation, which was stratified in the same way as for the first randomisation. 38 Those who consented were assigned to either (1) immediate LLETZ or (2) immediate biopsies with selective recall for LLETZ. Both randomisations were performed using a dedicated touch-tone telephone randomisation service provided by the Health Services Research Unit of Aberdeen University. To facilitate comparison with other studies in which a cross-sectional analysis of HPV test performance had been carried out, we examined the association between hrhpv status and the presence of CIN 2 or more severe disease (henceforth CIN2 or worse) in women who were randomised to initial colposcopy and underwent colposcopy. HPV testing The endocervical sample for HPV testing was obtained using a cytobrush. Human papillomavirus analysis was carried out in a single laboratory (Nottingham). Human and viral DNA was extracted from the swabs using the Qiagen UK kit (QIAampÒ DNA Mini Kit; Qiagen, Crawley, UK). Negative controls containing only the elution buffer AE were included in each batch. Extracted DNA was amplified and quantified using a type-specific real-time PCR for the human b-globin gene. Samples that did not contain sufficient DNA for analysis were classified as inadequate; this applied to less than 1% of the samples. Positive and negative controls from a commercial source were included in each batch of assays. Adequate samples underwent HPV polymerase chain reaction (PCR) using GP5+/6+ consensus primers, followed by enzyme immunoassay to amplify the LI (late gene) region of 14 high-risk HPV types that have been associated with increased risk of cervical cancer (types 16, 18, 31, 33, 35, 39, 45, 51, 52, 56, 58, 59, 66 and 68). 42 Women were classified as hrhpv positive (hrhpv+ve) if their sample had an ª 2010 The Authors Journal compilation ª RCOG 2010 BJOG An International Journal of Obstetrics and Gynaecology 647

4 Cotton et al. optical density three times greater than the optical density of the batch negative control, implying that they carried at least one of the hrhpv strains. Other women were classified hrhpv negative (hrhpv)ve). Samples were exchanged with a European laboratory (Amsterdam, the Netherlands) to confirm HPV positivity, and the laboratory participated in an International Quality Assurance Panel. Interventions and follow-up Women were randomised to either cytological surveillance in the community or hospital-based colposcopic examination. Women randomised to cytological surveillance underwent 6-monthly cytology tests, and were referred to colposcopy if any test showed moderate dyskaryosis or worse. Women randomised to colposcopy were further randomised in advance of the colposcopist viewing the transformation zone (TZ) to undergo, should the TZ be abnormal, either: (1) immediate LLETZ; or (2) targeted punch biopsies taken immediately with recall for treatment (by LLETZ) if these showed CIN2 or worse. If these biopsies showed no CIN or CIN1, women were managed by cytological follow-up. Colposcopic examination and histological diagnosis were blind to HPV status. Women were followed up for 3 years. During this time, the results of any cytology test taken, or colposcopy undertaken, in the NHS were recorded. Approximately 3 years after recruitment into the trial, all women were invited back to the TOMBOLA clinic for an exit examination, which included colposcopy, with LLETZ if the TZ was abnormal. The colposcopist was blinded to the women s cytology and HPV status at recruitment, the arm to which they had been randomised, and clinical outcomes (cytology and histology) since recruitment, to minimise potential bias in the assessment of the TZ. In addition, around the scheduled time of the exit appointment and irrespective of attendance, the woman s medical records, together with hospital and pathology databases, were reviewed to ascertain details of any additional relevant events and procedures (cytology, colposcopy, histology, treatment or admission to hospital). Outcomes In the longitudinal analyses, the primary outcome was the interaction between trial arms and hrhpv status at recruitment in relation to the cumulative incidence of CIN2, CIN3 or worse over the period from recruitment up to, and including, the 3-year exit examination. The secondary outcome was this interaction in relation to CIN3 or worse. Each woman was counted only once as a case at the date of first occurrence of CIN2/CIN3 or worse and was classified according to the highest grade of histology reported, whether on punch biopsy or LLETZ. Women with a normal TZ at initial colposcopy were assumed to have no CIN at that time. In the analysis of test performance within the colposcopy arm, the primary outcome was the performance of a single hrhpv test (sensitivity, specificity, PPV, NPV) in the detection of CIN2 or worse at the time of initial colposcopy, with performance in the detection of CIN3 or worse as a secondary outcome. Statistical analysis Women who did not have an HPV test or who had an insufficient sample were excluded from the analyses. Longitudinal analyses were conducted on the basis of intentionto-treat. To take account of variation between women in length of follow-up, each woman accrued person-years from the randomisation date until the date of their exit examination for those who attended or, for others, the date the exit appointment was scheduled, they requested to leave the trial, had a hysterectomy, died, or moved out of the area. In the analysis of the possible influence of HPV status on the comparative effectiveness of initial colposcopy versus cytological surveillance, relative risks and associated 95% confidence intervals (CI) were computed by Poisson regression for combinations of trial arm and HPV status, taking the hrhpv)ve women in the cytological surveillance arm as the reference group. Similarly, in the analysis of the possible influence of HPV status on the comparative effectiveness of immediate LLETZ versus biopsy and recall for treatment, relative risks and associated 95% CI were computed for combinations of treatment arm and HPV status, taking the hrhpv)ve women in the biopsy and selective recall arm as the reference group. In both sets of analyses, the test for a multiplicative interaction was the change in deviance ( 2*log likelihood) between a main effects model and one including a cross-product term for hrhpv status and management arm. In accordance with recommended practice, 43 the risk estimates were adjusted for the stratification factors (age group, recruitment cytology result and recruitment centre). In addition, we investigated the effect on the risk estimates for adjusting for the following factors: deprivation level, reproductive history, smoking status, hormonal contraceptive use, marital status, ethnic group and level of physical activity. Adjustment for these factors had minor, nonsignificant, effects on the risk estimates. The analysis was repeated, stratified by recruitment cytology result (BNA, mild dyskaryosis) and age (20 29, years). An analysis restricted to women who had no previous abnormal cytology was also carried out. In the analysis relating to the interaction between hrhpv and LLETZ versus biopsy and recall, the proportions of women assessed as having an abnormal TZ were compared between arms using the z-test. To facilitate comparison with other studies, we calculated hrhpv prevalence age-standardised to the truncated 648 ª 2010 The Authors Journal compilation ª RCOG 2010 BJOG An International Journal of Obstetrics and Gynaecology

5 HPV in the triage of low-grade cytological abnormalities (20 59) world standard population. 44 In the analysis of test performance within the colposcopy arm, women were included if they were randomised to colposcopy, attended for this and had an adequate colposcopy examination. Sensitivity, specificity, PPV and NPV were calculated for the sample overall and stratified by recruitment cytology result (BNA, mild dyskaryosis) and age (20 24, 25 29, 30 39, years). Statistical power When determining equipoise at the stage of designing the trial, it was apparent that a 50% difference in cumulative incidence in the smallest subgroup (here hrhpv+ve women in either the immediate LLETZ or biopsy and recall arms) would be sufficient for recommending a change in clinical practice. Thus, a priori, we calculated that a sample size of 2130 women provided 89% power to detect a relative risk of 1.5 in cumulative incidence of CIN2 or worse between immediate LLETZ and biopsy and recall in women who were hrhpv+ve (assuming an overall cumulative incidence of 15%, that 33% of women were HPV+ve and a = 0.05 for a two-sided test). 40 A sample size of 4500 women was calculated to provide 85% power to detect a relative risk of 1.3 in cumulative incidence of CIN2 or worse between cytological surveillance and colposcopy in women who were hrhpv+ve, with the same assumptions about the proportion hrhpv+ve, cumulative incidence and a level. Results Characteristics of participants A total of 4031 women were included in the longitudinal analysis of the interaction between hrhpv status and cytological surveillance versus initial colposcopy, and 1818 were included in the longitudinal analysis of the interaction between hrhpv status and immediate LLETZ versus biopsy and recall (Figure 1). The characteristics of the women included in the two randomised comparisons are summarised in Table 1 [further details in Refs (38,39)]. Overall, 43.5% of women from whom an adequate HPV sample was obtained were hrhpv+ve, 60.8% of those with mild dyskaryosis at recruitment and 34.0% with BNA. The hrhpv prevalence standardised to the world standard population was 37.8% per year (95% CI %). Among women with mild dyskaryosis it was 52.7% per year (95% CI %), and among women with BNA it was 30.5% per year ( %). Overall, 18.4% of women had CIN2 or worse detected up to and including the exit examination (cumulative incidence 69 per 1000 personyears); the figures for CIN3 or worse were 9.7% and 36 per 1000 person-years. A total of 11 women had diagnoses more severe than CIN3. Seven of these occurred in the cytological surveillance arm (one adenocarcinoma stage 1B1; one squamous carcinoma in large loop excision; one woman with glandular intraepithelial neoplasia; three women with microinvasion in large loop excision; one in which biopsy showed suspicion of microinvasion but no CIN in subsequent large loop excision). Four of these women were in the colposcopy arm (one squamous carcinoma; one adenocarcinoma in situ; one microinvasion in large loop excision; one in which biopsy showed suspicion of microinvasion but no CIN in subsequent large loop excision). Longitudinal analysis Of women who were hrhpv+ve at recruitment, 70% did not develop CIN2 or worse during the 3 years of followup. Twenty-nine percent of women who had CIN2 or worse detected over the 3 years of follow-up were hrhpv)ve at recruitment. The cumulative incidence of CIN2 or worse was more than three times higher in hrhpv+ve women (114 per 1000 person-years) than in hrhpv)ve women (36 per 1000 person-years). This pattern was similar between the cytological surveillance and initial colposcopy arms and there was no evidence of an interaction between HPV status at recruitment and management arm in relation to the cumulative incidence of CIN2 or worse (P = 0.76; Table 2). Of women who were hrhpv+ve at recruitment, 83% did not develop CIN3 or worse during the 3 years of follow-up. Twenty-three percent of women who had CIN3 or worse detected over the 3 years of follow-up were hrhpv)ve at recruitment. There was no evidence of an interaction between HPV status at recruitment and management arm in relation to the cumulative incidence of CIN3 or worse (P = 0.87; Table 2). These patterns did not differ by age, index cytology result, or after the restriction of analysis to women who had no previous abnormal cytology, for either CIN2 or worse or CIN3 or worse (Supplementary Table S1). Overall, 44% of women had a normal TZ at colposcopy, and therefore did not undergo any further procedures at that time [further details in Ref. (39)]. The proportion of women judged to have an abnormal TZ was higher in the biopsy and recall arm (60%) than in the immediate LLETZ arm (51%) (z = 5.04, P < 0.001). This difference was attributable to women who were hrhpv)ve (hrhpv)ve: immediate LLETZ = 42% abnormal TZ; biopsy and recall = 56%; hrhpv+ve: immediate LLETZ = 65%; biopsy and recall = 67%). Among women who were hrhpv+ve, the cumulative incidence of CIN2 or worse was slightly, but not significantly, higher among those allocated to immediate LLETZ than those randomised to biopsy and recall, whereas in hrhpv)ve women, the opposite was the case (P = 0.27, Table 3). There was no hint of an interaction for CIN3 or worse (P = 0.6; Table 3). These results ª 2010 The Authors Journal compilation ª RCOG 2010 BJOG An International Journal of Obstetrics and Gynaecology 649

6 Cotton et al women assessed for eligibility 4439 randomised to cytological surveillance/ initial colposcopy and confirmed to be eligible 9517 not included in comparison cytological surveillance and initial colpsocopy [4759 not eligible; 4515 did not attend for recruitment; 227 did not consent to participate; 16 other] 2223 randomised to cytological surveillance [875 (39%) high risk HPV positive; 1147 (52%) high risk HPV negative; 201 (9%) HPV status unknown] 2216 randomised to initial colposcopy [880 (40%) high risk HPV positive; 1129 (51%) high risk HPV negative; 207 (9%) HPV status unknown] 1983 had adequate colposcopy, consented to randomisation and randomised to immediate LLETZ/ biopsy and selective recall 233 not included in comparison of immediate LLETZ and biopsy and selective recall [152 did not attend for, or did not have, colposcopy; 26 did not consent to the randomisation or were not randomised; 55 had an inadequate colposcopy] 1296 (58.3%) attended exit examination CIN2/3 outcome information from exit examination and medical record abstraction 927 (41.7%) did not attend exit examination CIN2/3 outcome information from medical record abstraction only 983 randomised to immediate LLETZ [385 (39%) high risk HPV positive; 514 (52%) high risk HPV negative; 84 (9%) HPV status unknown] 1000 randomised to biopsy and selective recall [397 (40%) high risk HPV positive; 522 (52%) high risk HPV negative; 81 (8%) HPV status unknown] 1389 (62.7%) attended exit examination: CIN2/3 outcome information from exit examination and medical record abstraction 827 (37.3%) did not attend exit examination: CIN2/3 outcome information from medical record abstraction only Figure 1. Flowchart of recruitment, randomisation and follow-up of women. were similar by age, recruitment cytology result, or after the restriction of analysis to women who had no previous abnormal cytology (Supplementary Table S2). Analysis of test performance The participation of the 1831 women included in the analysis of test performance within the colposcopic arm of the trial is shown in Supplementary Figure S1 and their characteristics are summarised in Supplementary Table S3. The sensitivity, specificity, PPV and NPV of a single HPV test in detecting CIN2 or worse are presented in Table 4 (note that the numbers of women in this table are lower than those in Table 2 because this cross-sectional analysis does not include women with disease detected at the exit examination or during follow-up). The false-negative (HPV)ve but CIN2 or worse present) rate was 26.9% and the falsepositive rate (HPV+ve but no CIN2/3+) was 36.1%. Just under a third of hrhpv+ve women had CIN2 or worse (PPV 32.3%, 95% CI ). Nine percent of hrhpv)ve women had CIN2 or worse (NPV 91.0%, 95% CI ). Three hundred and forty-nine women (19.1%) had CIN2 or worse detected. The sensitivity of a single HPV test in detection of CIN2 or worse was 75.2% (95% CI ) in women with mild dyskaryosis and 69.9% (CI %) in those with BNA. The sensitivity decreased and the specificity increased with increasing age. The NPV exceeded 90% in women with mild dyskaryosis aged years and 650 ª 2010 The Authors Journal compilation ª RCOG 2010 BJOG An International Journal of Obstetrics and Gynaecology

7 HPV in the triage of low-grade cytological abnormalities Table 1. Characteristics of women included in the longitudinal analyses (a) randomised comparison of initial colposcopy versus cytological surveillance (b) randomised comparison of immediate large loop excision of the transformation zone versus biopsy and recall within colposcopy arm (a) Cytological surveillance Colposcopy n % n % Total Smear status at recruitment Mild, with previous BNA Mild, no previous BNA BNA, with previous BNA BNA, no previous BNA Age (years) HPV status* High-risk positive High-risk negative Not known** (b) Biopsy and recall Immediate LLETZ n % n % Total Smear status at recruitment Mild, with previous BNA Mild, no previous BNA BNA, with previous BNA BNA, no previous BNA Age (years) HPV status* High-risk HPV positive High-risk HPV negative Not known*** BNA, borderline nuclear abnormality; HPV, human papillomavirus; LLETZ, large loop excision of the transformation zone. *HPV+ve and HPV)ve percentages are of those for whom HPV status was known. **Includes women whose samples were inadequate for analysis (n = 28) and women who did not have an HPV test at recruitment (n = 380). The majority of the latter women declined to provide a sample because they were menstruating at the time of the recruitment appointment. ***Includes women whose samples were inadequate for analysis (n = 14) and women who did not have an HPV test at recruitment (n = 151). The majority of the latter women declined to provide a sample because they were menstruating at the time of the recruitment appointment. in women with BNA aged 25 or more. For women with mild dyskaryosis, the PPV was less than 40% for those aged and years. For women with BNA, the PPV was less than 30% for women aged 30 or more. Cervical intraepithelial neoplasia grade 3 or worse was detected in 177 women (9.7%). The sensitivity of a single HPV test in the detection of CIN3 or worse was 78.0%, specificity 60.6%, PPV 17.5% and NPV 96.3% (Table 5). The sensitivity was higher in women with mild dyskaryosis (80.6%) than in women with BNA (74.3%) but this difference was not statistically significant (P = 0.32). The specificity was significantly lower in women with mild dyskaryosis (43.8%) than in women with BNA (69.0%, P < 0.001). In women with mild dyskaryosis, the sensitivity was lower and specificity was higher in older women compared with younger women. A similar pattern in specificity was seen among women with BNA; however, sensitivity was highest in those aged (based on three women with CIN3, all of whom were hrhpv+ve). Discussion Strengths and weaknesses We assessed the potential value of HPV testing as a guide to management over the medium term (3 years) by means of a large population-based RCT with contemporaneous comparison of outcomes. Referral to colposcopy was not based on a positive hrhpv test. This avoided the possibility of management on the basis of an HPV test result having an effect on the ascertainment of CIN2 or worse. It also made possible the calculation of sensitivity, specificity, and NPV which are key measures when considering whether HPV testing can be used to rule out further investigation in addition to PPV. Women were enrolled in TOMBOLA on the basis of a low-grade abnormality detected in the period October 1999 to October In 2001, liquid-based cytology (LBC) was piloted in three health regions in Scotland, including Grampian and Tayside, where LBC was used in selected general practices. Liquid-based cytology was introduced across Scotland, in place of conventional cytology during 2003/ In England, LBC started to be rolled out in 2003, and the roll-out was completed in October Therefore, recruitment samples were predominantly conventional smears but included LBC samples. In the pilot implementations of LBC in England and Scotland, there were reductions in the proportions of cytological tests reported as inadequate, but there was no clear evidence of any impact of LBC on the detection rates of borderline abnormalities or of mild dyskaryosis. 46 Histology is the reference method for the diagnosis of CIN. We considered it inappropriate and unnecessary to ª 2010 The Authors Journal compilation ª RCOG 2010 BJOG An International Journal of Obstetrics and Gynaecology 651

8 Cotton et al. Table 2. Interaction between management policy (cytological surveillance versus colposcopy) and human papillomavirus status at recruitment hrhpv status Trial arm Disease detected* n/n (%) Cumulative incidence per 1000 person-years RR** 95% CI** CIN2 or more severe disease Negative Cytological surveillance 94/1147 (8) 30 1 (reference) Negative Colposcopy 126/1129 (11) Positive Cytological surveillance 222/875 (25) Positive Colposcopy 310/880 (35) P (interaction) = 0.76 CIN3 or more severe disease Negative Cytological surveillance 41/1147 (4) 13 1 (reference) Negative Colposcopy 50/1129 (4) Positive Cytological surveillance 134/875 (15) Positive Colposcopy 168/880 (19) P (interaction) = 0.87 CIN, cervical intraepithelial neoplasia; hrhpv, high-risk human papillomavirus. *Over 3 years following detection by management policy or at exit examination (for further details, see reference 38 ). **Adjusted for stratification variables (age group, index smear and recruitment centre). Table 3. Interaction between management policy (biopsy and recall versus immediate large loop excision of the transformation zone) and human papillomavirus status at recruitment hrhpv status Trial arm Disease detected* n/n (%) Cumulative incidence per 1000 person-years RR** 95% CI** CIN2 or more severe disease Negative Biopsy and recall 66/522 (13) 46 1 (reference) Negative Immediate LLETZ 55/514 (11) Positive Biopsy and recall 141/397 (36) Positive Immediate LLETZ 157/385 (41) P (interaction) = 0.27 CIN3 or more severe disease Negative Biopsy and recall 22/522 (4) 15 1 (reference) Negative Immediate LLETZ 25/514 (5) Positive Biopsy and recall 77/397 (19) Positive Immediate LLETZ 81/385 (21) P (interaction) = 0.66 CIN, cervical intraepithelial neoplasia; hrhpv, high-risk human papillomavirus; LLETZ, large loop excision of the transformation zone. *Over 3 years following detection by management policy or at exit examination (for further details, see reference 39 ). **Adjusted for stratification variables (age group, index smear and recruitment centre). undertake biopsies in women with an adequate colposcopy and normal TZ, although this has been done in other studies. 47,48 In line with similar studies, 23,49 women with a normal TZ were assumed to have no CIN. If some of these women had unrecognised CIN, the PPV calculated in these analyses will underestimate the true PPV of a single HPV test; the extent of that underestimation would depend on the prevalence of unrecognised CIN. The median (interquartile range) time from HPV testing to the date of colposcopy and samples being taken for histological examination was 56 (35 77) days. It is possible that, during this time-window, clearance or acquisition of HPV infection, or regression or progression of lesion, occurred. However, the average length of a hrhpv infection is 6 14 months Moreover, meta-analyses of women with ASCUS and LSIL cytology have reported rates 652 ª 2010 The Authors Journal compilation ª RCOG 2010 BJOG An International Journal of Obstetrics and Gynaecology

9 HPV in the triage of low-grade cytological abnormalities Table 4. Cross-sectional analysis of women in colposcopy arm: properties of a single human papillomavirus test in detecting cervical intraepithelial neoplasia grade 2 or worse N TP FP FN TN With CIN2 or worse (%) hrhpv +ve (%) Sensitivity (95% CI) Specificity (95% CI) PPV (95% CI) NPV (95% CI) Overall (age 20 59) ( ) 63.9 ( ) 32.3 ( ) 91.0 ( ) Women with a recruitment smear indicating mild dyskaryosis Overall (age 20 59) ( ) 46.9 ( ) 39.3 ( ) 80.5 ( ) Age ( ) 32.5 ( ) 38.2 ( ) 76.6 ( ) Age ( ) 43.2 ( ) 47.5 ( ) 72.7 ( ) Age ( ) 52.2 ( ) 43.9 ( ) 76.9 ( ) Age ( ) 63.6 ( ) 21.6 ( ) 92.1 ( ) Women with a recruitment smear indicating BNA Overall (age 20 59) ( ) 71.3 ( ) 25.3 ( ) 94.5 ( ) Current BNA: ( ) 46.3 ( ) 30.1 ( ) 88.7 ( ) Current BNA: ( ) 63.0 ( ) 30.2 ( ) 91.9 ( ) Current BNA: ( ) 73.4 ( ) 22.7 ( ) 95.0 ( ) Current BNA: ( ) 86.5 ( ) 8.8 ( ) 96.8 ( ) BNA, borderline nuclear abnormality; CIN, cervical intraepithelial neoplasia; hrhpv, high-risk human papillomavirus; NPV, negative predictive value; PPV, positive predictive value. TP true positives (HPV positive, CIN2 or worse), FP false positives (HPV positive, no CIN2 or worse), FN false negatives (HPV negative, CIN2 or worse), TN true negatives (HPV negative, no CIN2 or worse). Table 5. Cross-sectional analysis of women in colposcopy arm: properties of a single human papillomavirus test in detecting cervical intraepithelial neoplasia grade 3 or worse N TP FP FN TN With CIN3 or worse (%) hrhpv +ve (%) Sensitivity (95% CI) Specificity (95% CI) PPV (95% CI) NPV (95% CI) Overall age (20 59) ( ) 60.6 ( ) 17.5 ( ) 96.3 ( ) Women with a recruitment smear indicating mild dyskaryosis Overall (age 20 59) ( ) 43.8 ( ) 21.1 ( ) 92.4 ( ) Current mild: ( ) 31.1 ( ) 18.2 ( ) 95.3 ( ) Current mild: ( ) 40.4 ( ) 30.0 ( ) 86.4 ( ) Current mild: ( ) 48.6 ( ) 25.5 ( ) 88.5 ( ) Current mild: ( ) 61.2 ( ) 7.8 ( ) 97.4 ( ) Women with a recruitment smear indicating BNA Overall (age 20 59) ( ) 69.0 ( ) 13.9 ( ) 97.6 ( ) Current BNA: ( ) 43.7 ( ) 16.7 ( ) 95.3 ( ) Current BNA: ( ) 59.9 ( ) 17.4 ( ) 95.5 ( ) Current BNA: ( ) 70.8 ( ) 11.3 ( ) 95.9 ( ) Current BNA: ( ) 86.4 ( ) 5.3 ( ) ( ) BNA, borderline nuclear abnormality; CIN, cervical intraepithelial neoplasia; hrhpv, high-risk human papillomavirus; NPV, negative predictive value; PPV, positive predictive value. TP true positives (HPV positive, CIN3 or worse), FP false positives (HPV positive, no CIN3 or worse), FN false negatives (HPV negative, CIN3 or worse), TN true negatives (HPV negative, no CIN3 or worse). ª 2010 The Authors Journal compilation ª RCOG 2010 BJOG An International Journal of Obstetrics and Gynaecology 653

10 Cotton et al. of progression to high-grade squamous intraepithelial lesions or cancer at 24 months of 7.4% (ASCUS) and 21% (LSIL), and regression to normal cytology of 88.2% (ASCUS) and 47.4% (LSIL). 50 Therefore, the length of the interval between HPV test and histological examination is unlikely to have had a substantial impact on the analysis of test performance of a single HPV test. With regard to the longitudinal analysis, cost-effectiveness modelling of hrhpv testing in combination with conventional cytology in women with BNA or mild dyskaryosis in the Netherlands found that it was more cost effective to collect a sample for hrhpv testing 1 month after the index cytology than at the same time. 10 Given that the colposcopist was blinded to a woman s HPV status, it is intriguing that in women who were hrhpv)ve, the proportion of women in the biopsy and recall arm reported to have an abnormal TZ (56%) was higher than in the immediate LLETZ arm (42%), whereas there was no difference among hrhpv+ve women (abnormal TZ biopsy and recall: 67%; immediate LLETZ: 65%). This could be a chance finding, but does suggest that the HPV infection may change the appearance of the TZ. Flat condylomas, a term that has been used to describe the colposcopic features of HPV infection, may coexist with CIN. There is a degree of overlap in the abnormal features seen colposcopically that can suggest HPV infection or CIN. However, recent evidence from the ALTS study suggests that colposcopic features that are characteristic of high-grade CIN are significantly associated with HPV 16 positivity but not with other HPV types. 51 This suggests that those colposcopic features historically recognised as high-grade CIN, may in fact represent productive HPV 16 infection and colposcopists have been able to make use of this association with high-grade CIN in detecting disease. Our trial investigated the role of a single hrhpv test in the management of women with low-grade cytological abnormalities. It is established that persistent hrhpv infection causes almost all cervical cancer, 7 so an unresolved issue is the potential value of multiple hrhpv tests in management. 52 In cost-effectiveness modelling work in the Netherlands, there did not appear to be any advantage of a management strategy involving repeated hrhpv tests over strategies with single hrhpv tests. 10 Comparison with other analyses of test performance Meta-analyses of cross-sectional studies of HPV status and the results of histological examination of biopsy specimens in women with low-grade cytological abnormalities have shown substantial variation in the prevalence of HPV infection and CIN2 or worse, and consequently in the performance characteristics of the HPV tests used. 19,20,53 The sensitivity is lower and the specificity higher for detection of CIN2 or worse in the cross-sectional analysis than those reported in aggregate for studies which used Hybrid Capture 2, a test for a range of HPV types similar to the GP5+/6+ test used here: for women with LSIL cytology sensitivity was 97.2% and specificity was 28.2%; for women with ASCUS cytology sensitivity was 94.0% and specificity was 62.4%. 20 The PPV and NPV for women with BNA were similar to those observed for women with ASCUS in the meta-analyses. 20 The PPV for women with mild dyskaryosis was higher in our study (39.3%) than that for women with LSIL cytology (27.3%) in the meta-analysis; 20 conversely the NPV was lower (80.5% compared with 98.1%). The PPV for both CIN2 or worse and CIN3 or worse observed in our study was higher than found in the pilot implementation of HPV testing in the UK; 16 this may in part be the result of a lower prevalence of hrhpv in our study, or of differences in the underlying frequencies of CIN2, CIN3 or worse. The proportions of hrhpv+ve women who had CIN2 or worse were higher in TOMBOLA (39.3% for mild dyskaryosis; 25.3% for BNA) than in the pilot implementation (23.1% for mild dyskaryosis; 17.5% for BNA). Potential role of HPV testing: implications of our results It has previously been suggested that, among women with mild dyskaryosis or BNA, HPV testing could distinguish between those women with CIN2 or worse or at risk of progression to this and those who had no disease or were at low-risk of developing CIN2 or worse within a screening round. 8,33 The former group, identified by a positive HPV test, would be followed up colposcopically, and the latter group, identified by a negative HPV test, could be reassured and returned to routine screening or recommended a repeat cytology test in 6 12 months. In addition, it has been suggested that HPV testing might be useful in subgroups defined by severity of cytological abnormality or age. In TOMBOLA, HPV testing was not effective in determining the management either of women with mild dyskaryosis or of women with BNA. The ALTS trial also concluded that HPV testing was not effective in triage of women with LSIL (largely comparable with mild dyskaryosis), 21,22 but could be useful in deciding the management of women with ASCUS. 23 Our results indicate that a single HPV test could be useful in women aged 40 or more in determining which women with low-grade abnormal cytology should be referred for colposcopy, but not in younger women, or for determining the most effective management at colposcopy. The usefulness of a single HPV test in women aged 40 or more in the screening programme depends on the level of risk that policy-makers are prepared to accept. High NPVs 654 ª 2010 The Authors Journal compilation ª RCOG 2010 BJOG An International Journal of Obstetrics and Gynaecology

11 HPV in the triage of low-grade cytological abnormalities could be used to rule out further testing. Is an NPV of 97.4% for CIN3 in women over 40, in whom most CIN will not progress, 3 high enough? The possible value of a negative HPV test in women aged 40 or more in enabling cytology test intervals to be extended should be investigated. Much of the management controversy has been about whether HPV testing could identify a subset of women with low-grade abnormal cytology who have high-grade underlying lesions, for referral to colposcopy. Had women in the TOMBOLA trial been referred to colposcopy on the basis of a positive hrhpv test result, 39.5% (1755) would have been referred. Applying the detection rates for CIN2 or worse observed in hrhpv+ve women in the colposcopy arm, 464 women with disease would have been found on initial colposcopy. Assuming that the rest (1291) would have been referred for surveillance by repeat cytology, it would have been expected that a further 71 women with disease would have been detected. Assuming the women who were hrhpv)ve had cytological surveillance in primary care, 133 examples of disease would be expected to be detected on the basis of the detection rates for CIN2 or worse detected in hrhpv)ve women in the cytological surveillance arm, bringing the total number of women with disease to 668, 82% of the total detected in the trial. The detection rate for a policy of cytological surveillance was 77%, and for initial colposcopy 93%, with the proportions of women undergoing colposcopy 19% and 100%, respectively. For CIN3 or worse, the detection rate expected for a policy of referral for colposcopy on the basis of an hrhpv+ve test would be 86%, compared with 87% for cytological surveillance and 97% for initial colposcopy. Decisions on HPV testing require consideration of specificity as well as sensitivity. Decreased specificity usually accompanies increased sensitivity and it can be argued that screening is already too sensitive. 4 Hence, the flip side of the management controversy relates to minimising the risk of overtreatment of women who do not need therapy. The relatively high NPV found in our study (over 90%) and in previous cross-sectional studies 19,20,47,48,53 57 suggests that a negative hrhpv test may be of value in avoiding referral to colposcopy, particularly in women with BNA. Women who tested negative for hrhpv had a low rate of CIN2 or worse over 3 years (cumulative incidence 36 per 1000 personyears); their risk was less than one-third that of women who were hrhpv positive (114 per 1000 person-years). Nevertheless, using hrhpv positivity as a guide to management of women with mild dyskaryosis would result in 25% of CIN2 or worse being missed; for BNA the figure would be 30%. For CIN3 or worse, 19.4% of cases in women with mild dyskaryosis would be missed, and 25.7% in women with BNA. When CIN3 or worse was considered, the risk in hrhpv)ve women was one-fifth that in hrhpv+ve women (16 versus 74 per 1000). By way of comparison, the cumulative incidence of CIN3 or worse missed until the exit examination in the cytological surveillance arm was 4 per 1000, and of CIN3 or worse not detected at initial colposcopy in that arm of the trial 9 per 1000 [further details in Ref. (38)]. A policy of immediate LLETZ at initial colposcopy might have been expected to be more effective than biopsy and selective recall both because it treats CIN1, a proportion of which is the result of hrhpv infection, 58 and because it eradicates most hrhpv infection However, although we found at initial colposcopy that the frequency of CIN2 or worse was higher in the immediate LLETZ arm than in the biopsy and recall arm (37% versus 27%), this does not justify a policy of immediate LLETZ in hrhpv+ve women with an abnormal TZ, because there was no significant difference in cumulative incidence of CIN2 or worse (or CIN3 or worse) at 3 years. Furthermore, younger age is associated both with hrhpv infection and an abnormal TZ. A policy of immediate LLETZ in hrhpv+ve women would therefore most likely increase treatment of younger women without increasing detection rates of CIN3 or preventing cervical cancer. This is a particular concern because of the possible adverse effects of LLETZ on subsequent reproductive function. 59 A proportion of women examined at colposcopy, including those who are hrhpv+ve, will have a normal TZ. Despite the increased relative risk of CIN2 or worse over 3 years in hrhpv+ve women, among those hrhpv+ve women with a normal TZ on colposcopy (n = 269) the absolute risk up to, and including, the exit examination was low (8.9%; 24 women, 15 of whom developed CIN2, and nine who developed CIN3 or worse). This suggests that this subgroup of colposcopically normal women do not need more intensive investigation or follow-up than the cytological follow-up currently in place. Conclusions On the basis of these results from a large RCT, we conclude that a single HPV test would not be helpful in guiding management in women aged 40 or less, or for determining the most effective management at colposcopy. However, a single HPV test could be useful as a guide to management in older women. The TOMBOLA Group comprises: Grant-holders University of Aberdeen and NHS Grampian, Aberdeen, UK. Maggie Cruickshank (Principal Investigator November 2009 ), Graeme Murray, David Parkin, Louise Smart, Eric ª 2010 The Authors Journal compilation ª RCOG 2010 BJOG An International Journal of Obstetrics and Gynaecology 655