These comments are an attempt to summarise the discussions at the manuscript meeting. They are not an exact transcript.

Transcription

1 Dear dr. Weber, We would like to thank you for the review of our manuscript entitled Cervical screening with an interval beyond five years requires different rescreen times for HPV-negative and HPVpositive, triage negative women: fourteen years follow-up of the POBASCAM trial, written by Dijkstra et al. (reference BMJ ), and for giving us the opportunity to submit a revised version of our manuscript. We are grateful for the Reviewers expertise and appreciate the recommendations. We included a point-by-point response to the comments raised by the Editor and Reviewers, as well as the revised manuscript with and without track changes. We hope that we have addressed all issues to your satisfaction, and are looking forward to your decision regarding publication of our manuscript in The BMJ. Sincerely, Dr. Johannes Berkhof h.berkhof@vumc.nl **Report from The BMJ s manuscript committee meeting** These comments are an attempt to summarise the discussions at the manuscript meeting. They are not an exact transcript. Members of the committee were: José Merino (Chair), Angela Wade (Statistics advisor), Jessamy Bagenal, Elizabeth Loder, Amy Price, George Roeggla, Tiago Villanueva, Wim Weber. Decision: Put points Detailed comments from the meeting: We thought your study addresses an interesting and important research question. We also felt that the new version is much improved and much better readable. We had the following queries:

2 The finding that point estimates of RR for positive HPV subgroups range from 10.4 to 29.1 is given much coverage in the abstract, what the study adds, results and discussion. The statements that HPV positive women have at a least a 10-fold higher risk should be modified to state either that they have at least an almost 6-fold higher risk (lower confidence limit of estimate is 5.9) or that the point estimates show that the women have at least a 10-fold higher risk. The confidence interval for the 10.4 estimate should be given in the abstract. ANSWER: We agree that the lower bound of the confidence limit should be used as a basis for our statements. In order to retain a conservative and comprehensible statement, we rounded 5.9 down to the nearest integer. This means that we changed at least a 10-fold higher risk into at least a five-fold higher risk. We also added the confidence interval for the 10.4 estimate in the abstract. Another headline is that CIN3+ incidences amongst HPV negative women in the intervention group after 3 rounds (0.56%) are similar to the two round incidences amongst women with negative cytology in the control group (0.69%), but no formal test appears to have been done despite the confidence intervals of (0.43, 0.69%) and ((0.56, 0.81%) respectively. The difference should be presented with confidence interval and a p-value given to assess statistical significance. ANSWER: We now present the differences in the abstract and results as risk ratios including 95% confidence intervals as well as P-values. In the Material and Methods section we added the statistical tests used. The statistical (Wald) test relies on a normal approximation to the distribution of the logarithm of the cumulative incidence. For reasons of consistency, we have now also used a normal approximation to the distribution of the log incidence when constructing 95% confidence intervals whereas we earlier used a normal approximation to the distribution of the incidence itself. This means that the lower and upper bound of the cumulative incidence confidence intervals may have changed by a small amount. Modifications to the lower and upper bounds have been carried out in the results section whenever needed. Abstract, Results section: Cancer and CIN3+ risk ratios were 0.97 (95% confidence interval 0.41 to 2.31, P=0.95) and 0.82 (0.62 to 1.09, P=0.17) respectively. Materials and methods, Statistical analysis section: Subgroup differences between cumulative incidence curves were evaluated by log-rank testing and differences between incidence rates and between cumulative incidences at specific follow-up times were evaluated by Wald testing. Ninety-five percent confidence intervals were constructed using the normal approximation to the distribution of the logarithm of the incidence. Results: Two-and three-round cancer incidences were 0.03% (0.01 to 0.06%) and 0.09% (0.04 to 0.18%) among HPV negative women from the intervention group, 0.01% (0.00 to 0.05%) and 0.07% (0.03 to 0.17%) among double negative women from the intervention group, and 0.09% (0.05 to 0.14%) and 0.19% (0.12 to 0.28%) among cytology negative women from the control group. Two- and threeround CIN3+ incidences were 0.31% (0.24 to 0.41%) and 0.56% (0.45 to 0.70%) among HPV negative women from the intervention group, 0.27% (0.20 to 0.36%) and 0.52% (0.41 to 0.66%) among double negative women from the intervention group, and 0.69% (0.58 to 0.82%) and 1.20% (1.01 to 1.37%), among cytology negative women from the control group. The three-round cancer incidences among HPV negative and double negative women were similar to the respective two round incidences among cytology negative women (risk ratios 0.97; 0.41 to 2.31; P=0.95 and 0.83; 0.32 to 2.15; P=0.69), indicating that a negative HPV test provides longer reassurance against cancer than negative cytology. The CIN3+ incidence among HPV negative and double negative women from the

3 intervention group was slightly lower after three rounds of screening than the two round CIN3+ incidence among cytology negative women from the control group (risk ratios 0.82; 0.62 to 1.09; P=0.17 and 0.76; 0.57 to 1.03; P=0.07). Might it be possible to improve and explain figure 1 so it would be easier to understand? ANSWER: We have merged Figure 1A and 1B and adjusted the figure to improve it:

4 Figure 1: Overview of the POBASCAM population-based study cohort with 14 years follow-up including cumulative histology results in the first, second and third screening round ACIS = AdenoCarcinoma In Situ; AdCa = AdenoCarcinoma; BMD = Borderline or Mild Dyskaryosis or worse; CIN0/1 = no dysplasia or Cervical Intraepithelial Neoplasia grade 1; CIN2/3 = Cervical Intraepithelial Neoplasia grade 2 or 3; Cyt = Cytology; HPV = Human PapillomaVirus; NTD = Not To Determine; SCC = Squamous Cell Carcinoma

5 First, please revise your paper to respond to all of the comments by the reviewers. Their reports are available at the end of this letter, below. In your response please provide, point by point, your replies to the comments made by the reviewers and the editors, explaining how you have dealt with them in the paper. ** Comments from the external peer reviewers**

6 Reviewer: 1 Comments: Cervical cancer is caused by persistent infection of high risk human papillomavirus (HPV) and develops over several years through a series of precancers. In addition to HPV-vaccine, cervical cancer can be prevented by detecting and treatment of precancers before development into invasive cervical cancer. Usually it takes more than years from HPV-infection to development of cervical cancer. Cervical cytology (Pap-smears or liquid based cytology) is used in most countries with a cervical cancer screening program. Several countries recommend screening every three year for women years of age. In Finland and the Netherlands they recommend Pap-smear every five years for women years of age. In countries with an organized screening program, the cancer and CIN3+ incidence rates decrease with age because a high proportion of women with precancer are detected and treated during the three first screening rounds. Thus, the intensity of screening can be reduced for women 40 years and older. HPV-testing is more sensitive than cervical cytology and the screening intervals can be extended from three years to five years without increasing the number of cervical cancers. In the new HPV based screening program which will be implemented in the Netherlands in 2017, women will be invited for screening at 30, 35, 40, 50 and 60 years of age. Even though mathematical cost-effectiveness models supports 10 years intervals for HPV negative women above 40 years of age (especially if they have had negative HPV-tests at 30, 35 and 40 years of age), there is limited documentation that 10 years screening intervals is safe. Minor revisions: Page 2, line 28-34, Abstract: The cumulative cancer and CIN3+ incidence among hrhpv negative women in the intervention group, after three rounds of screening (0.009 and ) were similar to the corresponding cumulative incidences among cytology-negative women in the control group after two rounds (0.009 and ) The cumulative CIN3+ rate has to be higher than the cumulative cervical cancer rate. I think it is easier for the reader if the cumulative incidence rates are presented as XX per 100,000 women (similar to Table 1). ANSWER: We do agree that the notation of the cumulative incidence is not optimal. As suggested by another reviewer, we now present cumulative incidence rates as percentages. This is in accordance with the presentation in Figures 2/3, in which cumulative incidences are now presented as percentages. Abstract: The cumulative cancer and CIN3+ incidences among HPV negative women in the intervention group after three rounds of screening (0.09% and 0.56%) were similar to those among women with negative cytology in the control group after two rounds (0.09% and 0.69%).

7 Page 2, line 39-41, Abstract: The long-term CIN3+ and cancer risks among hrhpv negative women are low and are supportive of an extension of the screening interval up to 10 years add for women 40 years and older. ANSWER: The specification of the age group has been added. Page 3, line 25-29, What this paper adds: This study indicates that the 10-year cancer and CIN3 risks after a negative hrhpv DNA test are similar to the corresponding 5-year risks after negative cytology and is supportive of an interval extension to 10 years add for women 40 years and older. ANSWER: This bullet point has been changed, also based on comments by other reviewer(s), into: Age-dependent screening intervals, as implemented in the Netherlands where the five years interval is extended to ten years for HPV negative women of 40 years, are supported by CIN3+ risk estimates but data on cancer risk are inconclusive. Page 8, line 25-40, Results, Cumulative cancer incidences after two and three screening rounds (corresponding with nine and 14 years after baseline, respectively), were (95%CI ) and (95%CI ) among hrhpv negative women, (95%CI ) and (95%CI ) among double negative women, and (95%CI ) and (95%CI ) among cytology negative women. Two- and three-round CIN3+ incidences were (95%CI ) and (95%CI ) among hrhpv negative women, (95%CI ) and (95%CI ) among double negative women, and (95%CI ) and (95%CI ), among cytology negative women. I think it is easier for the reader if the cumulative incidence rates are presented as XX per 100,000 women. Page 19, Figure 1A, The y-axis could be presented pr 100,000 women ( ). Page 19, Figure 1B, The y-axis could be presented percent (0.0% 1.2%). Page 20, Figure 2, The y-axis could be presented as percent (0.0% 12.5%). ANSWER: In the result section, as well as in the figures, the presentation of the cumulative incidences has been changed. Cumulative cancer and CIN3+ incidences after two and three screens (corresponding with nine and fourteen years after baseline, respectively) are displayed in Figure 2. Two-and three-round cancer incidences were 0.03% (0.01 to 0.06%) and 0.09% (0.04 to 0.18%) among HPV negative women from the intervention group, 0.01% (0.00 to 0.05%) and 0.07% (0.03 to 0.17%) among double negative women from the intervention group, and 0.09% (0.05 to 0.14%) and 0.19% (0.12 to 0.28%) among cytology negative women from the control group. Two- and three-round CIN3+ incidences were 0.31% (0.24 to 0.41%) and 0.56% (0.45 to 0.70%) among HPV negative women from the intervention group, 0.27% (0.20 to 0.36%) and 0.52% (0.41 to 0.66%) among double negative women from the intervention group, and 0.69% (0.58 to 0.82%) and 1.20% (1.01 to 1.37%), among cytology negative women from the control group.

8 Figure 2: Cumulative incidence of A) cervical cancer and B) CIN3+ per trial group and baseline screening result A B

9 Figure 3: Cumulative CIN3+ incidence following different triage strategies in the intervention group

10 Reviewer: 2 Comments: GENERAL COMMENTS The authors provide new long term outcome data of the POBASCAM randomised trial which compares efficacy of HPV-based versus cytology-based cervical cancer screening. The results of this study are of high public health relevance since it provides evidence on crucial questions for new cervical cancer screening policies, such as determination of the screening interval and triage of HPV+ women. I had the opportunity to review an earlier version of this manuscript. The authors have implemented most reviewers remarks and have improved the quality of the paper considerably. Only minor comments are given. SPECICIF COMMENTS ABSTRACT Results. It would be good to mention the rates in the sentences CIN3+ incidence was 70% lower among HPV negative was and was 10.4 times higher than among HPV negative women ANSWER: The rates have been mentioned: Cancer and CIN3+ risk ratios were 0.97 (95% confidence interval 0.41 to 2.31, P-value = 0.95) and 0.82 (0.62 to 1.09, P=0.17) respectively. CIN3+ incidence rate was 72.2% (61.6 to 79.9, P<0.001) lower among HPV negative women aged 40 years compared to younger women, but for end-point cancer, no significant association with age could be demonstrated. The CIN3+ incidence rate among HPV positive women with negative HPV16/18 genotyping and repeat cytology was 10.4 (5.9 to 18.4) times higher than among HPV negative women. INTRODUCTION No comments MATERIALS AND METHODS No comments RESULTS Page 10 Among HPV negative women, similar incidence rate changes were observed (62.0% for end-point cancer and 72.2% for CIN3+). It is not clear what these % mean. ANSWER: This was indeed not clear. We have adjusted the manuscript: Among women with double negative test results pooled over the intervention and control group, the cancer incidence rate was 64.2% (-37.6 to 332) higher in women 40 years than in younger women. This increase was not statistically significant (P=0.32). The CIN3+ incidence rate was 72.1% (60.5 to 80.4, P<0.001) lower in women 40 years than in younger women. Similarly, among HPV negative women, the cancer incidence rate was 62.0% (-33.9 to 297, P=0.29) higher and the CIN3+ incidence rate was 72.2% (61.6 to 79.9, P<0.001) lower in women 40 years than in younger women.

11 DISCUSSION Page 11 A limitation to our study is that the presented cancer and CIN3+ incidence estimates were not adjusted for imperfections in screen detection and diagnostic work-up. This sentence needs explanation for the reader. ANSWER: This limitation links to other limitations mentioned in the discussion. Reviewer three also raised a question about the limitation in diagnosis. This part of the discussion has been rewritten as follows: A limitation to our study is that the presented cancer and CIN3+ incidence estimates were tracked through the nationwide histo- and cytopathology registry PALGA. This registry does not contain information on gynaecological procedures, and therefore, were not able to assess how many cases were missed because women did not comply with the colposcopy advice. The histological diagnoses themselves were performed by local pathologists. The use of local histological diagnoses may have led to misclassification, resulting in a dilution of true differences between study groups. However, in earlier publications we showed that inter-observer reliability of CIN3+ was very high (absolute agreement 0.97). 5,22 The absolute cumulative CIN3 incidences presented in our study may also be influenced by the screening protocol as the third screen after study entry uses cytology only, and thus newly developed CIN3 cases may have been missed. Page 12. Utility of registries. A ref recommending use of linked registries in Europe can be added. (Ann Oncol 2010; 21:448-58). ANSWER: This reference has been added.

12 Reviewer: 3 Recommendation: Comments: This manuscript considers the safety of using 10-year screening intervals with HPV based screening. Available analyses considered shorter intervals. This is an important topic. The trade off between level of protection, cost and women s discomfort can change between countries. It is important to make good risk estimates available. Results suggest that HPV screening every 10 years provides a protection similar to that provided by cytology every 5 year. The limited number of invasive cancers, despite the large study size, do not allow excluding an increased risk. Authors correctly recommend strict monitoring of interval cancers when this policy will be implemented in the Netherlands. The study is based of a further follow-up of the well designed and conducted POBASCAM randomised controlled trial. The follow-up, based on the nation-wide PALGA system is plausibly complete. Lack of review of histology can have resulted in some misclassification. However this would at most lead to an underestimate of differences between arms, not the opposite. My main doubt on the methods of the present analysis is about the choice to compare the cumulative incidence of invasive cancer after two and three screening rounds, corresponding to 9 and 14 years after baseline, respectively in the women screen-negative at baseline in the two study arms. Rijkaart et al (Lancet Oncol 2012) compared the study arms of POBASCAM as for the cumulative detection of CIN2+ and CIN3+ during the first and second screening round, defining the first round as up to 4 years from baseline and the second up to 9. This was reasonable because the interest was in detection. Indeed the assessment entailed in some case many test repeats and then diagnostic procedures. Thus, the detection of cases diagnosed up to 4 years after baseline was plausibly, in almost all cases, the result of the primary test performed at baseline. Aim of the present study is, conversely, to compare the risk of cancer with HPV at 10-year intervals to that with cytology at 5-year intervals. For this purpose, the approach employed implicitly assumes that no invasive cancer diagnosed up to 14 years after baseline would have been avoided by a test done after 10 years, i.e. that all cancers diagnosed 10 to14 years after baseline in the HPV arm were already invasive 10 years after baseline. A relevant proportion of them could well have progressed to invasion later and have been diagnosed because of symptoms. Same reasoning for cancers diagnosed 6 to 9 years from baseline in the cyto arm. It seems me that the most natural comparison is between cumulative incidence at 5.5 years (considering that in many women there is a slight delay in repeating) in the control arm and cumulative incidence at 10.5 years in the intervention arm (of course only considering women test-negative at baseline). ANSWER: We decided to include all cancers collected until the end of a screening round for the comparison of 5-year cancer risk after a negative cytology test and 10-year cancer risk after a negative HPV test. We set the cut-off at respectively 9 and 14 years for the following two reasons: 1. Women are invited for screening in the calendar year they turn 30, 35, 40, 45, 50, 55 or 60 years. This may lead to a variation in duration between two screens of maximum 0.5 year. Women that do not attend receive a reminder after 2 to 3 months. This also adds maximum 0.5 year to the variation in duration. 2. Management of positive screens is conservative in the Netherlands. Only women with moderate/severe dyskaryosis cytology are immediately referred for colposcopy. Other women are re-invited after 6-18 months. This adds maximum 2 year to the variation in duration between screens.

13 We agree that using cut-offs of 9 and 14 years may lead to positive bias in both the 5-year cancer risk after a negative cytology test and 10-year cancer risk after a negative HPV test. However, lowering the cut-off to 5.5 and 10.5 years may lead to a negative bias in the 5-year cancer risk after a negative cytology test and 10-year cancer risk after a negative HPV test. Cancers that have already developed may be detected at a later time point because of variation in the time of the screening appointment (reason 1) or the conservative management policy (reason 2). The onset time of cancer is latent because cancer has an asymptomatic phase, so that it is difficult to choose between the two approaches. Therefore, we have added a paragraph to the discussion in which we show that the results obtained with cut-offs 5.5 and 10.5 years, as suggested by the reviewer, are in accordance with results obtained with cut-offs 9 and 14 years: Another limitation is related to the time of diagnosing cancer. For the comparison of the threeround cancer risk after a negative HPV test in the intervention group and the two-round risk after a negative cytology test in the control group, we included all cancers collected until fourteen and nine years after enrolment respectively. These follow-up times are greater than the targeted screening times at ten and five years because of variation in the month of invitation, variation in the time between invitation and screening appointment, and because of conservative management of positive screening results. Regarding the latter, only women with moderate or severe dyskaryosis cytology are immediately referred for colposcopy. Other women are re-invited for a cervical smear after six and eighteen months. The adopted approach implicitly assumes that cancers are present at the beginning of the screening round, but that some of them are detected with delay. However, a proportion of them may actually have progressed during the screening round. To evaluate whether this alters our conclusions, we compared the 10.5 year cancer risk after a negative HPV test in the intervention group and the 5.5 year risk after a negative cytology test in the control group. The estimates were 0.05% for both subgroups with a risk ratio of 0.94 (0.37 to 2.43, P=0.91) and are in accordance with our findings. Minor Introduction page 4 line 12. The purpose of triage is not only to prevent overdiagnosis (for which its efficacy is not obvious) but to avoid overreferral to colposcopy and overuse of biopsies (which triage proved to reach, see Ronco et al Lancet 2014) ANSWER: We have changed the text into: HPV tests are more often positive than cytology in the general screening population and triage testing of HPV positive women by cytology and HPV16/18 genotyping has been recommended to avoid over-referral to colposcopy and overuse of biopsies. Methods pag.5 lines Better saying that at the second round women were screened following the protocol of the intervention arm and at the third round following the protocol of the control arm. Such protocols entail more than looking at the result on one HPV or cytology test. ANSWER: We have changed the text into: At the second screening round at five years, participants in both study groups were managed according to the protocol of the intervention group. At the third screening round at ten years, participants in both study groups were managed according to the protocol of the control group.

14 Discussion page 10 lines Ascertainment bias is usually employed to denote the bias in estimates of sensitivity and specificity deriving from the fact that not all study subjects had verification by the gold standard procedure. Using local histological diagnoses can have led to misclassification, resulting in a dilution of true differences between arms. ANSWER: The reviewer is correct. Reviewer two also raised a question about the limitation in diagnosis. This part of the discussion has been rewritten as follows: A limitation to our study is that the presented cancer and CIN3+ incidence estimates were tracked through the nationwide histo- and cytopathology registry PALGA. This registry does not contain information on gynaecological procedures, and therefore, were not able to assess how many cases were missed because women did not comply with the colposcopy advice. The histological diagnoses themselves were performed by local pathologists. The use of local histological diagnoses may have led to misclassification, resulting in a dilution of true differences between study groups. However, in earlier publications we showed that inter-observer reliability of CIN3+ was very high (absolute agreement 0.97). 5,22 The absolute cumulative CIN3 incidences presented in our study may also be influenced by the screening protocol as the third screen after study entry uses cytology only, and thus newly developed CIN3 cases may have been missed.