'The clinical effectiveness and cost-effectiveness of primary human papillomavirus cervical screening in England: extended follow-up of the ARTISTIC randomised trial cohort through three screening rounds

Study design

The original ARTISTIC study design has been previously described in detail. Briefly, 25,410 women aged 20–64 years undergoing routine cervical cytology screening in Greater Manchester between July 2001 and September 2003 underwent co-testing for HPV. They were randomised 3 : 1 to have the test either revealed and acted upon or concealed so as not to affect management. The intervention in the revealed arm was in those women who were cytology negative but HPV positive. Such women were invited back at 12 months and, if persistently HPV positive, were offered the choice between either colposcopy or a repeat HPV test at 24 months. If still positive at that time, all were offered colposcopy. In the event, two-thirds of those who chose selected colposcopy.

Women who had abnormal cytology at baseline were managed according to national guidance irrespective of HPV status. The hypothesis was that the addition of HPV testing would add sensitivity to detection of high-grade CIN in the first round and result in a reduction of high-grade CIN in round 2. The original primary outcome was therefore the incidence of detected high-grade CIN in round 2.

This extension of ARTISTIC involved extending follow-up of the study cohort through a third round of screening (Figure 1). All ARTISTIC women were invited to participate in a third round of screening 3 years following round 2. All women who participated in round 3 had LBC and HPV testing; however, for reasons explained below, a HPV-positive result only affected the management of those women with borderline changes or mild dyskaryosis who were then offered colposcopy, and this applied to both of the original arms. In all other respects routine clinical care applied.

FIGURE 1.

ARTISTIC protocol for rounds 1, 2 and 3.

As was the case throughout the ARTISTIC study, all colposcopy was performed by practitioners accredited in colposcopy through the British Society of Colposcopy and Cervical Pathology. Equally, all cytology procedures were done according to NHSCSP Quality Assurance (QA) standards and HPV testing was performed in a Clinical Pathology Accreditation-approved laboratory.

Setting and ethics committee approval for the extension

The original protocol did not include continued HPV testing after the second screening round. The trial extension with a further (third) round of HPV testing in addition to cytology was initially approved by the North West Multicentre Research Ethics Committee (ref 00/8/30) on the basis that there were no significant changes to the management of women involved in the 6-year follow-up and that all participants were reconsented to the study. It was anticipated by the investigators that this would be difficult to achieve. The patient information leaflets are included in Appendix 5 and Appendix 6 and consent forms are included in Appendix 7.

General practitioner and family planning involvement

The original primary care practices and family planning clinics maintained their involvement in the trial extension. As all clinics had been participating since 2001, and LBC roll-out and update training across the region had been completed, it was deemed unnecessary to hold any information or training sessions. All practices involved were sent information about the trial extension and spare copies of the patient literature to keep in the clinic rooms.

Reconsenting procedures

All previously enrolled women were sent a letter to their last known address, detailing the aims and procedures of the trial extension. The appropriate patient information leaflet and consent form were also included. Women were given stamped addressed envelopes to return the consent form to the trial office. Women who did not wish to continue their participation in the trial were asked to return their consent forms blank with just their unique trial number included to allow identification. Women who withdrew from the extension did not receive any further correspondence from the trial office. Non-responders to the initial information were sent a repeat letter containing the same information to give them an additional opportunity to reconsent.

General practitioners (GPs) and family planning clinics were sent lists of all ARTISTIC women so that they were able to check whether or not the women were still participating when they attended for screening. A reimbursement of £5 was offered to sample-takers to reconsent women at the clinics who had failed to respond to the letters from the trial office. As some women had changed GPs since the first round of screening in the trial began, practices that were not originally part of the trial were sent additional information about the trial and its extension, if they had ARTISTIC women registered with them.

Protocol amendments

Throughout the course of the third round extension, two major protocol amendments were necessary:

1. An amendment was submitted to the ethics committee to seek approval for triaging all ARTISTIC women on the basis of reflex HPV testing for low-grade abnormalities in line with the NHSCSP Sentinel Sites protocol. This amendment was approved on the basis that to deny the women participating in the trial access to the triage would have provided them with inferior management than that received by other women attending for routine screening in the Greater Manchester area.

2. Within 9 months of the process of reconsenting for the third round extension of ARTISTIC, it became clear, as had been anticipated, that this was not viable. Fewer than one-quarter of women had returned consent forms, and not all of these women actually underwent screening. It was obvious that this would result in insufficient data to allow any useful conclusions to be drawn. A substantial amendment was submitted to the ethics committee, who agreed that under the circumstances it was in the public interest for approval to be given to the use of unconsented samples to be tested for HPV and recommended the following:

   1. to obtain samples of residual material from routinely performed cytology (ARTISTIC participants)

   2. to ensure that the HPV test result would not generate any non-routine clinical action

   3. to ensure that data were anonymised by removing the use of the patient NHS number on the study database and instead use the unique study identifier to make it impossible for researchers to link HPV data to named individuals.

This amendment meant that colposcopy could not be offered to cytology-negative women who were HPV positive at the third screening round. Consequently, the positive and negative predictive values could be affected by the lack of colposcopy verification in cytology-negative/HPV-positive women. On the other hand, the triage of low-grade abnormalities would have resulted in rapid colposcopic verification of underlying CIN. The impact of the protocol amendments on the management of women in the trial is summarised in Table 1. Round 3 management was based on the post-amendments management recommendations.

Impact of the NHS Cervical Screening Programme sentinel sites for human papillomavirus triage

On 1 March 2008, the Manchester Cytology Centre became one of six sentinel sites for HPV triage in England. The aim of the sentinel sites project was to evaluate the roll-out of HPV triage of low-grade cytological abnormalities. Women who tested HPV positive were referred to colposcopy, while those who were negative were returned to routine recall. This process expedited the referral to colposcopy for those who were at risk of an underlying abnormality, while women who were negative, and therefore at low risk, were spared repeat cytological testing.

In order to keep the management of the ARTISTIC women in line with local policies as required by the ethics committee, women in both revealed and concealed arms were entered into the triage protocol. 12 Women entered into the triage protocol received both a combined cytology/HPV result letter from the screening agencies as well as a results letter from the trial office. The changes in the HPV testing protocol were explained in the patient information leaflets. Women living in the Stockport area were also sent a leaflet explaining HPV triage, as this would not have been sent to them by their local screening agency.

The trial protocol originally meant that only the HPV result in women with negative cytology influenced management in the revealed arm and differentiated the effect of HPV testing between the two arms. The maintenance of the revealing and concealment of HPV results in round 3 was rendered impossible by the consequences of the ethics committee’s decision.

Clinical samples

All samples were collected as previously described. 1 Samples that were reported as showing low-grade abnormalities were HPV tested as part of the sentinel sites protocol. All HPV triage testing was carried out at the Manchester Virology laboratory alongside the ARTISTIC HPV testing. Samples were reported as positive and acted upon at the recommended cut-off of 1 RLU/Co. During the course of the ARTISTIC study, the following QA measures were followed: retesting of every 50th sample and participation in external QA schemes when they became available. External QA schemes included National External Quality Assessment Service (NEQAS), Quality Control for Molecular Diagnostics (QCMD) and samples sent from the Scottish HPV Reference laboratory.

Reading of cytology slides and samples

All samples were read and reported in line with current laboratory and NHSCSP guidelines as previously reported. 1 Samples were reported blind to both the allocation of the sample within the trial and the HPV test results, with the exception of slides showing low-grade abnormalities which were sent for HPV testing as part of the NHSCSP Sentinel Sites project.

Human papillomavirus testing

Statistical analysis

Overview of model platform used for the evaluation

An existing model platform was adapted for the economic analysis, and the components of this platform are summarised in Figure 2. This model platform has been previously used to evaluate changes to the cervical screening interval in Australia and the UK,15,16 the role of alternative technologies for screening in Australia, New Zealand and England,17–19 the role of HPV triage testing for women with low-grade cytology in Australia and New Zealand,20 and the cost-effectiveness of alternative screening strategies, combined screening and vaccination approaches in China. 21,22 The model platform has three main components:

1. A dynamic model of sexual behaviour and HPV transmission (and vaccination as appropriate) in England. This incorporates other data from local sexual behaviour surveys,23 and information on vaccination catch-up and coverage rates in England. 24

2. A multitype Markov cohort model of the natural history of CIN and invasive cervical cancer (referred to as the ‘natural history model’). The natural history model takes into account differing clearance and progression rates associated with HPV 16 and HPV 18, compared with other oncogenic types. The model also explicitly simulates post-treatment recurrence in women previously treated for high-grade lesions.

3. A cohort and multicohort model of screening, diagnosis, treatment and post-treatment management.

FIGURE 2.

Schematic diagram of model platform.

Methods for model validation against ARTISTIC data

Reanalysis of ARTISTIC was used to support validation of the existing model platform. The steps involved in ARTISTIC validation can be summarised as follows:

1. Set up and simulate a ‘virtual’ cohort based on the age distribution and screening behaviour of the ARTISTIC trial cohort.

2. Run a multicohort simulation (i.e. a subsimulation for each age group in the cohort) to predict the number of screen-detected low- and high-grade cervical abnormalities that would be observed in each age group and overall, over the duration of trial follow-up.

3. Validate the predictions obtained from the model with the data from the ARTISTIC trial.

The model platform used for the main evaluation has three main components. These three components consider:

1. HPV transmission and vaccination

2. the natural history of cervical CIN and cancer in HPV-infected women; and

3. cervical screening.

The methods by which the model components two and three above were adapted for modelling the ARTISTIC cohort are described below.

Human papillomavirus transmission and dynamic modelling

Age-specific pre-vaccination HPV incidence was obtained from a dynamic model of HPV transmission. As the model is a dynamic transmission model, the effects of herd immunity are taken into account. Data on sexual behaviour from the UK’s National Survey of Sexual Attitudes and Lifestyles II (NATSAL II)25 (data collected between 2000 and 2001) were incorporated into the model. (See Appendix 16 on sexual behaviour data for the population model for more details on the simulation of sexual behaviour in England.)

Model of the natural history of cervical intraepithelial neoplasia and cancer

The model assumes that once a woman becomes infected with HPV, the virus can be cleared, or can persist and eventually cause CIN or cancer of the cervix. Six different health states in which a woman can exist were considered: well (or uninfected with HPV); HPV; CIN1 (persistent HPV infection); CIN2; CIN3; and cancer. The cancer state was further divided into stages to reflect International Federation of Gynecology and Obstetrics (FIGO) clinical stage at diagnosis and for each stage, diagnosed and undiagnosed cancer. Women who are diagnosed with invasive cervical cancer have an increased, stage-specific rate of death. Women who have survived cervical cancer for over 5 years are considered cancer survivors in the model, and their rate of survival is assumed to be the same as that of the normal population. Women who had HPV, CIN1, CIN2 or CIN3 were further categorised into three pathways associated with HPV 16, 18 or other HR types. The initial prevalence of each HPV type for each CIN state and for each age group was weighted to be consistent with the type-specific prevalence observed at enrolment in the ARTISTIC trial.

The natural history aspect of the model captures the rates at which women in different health states can progress to a more advanced stage (e.g. CIN2 to CIN3), remain in the same health state (e.g. remain in the CIN2 state), or regress (e.g. CIN2 to CIN1). This was captured in a HPV-type-specific matrix of progression, regression and stasis rates between HR HPV infection, CIN1, CIN2, CIN3 and invasive cervical cancer. The model captured the higher rates of disease progression and lower rates of regression in women infected with HPV 16 and 18 compared with other HR HPV types.

The rates of progression, regression and stasis between the different health states have been calibrated and validated previously. However, the ARTISTIC data were used to further validate the natural history model.

Modelling screening

The natural history aspect of the model dealt with modelling the actual underlying health state distribution of women in the population. However, the ARTISTIC trial reported histology results from women who attended screening and were referred for colposcopy and biopsy. Thus, in order to compare the natural history model results with ARTISTIC results, screening behaviour as observed in the trial was simulated. Management of the two different screening arms of the trial were simulated, and the ARTISTIC data were used to determine the rates of non-attendance, early rescreening and late rescreening.

We then compared model predictions for histology detected CIN cases with observed data from the trial.

Fitting test characteristics for cytology and the underlying health state in the ARTISTIC cohort

In order to simulate the ARTISTIC cohort, the actual underlying health states of women entering the trial were estimated. To do this, we also needed to estimate the test characteristics of LBC and HPV testing in the ARTISTIC trial.

The test characteristics for LBC describe the probability of receiving any test result for each possible health state. For each underlying health state (of which there were six: well, HPV, CIN1, CIN2, CIN3 and cancer), there were five possible LBC test results (given an adequate test result): negative, borderline, mild, moderate and severe. The probabilities of receiving each test result for each possible health state form a table called the ‘test probability matrix’, or TPM. The TPM gives complete information on test characteristics and the test parameters for sensitivity, specificity and positive predictive value (PPV). Negative predictive value can be secondarily derived for thresholds at any health state [such as high-grade lesions (CIN2+) or all CIN lesions] and for any testing threshold (such as borderline or mild dyskaryosis for cytology).

A Gibbs sampler was used to simulate the posterior distribution of the health state and LBC test probability parameters using the cytology and HPV data observed at baseline in ARTISTIC. 26 Multiple chains were run and the mixing of the different sequences was monitored. The sampler was stopped when the potential scale reduction was near one for all the estimated parameters. The LBC TPM determined from this sampling procedure was further modified to match histology outputs from the ARTISTIC trial (see Appendix 9 for more information on the Gibbs sampling and fitting procedure).

Construction of the virtual ARTISTIC cohort and validation of model outcomes to ARTISTIC data

The procedure for simulating the ARTISTIC cohort can be summarised in the following steps:

1. A ‘virtual’ ARTISTIC cohort was configured to represent the trial population as at ARTISTIC enrolment. The enrolled virtual cohort represented the actual ARTISTIC cohort in terms of age profile and HPV test positive (type-specific) rates at baseline. We assumed that all women attended their baseline screen at the same time, and so screening events are synchronised (as opposed to simulating the actual date of attendance, in line with the way such trial and cohort data are routinely reported). All women had a baseline cytology and HPV test, with follow-up management dependent upon their trial arm, baseline test results and compliance with guidelines as observed in the ARTISTIC trial. The actual rates of colposcopy referral, delays in colposcopy attendance and screening attendance rates that occurred throughout ARTISTIC were captured in the model.

2. Each woman’s true underlying health state was updated at 6-monthly intervals throughout the simulation according to the parameters from the natural history model and the HPV transmission model. The number of women attending routine screening, follow-up screening or colposcopy was also updated at 6-monthly intervals.

3. For the modelled cohort, for woman with histologically detected CIN2+, any subsequent screening visits were censored, and this group were removed from the simulation. If the histology result was less than CIN2 (‘< CIN2’: CIN1 or histological manifestations of HPV effect), women remained in the simulation.

The simulation predicted the number of histology-detected CIN cases at 6-monthly intervals, and these model predictions were compared with the data from the ARTISTIC trial at 6-monthly intervals. As the natural history simulated in the model captured disease progression and regression at various rates, the actual underlying health states of women throughout the trial changes over time. Thus, the model outputs at later times in the trial give an indication of how well the natural history model captured the rates of disease progression, regression and stasis (after taking into account the ‘overlay’ of screening, i.e. the influence of screening behaviour and test characteristics on the observed outcomes in the trial).

Methods for the evaluation of primary human papillomavirus screening in England

The validated natural history model of CIN and invasive cervical cancer was then used to evaluate health outcomes under different screening strategies on a population basis for England. The cost-effectiveness of various HPV primary screening strategies were evaluated against a comparator of current screening practice, and cross-sectional outcomes, including estimated rates and case numbers of cervical cancer cases and deaths, detected high-grade abnormalities, colposcopies and numbers of screening tests, were predicted after incorporating information on the population age structure. Data from local sexual behaviour surveys, screening compliance information, and information on vaccination catch-up and coverage rates were also incorporated into the England-wide model.

A health-services perspective was used for the collection of costs. The economic evaluation also took a health-services perspective, taking into account the health-services costs associated with population-based screening, management, diagnosis, follow-up, and CIN and cancer treatment. A discount rate of 3.5% was used for costs and effects (the 3.5% discount rate for both costs and effects is recommended by the National Institute for Health and Care Excellence based on the recommendations of the UK Treasury, 2008 guide). 27

Primary human papillomavirus screening strategies considered in the evaluation

A number of potential future primary HPV screening strategies were simulated. These are described and detailed diagrammatically in the following pages. Many of the strategies chosen were thought to represent possible candidates for effective, and cost-effective, strategies in England. Other, more exploratory, strategies (as noted) were included for comparative purposes.

Comparator strategy (current screening practice)

The comparator for the evaluation is current screening practice, with and without the effects of HPV vaccination, and is depicted in Figure 3.

FIGURE 3.

Comparator strategy (current screening practice).

The comparator strategy involves cytology-based screening with LBC and HPV triage testing of low-grade abnormalities (borderline changes and mild dyskaryosis). The recommended screening interval is 3 years for women aged 25–49 years and 5 years for women aged 50–64 years.

Basic strategies

Strategy 1 (Figure 4)

The strategy can be summarised as follows: women positive for any oncogenic infection receive reflex cytology triage testing. Cytology-positive women (borderline dyskaryosis or worse) are referred to colposcopy. Cytology-negative women are sent for repeat HPV testing with cytology triage in 24 months, and any women who are HPV positive and borderline dyskaryosis or worse are referred to colposcopy at that point. HPV-positive, triage-negative women are sent for a repeat HPV testing in another 24 months, and HPV-positive women at that point are referred to colposcopy. HPV-negative women are returned to routine screening at each stage.

FIGURE 4.

Screening and follow-up management of primary HPV screening for strategy 1. a, At the second 24-month follow-up, women negative for HPV are returned to routine screening and HPV-positive women are referred to colposcopy.

Strategy 2 (Figure 5)

The strategy can be summarised as follows: women with any oncogenic HPV-positive infection have reflex cytology triage. Cytology-positive women (borderline dyskaryosis or worse) are referred to colposcopy. HPV-positive, cytology-negative women have repeat HPV and reflex cytology in 24 months with partial genotyping and any HPV 16/18 positive or borderline dyskaryosis or worse are referred to colposcopy at that point. Cytology-negative women and/or women negative for HPV 16/18 are sent for a repeat HPV test in another 24 months, with any HPV-positive women referred to colposcopy at that point. HPV-negative women are returned to routine screening.

FIGURE 5.

Screening and follow-up management of primary HPV screening for strategy 2. a, Using a HPV test that has HPV 16/18 partial genotyping. b, At the second 24-month follow-up, any oncogenic HPV positive is referred to colposcopy. HPV negatives return to routine screening. OHR, other high risk.

Strategy 3 (exploratory) (Figure 6)

The strategy can be summarised as follows: a primary screening test with partial genotyping is used and women with HPV 16/18 are referred directly to colposcopy. Cytology triage is performed on other oncogenic HPV-positive women, with those cytology-positive (borderline dyskaryosis or worse) referred to colposcopy. Other oncogenic HPV-positive women who are cytology negative are then followed up as in strategy 2.

FIGURE 6.

Screening and follow-up management of primary HPV screening for strategy 3. a, Using a HPV test that has HPV 16/18 partial genotyping. b, At the second 24-month follow-up, any oncogenic HPV positive is referred to colposcopy. HPV negatives return to routine screening. OHR, other high risk.

Strategy 4 (exploratory) (Figure 7)

This ‘co-testing’ exploratory strategy can be summarised as follows: women receive HPV and cytology co-testing. Women with moderate or severe dyskaryosis are referred to colposcopy. Women who are HPV positive with a cytology result of borderline or mild dyskaryosis are also referred to colposcopy. Women who are HPV positive with a normal cytology result have repeat testing in 24 months, and are thereafter managed as in strategy 2. Women who are HPV negative with a cytology result of mild dyskaryosis also have repeat testing in 24 months, and are thereafter managed as in strategy 2. Women who are HPV negative with a cytology result of borderline dyskaryosis or normal are returned to routine screening.

FIGURE 7.

Screening and follow-up management of primary HPV screening for strategy 4. a, Using a HPV test that has HPV 16/18 partial genotyping. b, At the second 24-month follow-up, any oncogenic HPV positive is referred to colposcopy and HPV negatives return to routine screening. OHR, other high risk.

Strategy variations

Diagnosis and treatment procedures

The diagnosis and treatment procedures simulated in the model are detailed in Figures 8–10.

FIGURE 8.

Colposcopy management for women referred with borderline/mild cytology and/or a HPV positive test. a, Women diagnosed with invasive cancer undergo cancer treatment. Women treated for CIN undergo HPV and cytology testing 6 months post treatment. If both cytology and HPV negative, women are returned to routine screening. Women with abnormal cytology or positive HPV result are referred to colposcopy and are subsequently followed up under current practice post-treatment management. b, At the 12-month follow-up, any abnormal cytology result is referred to colposcopy. Women with normal cytology are referred to routine screening.

FIGURE 9.

Colposcopy management for women referred with moderate/severe cytology. a, Immediate in women over 50 years; delay 12 months in women younger than 50 years. b, Women diagnosed with invasive cancer undergo cancer treatment. Women treated for CIN undergo HPV and cytology testing 6 months post treatment. If both cytology and HPV negative, women are returned to routine screening. Women with abnormal cytology or positive HPV result are referred to colposcopy and are subsequently followed up under current practice post-treatment management.

FIGURE 10.

Management for women undergoing biopsy. a, A small proportion of CIN1 are assumed to be treated along with CIN2+. b, Women undergo HPV and cytology testing 6 months post treatment. If both cytology and HPV negative, women are returned to routine screening. Women with abnormal cytology or positive HPV result are referred to colposcopy and are subsequently followed up under current practice post-treatment management. c, At the 12-month follow-up, any abnormal cytology result is referred to colposcopy. Normal cytology is referred to routine screening.

Data sources

Relationship between human papillomavirus, cytology and age over three rounds

Figure 14 depicts the comparable proportion of women who tested cytology negative and HPV positive according to their age at that round, and the proportions of low-grade and high-grade cytology, according to age at round 1, 2 and 3, are shown in Figures 15 and 16, respectively. The numbers of abnormalities in round 3 are small due to the effect of previous screening and the reduced proportion of the original cohort being screened. The number of women under age 25 years fell as the cohort became older, with almost none by round 3. By contrast, the number of women aged over 64 years increased.

FIGURE 14.

Proportion of normal cytology with a positive HPV result according to age at round. R, round.

FIGURE 15.

Proportion of borderline/mild cytology with a positive HPV result according to age at round. R, round.

FIGURE 16.

Proportion of moderate+ cytology with a positive HPV result according to age at round. R, round.

As has been found elsewhere, the proportion of HPV-positive women overall fell rapidly with age. This is most obvious in women with normal cytology, where HPV prevalence decreased from around 25% in those under 25 years to around 5% in those over 50 years; this remained relatively unchanged over the rounds. There was also a marked fall in the HPV positivity rate by age among women with abnormal cytology in round 1. In rounds 2 and 3, the proportion of women with cytological abnormalities was much lower and the data by age became sparse.

The numerical data (see Tables 62–71 in Appendix 8) allow an indication of primary screening with HPV triaged with cytology to be shown over repeated rounds of screening. The round 1 data are broadly representative of what would happen in routine population-based screening, and rounds 2 and 3 reflect the impact of the detection of disease in the first (prevalent) round.

Patterns of any lesion of CIN2+ over three rounds of screening

The impact of screening in the initial (prevalence) round is shown in Table 7, with a marked fall in the proportion of women with CIN2+ between rounds 1 and 2, but not between rounds 2 and 3. The cumulative risk of CIN2+ was 3.9% (95% CI to 3.6% to 4.3%) in the revealed arm and 3.7% (95% CI 3.2% to 4.3%) in the concealed arm. There was no significant difference between the arms [odds ratio (OR) 1.1, 95% CI 0.9 to 1.3; p > 0.1]. The corresponding figures for CIN3+ were 2.1% (95% CI 1.7% to 2.6%) in the revealed arm and 1.9% (95% CI 1.7% to 2.2%) in the concealed arm (OR: 0.9, 95% CI 0.7 to 1.2, p > 0.1).

Logistic regression analysis of abnormality rates adjusted for age and randomisation arm for HPV and cytology (also adjusting for HPV) is shown in Table 8. Both HPV and cytology showed a significant decrease between rounds 1 and 2. HPV was significantly more prevalent in round 3 than in round 2.

Those who had negative cytology but were HPV positive at entry continued to develop CIN2+ at a higher rate than the overall proportion of CIN2+ in round 1, and at a higher rate than women with abnormal cytology who were HPV negative (Table 9).

CIN2+ detected in rounds 1, 2 and 3 according to HPV status and cytology grade in the corresponding round is shown in Table 9. Although the numbers are smaller, the relationship between HPV, cytology and CIN2+ remain essentially the same when compared with the round 1 data. Around 20–30% of women with abnormal cytology who are HPV positive had CIN2+ detected in each of the rounds.

The abnormality rates in the concealed arm in terms of histopathology, cytology and HPV detection over the three rounds are shown in Table 10. Largely similar falls proportionately were seen with CIN2+ rates and high-grade cytology and low-grade cytology, all of which reflect to some extent treatment of detected disease and the reduced prevalence of disease in serial rounds of screening. The rate of HPV infection also fell reflecting both treatment of CIN and the ageing cohort.

The impact of being screened or not in round 2 on round 3 outcomes is shown in Table 11. Among the round 3 women who had not attended in round 2, the overall proportion of women with CIN2+ was 1.11% and 1.42% in the concealed and revealed arms, respectively, compared with 0.5% for both arms in round 3 women who had been screened in round 2. The same effect was seen whether HPV positive or HPV negative at entry to the study. This confirms the protective effect of regular repeated rounds of screening.

Cumulative detection of any lesion of CIN2+ over three screening rounds according to baseline human papillomavirus test and cytology results

The cumulative risk of being diagnosed with CIN2+ over approximately 6 years is shown in Tables 12–16. The cumulative rate of CIN2+ was 0.67% (95% CI 0.51% to 0.87%) for women who were both cytology negative and HPV negative at baseline compared with 37.4% (95% CI 34.91% to 40.04%) for women who were cytology positive/HPV positive (see Table 14). The cumulative risk in the HPV-positive group [20.12% (95% CI 18.68% to 21.61%)] is identical to that in the abnormal cytology group [20.12% (95% CI 19.04% to 22.08%)]. This reinforces the relative similarity in sensitivity between LBC and HPV testing already reported in the ARTISTIC study. 1 There is, however, a significantly lower cumulative risk (p = 0.0002) of developing a CIN2+ lesion following a negative HPV test [0.87% (95% CI 0.70% to 1.06%)] compared with a negative cytology result [1.41% (95% CI 1.19% to 1.65%)] (see Table 14).

An important group is women who were HPV positive/cytology negative at entry. This group had a cumulative CIN2+ rate over 6 years of 7.73% (CI 6.39% to 9.36%), significantly higher than that of those with abnormal cytology/HPV negative at entry at 3.24% (CI 2.32% to 4.38%) and twice that of the screened population overall (3.9%). It is quite clear that the risk to these women persists, presumably due to the proportion who develop long-standing persistent infection.

Table 16 shows cumulative CIN2+ rates over three rounds by both HPV status and age. Within the three age bands of 20–34, 35–49 and 50–64 at entry, there was a 13-, 33- and 40-fold difference, respectively, between women who were HPV positive and those who were HPV negative at entry. Women who were HPV positive and aged between 20–34 years at entry had a cumulative CIN2+ rate of 23.95%, whereas women who were HPV negative and aged 50–64 years at entry had a cumulative rate of only 0.16%.

Interval-specific findings for detected CIN2+

Model outputs for the predicted histology confirmed that cases of CIN over time were compared with observed histology-confirmed CIN cases from the ARTISTIC trial. The model predicted the numbers of histology-confirmed CIN cases at 6-monthly intervals, for both the concealed and the revealed arms of the trial, by 10-year age groups. In this section we produce summarised versions of our results. (For more details of the model predictions by age group see Appendix 18.)

We began by investigating how well the model predicted histology-confirmed CIN2+. We assumed that once women had a histology result of CIN2+ they received treatment, and censored the data thereafter.

Figure 17 shows the cumulative rate of CIN2+ for both the model (solid green diamonds) and the ARTISTIC data (hollow blue squares). These cumulative rates were generated using Kaplan–Meier methods that take into account censoring after the occurrence of histology-confirmed CIN2+ or loss to follow-up. The horizontal axis shows the time since the first screening round in the trial.

FIGURE 17.

Cumulative rate of histology detected CIN2+ by time since enrolment (months), expressed as a percentage for each trial arm and each age group. The model results are shown as the solid green diamonds and the data are shown as the hollow blue squares. 95% CIs are also shown. The value ‘n = ’ in each graph represents the number of women who entered the trial in that age group and trial arm.

Figure 17 indicates that the model-predicted cumulative rate of CIN2+ at 6-monthly intervals compares very well with the ARTISTIC data in both trial arms, with most model predicted cases of cumulative CIN2+ remaining well within, or just outside of, the 95% CI for each 6-monthly observation. For women aged 40–49 years, the predicted cumulative detection of CIN2+ by 6 years is slightly higher than the 95% CI for the revealed arm; however, rates of cumulative detection in this age group for the concealed arm and in both arms for younger and older age groups (50+ years) showed adequate fit to the observed data.

Results for histologically confirmed low-grade disease

The model-predicted incidence of histologically confirmed low-grade cervical disease was compared with data from the ARTISTIC trial. In the ARTISTIC trial, women without CIN2+, but who still had a histological abnormality, were said to have a histology result of ‘< CIN2’. Thus, ‘< CIN2’ captured women with detected CIN1, and women with histological evidence of HPV infection. In the model we assumed that all women with a true underlying health state of CIN1 will have a histology result of ‘< CIN2’ if they have a biopsy. As women were not censored from further analysis after a ‘< CIN2’ histology, women can have ‘< CIN2’ results if subsequent biopsies are taken throughout the trial (in fact, the ARTISTIC data show that many women who have detected < CIN2 will have another biopsy within 2 years). By contrast, as mentioned previously, once a woman has a histology result of CIN2+, any subsequent results were censored.

Figure 18 shows the cumulative detected < CIN2 cases for both the model (solid green diamonds) and the ARTISTIC observed data (hollow blue squares). The cases are presented as a percentage of the starting total number of women in the corresponding trial arm and age group. Overall, the model results compared well with the ARTISTIC data throughout the trial and the number of ‘< CIN 2’ cases detected by the model compared well with the number of cases as detected in the trial. In this case, for women aged 40–49 years, the predicted cumulative detection of < CIN2 by 6 years is slightly lower than the 95% CI for the concealed arm; however, rates of cumulative detection in this age group for the revealed arm and in both arms for younger and older age groups (50+ years) showed adequate fit to the observed data.

FIGURE 18.

Cumulative ‘< CIN2’ cases by time since enrolment (months) expressed as a percentage of the total number of women who entered the trial in the corresponding trial arm and age group. The model results are shown as the solid green diamonds and the data are shown as the hollow blue squares. The value ‘n = ‘ in each graph represents the number of women who entered the trial in that age group and trial arm.

Validation of model outcomes by human papillomavirus type

Cumulative outcomes by human papillomavirus type at enrolment (baseline)

For this analysis it was assumed that the underlying HPV type of a CIN diagnosis was the HPV type detected at baseline. For the small number of HPV-negative women at enrolment with a non-negative histology result within 12 months of enrolment, the assumption was made, for the purposes of this exercise, that the baseline HPV status was positive but with a false-negative HPV test.

Summary results for all ages and across each trial arm are presented in Figure 19, which shows the cumulative incidence of CIN2+ by assumed baseline HPV type over 5 years. This representation shows that the model captures the cumulative incidence of CIN2+ by HPV type at baseline over the first 5 years of the ARTISTIC trial.

FIGURE 19.

Cumulative incidence of CIN2+ at 6-monthly intervals by HPV type status at baseline. OHR, other high risk.

Figure 20 shows incident CIN2+ in women who present as negative at baseline or HPV positive for any type. More detailed model outputs on cumulative incidence of CIN2+ by age group is shown in Figure 21, which demonstrates that the model is capturing the incidence of CIN2+ by HPV type for different age groups. The model predicted rate of CIN2+ in HPV negative women is within the 95% CI of the observed data at 5 years.

FIGURE 20.

Overall cumulative incidence of CIN2+ by HPV positivity at baseline.

FIGURE 21.

Cumulative incidence of developing CIN2+ for each age group at 6-monthly intervals by HPV type status at baseline. OHR, other high risk.

Summary of the findings of the ARTISTIC validation procedure

The purpose of simulating the ARTISTIC trial cohort was to determine whether the natural history model or the test probability matrices required recalibration. We found that the model results were in close agreement with ARTISTIC trial results. We concluded that the natural history model was accurately capturing the true underlying rate of disease progression and regression, in time steps of 6-monthly intervals and when considering the underlying HPV type. In conjunction with prior validation exercises for previously reported evaluations, this finding supported the use of the current model platform for use in a national assessment of the role of primary HPV testing in England.

Calibration and validation of the England-wide model under current screening practice in an unvaccinated cohort

In this section we report on the outcomes of an England-wide evaluation of primary HPV screening. The model used for this evaluation incorporated the natural history model prevalidated in other countries and then validated against the ARTISTIC data. We also modelled sexual behaviour assumptions, which were calibrated to the NATSAL II report25 using a dynamic simulation of sexual behaviour and HPV transmission. We predicted HR HPV prevalence in cytologically normal women by age in England, comparing the results to the prevalence reported in the ARTISTIC trial (in this instance using the ARTISTIC prevalence data as a proxy for all-England prevalence). 3 We also validated the model against England data for age-specific cervical cancer incidence, cervical cancer mortality, and a number of other outcomes.

The England-wide model predicted age-specific and type-specific HPV prevalence rates across the population (including those women subsequently found to have confirmed CIN) similar to those observed in the baseline round of the ARTISTIC trial3 (Figure 22).

FIGURE 22.

Overall HPV prevalence (OHR) by age as predicted by the model (blue dashed curve). We assume an age structure in England as observed in the 2011 census. 48 The HPV prevalence, as observed in ARTISTIC, is also shown with 95% CIs for the estimated prevalence at each age (ARTISTIC prevalence is used as a proxy for England-wide prevalence here). OHR, other high risk.

Cervical cancer incidence

The model-predicted cervical cancer incidence compares well with recent national observed cervical cancer, as shown in Figure 23. The data used for comparison are 2010 data from the Office for National Statistics. 49

FIGURE 23.

Cancer incidence rates per 100,000 women as predicted by the model are shown as the black dashed curve. We assume an age structure in England as observed in the 2011 census. 48 Data on England-wide cervical cancer incidence in 2010 are shown as the solid green curve. 49

The predicted age- and type-specific incidence of invasive cervical cancer shown in Table 31 indicates that 52.5% and 13% of cervical cancers are predicted to be attributable to HPV to types 16 and 18, respectively. This compares well with observed international data. 50

The proportion of cancers detected at each of the four FIGO stages is also shown in Figure 24, and this compares well with observed data from the West Midlands. 51

FIGURE 24.

The stage distribution of invasive cervical cancer as predicted by the model is shown as the black bars. We assume an age structure in England as observed in the 2011 census. 49 White bars show observed cancer staging from the West Midlands.

Cervical cancer mortality

Age-specific rates of cancer mortality, as predicted by the model, are shown in Figure 25. The model age-specific predictions compare relatively well with 2008 mortality data (predicted results within the 95% CI of observed data for women aged > 45 years; rates are slightly higher than observed values for younger women).

FIGURE 25.

Age-specific cancer mortality rates per 100,000 women as predicted by the model are shown as the black dashed curve. We assume an age structure in England as observed in the 2011 census. 49 Observed mortality rates in the UK in 2008 are shown as the solid green curve, with 95% CIs. 16

Predictions of cost, effectiveness and resource utilisation outcomes for current practice

A summary of resource utilisation, cancer incidence and mortality and costs as predicted by the model for current practice is given in Table 32. Note that the life-years presented in the table are a somewhat counterintuitive measure to which discounting has been applied and which reflects the discounted additional number of years a woman is predicted to live given that she has already survived to 10 years of age (the model simulation commences at 10 years of age). Compared with observed data, the model predicts very similar cervical cancer case numbers and cervical cancer deaths overall (although the age-standardised mortality rate is slightly higher than observed).

The model was also used to calculate the predicted annual number of screening events in England under current practice and the results were compared with values reported in the 2010–11 report of the NHSCSP. Compared with these data, the model predicts very similar numbers of cytology tests overall (within 2%). Other predicted model outcomes, such as the number of colposcopies, biopsies and treatments for CIN, were calculated although it was not always possible to directly compare these with the reported observed data which were often available as a per-woman versus per-test number (as noted in Table 32).

Human papillomavirus prevalence

The predicted HPV prevalence for all oncogenic types in a vaccinated cohort (vaccinated as 12- to 13-year-olds) is shown in Figure 26. This indicates that vaccination against 16/18 with current vaccine coverage rates will reduce the overall oncogenic HPV population prevalence in England by about 25% (assuming current generation vaccines do not confer significant long-lasting cross-protection against non-vaccine-included types). This is in accordance with expectations, as HPV 16/18 infections are expected to comprise ∼ 30–40% of all oncogenic type infections in the population, and vaccination coverage in England is approximately 84%.

FIGURE 26.

Predicted overall HPV prevalence for all oncogenic types in England, assuming 84% vaccination coverage (green dashed curve) and HPV prevalence predicted by the model in an unvaccinated setting (black dashed curve).

Cervical cancer incidence

The model-predicted cervical cancer incidence in vaccinated cohorts is depicted in Figure 27. Vaccination against HPV 16/18 at the current coverage rate is predicted to reduce the incidence of cervical cancer by close to 56% in women across all ages. This is in accordance with expectations as HPV16/18 infections are implicated in ∼ 65–70% of invasive cervical cancers and vaccination coverage is assumed to be 84% (and there is an additional effect of herd immunity).

FIGURE 27.

Cancer incidence per 100,000 women in a vaccinated cohort in England is shown as the green dashed curve. Model predictions in an unvaccinated setting are shown as the black dashed curve.

The predicted age- and type-specific incidence of invasive cervical cancer after vaccination is shown in Table 33. This table shows that after vaccination, HPV 16 and 18 associated cancers decrease significantly in relation to the other oncogenic types. Post vaccination, the relative proportions of HPV 16 and HPV 18 in cancer are predicted to be 12% and 3%, respectively.

Cervical cancer mortality

Mortality rates from cervical cancer in the vaccinated cohort are shown in Figure 28. The mortality rate is predicted to decrease by a similar ratio to that for cancer incidence in the vaccinated cohort.

FIGURE 28.

Cervical cancer mortality rates as predicted by the model in cohorts offered vaccination are shown as the green dashed curve. Model predictions in an unvaccinated cohort are shown as the black curve.

Predicted resource utilisation, cancer incidence and mortality, and costs associated with current screening practice in a vaccinated setting are summarised in Table 34.

Table 34 summarises the model-predicted outcomes in the context of current screening practice, with and without the effect of HPV vaccination. The total annual cost of cervical screening, management and diagnosis and cancer treatment under current practice is estimated at £155M per annum; and this would reduce to £129M per annum after vaccination takes effect. This reflects the fact that without a change in screening recommendation, only a modest reduction in the number of cytology and HPV tests are predicted, although vaccination will have a considerable influence on the predicted number of colposcopies, biopsies and treatments for CIN, due to the falling rates of cervical abnormalities in the population.

Cost, effectiveness, life-years predictions and quality-adjusted life-years predictions in unvaccinated cohorts

Model predictions for the predicted effectiveness of each of the strategies and their variants are presented in Table 37 and cost and cost-effectiveness outcomes are summarised in Table 38.

These results predict that most of the strategies and the variants considered will be less costly than current practice. Table 37 also shows the difference in effectiveness (in terms of life-years saved) between the variant strategies and the corresponding basic strategy without the variant. For each basic strategy, decreasing the follow-up interval for intermediate risk women from 24 to 12 months was the most effective variant, resulting in 73–113 more life-years per 100,000 women than with the corresponding basic strategy. The next most effective strategy was decreasing the routine screening interval from 6- to 5-yearly, which resulted in 40–45 more life-years per 100,000 women than with the corresponding basic strategy.

These results for cost and life-years saved are also summarised graphically in a cost-effectiveness plane, shown in Figure 31.

FIGURE 31.

Cost-effectiveness plane for unvaccinated women (life-years saved outcome). CP, current practice; FU, follow-up; S, strategy.

Compared with current practice, and considering the primary effectiveness outcome of life-years saved, almost all primary HPV screening strategies are in either the ‘south-west’ quadrant of the cost-effectiveness plane, (i.e. cost saving but less effective) or the ‘south-east’ quadrant (i.e. cost saving and also more effective). Almost no strategies were predicted to be more costly than current practice, with the notable exceptions of certain variants of co-testing such as 5-yearly screening and 12-month follow-up of intermediate risk women (strategy 4) and strategy 3 (partial genotyping) combined with 12-month follow-up of ‘other oncogenic’ positive women.

It should be noted that the most typical situation for cost-effectiveness analysis is that new strategies are more costly and more effective compared with current practice (i.e. in the ‘north-east’ quadrant of the cost-effectiveness plane). For the current evaluation, in which this generally does not apply, incremental cost-effectiveness ratios are not appropriate or informative53 and, therefore, results for costs, effects and resource utilisation are presented in a disaggregated form.

As discussed in Chapter 2 [see Methods, Cost and utilities (quality-adjusted life-year weights)], we considered two sets of QALY weights for exploratory analysis, and the findings are presented in the cost-effectiveness planes in Figures 32 (QALY weight set 1) and 33 (QALY weight set 2).

FIGURE 32.

Cost-effectiveness plane for unvaccinated women (QALY outcome) using QALY weight set 1. CP, current practice; FU, follow-up; S, strategy.

FIGURE 33.

Cost-effectiveness plane for unvaccinated women (QALY outcome) using QALY weight set 2. CP, current practice; FU, follow-up; S, strategy.

The ranking of strategies when QALYs were considered as the effectiveness outcome differed considerably from the life-years saved outcome (see Figure 31) for both QALY weight sets 1 and 2. QALY weights set one was used to calculate the outcomes depicted in Figure 32, and because this applies a disutility to the experience of being screened, the primary HPV screening strategies involving more screening (i.e. decreased screening interval) or more follow-up resulted in lower effectiveness. When the alternative set of QALY weights two (shown in Figure 33) were used, the position of strategies relative to each other on the cost-effectiveness plane was different again. QALY weight set two does not apply a disutility to being screened; however, this QALY weight set applies a higher disutility to having positive HPV test, and so all of the primary HPV screening strategies appear less effective than current practice. The diversity of the findings when different QALY weights are used emphasises the uncertainty involved in the selection and application of these weightings. Therefore, this report focuses on the life-years saved outcome as the main outcome. This result also highlights a need for further investigation into utilities relevant to primary HPV screening.

When considering the primary outcome of LY saved (see Figure 31), the basic strategies 2 and 3 both involved cost savings in the context of increased life-years saved. In both cases, increasing the frequency of screening to 5-yearly and/or decreasing the follow-up interval in HPV-positive, triage-negative women, improved the effectiveness (and increased costs) of the strategies. Either of these actions applied to strategy 1 also made this strategy both cost and life-year saving. The increased costs were primarily driven by increasing numbers of screening tests (for 5-yearly vs. less frequent screening strategies) or by increasing numbers of diagnostic referrals (for 12-month follow-up of HPV-positive, triage-negative women).

Most of the primary HPV strategies examined were cost saving and many of the strategies also resulted in an increase in predicted life-years saved in the population. The most effective strategy involving HPV as the sole primary screening test incorporated genotyping for HPV 16/18 and direct referral of women positive for these types to colposcopy, but although this is predicted to be cost saving compared with current screening practice, it would substantially increase the number of colposcopy referrals.

In general, strategies for which HPV is used as the sole primary screening test were found to be both cost and life-years saving, if attention was paid to the optimal combination of screening interval (5- or 6-yearly) and follow-up interval (12 or 24 months) for HPV-positive, triage-negative (‘intermediate risk’) women. Decreasing the follow-up interval for intermediate risk women from 24 to 12 months increased the overall effectiveness of primary HPV screening. If 24-month follow-up is retained, the (relative) loss of effectiveness can be partly compensated for by decreasing the screening interval from 6-yearly to 5-yearly, although this was generally a less effective approach than optimising the follow-up of intermediate risk women.

The exploratory strategies for which cytology screening was retained until either age 30 or 35 years, and for which a switch to primary HPV screening was implemented in older women, were predicted to be of somewhat higher costs than those associated with full implementation of primary HPV screening from age 25 years. Findings for this variant of strategy 1 are somewhat difficult to interpret, because the basic version of this strategy was slightly less effective than current practice. However, for strategy 2, the mixed strategy of cytology in younger women switching to HPV screening in older women was found to be of intermediate effectiveness to the related ‘pure’ strategies (i.e. less effective than full implementation of primary HPV screening but more effective than current practice).

Health outcomes, cervical cancer and mortality rates in an unvaccinated setting

Model predictions for health outcomes in an unvaccinated setting (i.e. prior to the impact of vaccination) are summarised in Table 39, and the related data for predicted age-standardised incidence and mortality rates for cervical cancer in England under current practice and for each of the potential primary HPV screening strategies are shown graphically in Figure 34. Each point on the graph represents the predicted incidence and mortality rate associated with a particular strategy (as shown in the figure). In general terms, most of the primary HPV screening strategies resulted in a predicted equivalence (or close-to-equivalence) or improvement in the predicted rates of cervical cancer incidence and mortality, compared with current practice, with the notable exception of strategies involving 10-yearly screening in women over 50 years of age.

FIGURE 34.

Age-standardised rate (ASR) (assuming the European Standard Population52) predictions for cancer incidence and mortality are shown for each strategy and the variants as the black circles and green squares, respectively. Current practice (CP) predictions for cancer incidence and mortality are shown as the black and green lines, respectively, for ease of comparison with the other strategies.

Resource utilisation in an unvaccinated setting

Table 40 shows predicted annual outcomes of resource utilisation in England (assuming an unvaccinated setting, i.e. prior to the impact of vaccination). The annual number of HPV tests and cytology tests are shown graphically in Figure 35. This figure shows that the number of screening tests in the majority of strategies is predicted to be lower than current practice, although in the co-testing strategies (strategy 4 and variants), the number of overall number of HPV and cytology tests is, not unexpectedly, predicted to increase.

FIGURE 35.

Annual number of cytology and HPV tests performed in England as predicted by the model for each of the strategies and the variants in an unvaccinated setting. CP, current practice; S, strategy.

The annual number of colposcopies performed in England is shown in Figure 36. The strategies varied in respect to whether a greater or lower number of colposcopies were predicted in relation to current practice. In general terms, more colposcopies were predicted for strategy 3, which involved genotyping and thus referred HPV 16/18 positive women to colposcopy immediately. In general terms, the number of biopsies was predicted to be proportional to the number of colposcopies performed – the model generally predicts that about 48% of women who attend a colposcopy will have a biopsy taken.

FIGURE 36.

Annual number of colpsocopies performed in England as predicted by the model for each of the strategies and the variants. CP, current practice; S, strategy.

Summary of the effectiveness and costs of the strategies in an unvaccinated cohort

Overall, when considering unvaccinated women, strategies 2 and 3 (and most of their variants) were cost and life-years saving, but strategy 1 could also be tailored to be cost and life-years saving by decreasing the follow-up time for intermediate risk women to 12 months.

Strengths of the current analysis

The analysis presented in this report used a model with extensive type-specific natural history calibration to ARTISTIC data and shows close duplication of the actual findings in the ARTISTIC trial. The England-wide model, which was used for the overall evaluation of the costs and effects of primary HPV screening in England, incorporated a careful and detailed simulation of sexual behaviour in England, the natural history of type-specific HPV natural history (progression and regression), and a detailed simulation of cervical screening, follow-up and management. This incorporated information on compliance with screening recommendations and also explicit modelling of post-treatment recurrence and test-of-cure management for England. We performed extensive calibration of the England-wide model, including fitting of the sexual behaviour assumptions to national survey data for sexual behaviour (NATSAL II), validation to ensure that the model accurately predicted type-specific HPV infection rates (pre-vaccination) and accurate predictions, under conditions of current screening practice, of age-specific cancer incidence and mortality rates, as well as stage-specific survival. In addition, we validated model predictions for the predicted number of cervical screening tests and other outcomes against current data.

Unlike many previous models of cervical screening, we have also taken into account detailed information on age-specific compliance with cervical screening, incorporating not only overall participation rates, but also the timing of reattendance after a negative screening test.

Other cost-effectiveness evaluations of primary HPV screening has been performed in the Netherlands,56 Germany,57 Norway58 and Canada,59 and each of these analyses concluded in favour of switching to longer-interval HPV-based screening over cytology-based screening. However, our modelled analysis involved a significantly more detailed validation of the modelled natural history assumptions and significantly more detail in terms of the specification of how cervical screening is modelled, as well as a detailed simulation of the effect of vaccination. In addition, we performed extensive analysis of a large number of potential strategies, allowing overall patterns and generalisable model conclusions to be analysed.

Limitations of the current analysis

The main limitations of the current analysis relate to the fact that a number of assumptions about future population behaviour were made, by necessity, and to some extent the conclusions of the current analysis are dependent on whether the assumptions made reflect future screening and vaccination practice.

For example, a key assumption in this modelled analysis was that compliance with follow-up for HPV-positive, triage-negative women will involve approximately 80% of women attending within the recommended follow-up time. However, in sensitivity analysis we demonstrated that the overall outcomes of the analysis are highly dependent on such assumptions, and if actual compliance is less (or more) than 80%, this will have a substantial impact on the findings for the relative costs and effects of primary HPV screening. We also assumed independence of screening and vaccination compliance rates in the population. In the context of high coverage rates for vaccination, as observed in England, heterogeneities between vaccination and screening uptake may not be a major driver of outcomes. There exists high coverage of the vaccine in the school-based system, and there is not strong evidence to suggest that vaccination uptake in individual 12- to 13-year-olds will be correlated with their screening compliance as adults in England. Therefore, we believe this assumption is reasonable at the current time.

Secondly, there are major uncertainties in relation to the costs of primary HPV screening in the future. Although we used the findings of a dedicated costing exercise in relation to current test prices, there has been a general move towards automation of HPV testing (which may drive down test-associated costs) and there is an active competitive environment, with several new HPV test technologies entering the market in recent years. If HPV test costs eventually drop to lower values than those considered here, this will further increase the favourable cost-effectiveness profile of primary HPV testing relative to cytology. In this sense the current evaluation should be considered conservative in relation to the relative costs of primary HPV testing compared to current screening practice. However, we did consider a range of potential future costs in a PSA for costs, and found that HPV screening (if HPV is used as the sole primary test) is predicted to be cost saving under many combinations of projected costs.

Thirdly, the baseline analysis assumed that current levels of HPV vaccination coverage rates will be sustained, and that the vaccine-induced duration of protection will be lifelong. Both of these are likely to be reasonable assumptions given the school-based programme and an absence of evidence that vaccine efficacy substantively decreases over time. We presented our findings for the costs and effects of new potential screening strategies in both unvaccinated and vaccinated cohorts, and it should be noted that both are important in interpretation of the overall results. The results for unvaccinated and vaccinated cohorts (with high coverage rates) should be considered to encompass a range of possibilities with respect to the influence of vaccination coverage on the finding for primary HPV screening. Strategies that were found to have a favourable balance of costs and effects in both situations are likely to also have a favourable balance with respect to intermediate vaccination coverage situations. We also considered vaccination coverage in sensitivity analysis, and found that this factor was not a major influence on the findings for the relative costs and effects of primary HPV screening strategies in relation to current practice.

Fourthly, we made several assumptions about vaccine efficacy. The potential for type-replacement of non-vaccine HPV types has been raised as a concern, although substantive evidence for this phenomenon has not emerged. 60 It is thought to be unlikely as there is little evidence for interaction between HPV types. If substantive evidence type replacement were to be identified then our findings for a vaccinated setting would require re-evaluation. Also, because we used a hierarchical approach to lesion type-assignment when fitting the model to data (HPV 16, then HPV 18, then other oncogenic infections), the rate of other oncogenic type infections may be underestimated, and thus there is potential for HPV 16/18 vaccination to ‘unmask’ underlying co-infections with other oncogenic types. In terms of the effect on the analyses presented here, it is possible that we have somewhat overestimated the efficacy and efficiency (in terms of colposcopy referrals) of partial genotyping strategies in a vaccinated population.

In terms of the partial genotyping strategies considered, we assumed that HPV 16/18 women would be managed in the same way, and did not consider differential management strategies for HPV 16- and HPV 18-positive women. Further work would be needed to assess the potential for strategies involving differential consideration of women according to their infection status with respect to these types or with respect to type 45. In addition, although we did perform initial exploratory analysis of cytology-to-HPV ‘switching’ strategies (where the choice of primary screening test is dependent on a women’s age at the time of screening), the results of such analyses must be considered preliminary as they critically depend on a range of assumptions made about the management of women in follow-up and about the management of underscreened women around the time of the primary test ‘switch’.

In general terms, we have attempted to deal with the uncertainties inherent in modelled analysis via a range of methods. Firstly, we used a modelling platform that was extensively pre-validated against data from several other countries, and was then specifically revalidated against both ARTISTIC data (via a specific simulation of the ARTISTIC cohort) and, for the England-wide model, several observational data sets. Secondly, we have performed extensive sensitivity analysis using a range of techniques including exploring the impact of using different utility weight sets, performing one-way sensitivity analysis of the main results of the analysis, and performing PSA for costs.

The consistency of our findings overall is an important outcome. We found, for a range of strategies, that primary HPV screening is likely to be more effective and less costly than current practice for cervical cytology, provided that attention is paid to the optimal management of HPV-positive, cytology-negative intermediate risk women. Ultimately, the final choice of the particular strategy to be used will consider a trade-off between effectiveness, costs, colposcopy referrals, acceptability to women, and practicalities.

The ARTISTIC protocol originally defined definitions of the duration of time for the second and third screening rounds. This was necessary because women attend following recall for screening over a continuum of time following the invitation. As it turned out, this continuum was of longer duration than anticipated, and it was felt that to avoid lesions being attributed to the wrong screening round, the definitions should be widened. This would of course not affect the combined results for rounds 2 and 3, which is of greatest relevance to this report, and its transfer of some lesions that would have been attributed to round 3, to round 2, is not considered to be of great significance.

The ARTISTIC Trial study group

Acknowledgement

We thank Dr Rosa Legood of the London School of Hygiene and Tropical Medicine for supplying some of the cost data used in the economic analysis.