THE UNIVERSITY OF SHEFFIELD

Appendix 5 The Sheffield model THE UNIVERSITY OF SHEFFIELD Modelling the cost-effectiveness of human papillomavirus (HPV) testing for triage of women with low-grade abnormal cervical smears : a study within the TOMBOLA trial A dissertation submitted in partial fulfillment of the requirements for the degree of Master of Health Economic and Decision Modelling by Eun-Ju Kim 09/2010

Modelling the cost-effectiveness of human papillomavirus (HPV) testing for triage of women with low-grade abnormal cervical smears : a study within the TOMBOLA trial A dissertation submitted in partial fulfillment of the requirements for the degree of Master of Health Economic and Decision Modelling at The School of Health and Related Research, The University of Sheffield by Eun-Ju Kim 09/2010 Supervisor: Hazel Squires Word count: 14,832

TABLE OF CONTENTS ABSTRACT... 1 1. INTRODUCTION... 1 1.1. Aims and objectives... 1 1.2. Cervical cancer and HPV... 3 1.3. The UK cervical screening programme... 5 1.4. The role of HPV test... 6 2. LITERATURE REVIEW... 7 2.1. Search strategy and results... 7 2.2. Summary of included studies... 8 3. METHODS... 12 3.1. Description of model... 12 3.2. Screening pathway model... 13 3.3. Natural history model... 17 3.4. Model data... 18 3.5. Sensitivity analysis... 22 4. RESULTS... 28 4.1. Model validation... 28 4.2. Base case... 29 4.3. One-way sensitivity analysis... 35 4.4. Probabilistic sensitivity analysis... 37 5. DISCUSSION... 38 6. CONCLUSION... 40 REFERENCES... 42 APPENDIX... 46 1. Ethics application form and the approval letter... 46 2. Search Strategies and the results... 48 3. Summary of systematic review... 50 1

Tables <Table 1> Referral policy to colposcopy... 16 <Table 2> Key parameters and sources... 23 <Table 3> Baseline estimates of costs and effectiveness... 30 <Table 4> Long-term effectiveness by strategy... 31 <Table 5> Long-term costs and resource use by strategy... 31 <Table 6> Cost-effectiveness compared by the type of cytology-based strategy... 34 <Table 7> Results of one-way sensitivity analysis compared with Strategy A... 36 Figures <Figure 1> The number of deaths by cervical cancer in 1999-2008... 2 <Figure 2> Numbers of new cases and age-specific incidence rates of cervical cancer... 3 <Figure 3> Major steps in the development of cervical cancer... 4 <Figure 4> Flow-chart of study selection... 8 <Figure 5> Conceptual framework of the screening pathway model without HPV test... 13 <Figure 6> Conceptual framework of the screening pathway model with HPV test... 14 <Figure 7> Natural history model... 17 <Figure 8> Estimated age-specific prevalence of HPV and CIN... 28 <Figure 9> Estimated age-specific incidence of invasive cancer... 29 <Figure 10> Estimated age-specific cervical cancer death rate... 29 <Figure 11> The number of CIN1... 32 <Figure 12> The number of CIN2/3... 33 <Figure 13> The number of invasive cancer stage I... 34 <Figure 14> CEACs of strategies with (D1-F3) or without (A-C) HPV testing... 37 <Figure 15> CEACs of all strategies... 37 2

Abstract Objective: The aim of this study is to evaluate cost-effectiveness of a single HPV testing for women with a borderline or mild smear result in NHS perspective. Methods: A Markov model incorporating a screening pathway model and a natural history model of cervical cancer was developed based on the ScHARR colposcopy model. Data were primarily derived from the TOMBOLA trial and the ScHARR model. 12 screening strategies depending on the referral policies to colposcopy were compared. The key outcomes were the number of deaths from cancer, the detection rate of high-grade disease, QALYs and costs per QALY gained. Results: Compared with the current recommendation based on conventional cytology, HPV testing did not reduce cancer incidence due to the low sensitivity for CIN2 or worse in the TOMBOLA trial. However, a single HPV testing was estimated to save QALYs reducing utility decrements associated with routine cytology-based screening. Among various strategies using HPV test, the strategies providing a single HPV test for women over 30 (E1-E3) were the most costeffective strategies. Conclusion: A single HPV testing for women with borderline or mild smears is cost-effective in the UK. However, the performance of HPV test in detecting highgrade disease should be clearly assured in order to prove the value of HPV testing. 1. Introduction 1.1. Aims and objectives Since the National Health Service (NHS) cervical screening programme was launched in 1988, it has contributed to reduction in mortality rate of cervical cancer significantly through detecting and treating early-stage precancerous lesions in the UK (Figure 1) (NHS, 2009). However, the management of women, diagnosed as having low-grade abnormalities including borderline nuclear change or mild dyskaryosis is controversial, as, though a proportion of these women have precancerous lesions and need treatment, many do not. Many studies have been conducted to evaluate the cost-effectiveness of various management strategies for this group of women. Based on these evidences, recent NHS cervical cancer guideline recommends a referral to colposcopy after 3 repeated results of borderline change or after 1 mild dyskaryosis (NHSCSP, 2010). 1

<Figure 1> The number of deaths by cervical cancer in 1999-2008 in the UK (NHS, 2009) Currently, interests in the role of human papillomavirus (HPV) test to triage low-grade abnormalities have been increased. The strong causation between HPV and cervical cancer supports inclusion of HPV test within the national screening programme. As HPV testing appears to be more sensitive than repeating cervical cytology for identifying women with low-grade abnormality who might have cervical intraepithelial neoplasia (CIN) 2 or 3, it seems to help decide whether a woman with low-grade abnormal smears needs treatment or not. However, the TOMBOLA trial (Trial Of Management of Borderline and Other Lowgrade Abnormal smears) has recently shown that a HPV test would not be effective in triage of all women, although the test has a very high negative predictive value in women over 40 years suggesting that it might be of use in this age-group (Cotton et al, 2010). An economic evaluation alongside the TOMBOLA trial has been undertaken (TOMBOLA Group, 2009), but the long-term costs and outcomes associated with HPV testing are unknown. Thus this study aims to analyze the long-term cost-effectiveness of a single HPV testing in triage of women with low-grade abnormalities within the TOMBOLA trial. This study has been approved ethically by the Ethics Committee of Sheffield University (Appendix 1). The objectives of this study are; 1) To design a screening pathway model based on various screening strategies 2) To construct a natural history model of cervical cancer 3) To evaluate the long-term cost-effectiveness of HPV testing for women with lowgrade abnormalities through a mathematical model linking the screening pathway model and the natural history model 2

1.2. Cervical cancer and HPV Cervical cancer is malignant neoplasm of the cervix connected by vagina. As cervical cancer is one of the most common cancers in women, it accounts for approximately 2% of all female cancers in the UK. In 2007-08, there were 2,828 new cases of cervical cancer and 957 deaths from cervical cancer in the UK. The European age-standardized annual incidence rate was 8.4 per 100,000 females and the death rate was 2.4 per 100,000 females (Cancer research UK, 2010). The prevalence of HPV reached a peak around age 30-34 in 2007 (Figure 2). <Figure 2> Cervical Cancer (C53), Numbers of new cases and age-specific incidence rates of cervical cancer (C53) in the UK (Cancer research UK, 2010) Cervical cancer is mainly caused by a venereal transmission of HPV, and more common in women who had sexual intercourse at an early age or having multiple sexual partners (Fauci et al, 2008). HPVs, are members of Papovaviridae family, and selectively infect the epithelium of skin and mucous membranes. The causal association between HPV infection and cervical cancer is well-known, as HPV DNA has been recovered from over 95% of all cervical tumours (Cuzick et al, 2000). Particularly HPV16 and HPV18 are known to be responsible for 70% of cervical cancers and about 50% of cervical intraepithelial neoplasia (CIN) grade 3 (Smith et al, 2007). These carcinogenic HPVs invade the cervical 3

transformation zone through direct contact with infectious lesions, and then cause dysplasia gradually stimulating the normal metaplasia over years. Most infections are asymptomatic and transient and cleared within 2 years, but 10% remains persistent (Stanley, 2006). The persistent HPV infection progresses to cervical cancer in 10-20 years on average (Schiffman et al, 2007). <Figure 3> Major steps in the development of cervical cancer (Schiffman et al, 2007) According to the European Commission Training Programme (ECTP) terminology, CIN is classified into 3 grades from CIN1 to CIN3 depending on the risk of progression to cervical cancer (Coleman et al, 1993). In terms of histopathology, CIN3 represents precancer, and CIN2 is equivocal in cancer potential. People with CIN1 have a lower risk of progression to cervical cancer (Schiffman et al, 2007). Cervical cancer is generally detected on routine cervical screening. According to the International Federation of Gynecology and Obstetrics (FIGO) classification, Stage I is a disease confined to the cervix, stage II, one which invades beyond the cervix but not the pelvic wall or lower third of the vagina, stage III is a disease to pelvic wall or lower third of the vagina and Stage IV is present when the tumor invades the mucosa of bladder or rectum or extends beyond the true pelvis. Five-year survival rates are 85% for Stage I, 65% for Stage II, 35% for Stage III and 7% for Stage IV (Fauci et al, 2008). 4

1.3. The UK cervical screening programme Cervical cancer screening programme includes the whole process of cytological screening, triage of equivocal results, colposcopy and biopsy of abnormal screening results, treatment and follow-up. According to the recent guideline for the NHS cervical screening programme, all women at ages 25-64 are invited for screening and should be screened triannually between the ages of 25 and 49 and five-yearly thereafter (NHSCSP, 2010). Women over age 65 are only called for screening if they have not been screened since age 50 or have had recent abnormal test results. Women under age 25 are not called for screening anymore because much of the prevalence of HPV infection is transient and disappear spontaneously. In 2009, the number of women invited for screening reached 4.0 million, and the coverage rate was 78.9% in the UK (NHS, 2009). The current standard method of screening is liquid-based cytology (LBC). Compared with the conventional cytological test, it is thought to improve sensitivity and laboratory efficiency. It improved the readability of the slides reducing the proportion of inadequate smears from over 9% until 2006 to 2.5% in 2008-09 (NHS, 2009). Cytological results are classified into borderline, mild, moderate or severe dyskaryosis depending on the risk of cervical cancer. In 2008-09, 22% had borderline changes, 39% had mild dyskaryosis, 16% had moderate dyskaryosis and 21% had severe or worse dyskaryosis among 107,354 screening samples (NHS, 2009). Women who have the cytological results showing moderate or severe abnormalities are recommended to be referred for a colposcopy within four weeks. It involves a detailed visual assessment of the cervix and may incorporate a biopsy when endocervical tumor is suspected. Current guideline recommends that women with mild dyskaryosis should also be referred for a colposcopy immediately, although a repeat test after 6 months is still acceptable. For women with borderline change, a repeat test is recommended in six months, and colposcopy is expected after three consecutive borderline changes. Following a mild or borderline smear change, women are recommended to return to routine screening after three negative test results at least every six months. Women having an inadequate result are also called for rescreening and referred for a colposcopy after three consecutive inadequate samples. Women are returned to regular screening in 3 or 5 years after three negative tests or colposcopic assessment presenting no abnormality. Treatment may be carried out at the first visit, or at later date. However, immediate treatment for borderline or mild dyskaryosis should be decided carefully and only in 5

exceptional cases because of the potential for over-treatment. There are several surgical techniques to treat precancerous lesions such as knife cone biopsy, laser conisation, large loop excision of the transformation zone (LLETZ), laser ablation, cryocautery, cold coagulation, and radical diathermy. For management of CIN1, treatment is not necessarily required, but cytological rescreening should be followed up instead until cytological regression occurs or treatment is undertaken. Cryocautery with a double freeze thaw freeze technique is recommended only for low-grade CIN. Hysterectomy is recommended for women with high-grade CIN. All women must have had histological diagnosis established before destructive therapy. Unless an excisional treatment is planned, biopsy should be carried for moderate dyskaryosis or worse. 1.4. The role of HPV test While cytological tests such as Papanicolaou (Pap) and LBC test have significantly contributed to detection of precancerous lesions, the weaknesses in sensitivity and specificity led to the interest in the use of HPV testing as a screening method. This interest is based on the evidence that HPV DNA is found in a large proportion of cervical cancer, and the ease of use with higher sensitivity to detect high-grade CIN than cytology. The higher sensitivity can contribute to reduction of cancer rates and longer screening intervals. Therefore, HPV testing is expected to improve the effectiveness of cervical cancer screening programme providing much precise diagnosis of those women at risk of progression to CIN2 or worse. HPV-DNA is a molecular assay that can be used to detect the presence of high-risk oncogenetic strains of HPV. Although HPV testing is currently not recommended for routine use in the UK, its addition for the test to triage equivocal cytology has been considered in many other countries. According to the HPV test result, HPV positive group would be referred to colposcopy, and HPV negative group could be reassured and returned to routine screening. The role of HPV testing for the management of women with borderline or mild dyskaryotic smears is considered to be clear at present (Cuzick et al, 2000). The International Agency for Research on Cancer (IARC) concluded that there is sufficient evidence to support that HPV testing can reduce the incidence and mortality from cervical cancer and that it is likely to be at least as effective as cytology. Moreover, the English 6

Cervical Screening Advisory Committee has recently accepted the IARC statement and agreed that an HPV test is a valid screening method, which could be used by the national screening program (Cuzick et al, 2006). However, there are still issues with regard to screening ages and appropriate test intervals, which could be age-dependent and the best combination with cytology (Cuzick et al, 2000). Thus this study focused on finding the best screening strategy integrating conventional screening and a single HPV test for the different age groups of women with low-grade abnormalities. 2. Literature review 2.1. Search strategy and results A literature review was conducted on previous modelling studies on the costeffectiveness of HPV testing within a cervical screening programme. The Cochrane Library, Medline and Embase were searched in July 2010, and additional papers were included through reference tracking. Search keywords were determined by population, interventions, comparators and outcomes (PICO), and the combinations of each term were used as a search strategy. Full search strategies and search results are provided in Appendix 2. Through extensive literature search, 216 papers were identified. After excluding duplications and checking abstracts and original papers, potential papers were selected. Vaccine studies and studies not including any cost-effectiveness analysis (CEA) modelling technique were excluded. Each stage of the process is presented in Figure 4. In total, 16 relevant modelling papers were included, and each study was reviewed in terms of study characteristics, natural history model, screening strategy, performance of HPV test and CEA result (Appendix 3). 7

<Figure 4> Flow-chart of study selection 2.2. Summary of included studies Four studies were based on the US populations, and two studies, on the UK population. Most studies used a Markov model with Monte Carlo simulations 6-monthly or yearly, and cohorts were followed up over lifetime. Legood et al. conducted a cost-effectiveness modelling study on HPV test conducted as part of the NHS HPV/LBC pilot studies in the UK (Legood et al, 2006). The detailed methodology and results were reported in the original NHS HPV/LBC pilot studies (Moss et al, 2004). A Markov model following up a cohort of women aged 25-64 with a borderline or mild dyskaryosis was constructed. Without any relevant UK-based modelling study on HPV 8

infection, they chose the Myers et al. model as a natural history model because it incorporates the history of HPV explicitly; it is validated; and all the parameters used are in the public domain (Moss et al, 2004). They compared 5 screening strategies, including LBC only (Strategy A), combined LBC and HPV testing for all women (Strategy B) or women aged over 35 (Strategy C), combining LBC and adjunctive HPV testing for women aged 35 or more (Strategy D), and combining LBC and adjunctive HPV testing but no testing for HPV in repeat tests context (Strategy E). Compared with Strategy A, Strategy D cost 3735/LYG, Strategy E cost 4233/LYG and Strategy B cost 18605/LYG while Strategy C was dominated. The other UK-based study is a study by Sherlaw-Johnson and Philips comparing repeating cytology, HPV triage, primary HPV and combined cytology and HPV with 3- or 5- year routine intervals and both with and without LBC (Sherlaw-Johnson and Philips, 2004). They modified the previous Sherlaw-Johnson's model (Sherlaw-Johnson et al, 1994), and used a number of plausible ranges of input values from various literatures. The results presented that the strategies with HPV triage consistently lie on, or close to the efficiency frontier. Sherlaw-Johnson et al. also investigated the feasibilities of HPV testing within Eastern Europe with the same model above and strategies, comparing smear testing and 5-yearly or 10-yearly HPV testing (Sherlaw-Johnson et al, 2000). Although it showed that 10-yearly HPV testing would be a cost-effective option, the results were limited to interpret because the model had not been locally validated. Kim et al. examined the cost-effectiveness of HPV testing in European countries including the UK, the Netherlands, France, and Italy (Kim et al, 2002). Each country s current screening policy was compared with HPV DNA testing and combination testing, that is, cytology for women up to the age of 30 and HPV DNA testing for those over 30 years of age every 3- or 5-yearly. In the UK, both strategies of HPV triage or combination testing every 5 years cost less than $15000 per year of life saved. As the screening interval is shorter, the costs per year of life saved were increased. All other strategies were strongly or weakly dominated. The rank ordering of strategies was similar in the other countries. Kim et al. also examined the cost-effectiveness of HPV testing as an alternative triage strategy for atypical squamous cells (ASC) results in the US (Kim et al, 2002). The study showed that, compared with HPV DNA testing, a strategy of repeat cervical cytology or delayed HPV testing costs more but less effective. There were two more US-based studies using Markov model (Mandelblatt et al. 2002; 9

Maxwell et al, 2002). Unusually, Mandelblatt et al designed a deterministic semi-markov model and combined low-grade squamous intraepithelial lesion (LSIL) and HPV status in their model (Mandelblatt et al. 2002). Moreover, they assumed that HPV negative can develop invasive cancer. Maxwell et al. constructed a Markov model with a cohort of military beneficiaries aged 18-85, and compared conventional cytology, LBC and LBC with HPV triage at 1-, 2-, and 3-year intervals (Maxwell et al, 2002). These studies also concluded that the combination strategies between cytology and HPV test are more effective and less costly than conventional cytology repeating cytology in the US. Two studies showed the cost-effectiveness of HPV testing in Canada (Kulasingam et al, 2009; Chuck et al, 2010). Both studies defined 27 and 21 strategies, respectively depending on the age at which to begin screening, frequency and type of screening. In the comparison of various strategies, they showed that there were trade-offs in costs and benefits associated with each screening strategy, and suggested the importance of finding the combination of treatment type and screening process to provide the best balance between costs and benefit. Berkhof et al. designed a Markov model with a Netherlands-based cohort (Berkhof et al, 2006). Age-specific incidence rates were derived from the POBASCAM (the Population Based Screening Study Amsterdam) and transition probabilities came from recent Dutch longitudinal datasets. The classical strategy for cytological follow-up after a borderline or mild dyskaryosis was compared with adjunct HPV testing at different recall intervals. They did not differentiate between borderline and mild dyskaryosis because the management is the same for these classes in the Netherlands. The results showed that HPV testing strategies were more effective in terms of cancer-free life expectancy and less expensive than the classical strategy. Cost savings were mainly due to a drop in the number of colposcopy referrals. The study by Bistoletti et al. compared cervical cytology screening, same strategy with a single HPV test at age 32, screening with combined cytology and HPV at every 9 years and no screening in Sweden (Bistoletti et al, 2008). The cohort consisted of women aged 32, and the cycle length was 3-5 years which is the length of the screening intervals in Sweden. Transition probabilities were mainly derived from a randomized population-based study in Sweden. According to the results, a single HPV test was dominated by cytology screening, while combined cytology with HPV testing at least 9 years was dominating the conventional cytology screening strategy. The studies by Goldie et al. and Vijayaraghavan et al. were conducted in South Africa 10

(Goldie et al, 2001; Vijayaraghavan et al, 2009). They had a characteristic of considering HIV in the natural history model, as HIV increases the risk of HPV infection. The screening strategies also had differences with other studies; for example, including direct visual inspection (DVI) or long screening interval of 10 years, as they had to reflect low-resource settings. However, they also showed that incorporating HPV testing could be an attractive alternative to cytology-based screening programmes where resources are limited. Andrés-Gamboa et al. assessed the cost-effectiveness of conventional cytology and HPV testing with country-specific data in Colombia (Andrés-Gamboa et al, 2008). The strategies include a primary HPV testing in three or five years followed by cytology in positive results and HPV testing for women receiving Atypical Squamous Cells of Undetermined Significance (ASCUS). The results showed that HPV screening dominates cytology-based screening, and the primary HPV testing was the least costly and the most cost-effective strategy. Kulasingam et al. designed a decision-analytic model based on ALTS trial which is a multicenter randomized trial that was designed to evaluate three management strategies for detection of CIN3 or cancer in the US (Kulasingam et al, 2006). It compared repeating Pap smear, triage with HPV testing and immediate treatment for 2 years, and showed that the HPV triage strategy would be cost-effective with an ICER of $3517 per CIN3+ detected compared with the conservative strategy. Sheriff et al. also compared repeating Pap smear, triage with HPV testing and immediate treatment for the management of atypical and abnormal Pap smear results in a decision-analytic model (Sheriff et al, 2007). For patients with initial PapIIw, III, and IIId results, ICERs for HPV triage versus repeat Pap smears are 2,232, 815, and 487 per additional case of CIN2+ detected and treated 1. The published results varied due to various populations and strategies, different modelling assumptions and input data. While the structures of natural history model were similar, input data such as HPV incidence and transition probabilities tended to be countryspecific. Screening strategies also varied according to eligible ages of screening, screening interval, type of cytological test and whether it is a primary or adjunctive HPV test. 1 In Germany, which grades Pap smears in accordance with the Second Munich Cytological Classification, such smears include PapIIw (inadequate specimens, minimal dysplastic changes, equivalent to atypical squamous cells of undetermined significance unofficial category), PapIII (important degenerative, inflammatory or iatrogenic changes of the cells where benignity or malignancy cannot be diagnosed with certainty even if the smear is adequately prepared), and PapIIId [mild-to-moderate dysplasia, cervical intraepithelial neoplasia (CIN) I and II]. (Sheriff et al, 2007). 11

However, most studies found the similar results that HPV testing would be a costeffectiveness option for management of women with borderline or mild dyskaryosis within an acceptable range of incremental cost-effectiveness ratio (ICER). 3. Methods 3.1. Description of model The model was designed to analyze the lifetime cost-effectiveness of a single HPV testing for a triage of women with borderline and mild dyskaryosis within the NHS screening programme. It includes a Markov model that simulates a screening pathway model and a natural history model of cervical cancer in a hypothetical cohort. The cohort consisted of 100,000 women aged 15 and those with no previous precancerous disease and cervical cancer. The key outcomes are the number of deaths from cancer, the detection rate of high-grade disease (CIN2/3 and invasive cancer), life-years gained (LYG), qualityadjusted life years (QALYs) and costs per QALY gained in NHS perspective. All costs per QALY gained were compared with those of Strategy A reflecting the current practice for generalization. A strategy was considered to be cost-effective if the ICER is less than 30,000 which is generally accepted as an ICER threshold in the UK. The ScHARR colposcopy model was reviewed and modified as appropriate (Eggington et al, 2006). As the ScHARR model was initially designed to evaluate the cost-effectiveness of the change to mild referral guidelines to individual colposcopy clinics in England, it provides not only validated transition probabilities and colposcopy findings, but also a wide range of national colposcopy practice data (Eggington et al, 2006). The model was constructed using Excel. Conventional cytology was assumed as the comparator in base-case, and LBC was considered in sensitivity analysis. Costs and benefits were discounted at 3.5% a year. To test the effect of parameter uncertainty, both one-way deterministic and probabilistic sensitivity analysis were conducted. Model validity was assessed by comparing the predictions with published data on incidence of HPV, CIN and cancer in the UK. 12

3.2. Screening pathway model The screening pathway model represents the process from an initial cytology test to colposcopy, and predicts the number of women from each smear state and being referred to colposcopy at every 6 months. Figures 5 and 6 present the conceptual frameworks of the screening pathway model with or without HPV test. <Figure 5> Conceptual framework of the screening pathway model without HPV test 13

<Figure 6> Conceptual framework of the screening pathway model with HPV test Following the current UK guideline, women were assumed to have routine screening between the age of 25 and 64 at age-related interval, that is women between the age of 24 and 49 are screened every three years and women between the age of 50 and 64 are screened every five years. Screening was assumed to be taken up by a certain percentage of women in the cohort. It was defined as the coverage of screening. In 2008-09, coverage rate was 78.9% within the previous five years and 69.3% with the previous 3 years in the UK (NHS, 2009). Women who do not attend for a routine screening were assumed to follow the natural history model only. HPV test could be used as the primary screening modality, an adjunct to cytology or a single HPV testing for the triage of women with borderline and mild dyskaryosis. However, this study focused on the third role, because the clearest role for HPV testing is in the management of women with borderline or mild dyskaryosis at the moment (Cuzick et al, 2000). To examine the difference in the cost-effectiveness of HPV testing by age group, age restrictions of over 30 and 40 were set and compared with the strategies providing HPV testing to all women. 12 screening strategies were distinguished by referral policy to colposcopy, age group providing HPV testing and management policy for women with HPV negative result (Table 1). 14

The strategies were as follows; 1) Strategy A: Women with a smear test result of mild are referred for immediate colposcopy. Women with a borderline smear result are called for rescreening at 6 months, and referred for colposcopy after a third borderline result. 2) Strategy B: Women with a borderline or mild smear result are called for rescreening at 6 months, and referred for colposcopy after a third borderline or a second mild result 3) Strategy C: All women with a smear result of borderline and mild are referred for immediate colposcopy. 4) Strategy D1: All women with a borderline or mild smear result are called for a single HPV DNA test. Women with HPV positive are referred for colposcopy, but women with HPV negative return to routine screening with Strategy A. 5) Strategy D2: All women with a borderline or mild smear result are called for a single HPV DNA test. Women with HPV positive are referred for colposcopy, but women with HPV negative return to routine screening with Strategy B 6) Strategy D3: All women with a borderline or mild smear result are called for a single HPV DNA test. Women with HPV positive are referred for colposcopy, but women with HPV negative return to routine screening with Strategy C 7) Strategy E1: Women over 30 years old with a borderline or mild smear result are called for a single HPV DNA test. Women with HPV positive are referred for colposcopy, but women with HPV negative return to routine screening with Strategy A. Women under the age of 30 follow Strategy A. 8) Strategy E2: Women over the age of 30 with a borderline or mild smear test result are called for a single HPV DNA test. Women with HPV positive are referred for colposcopy, but women with HPV negative return to routine screening with Strategy B. Women under 30 years old follow Strategy B. 9) Strategy E3: Women over 30 years old with a borderline or mild test smear result are called for a single HPV DNA test. Women with HPV positive are referred for colposcopy, but women with HPV negative return to routine screening with Strategy C. Women under the age of 30 follow Strategy C. 10) Strategy F1: Women over the age of 40 with a borderline or mild smear test result are called for a single HPV DNA test. Women with HPV positive are referred for colposcopy, but women with HPV negative return to routine screening with Strategy A. Women under the age of 40 follow Strategy A. 15

11) Strategy F2: Women over the age of 40 with a borderline or mild smear test result are called for a single HPV DNA test. Women with HPV positive are referred for colposcopy, but women with HPV negative return to routine screening with Strategy B. Women under 40 years old follow Strategy B. 12) Strategy F3: Women over the age of 40 with a borderline or mild smear test result are called for a single HPV DNA test. Women with HPV positive are referred for colposcopy, but women with HPV negative return to routine screening with Strategy C. Women under the age of 40 follow Strategy C. Strategies A, B and C are cytology-based strategies which refer women with borderline or mild smear to colposcopy after the specified number of smears above. Strategy A reflects the current recommendation of the management for women with low-grade abnormal cervical smears. Strategy B had been recommended previously to refer the women after a second mild and a third borderline result, and Strategy C assumed that all women with any abnormality would be referred immediately. Strategies D, E and F include a single HPV testing for triage of women with borderline and mild dyskaryosis. Age restrictions were applied providing HPV testing to all women (D), to women over 30 (E), and to women over 40 (F). These were subdivided into D1-F3 by the cytological management for women with HPV negative and in age-restricted groups. <Table 1> Referral policy to colposcopy Strategy Referral for colposcopy Management for HPV- Management for age restriction A B C D1 D2 D3 E1 E2 E3 F1 F2 F3 Based on the current guideline after 1 mild smear after 3 borderline smears Based on the previous guideline after 2 mild smears after 3 borderline smears Immediate colposcopy after 1 mild smear after 1 borderline smears HPV test for all women HPV test+ HPV test+ HPV test for women over 30 <30: after 1 mild smear 30: HPV test+ <30: after 3 borderline smears 30: HPV test+ HPV test for women over 40 <40: after 1 mild smear 40: HPV test+ <40: after 3 borderline smears 40: HPV test+ NA NA NA A B C A B C A B C NA NA NA A B C A B C A B C 1) For the management of HPV- and in age-restricted group, A, B and C refer to the referral policies for colposcopy of cytology-based strategies. 16

In all strategies, women having moderate or severe dyskaryosis were assumed to be referred to colposcopy immediately, and women showing negative smear results were recalled for routine screening at age-related interval. Women with an inadequate slide were rescreened after 6 months and referred for colposcopy after three consecutive inadequate samples. 3.3. Natural history model Natural history model consists of 9 states - clear, HPV infection, CIN1, CIN2/3, 4 invasive cancer stages from I to IV and death (See Figure 7). <Figure 7> Natural history model HPV infection state was defined as the state having the detectable HPV virus but do not have histologically defined symptoms of preinvasive cancer. Preinvasive cancer was graded into CIN1 and CIN2/3. Women were allowed to progress and regress between these states in every 6 months. Invasive cancer was graded by the FIGO classification of stage I IV. Once women diagnosed with invasive cancer (stage I), they were assumed to enter tunnel states for a period of 5 years. They can either progress to higher an invasive cancer stage or continue at that stage during the time or die from cervical cancer as determined by survival probability of each stage. 17

To simplify the model, following assumptions were made. 1) Precancer states begin with HPV infection. Thus HPV infection is a prerequisite for precancer states. 2) The incidence of HPV infection, and the probabilities of progression and regression of HPV infection are age-dependent, but other transition probabilities are not agedependent. 3) Precancerous states are allowed to regress to the previous states, ie CIN1, HPV+ or clear. Once invasive cancer develops, however, the stage does not regress. 4) Colposcopy and biopsy is the diagnostic gold standard for confirming the presence and grade of CIN and cervical cancer. 5) Attendance rates of initial screening, treatment and follow-up were assumed to be the same for every cycle and not dependent on the previous attendance. 6) Women diagnosed with negative or HPV infection states without the evidence of CIN are assumed to be discharged. 7) Women diagnosed with CIN1 are recalled for repeating colposcopy at 6-monthly interval and a certain percentage of them are treated at 24 months. 8) Cancer-related death occurs in 5 years after diagnosis. 9) Invasive cancer is detected by symptoms or by screening, and all symptomatic cancers in survivors are treated. 10) After successful treatment, women are transferred to the clear state. 3.4. Model data 3.4.1. Transition probability The model included three major transition matrixes 1) the probability of routine smear test results given previous result, 2) colposcopy results by smear test result and 3) transitions of natural history model based on colposcopy results. Transition data between cytological results of routine screening were obtained from the ScHARR colposcopy model (Eggington et al, 2006). This study did not update the transition probabilities of the ScHARR colposcopy model as these were constructed as a validated transition matrix. The ScHARR model defined seven cytological states negative, inadequate, borderline changes, mild dyskaryosis, moderate dyskaryosis, severe 18

dyskaryosis, suspected invasive cancer and glandular neoplasia. Overall transition probabilities from a negative and inadequate result were based on the KC61 returns which were data on screening samples examined by pathology laboratories in 2003-04. These were amended as state transitions over 6 months in the ScHARR model. Transition probabilities from a borderline and mild came from other studies conducted by Rawal et al (2003) and Woodward et al (1999). No transitions from moderate and severe smear results were applied because these patients were referred to colposcopy immediately. Cytological states are linked to the histologically confirmed disease states through colposcopy and biopsy. In this study, the matrix of colposcopy findings by the severity of referral samples takes a role in linking cytological and histological states. The matrix also relied on the ScHARR model. The ScHARR model used KC61 Returns in 2002-03 to estimate the proportion of colposcopy results from inadequate, borderline, moderate and severe dyskaryosis. These data were combined with the proportion of colposcopy results where referral smear is negative. The colposcopy results of mild smear were differentiated between the results after first and second mild smear. While the transition probabilities in the ScHARR model was mainly based on the Sherlaw-Johnson s model (Sherlaw-Johnson et al, 1994), this study choose the transition probabilities of Myers et al s study (Myers et al, 2000). There were several reasons for the decision. Because the Sherlaw-Johnson s model did not incorporate HPV state in their natural history model, the ScHARR model added HPV state into the underlying pathway of cervical cancer, and then adjusted the transition probabilities with age-specific HPV incidence rates and progression and regression rates of HPV infection of the Myers s model (Eggington et al, 2006). Moreover, the Sherlaw-Johnson s model did not assume the regression of CIN2 or 3 to the less severe states. Considering there has been good evidence of regression of CIN2/3 to CIN1 or clear, and most recent modelling studies of HPV test allow the regression of CIN2/3 to CIN1 or clear, the regression was needed to be allowed in this study. The Myers s model has a merit of considering these limitations of the Sherlaw- Johnson s model integrating evidence from an extensive systematic review. It allows the regression of CIN2/3 to CIN1 or well and provides age-specific prevalence, incidence, progression and regression probabilities of HPV and CIN states. Therefore, this study decided to adapt all transition probabilities of the Myers s model to keep consistency. The Bethesda system terminologies were converted to CIN classification as follow; LSIL is equivalent to CIN1 and HSIL is equivalent to CIN2/3. To model the invasive cancer states, this study also adapted Myers et al s estimates of progression between cervical cancer 19

stages and annual survival rates for 5 years by cancer stage I-IV. Age-specific all-cause mortality was estimated from UK life table for the years 2008-09 for females in England and Wales (ONS, 2009). Attendance rates of initial colposcopy, treatment and follow-up were used as the proxies of compliance figures. It was assumed that appointments cancelled by the service or by the patient when prior notice was given would be rescheduled. Thus, the attendance rates including the cancelled appointments were 92% at initial visit, 95.10% at treatment and 85.40% at follow-up in 2008-09 (NHS, 2009). 3.4.2. Performance of HPV test Performance of HPV test was defined in terms of sensitivity and specificity to detect CIN2 or worse. Sensitivity is the proportion of true positives and specificity is the proportion of true negatives of CIN2 or worse. The aim of this work is to assess the costeffectiveness of HPV testing based upon the TOMBOLA trial. Thus, these data were based on the TOMBOLA trial which presented the results of cross-sectional analysis of a single HPV testing in detecting CIN2 or worse by smear state and age group (Cotton et al, 2010). The TOMBOLA trial is a multicentre randomized clinical trial comparing different methods of management for women with low-grade cervical abnormalities within the NHS cervical screening programme in the UK. The trial invited women aged 20-59 and having a cervical smear that showed borderline or mild changes in the period October 1999 to October 2002, and followed up for 3 years. To determine whether repeating cytological tests or immediate colposcopy is more effective, the women were randomly allocated to one of the arms. Cytological surveillance involves repeat smears at six-monthly intervals in primary care. If a woman had three consecutive normal smears, she was returned to regular screening every three years. If any of results showed moderate dyskaryosis or worse, or three consecutive inadequate smears, the woman were arranged to have a colposcopy examination at an NHS clinic. Otherwise, she remained on 6-monthly recall for smears. In the immediate colposcopy arm, the woman did not need any further tests or treatment, when the cervix looked normal. If the doctor saw an abnormal area on the cervix, the women were invited to participate in a second part of the TOMBOLA trial. The second part of the TOMBOLA aimed to determine whether immediate treatment (using large loop excision of the transformation zone; LLETZ) or biopsies and recall if necessary for LLETZ are more effective and efficient. 20

To evaluate the contribution of HPV testing to these policies, samples for HPV testing were taken from women at recruitment and at the end of the 3-year follow-up. The woman's HPV status was determined using the GP5+/6+ consensus primers. Women were classified as high-risk HPV positive if their sample had an optical density (OD) reading of 3 times greater than that of the OD of the negative controls included in each assay. Other women were classified as high-risk HPV negative. Samples without recordable levels of DNA were classified as inadequate. It was less than 1% of the recruitment HPV samples. 3.4.3. Costs The unit costs of conventional cytology (Pap smear), LBC and HPV test were obtained from HPV/LBC Cervical Screening Pilot Studies (Moss et al, 2004). The unit costs include 1) the costs incurred in primary care, including taking smears and collecting HPV samples 2) equipment, consumables and labour involved in the laboratory to process slides or HPV tests. The adjusted unit costs of conventional cytology, LBC and HPV test were 30.70, 33.43, and 26.66 in 2008-09, respectively. The unit costs of diagnostic and therapeutic colposcopy examinations were taken from an economic evaluation alongside the TOMBOLA trial (TOMBOLA Group, 2009). The costs were based on the NHS reference costs of outpatient colposcopy in 2003. As the reference costs included all costs of colposcopic examination, biopsy, excision, or ablation if carried out, these were adjusted properly on the basis of the proportion of colposcopies followed by each of these other procedures in the UK programme and the unit costs of histopathology, excision, and ablation. The adjusted cost of an examination alone was 136.57 and the cost of a combined examination, smear test, and large loop excision were 250.92 in 2008-09. It was assumed that all high-grade lesions and 30.1% of patients with CIN1 are treated. The proportion of women treated with CIN1 was based on the BSCCP/NHSCSP National Colposcopy Questionnaire. Treatment of CIN was either carried out at a single visit with the colposcopy ( see and treat ), or later. According to the BSCCP/NHSCSP National Colposcopy Questionnaire, the proportion of treatment on a see and treat basis was 15.7% for the women treated with low-grade disease and 61.0% for those with high-grade disease. For the unit costs of treatment later, the same unit costs as therapeutic colposcopy were assumed, but it is also assumed to incur additional costs of diagnostic colposcopy. For the costs of invasive cancer treatment following detection, estimates of 5-year 21

treatment costs by invasive cancer stage were applied. These costs were originally derived from Wolstenholme et al. s study which reported detailed cost estimates of treating invasive cancer including costs of diagnosis, treatment and palliative care over a 5-year time horizon in the UK (Moss et al, 2006). Time and travel costs were considered in sensitivity analysis. The TOMBOLA trial provided these unit costs combining the reported duration of an event with the prevailing average wage. All time costs were adjusted by the length of visit. All costs were uplifted to 2008-09 prices using the HCHS (Hospital and community health services) pay and prices inflator. 3.4.4. Utility In order to determine the quality-adjusted health gains associated with each screening strategy, utilities of 0.91 for CIN1 and 0.87 for CIN2/3 were assumed (Insinga et al, 2007). For the utilities of cervical cancer by stage, 0.67-0.48 without treatment and 0.86-0.63 with treatment were assumed (Chuck et al, 2010). Minor utility decrements associated with colposcopy and with a smear test result of borderline abnormality but not referred to colposcopy were assumed to be 0.03 and 0.02, respectively (Karnon et al, 2004). The utility decrement associated with time spent receiving routine cervical cancer screening was assumed to be 0.02 (Insinga et al, 2007). 3.5. Sensitivity analysis 3.5.1. One-way sensitivity analysis One-way sensitivity analyses were conducted to observe the robustness of the costeffectiveness of each strategy in the range of minimum and maximum values of transition probabilities, sensitivity and specificity of HPV test, costs of HPV test and cervical cancer treatment and utilities. The ranges are presented in Table 2. CIN1 time to treatment and patient s time and travel costs were also considered in one-way sensitivity analysis. 3.5.2. Probabilistic sensitivity analysis 22

Probabilistic sensitivity analysis was undertaken to explore the uncertainty surrounding the variables. Beta distribution was assigned to transition probability, sensitivity and specificity and utility; and gamma distribution was assigned to costs. During the simulation of 1,000 times, the specified parameter distributions generated the estimates of cost-effectiveness of each strategy. It reflects the overall uncertainty defined in this model. Comparing the net benefit of each strategy incorporating willingness to pay per QALY in society (Lamda; λ) provides the probability of a strategy being cost-effective compared with another strategy. Cost-effective acceptability curves (CEACs) represent the probability of a strategy being optimal for different values of Lamda. <Table 2> Key parameters and sources Population and screening parameters Total population (female) 100,000 Start age 15 First screen 25 Last screen 64 Screening interval Age-related Coverage <5 years since last test 78.90% Statistical Coverage <3 years since last test 69.30% Bulletin (KC53, 2008-09) Attendance at initial visit 92.00% Statistical Attendance at treatment 95.10% Bulletin Attendance at follow-up 85.40% (KC61, 2008-09) Cytological results If cytology results are inadequate Conventional LBC cytology - Inadequate 18.4% 1.91% Eggington et - Negative 75.92% 91.31% al, 2006 - Borderline 3.18% 3.82% - Mild 1.59% 1.91% - Moderate 0.50% 0.61% - Severe 0.32% 0.38% - Invasive cancer 0.05% 0.06% If cytology results are negative - Inadequate 7.1% 1.91% Eggington et - Negative 85.9% 90.72% al, 2006 - Borderline 4.21% 4.45% - Mild 1.86% 1.96% 23

- Moderate 0.52% 0.55% - Severe 0.34% 0.36% - Invasive cancer 0.05% 0.05% If cytology results are borderline - Inadequate 8.8% 1.91% Eggington et - Negative 54.6% 58.72% al, 2006 - Borderline 24.60% 26.46% - Mild 8.40% 9.03% - Moderate 2.60% 2.80% - Severe 1.00% 1.08% - Invasive cancer 0.00% 0.00% If cytology results are mild - Inadequate 0.4% 1.91% Eggington et - Negative 42.0% 41.36% al, 2006 - Borderline 13.38% 13.18% - Mild 31.60% 31.11% - Moderate 8.18% 8.05% - Severe 4.46% 4.39% - Invasive cancer 0.00% 0.00% Colposcopy results by cytology results proportion of inadequate of Pap test 8.55% Eggington et al, 2006 proportion of inadequate of LBC 1.40% Eggington et al, 2006 If colposcopy results are clear conventional LBC cytology - Negative 90.93% 98.04% Eggington et - Borderline 0.20% 0.22% al, 2006 - Mild 0.24% 0.25% - Moderate 0.05% 0.05% - Severe 0.02% 0.03% If colposcopy results are HPV - Negative 51.30% 55.32% Eggington et - Borderline 24.69% 26.63% al, 2006 - Mild 7.15% 7.71% - Moderate 6.02% 6.49% - Severe 2.29% 2.47% If colposcopy results are CIN1 - Negative 39.77% 30.64% Eggington et - Borderline 20.48% 26.24% al, 2006 - Mild 15.93% 21.33% - Moderate 11.91% 14.81% - Severe 3.36% 5.59% 24

If colposcopy results are CIN2/3 - Negative 44.69% 35.94% Eggington et - Borderline 4.09% 8.56% al, 2006 - Mild 7.87% 12.64% - Moderate 15.51% 18.69% - Severe 19.28% 22.76% If colposcopy results are Invasive cancer - Negative 0.00% 0.00% Eggington et - Borderline 9.74% 8.53% al, 2006 - Mild 14.54% 13.70% - Moderate 12.65% 15.61% - Severe 54.52% 60.76% Improvement in sensitivity of LBC (low grade) Improvement in sensitivity of LBC (high grade) Baseline Min Max 0.0842 0.0421 0.1263 Eggington et al, 2006 0.04 0.02 0.06 Improvement in specificity of LBC 0-0.01 0.01 Transition probability of natural history model Prevalence of HPV infection age 15 0.1 Myers et al, Prevalence of CIN1 age 15 0.01 2000 Age-specific incidence of HPV infection 15 0.050 16 0.050 17 0.060 18 0.075 19 0.085 20 0.075 21 0.060 22 0.050 23 0.050 24-29 0.025 30-49 0.005 >50 0.003 Age specific regression rate of HPV infection (15-24) Age specific regression rate of HPV infection (25-29) Age specific regression rate of HPV infection (30+) Myers et al, 2000 0.33 0.26 0.54 Myers et al, 2000 0.21 0.18 0.26 0.05 0.03 0.07 Progression rate of HPV to CIN1 0.04 0.03 0.06 Proportion of HPV infections progressing to CIN2/3 0.1 0.05 0.5 25

Regression rate of CIN1 to HPV or well (15-34) 0.084 0.074 0.126 Regression rate of CIN1 to HPV or well (35+) 0.042 0.029 0.074 Proportion of CIN1 reverting to well 0.9 0.5 1 Progression rate of CIN1 to CIN2/3 (15-34) 0.0087 0.0087 0.029 Progression rate of CIN1 to CIN2/4 (35+) 0.035 0.029 0.056 Regression rate of CIN2/3 to CIN1 or well 0.035 0.0292 0.056 Proportion of CIN2/3 reverting to well 0.5 0 0.5 Progression rate of CIN2/3 to invasive cancer 0.025 0.018 0.034 Sensitivity and specificity of HPV test Sensitivity of HPV test for BNA aged 20-24 0.797 0.672 0.89 Cotton et al, aged 25-29 0.743 0.567 0.875 2010 aged 30-39 0.667 0.482 0.82 (TOMBOLA) aged 40-59 0.313 0.11 0.587 Sensitivity of HPV test for Mild aged 20-24 0.808 0.703 0.888 Cotton et al, aged 25-29 0.76 0.618 0.869 2010 aged 30-39 0.705 0.574 0.815 (TOMBOLA) aged 40-59 0.647 0.383 0.858 Specificity of HPV test for BNA aged 20-24 0.463 0.393 0.534 Cotton et al, aged 25-29 0.63 0.55 0.704 2010 aged 30-39 0.734 0.678 0.785 (TOMBOLA) aged 40-59 0.865 0.827 0.898 Specificity of HPV test for Mild aged 20-24 0.325 0.251 0.405 Cotton et al, aged 25-29 0.432 0.318 0.553 2010 aged 30-39 0.522 0.427 0.616 (TOMBOLA) aged 40-59 0.636 0.539 0.726 Treatment and effectiveness Treatment success effectiveness of conization 0.9 0.8 1 Eggington et screen detected suitable for conization 0.3 0.1 0.5 al, 2006 effectiveness of hysterectomy 0.85 0.75 0.95 Treatment parameters (low-grade) - Proportion of patients in low-grade 30.10% 2004 NHSCSP - Proportion of patients treated 15.66% / BSCCP on a "see & treat" basis Questionnaire - Proportion of patients treated at a later visit 84.34% Treatment parameters (high-grade) - Proportion of patients treated 60.96% 2004 NHSCSP 26

on a "see & treat" basis - Proportion of patients treated at a later visit 39.04% Utility / BSCCP Questionnaire CIN1 0.91 0.819 1.001 Insing et al, CIN2/3 0.87 0.783 0.957 2007 StageI with treatment 0.86 0.73 0.99 Chuck, 2010 StageII with treatment 0.83 0.68 0.98 StageIII with treatment 0.83 0.68 0.98 StageIV with treatment 0.63 0.47 0.78 StageI without treatment 0.65 0.49 0.81 StageII without treatment 0.67 0.44 0.9 StageIII without treatment 0.56 0.42 0.7 StageIV without treatment 0.48 0.36 0.6 Utility decrement associated with time spending on routine screening Utility decrement associated with colposcopy (not IC) 0.02 0.018 0.022 Insing et al, 2007 0.03 0.027 0.033 Eggington et al, 2006 Utility decrement associated with BNA smear (no colposcopy) 0.02 0.018 0.022 Costs costs of Pap test 30.70 30.44 30.96 NHS pilot costs of HPV test 26.66 13.40 44.88 study, 2004 costs of LBC 33.43 30.57 36.68 costs of treatment colposcopy 250.92 135.00 235.00 TOMBOLA, costs of diagnostic colposcopy 136.57 116.57 156.57 2009 5 year treatment cost of stage I 13920.37 12166.99 15662.05 NHS pilot 5 year treatment cost of stage II 22930.51 18573.07 27291.84 study, 2004 5 year treatment cost of stage III 22779.62 18054.08 27501.26 5 year treatment cost of stage IV 24244.24 10597.02 37888.86 time and travel costs for colposcopy 35.85 TOMBOLA, time and travel costs for smear test 12.07 2009 Discount rate of costs and life 3.50% 27

4. Results 4.1. Model validation The model was validated by comparing the predictions of the prevalence and incidence of HPV, CIN and cervical cancer against the UK and other population-based figures. Figure 8-10 present the model predictions for an unscreened population. The natural history model predicted a peak prevalence for HPV at the age of 20 years and tailed off significantly thereafter in an unscreened population. It was consistent with general epidemiology of HPV infection that reaches a peak in the age group 20-24 years and declines thereafter (Cuzick et al, 2006). The predicted prevalence of CIN1 reached a peak at 25 years of age and CIN2/3 prevalence between the ages of 30-50 at around 3%. The age-specific incidence of invasive cancer was similar to that of women born in the 1980s peaking at 40-50 (Quinn et al, 1999). The age-specific death rate from cervical cancer showed a similar distribution to the lifetime risks of death from invasive cancer for women born in 1932 (Peto et al, 2004). <Figure 8> Estimated age-specific prevalence of HPV and CIN in an unscreened population 28

<Figure 9> Estimated age-specific incidence of invasive cancer in an unscreened population (per 100,000) <Figure 10> Estimated age-specific cervical cancer death rate in an unscreened population (per 100,000) 4.2. Base case Table 3 reports the cost-effectiveness results of the base case in comparison with the current recommendation of cervical cancer screening (Strategy A). QALY and costs are the means of your PSA runs. As presented in Table 3, Strategy E1 and E2 were the most costeffective strategies dominating Strategy A. Other strategies including HPV testing slightly saved QALYs, while increasing total lifetime costs compared with Strategy A. Strategy C was dominated by Strategy A. 29

<Table 3> Baseline estimates of discounted costs and effectiveness compared with Strategy A Strategy QALY Costs ( ) Comparison with A A 25.918 269.60 Inc QALY Inc Costs ( ) CE ratio B 25.915 261.03-0.003-8.58 2729.26 C 25.905 298.40-0.013 28.80 dominated D1 25.921 284.23 0.002 14.62 5877.37 D2 25.929 280.89 0.011 11.29 1061.74 D3 25.925 280.34 0.006 10.73 1707.84 E1 25.929 254.44 0.011-15.16 dominate E2 25.931 262.34 0.013-7.26 dominate E3 25.927 278.30 0.009 8.70 1015.60 F1 25.929 290.45 0.011 20.85 1907.55 F2 25.919 298.38 0.000 28.78 60801.99 F3 25.922 306.25 0.004 36.64 9231.56 Table 4 and 5 present the detailed outcomes relating to costs and effectiveness of each strategy. HPV testing increased the incidence of invasive cancer, and the detection rates of strategies including HPV testing were approximately 10% lower than cytology-based strategies. Thus, the number of cancer cases was higher in the strategies including HPV testing, and it led to more deaths from cancer in the strategies. The cause could be found on the low sensitivity of HPV test in the TOMBOLA trial which was referred in this study. According to Cuzick et al. s study which reviewed the European and North American studies on HPV testing in primary cervical cancer screening (Cuzick et al, 2006), HPV testing was consistently very sensitive in all studies, with the sensitivity for CIN2 and CIN3 both being 96.1% overall (94.2 97.4% and 93.6 97.6%, respectively). Also, the sensitivity was unaffected by age (Cuzick et al, 2006). In the TOMBOLA trial, however, the overall sensitivity of HPV testing in detecting CIN2 or worse was 69.9% for the women indicating borderline and 75.2% for the women indicating mild. In particular, the sensitivity was much lower for women over the age of 40, showing only 31.1% (95% CI 11.0-58.7%) for women with borderline change and 64.7% (95% CI 38.3-85.8) for women with mild dyskaryosis. Considering sensitivity of cytology to detect CIN2/3 or worse has been reported in the range between 40 and 80% (Cuzick et al, 2000), improvement in sensitivity by adding a single HPV test may not so great. In spite of lower life-year gains in the strategies including HPV testing, increase in QALYs from HPV testing is explained by the higher utility decrements in cytology-based strategies. These utility decrements are associated with the time spent on routine screening, 30

undergoing a colposcopy and anxiety over the borderline abnormality until they receive a colposcopy in the cytological screening strategies. Thus, the reduction in the number of smears and colposcopy via HPV testing contributed to a decrease in the utility decrements in strategies including HPV testing. <Table 4> Long-term effectiveness by strategy Discounted Life Years (per person) Utility decrement (per person) Rates of detection of CIN2/3 & IC Effectiveness Cancer case Incidence of cancer Death from cancer % deaths from cancer A 26.0738 0.102 43.39% 2,896 0.00111 547 0.55% B 26.0723 0.103 40.16% 3,025 0.00116 574 0.57% C 26.0766 0.108 49.20% 2,585 0.00099 487 0.49% D1 26.073 0.097 35.85% 3,052 0.00117 576 0.58% D2 26.0728 0.098 35.69% 3,053 0.00117 577 0.58% D3 26.0726 0.1 35.16% 3,047 0.00117 578 0.58% E1 26.073 0.097 36.13% 3,053 0.00117 576 0.58% E2 26.0725 0.098 35.50% 3,066 0.00118 579 0.58% E3 26.0727 0.1 35.10% 3,030 0.00116 574 0.57% F1 26.073 0.097 35.85% 3,052 0.00117 576 0.58% F2 26.0736 0.1 36.71% 3,017 0.00116 570 0.57% F3 26.0729 0.102 36.10% 3,047 0.00117 579 0.58% <Table 5> Long-term costs and resource use by strategy Costs Resources use Routine smears Diagnostic colposcopy See and treat Treat later HPV test Cancer treatment Number of smears Number of colposcopy A 129.89 27.59 2.77 7.58 0 118.29 959,387 45,842 B 131.76 23.84 2.56 6.56 0 124.35 972,591 40,201 C 132.39 48.84 3.19 10.91 0 101.87 976,973 78,460 D1 125.15 26.05 2.88 6.71 8.65 126.03 925,257 43,393 D2 126.73 26.29 2.87 6.79 8.64 125.6 933,069 43,652 D3 128.29 27.94 2.77 7.12 8.6 124.1 943,000 45,750 E1 125.15 17.28 2.45 6.7 8.81 126.11 925,257 28,495 E2 127.4 19.12 2.5 6.44 8.81 125.28 936,359 30,719 E3 129.57 22.85 2.53 6.96 8.73 121.72 950,135 36,004 F1 125.15 31 3.13 6.71 8.65 126.03 925,257 50,786 F2 127.69 36.43 3.2 8.24 8.53 121.91 937,853 56,434 F3 130.07 39.07 3.1 8.39 8.51 122.2 953,094 59,979 All strategies using HPV testing greatly reduced the number of routine smears. The rationale behind it is that women with borderline and mild smear results were more often 31

referred immediately to colposcopy rather than waiting for two or three consecutive mild and borderline smear results. Moreover, for the women with HPV negative result, they were not required to wait for three consecutive negative smear results until returning to routine screening test. Strategies including HPV testing also reduce the number of colposcopy, while HPV testing has shown an increase in the number of colposcopy in previous studies. It means that the proportion of women with HPV negative was higher than the previous studies, so more women were returned to routine smear screening rather than referred for colposcopy. In this study, only 30% of the women receiving HPV testing had HPV negative and were referred for colposcopy. The percentage was particularly low for women over the age of 40. It is also because of the lower sensitivity of HPV testing in the TOMBOLA trial. However, the impact of HPV testing on the cost of diagnostic or treatment colposcopy was varied depending on the employed strategy. In particular, costs of treatment colposcopy were slightly higher in strategies including HPV test. Higher costs of treatment colposcopy mean a greater number of CIN that requires any type of treatment. As can be seen from Figures 11 and 12, the number of CIN was higher in HPV testing strategies. <Figure 11> The number of CIN1 32

<Figure 12> The number of CIN2/3 Previous studies showed that the more the treatment colposcopy, the less the diagnostic colposcopy. However, strategies including a HPV testing did not reduce the number of diagnostic colposcopy in this study due to more colposcopies during the followup period of CIN1. This study assumed that women with CIN1 repeat colposcopy in every 6 months, and a certain percentage of them are treated at 24 months. As mentioned above, more CIN1 were observed in the strategies including HPV test, so it caused an increase in the number of diagnostic colposcopy. However, the increase in total costs was mainly driven by the costs of invasive cancer treatment. Figure 13 presents that HPV testing developed the more invasive cancer Stage I. The gap in the number of invasive cancer Stage I between cytology-based strategies and HPV testing strategies were much larger after the age of 40. The reason also can be found on the low sensitivity of HPV testing. The number of Stage I had a direct impact on the number of invasive cancer Stages II-IV and caused a rise in the cost of cancer treatment. As this study applied 5-year treatment costs following detection of invasive cancer, the impact on total costs is considerable. 33

<Figure 13> The number of invasive cancer stage I Table 6 shows the incremental cost-effectiveness of a single HPV testing by age restriction compared with a cytology-based strategy which were added on. For example, D1, E1 and F1 are the strategies adding a single HPV test for all women, aged over 30 or 40 onto Strategy A. In all comparisons, strategies providing a single HPV test for women over 30 (E) were the most cost-effective strategies and those over 40 (F) were the least cost-effective strategies. ICERs of Strategy E1-E2 were 91.96 and 90.26 compared with Strategy A and B, respectively, and Strategy E3 dominated Strategy C. However, Strategy F1-F3 were consistently dominated in the comparisons with Strategy A-C. <Table 6> Cost-effectiveness compared by the type of cytology-based strategy QALY Costs ICER Compared with strategy A E1 25.9413 286.49 91.96 D1 25.9412 295.48 Dominated by E1 F1 25.9412 300.67 Dominated by E1 A 25.9372 286.12 Compared with strategy B D2 25.9402 296.91 15220.85 E2 25.9397 289.55 90.26 F2 25.9396 305.99 Dominated by E2 B 25.9345 289.08 Compared with strategy C D3 25.9383 298.82 24886.25 E3 25.9381 292.36 - F3 25.9365 311.35 Dominated by D3 C 25.9347 297.20 Dominated by E3 34