Human papillomavirus testing with Pap triage for cervical cancer prevention in Canada: a cost-effectiveness analysis

We used a Markov model that simulates the natural history of cervical cancer in a theoretical cohort of women to estimate lifetime costs and life expectancy of different screening strategies. This Markov model has previously been described in detail [10‐12]. Briefly, a population advances from one health state to another based on predefined probabilities. These health states are chosen to reflect the progression from healthy and disease-free through pre-cancer to cancer and death. Each year, women can either remain in the same state, progress to a more advanced disease state, or regress to a less severe disease state. Each year, women are also at risk of dying from causes other than cervical cancer or having a hysterectomy for a non-cancerous uterine condition.

The model was calibrated to obtain an age-specific HPV prevalence curve based on large HPV epidemiology studies conducted in different Canadian settings; this was accomplished by varying the age-specific HPV incidence rates [13‐16]. Progression and regression rates between the different states were from the original model by Myers et al. [10]. Benign hysterectomy rate was estimated from the Canadian Community Health Survey [17]. Death from non-cervical cancer causes was estimated from the Canadian Incidence and Mortality data set [18]. Cancer progression rates and the probability of symptoms were from Myers et al. [10]. Stage-specific survival probabilities were based on United States. Surveillance Epidemiology and End Results data from the National Cancer Institute [19]. The same underlying natural history was assumed for Canada as a whole and for each province.

We compared 27 different screening strategies as summarized in Table 1. Strategies differed by the age at which to begin screening (either age 18 or age 25) and whether the most widely recommended strategy for cervical cancer screening in Canada was conducted prior to age 25 [20]. This baseline recommended strategy for Canada (screening every year from age 18 until 21 and then every three years afterwards until age 70 with conventional Pap tests) is hereinafter designated the 'Miller' strategy. The choice of age 25 was based on the fact that HPV testing prior to that age suffers from an unacceptably low specificity (to detect high grade cervical lesions) as a consequence of the high HPV prevalence following the first few years after the onset of sexual activity. This age may be an appropriate age to implement HPV DNA testing because HPV infections among such women are more likely to reflect underlying lesions [14]. In addition to single test (HPV or Pap) strategies, we examined a combined test strategy (HPV and Pap), in which women who were HPV positive or who had smears with findings of atypical squamous cells of undetermined significance (ASC-US) or more severe (≥ ASC-US) abnormalities were referred to colposcopy. We also examined two triage strategies: 1) HPV followed by Pap testing for HPV positive women and 2) Pap followed by HPV testing for women with ≥ ASC-US Pap test results. Adherence to screening recommendations by age was varied based on province-specific estimates provided by the 1998 Surveillance Report on Cervical Cancer Screening in Canada [2]. All women who received an initial screening test were also assumed to receive the triage test if a triage test was included as part of the strategy. Women with histologically confirmed CIN of grade 1 (CIN 1) were assumed to be followed with repeat Pap tests and to be treated if they had another abnormal test result (≥ ASC-US). Women with biopsy-confirmed CIN of grades 2 or 3 (CIN 2+) were assumed to receive loop electrosurgical excision procedure (LEEP). Adherence to follow-up and treatment was assumed to be 100%. Women with cancer were assumed to receive stage-specific treatment. Selected parameters used in the model are presented in Table 2.

Estimates of sensitivity and specificity were from a recently published randomized trial comparing HPV and conventional Pap tests [8, 9]. Since the trial enrolled women aged 30+, we used estimates from another study that corrected for verification bias and provided estimates of test accuracy stratified by age <30 or 30+ to determine the sensitivity and specificity in women aged <30 years [21] Given a move towards considering CIN 2-3 as the disease state that is more likely to be a cancer precursor, and CIN 1 as a marker of HPV infection, we examined two different thresholds for disease: CIN 1+ and CIN 2+ [22]. Conventional Pap smears were assumed for the base case analysis; in sensitivity analyses we assumed that liquid-based Paps would have approximately the same sensitivity and lower specificity compared to conventional Paps, based on the results of a randomized controlled trial [23]. We did not model the impact of either type of Pap smear on inadequate results. To address concerns that the estimates of sensitivity for CIN 2+ for both Pap and HPV reported in CCCaST may be lower than those based on actual practice patterns in Canada, in sensitivity analyses, we examined the impact of using more commonly accepted estimates of sensitivity of Pap and HPV tests for detection of CIN 2+: 55.6% and 95% respectively [24, 25].

The costs (Table 2) for Pap and HPV tests including laboratory costs were obtained from provincial fee schedules and communications with selected laboratories [26‐28]. Professional costs for diagnostic procedures were also obtained from provincial fee schedules, while technical costs were based on data from the Ontario Case Costing Initiative website [26‐29]. These were then adjusted for Alberta and Newfoundland by using provincial costs per weighted case (CPWC) as reported by the Canadian Institute of Health Information (CIHI) [30]. Costs for cancer care by stage were based on figures from Brisson et al. (2007), and were also adjusted for each province using CIHI's CPWCs [31]. Costs from Ontario were assumed to reflect those for the country as a whole. All estimates were measured in 2006 Canadian dollars. The triage test strategies were assumed to be conducted at two separate visits. Both tests were assumed to be collected at a single visit for the combined test procedure.

For the model, previously published utility estimates for a false positive Pap screening test result and cancer (Stage I, Stage II-IV) were used (Table 2); these estimates were assumed to be constant by age [32]. The false-positive test result was assumed to be applicable to both Pap and HPV tests. Since these estimates were derived from a small study of college-aged women in the U.S., we examined the impact of quality-adjusted life years on cost-effectiveness results in sensitivity analyses only.

The analysis was conducted from a health-system perspective. We calculated incremental cost-effectiveness ratios (ICERs) in which the average lifetime costs and life-expectancy or quality-adjusted life-expectancy of a strategy were compared with an adjacent strategy. Strategies that were more costly and less effective or less costly and less cost-effective than an adjacent strategy were considered to be dominated. We adjusted future costs and life expectancy to current values by discounting them at 3% annually. In sensitivity analyses, we calculated cost per quality-adjusted life year. In addition to one-way sensitivity analyses, in which key parameters, including test accuracy estimates, adherence to screening and costs were varied, a probabilistic sensitivity analysis was also performed on strategies consistently identified as cost-effective to explore the impact of the joint uncertainty surrounding model parameters. This analysis was undertaken by assigning probability distributions to key model parameters (costs and estimates of sensitivity and specificity) and subsequently propagating this uncertainty through the model using Monte Carlo sampling techniques to produce information on the likelihood that each intervention produces the greatest amount of net benefit at different willingness to pay thresholds. The results of these simulations are presented as cost-effectiveness acceptability curves (CEACs).