Human Papillomavirus (HPV) DNA testing has been accepted as an adjunct to cytology for primary cervical screening in women over 30 years of age.1, 2 The combination of HPV testing and cervical cytology is more sensitive in detecting cervical intraepithelial neoplasia (CIN) lesions with risk for progression to invasive cancer than either method alone, with extremely high negative predictive values.3, 4, 5 Several studies have evaluated the economic impact of cervical screening with and without HPV testing, but have reported variable results.6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22 Possible reasons for discrepancies include different modeling methodologies, use of surrogate endpoints and use of input data based on clinical studies that may not have been representative of the population as a whole. However, most modeling studies have consistently found similar reductions of the cumulative lifetime risk of invasive cervical cancer as a result of cytology and HPV DNA screening.6, 11, 12, 19, 23, 24, 25
State transition models are used in more than 85% of all cost-effectiveness analyses and several cervical cancer state transition models have been reported previously.26, 27 The long-term effects and cost of cytology and HPV screening cannot be captured in clinical trials, where all studies have used surrogate endpoints, such as CIN grade 2 or 3 (CIN2-3).28 Using surrogate endpoints alone for comparing policies is problematic29 and needs to be complemented with state transition modeling efforts on clinically important outcomes such as life expectancy and life-time resource utilization. The present study uses a state transition model based on actual data from a population-based screening program to simulate the cost-effectiveness of different conceivable strategies in the setting of an organized screening program.
A computer-based probabilistic Markov model was constructed and analyzed with second order Monte Carlo simulations and was run from 32 years of age in 3–5 year cycles. The cycle length used is the same as the length of the screening intervals in the current program (ages 32–50) and was 5 years also after screening was finished (from age 50 to death). Women in the cohorts could remain in the same health state or move between states, as a result of being screened, tested, having treatment for CIN2-3 or dying from cervical cancer or other causes (Fig. 1).
Strategy 2: HPV DNA testing as an add-on to strategy 1 at age 32.
Strategy 3: Combined HPV DNA testing and cervical cytology screening 3 times during life-time (only at ages 32, 41 and 50).
Strategy 4: No screening.
Women in Sweden are invited to screening from age 23 to 50 at 3 year intervals and from 50 to 60 at 5 year intervals, according to the national guidelines. Cytology samples are usually obtained by trained midwives.30 The comparator with screening start at age 32 was chosen because HPV DNA screening before age 30 is not accepted as a screening test2 and because start of conventional cytology screening at age 30 is minimum practice according to WHO guidelines. The time horizon of the analysis was from age 32 until end of life, which was considered the required length to capture streams of health and economic outcomes. The perspective of the analysis was that of the national health service.
Primary data on HPV testing were obtained from a prospective, randomized population-based study of combined cytology and HPV testing that was nested entirely within the routine population-based organized cervical cancer screening program in Sweden.31, 32 The trial targeted women at ages 32–38 and found that PCR-based HPV DNA testing for identification of type-specific persistent infection with oncogenic HPV types with subsequent referral of women with HPV persistence resulted in an increased detection rate of CIN2-3.31, 32
Primary data on conventional cytology and cervical cancer were obtained from comprehensive population-based databases held by the National Board of Health and Welfare in Sweden and from population-based regional screening registries. The regional screening registers contain data on all smears taken (both in organized screening and opportunistic testing) and have been found to be comprehensive and accurate.33 For the present study, regional screening registers together covering about half of Sweden (including 6 million cytology samples, 33,348 cases of cancer in situ and 3,751 cases of invasive cervical cancer) could be accessed. Probability estimates for attendance rates were derived from national surveys. The prevalence and predictive values of cytology- and HPV DNA-test for biopsy-confirmed CIN2-3 were derived from results from a population-based study that included random colposcopies and blind biopsies to avoid verification bias.32 Probabilities of invasive cervical cancer and mortality rates for cervical cancer and other causes were derived from the Swedish Cancer Registry34, 35 (Table I). The background risk for cervical cancer in the absence of screening was assumed to be the same as the cumulative life-time risk for cervical cancer before screening started in 1965, namely at 2%.34, 35
Table I. Input Probability Values
Swedish National Cytology Registry, data excerpt.
Threshold for positive CYT: ASCUS/ LSIL⇒.
Threshold for HPV positive: persistent, type specific HPV infection.
Assuming that CIN2/3 detected by HPV screening has the same probability to progress to invasive cancer as CIN2/3 detected by cytology.
Life-time risk for invasive cervical cancer in Sweden before screening started in 1965 (Cancer Registry of Sweden statistics), assuming that background risk has not changed over time.
Survival after 3 years (between 32 and 50 years of age) and survival after 5 years (from 50 years and onwards). The same modelling cycles for survival were used in all strategies, i.e., also in the strategy with extended screening interval.
Table II. Modeling Assumptions Regarding Structure and Parameters of the Model
Parameters such as cure rates, treatment methods and treatment compliance are used only for estimating costs. Progression probabilities in the transition state model used only the actual risks of disease after positive and negative tests.
Modeling assumptions regarding the structure of the model:
•The attendance rate (75%) was assumed to be the same for every cycle and not dependent on previous attendance.
•The cycle length of the Swedish organised program was used (every 3 years until age 50 and then every five years until age 60).
•Nonattendees for HPV-DNA testing at age 32 continue in the cytology cycles.
Modeling assumptions regarding the parameters of the model:
•Cure rate for treatment of biopsy-confirmed CIN2-3 was 90%.1
•The proportion of organized and opportunistic screening was set at 50% each.
•Biopsy confirmed CIN1 was not treated, but followed-up after three years with cervical cytology testing.1
•Perfect (100%) compliance with the follow-up after positive test results and after treatment was assumed.1
Outcomes are presented as average and incremental costs and life-years gained or lost over the lifetime of women aged 32 and older and who were exposed to each screening option, although not necessarily attending.
Future costs and benefits were discounted at a 3% and 5% annual rate along with the consequences of no discounting. Only direct medical costs for screening, interventions, treatment and follow-up were included as this reflects the true costs of providing the services. Cost estimates were based on reimbursed charges and on resource use-based microcosting. Resource based microcosting is considered as the most reliable method for cost estimation.38 Cost data were provided by the health care providers (county councils) and were standardized to US Dollars at 2005 conversion rates. Sensitivity of the model to changes in cost data was investigated by assigning a gamma distribution to the data, varying the input from −10% to +10% of the actual cost.
All modeling and simulations were performed using a commercially available decision analysis program (DATA TM version 3.5; Decision analysis, TreAge professional software for Windows, Williamstone, MA). The complete structure of the model used is available on request.
With a time horizon from age 32 until end of life and assuming that the background risk for invasive cervical cancer without screening was 2%, the undiscounted mean life expectancy without screening (strategy 4) was 50.28 years of life remaining after age 32. Undiscounted mean life expectancy was 51.0 years with cytology screening (strategy 1), 51.0 years with addition of HPV DNA testing (strategy 2) and 51.04 years with combined cytology and HPV DNA screening at age 32, 41 and 50 (Strategy 3) (Table IV).
Mean costs per woman during remaining lifetime were after discounting at a 3% annual rate 245 US Dollars with cytology screening at ages 32, 35, 38, 41, 44, 47, 50, 55 and 60 (strategy 1), 284 US Dollars with addition of testing for HPV-DNA persistence (strategy 2) and 210 US Dollars with combined cytology and testing for HPV-DNA persistence at age 32, 41 and 50 (strategy 3) (Table III). Without screening, mean costs per woman during remaining lifetime after discounting at a 3% annual rate were 523 US Dollars, based on a 2% lifetime risk for invasive cervical cancer without screening (strategy 4) (Table IV).
Table III. Direct Medical Mean Costs, 2005 US Dollars
Cytology and HPV test combined
Colposcopy and biopsy
Followup, doctor visits CIN2-3 year 1–3
Followup, midwife visit CIN2-3 year 4–6 and 7–9
Followup, doctors visit CIN2-3 year 4–6 and 7–9
Locally invasive cancer, stage 1A
Invasive cancer, stage 1B or more
The variables attendance rate, probability of a positive test result, the predicitive values of the screening tests for CIN2-3 and invasive cervical cancer, survival of invasive cervical cancer with and without screening were varied simultaneously over a plausible range according to a predefined (beta) distribution. A gamma distribution was used for cost data. Effects and costs results of the 4 screening strategies were also analyzed undiscounted and with a 5% annual discount rate (Table IV).
Table IV. Remaining Life Years and Remaining Lifetime Health Care Costs Per Woman for Four Cervical Cancer Screening Strategies, Undiscounted and with 3% and 5% Annual Discount Rates [US Dollars (2005)]
Strategy 1: cytology screening at ages: 32, 35, 38, 41, 44, 47, 50, 55, 60. Strategy 2: HPV DNA test at age 32 in combination with strategy 1. Strategy 3: combined cytology and HPV DNA testing at ages: 32, 41 and 50. Strategy 4: no screening.
In comparison with cytology screening (strategy 1), the addition of testing for HPV DNA persistence (strategy 2) was dominated by strategy 1 (Table IV). Nevertheless, combined cytology and HPV DNA screening with extension of the screening interval to only 3 screenings during lifetime (at age 32, 41 and 50) (strategy 3) was dominating the conventional cytology screening strategy (Table IV).
Our analysis of 4 different cervical screening strategies for cost-effectiveness advances current knowledge as we used a probabilistic state transition model that was based on real-life population-based input data.
Because randomized trials with cancer endpoints would not be practical or ethical, modeling is probably the only way to analyze cervical screening using clinically important outcomes such as life expectancy and costs. Modeling is also fast, cheap, imposes no ethical problems and is not biased because of losses to follow-up.27, 28, 39 A model should be kept as simple and as transparent as possible.40, 41, 42 With respect to model structure, our model is more simple and depends on fewer inputs than previously reported models.6 The challenge of model construction lies in obtaining high quality primary data representative for the population as a whole. Inaccurate results may follow if too many minor variables are included, especially if not based on firm grounds. Our primary data were derived from a prospective, population-based randomized clinical trial nested entirely within the routine population-based screening as well as data from mandatory and comprehensive cancer registries and cytology registries.32 Verification bias could be eliminated using predictive values as parameters, which are based not only on sensitivity and specificity of the diagnostic tests but also on knowledge of the true prevalence of CIN in the population.43 Colposcopy was performed on a population-based, random sample of negative screened women with the investigators blinded for test results. In all those women, a blind cervical biopsy was taken if colposcopy was negative.32 All cytological data in the population could be individually linked to the national population registry and to the cancer registry of Sweden to obtain real-life population-based transition probabilities based on actual data of whole populations followed over a long period of time.36, 44 Thus, the only assumptions required for the model were: (i) that high grade CIN has the same biologic behavior regardless of whether it was detected by HPV screening or cytology screening and (ii) the underlying background life-time risk for invasive cervical cancer in the absence of screening is the same as it was in 1965 when the screening program started (2% according to official registry statistics). Regarding the first assumption, high grade CIN is a histopathologically defined entity and we therefore consider it unlikely that different detection methods would detect biologically different entities. Regarding background risk, there is indeed some evidence that HPV infections (and thus cervical cancer background risk) may have increased since 1965. However, this would tend to cause a conservative bias (i.e., benefits of screening would be somewhat underestimated).
The input value that was most problematic was the cervical cancer survival rate, as this rate has a biphasic pattern (with substantial mortality in the first years after diagnosis but subsequently much lower disease-specific mortality) and since it was not possible to construct a probabilistic model accounting for biphasically changing survival rates. The input value used (80% survival per cycle) fits well to the disease-specific survival during 2 modeling cycles (a modeling cycle was 3 years during ages 32–50, subsequently 5 years) but overestimates mortality after that. As a result, our estimates of life years gained represent an overestimate compared to models that are able to accommodate the actual cervical cancer survival rates during a whole life span. However, as the same survival rates with the same intervals were applied to all screening strategies compared, our tendency to overestimate life expectancy gains does not affect the relative comparison between the 4 different screening strategies evaluated.
We used probabilistic multivariate sensitivity analysis to address the impact of parameter uncertainty. The model is run 100.000 times, each time varying the input data according to a beta distribution. When one strategy is found to be better every time, in spite of the fact that input data is varied, this addresses the same issue as a sensitivity analysis is addressing in a deterministic model. Today, the use of a probabilistic multivariate sensitivity analysis is the method recommended by a number of good-practice guidelines and regulatory agencies,45 as uncertainties in the input data need to be propagated through the model in order to provide a more realistic representation of uncertainty in the models results.
The results from our study are most closely applicable to countries with nationally organized cervical cytology screening from age 30 to 60 with 3–5 year intervals,9 which is similar to strategy 1 in our model. Indeed, some cost-effectiveness studies that have used a similar setting have found similar results. Thus, a cancer registry-based study from the Nordic countries found that cytology screening not only prevents 80–91% of death from cervical cancers but also saved as much as 11,700 US Dollars per cancer prevented.9 Our results on preventive effect are also quite similar to a Dutch modeling study that found combined cytology and HPV DNA screening every 10 years from age 30 to 60 to reduce cervical cancer mortality with 91%, whereas cytology screening every 3 years from age 30 to 60 yielded a mortality reduction of 79%.46 By contrast, studies using liquid based cytology and HPV DNA testing have reported costs from 121 USD to 184,000 USD per life year saved or per quality-adjusted life-year (QALY).11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22 Possible reasons for discrepancies include the fact that we did not include screening of young women in our model, the setting with organized screening, the setting with midwives (who have lower salaries than doctors) for taking smears, population-based real-life input data and analysis using a probabilistic state transition model.
A policy where 94% of the women over age 50 (those having consecutive normal smears) can be withdrawn from screening has been proposed.47, 48, 49 The extremely high negative predictive value of a negative combined cytology and HPV DNA test is supportive of stopping screening at age 50, particularly if there have been multiple negative combined tests before age 50. The incidence of carcinoma in situ decreases considerably after age 50.50 Indeed, screening has not measurably reduced the age specific incidence or mortality in women over age 55 in Sweden.37 Overall, previous reports and the results of our model support a policy of not screening women after age 50, if they have had multiple negative screening tests previously.49, 51
Adding HPV-DNA testing to cytology improves the negative predictive value from 96% to >99% and a double negative test confers an almost 100% assurance of not having CIN2-3 or invasive cancer, allowing an extension of the screening interval.52 Screening frequency is primarily determined by the probability of a test being negative in a woman with disease and the average time required for a precursor to progress to invasive cancer.53 The sojourn time with screen-detectable HPV infection and CIN is on average 15–20 years on average before invasive cancer.34, 54, 55 Because histologic diagnosis of CIN2-3 may vary5, 56 and since most CIN 2-3 lesions will never progress to invasive cancer,8, 57 the majority of women diagnosed with CIN2-3 will have unnecessary treatment. The number of diagnostic procedures, over-treatments and follow-up examinations is reduced by extending screening intervals, resulting in cost savings without changes in the clinical outcome, as found both by us and others.46
Our study did not include hysterectomy rates in the model. Several studies have shown that about 10%–15% of the decrease in the cervical cancer incidence might be attributed to hysterectomies decreasing the population at risk.58, 59 We have opted to use actual population-based transition probabilities as input data, as this reflects the impact of the screening program as a whole (i.e., not only the effect of the screening test per se, but also of the induced procedures).
Another limitation of the study is that only the provider/health service perspective was chosen. However, no significant cost shifting can be expected between health care providers. Also, the uncertainty regarding the societal perspectives of cervical cancer screening make economic analyses attempting to capture all the effects and consequences of screening unreliable. Notable adverse effects are complications after conization like preterm delivery, anxiety caused by false positive test results and over treatment of nonprogressive CIN lesions.46 The use of QALYs as outcome measures in cost-effectiveness modeling is recommended by the panel of cost-effectiveness in health and medicine60 but no reliable studies are available allowing quantitation of quality of life aspects of screening and cervical cancer.
Our cost estimates are on average slightly lower than cost estimates from the US, the UK and the Netherlands.38 Costs for screening and follow up after treatment for CIN2-3 are in comparison with other studies comparatively low, because these activities are in Sweden mainly performed by midwives. A higher prevalence of positive cervical cytology screening tests can be observed in the UK and the USA than in Sweden,61 which may have contributed to a better cost-effectiveness observed in our model compared to studies from the UK and the USA. Exchange of reference slides has shown that this difference is attributable to differences in cytological diagnostic practices, with most smears classified as ASCUS in UK and the USA being classified as normal in Sweden.61 Because also the costs for treatment of early and curable invasive cancer as well as for continuing care and terminal care are also slightly lower in Sweden than previously reported from other countries,38 the overall lower costs in Sweden are not generally biasing the results in favor of or against screening.
Another drawback of our model is that it starts at age 32 and therefore the costs and effects of screening before that age were not analyzed. HPV-DNA testing before age 30 is not an FDA approved strategy and the benefits and costs of conventional cytology screening before age 30 are also unclear.25, 34, 36, 37 Globally, policies on when to start and stop screening, and screening intervals, vary considerably.62, 63
Implementation of screening policies based on evidence from cost-effectiveness studies remains not widely used, but the increasing quality and consistency of modeling studies is likely to eventually have impact on screening policies.
None of the funding organizations had any role in design and conduct of the study, collection, management, analysis, and interpretation of the data, nor in preparation, review or approval of the manuscript.