Is Age-targeted full-field digital mammography screening cost-effective in emerging countries? A micro simulation model

Mathematical model

The developed mathematical model was constructed using decision analysis software (TreeAgePro2009 Suite, release 1.0.2, Tree Age Software, Inc., Williamstown, MA). Specifically, a Markov model was used to compare populations of young women in Brazil. The structure of the model (Figure 1) is similar to other models used for BC screening programs and characterizes the complexity of the natural history of the disease (e.g., invasive stages are defined following the tumor–node–metastasis classification (Edge et al. 2010; van Oortmarssen et al. 1990; Szeto & Devlin 1996; Rojnik et al. 2008).

In the real world, women diagnosed with BC have a relatively high risk of developing a new cancer or a recurrent disease. Women who develop a recurrent disease will create extra cost and utilities for the state in line with the basal risk of recurrence of the higher Markov state.

The micro simulation approach with a cycle length of 1 year with half-cycle correction was chosen for this study. BC incidence, mammography sensitivity, mortality, and relative survival rate were modeled as time-dependent transition probabilities.

We considered the following seven BC screening strategies for women aged 40–49 years: (A) usual care; (B) annual SFM; (C) SFM every 2 years; (D) annual FFDM; (E) FFDM every 2 years; (F) “age-targeted digital” (i.e., annual FFDM for the 40–49 age group and annual SFM for the 50–69 age group); and (G) “age-targeted digital” (annual FFDM for the 40–49 age group and SFM every 2 years for the 50–69 age group). These strategies were based on the findings of previous studies (Schopper & de Wolf 2009; US Preventive Services Task Force 2009) and they included the current status of the Brazilian public health system (Strategy A “no formal BC screening” as the base case). According to DATASUS, the Brazilian public health system database, the annual utilization of SFM is approximately 17.5% in women above 50 years (Ministério_Saúde_Brasil, DATASUS 2011).

To determine whether the increased costs of screening strategies are warranted by health gains compared with usual care, we assessed the cost-effectiveness of these seven screening strategies from a public healthcare perspective. The time horizon covered the full lifetimes of the sample population from age 40 onward. Mammography screening stops after 69 years as recommended by the Brazilian National Cancer Institute (INCA) (INCA 2007). The starting age for the micro simulation ranged from 40 to 49 years based on the Brazilian population census (IBGE 2010).

Model calibration

Transitions to Markov states are governed by the rate of incidence, clinical-stage distribution data, and sojourn time. We modeled for an increase in the incidences of in situ carcinomas through the introduction of screening mammography (Kerlikowske 2010). Ten years after the introduction of these screening programs, ductal carcinoma in situ (DCIS) incidence rates were assumed to have stabilized.

Moreover, evidence of better prognostic screening compared with the pre-screening era for in situ cancer cases was incorporated into our model (Ernster et al. 2000), while over diagnosis was adjusted for confounding and lead time bias according to the findings of Smith and Duffy (Smith & Duffy 2011). The risks and benefits of screening schedules were also adapted from Mandelblatt et al. (Mandelblatt et al. 2009). BC prognosis subgroups were then determined for advanced disease status: Luminal A, Luminal B, HER2-enriched, and triple negative (Kennecke et al. 2010). Screened BC has been shown to have independently lower mortality rates compared with non-screened BC (Mook et al. 2011). Finally, we also adjusted BC recurrence risk based on BC subgroup, exposure to adjuvant/palliative chemotherapy, adjuvant/palliative trastuzumab, and adjuvant/palliative endocrine therapy based on previous findings (Perez et al. 2011; Hortobagyi 1998; Mauri et al. 2006; Mouridsen et al. 2003; Mouridsen et al. 2001; Slamon et al. 2001).

Patterns of stage-specific treatments were adapted from DATASUS and from the literature. For DCIS, two treatments were possible: surgery with or without radiotherapy (Wapnir et al. 2011; Baxter et al. 2004; Meijnen et al. 2008). For invasive BC, five treatments were possible: surgery, anti-HER2 adjuvant biologic therapy, radiotherapy, chemotherapy, and endocrine therapy. The last three treatments could be used in an adjuvant as well as in a palliative setting, while anti-HER2 therapy was only allowed in adjuvant stage 2 and 3 settings for HER2-positive patients for 1 year (Perez et al. 2011). Further, the sensitivity of the mammography in the base case was adapted from Breast Cancer Surveillance Consortium (BCSC) data (Kerlikowske et al. 2011). The BCSC database was chosen because it is the largest source of effectiveness data from population screening using both film and digital technology.

The deployment of an FFDM screening program may cause higher recall rates compared with an SFM-based program (Bluekens et al. 2010). For the first round of screening in the present study (prevalence screening), recall rates were 4.29% and 3.41% for FFDM and SFM, respectively. For the second round (incidence screening), recall rates were 1.69% and 1.01% (Bluekens et al. 2010). We assumed that all recalled women would undergo another mammography and/or ultrasound. Approximately 5% and 3% of the recalled women underwent fine needle aspiration and surgical biopsy, respectively (Moss et al. 2006).

All death rates were adapted from the Brazilian Institute of Geography and Statistics Census (2010) and BC deaths were calibrated based on the Mortality Information System of Brazil (DATASUS 2000).

We tested whether the model was calibrated according to the life expectancy of Brazilian women (IBGE 2010). The model does not include input parameters for life expectancy, which is estimated indirectly as a function of the parameters for relapse rates, progression, and overall and BC deaths. Thus, the life expectancy of Brazilian women was defined as an appropriate parameter to validate the model in Brazil. Figure 2 presents the life expectancy predicted in the model at a 95% confidence interval.

Screening and participation rates

There is no formal screening activity in Brazil (despite some isolated initiatives at a regional level). Therefore, opportunistic screening is considered to be usual care in the Brazilian public health system. DATASUS shows that approximately 18% of women above 50 have undergone SFM (Ministério_Saúde_Brasil, DATASUS 2011). A Brazilian prospective cohort achieved a similar opportunistic screening rate of approximately 24% (Marchi & Gurgel 2010). Finally, annual participation rates ranged from 18% to 70%.

Costs and health outcomes

Table 1 presents the costs and utilities applied in the model. Total costs consist of the costs of primary care consultancy, mammography screening, additional work-up exams (when required), cancer diagnostic procedures (images, biopsy, pathology), cancer staging (images), cancer treatment (surgery, radiotherapy, chemotherapy, anti-HER2, and endocrine therapy), and cancer follow-up. Costs were obtained from Ministério_Saúde_Brasil, DATASUS (2011) and the BC database of resource utilization in the public healthcare system in Brazil (Souza 2012). All costs are expressed in 2010 Brazilian Real (US$ 1 = R$ 1.67). Quality-adjusted life years (QALYs) were estimated based on the patient’s SF-6D scores (Souza 2012; Cruz 2010).

Base case analysis

Using a set of natural history input parameters, we calculated the expected costs and effectiveness of each strategy in base case and sensitivity analyses. The costs and effects of each simulated screening program were then assessed. Future costs and health effects (e.g., life years and utilities losses) were discounted at a rate of 5% according to the Brazilian Guidelines for Health Technology Assessment (Ministério_Saúde_Brasil 2009). After ranking them in order of increasing costs and eliminating all dominant strategies (greater cost and fewer benefits than any other combination of strategies), we calculated incremental cost-effectiveness ratios (ICERs).

Since there is no recommended threshold to determine whether an intervention is cost-effective in Brazil (Ministério_Saúde_Brasil 2009), we adapted the recommendations of the World Health Organization, which suggests that a cost-effective intervention would avert one additional disability-adjusted life year for less than three times average per capita GDP (World Health Organization 2001). We assumed that society’s willingness to pay (WTP) for one additional disability-adjusted life year was equivalent to its WTP for one QALY. This approach has been used in previous economic evaluations performed in Brazil and in other middle-income countries (Goldie et al. 2007; Goldie et al. 2008; Vanni et al. 2010; Vanni et al. 2012). Programs that were more costly and less effective than other programs were immediately ruled out as inefficient (i.e., according to the simple dominance principle). The remaining programs constituted the frontier of efficient screening programs.

Sensitivity analysis

To assess uncertainty in the model, one-way, scenario, and probabilistic sensitivity analyses were conducted. In one-way sensitivity analysis, the key parameters were varied using minimum and maximum values, as shown in Table 1. A probabilistic sensitivity analysis was also performed to explore joint uncertainty across parameters. By sampling the distribution of the model parameters, we generated 10,000 estimates for the costs and effects of each strategy. These estimates were plotted on a cost-effectiveness plane and cost-effectiveness acceptability curves were used to depict the level of uncertainty for the optimal strategy at different WTP thresholds for an additional QALY (Barton et al. 2008).

Base case analysis

Thus, the costs and effectiveness of the strategies were considered to identify which strategy represented better value for money. Figure 3 and Table 2 show that usual care (Strategy A) was the cheapest but least effective strategy. Adopting SFM every 2 years (Strategy C) was slightly more costly but also more effective, yielding an ICER of R$ 1,509 per QALY gained. The next best alternative also adopted SFM, but now annually (Strategy B), which was cost-effective at an additional R$ 13,131 per QALY gained. FFDM annual screening (Strategy E) was dominated by Strategy F, which is an age-targeted option, with FFDM used annually until 49 years and SFM annually from 50 to 69 years. For younger women (<50 years), this is the most effective strategy. With an ICER of R$ 30,520, Strategy F could also be considered to be a cost-effective strategy for an emerging country such as Brazil.

Sensitivity analysis

In the one-way sensitivity analysis, the ranking of the seven strategies remained unchanged for most model parameters. The results were most sensitive to changes in the coverage of opportunistic screening under usual care (Strategy A). At a coverage rate of approximately 30%, SFM every 2 years showed cost savings. Although the discount rate and BC incidence seemed to play an important role in determining the magnitude of ICERs, they did not change the order of the strategies that composed the cost-effectiveness frontier.

Figure 4 presents the range of ICERs according to BC incidence, age distribution, and mammography coverage by Brazilian region (INCA 2010; MS 2011). In regions that have a lower BC incidence (e.g., Belem and Cuiaba city), ICERs have a higher probability of not being cost-effective (R$ 257,889 and R$ 49,362, respectively). On the contrary, for regions that have a higher BC incidence (e.g., São Paulo and Recife), the ICER is approximately R$ 21,000. The best scenario was in Porto Alegre, with an ICER of R$ 15,300.

Figure 5 reports the results of the probabilistic sensitivity analysis. By adopting the threshold suggested by the Commission for Macroeconomics in Health for cost-effectiveness interventions (R$ 17,869/QALY), and by considering both SFM strategies (annual and every 2 years), we found a high probability that SFM is a cost-effective approach for the Brazilian public health system (approximately 70% of the simulations). Moreover, at a much lower ICER of R$ 6,000/QALY, SFM every 2 years was cost-effective in more than 95% of the simulations.

Nonetheless, it is important to recognize that FFDM is increasingly used by institutions in Brazil. When we consider a cost-effective threshold of three times national GDP (R$ 53,607/QALY), we found a high probability that an age-targeted strategy (e.g., Strategy F) would be cost-effective (approximately 60% of the simulations). Considering a WTP of R$ 20,000/QALY, 10% of the simulations with Strategy F would be considered to be cost-effective. In the case of a WTP of R$ 100,000/QALY, this figure grew to 70% of the simulations.