Costs for Breast Cancer Screening After Digital Mammography
Costs for Breast Cancer Screening After Digital Mammography
We considered eight screening scenarios. First, the comparative effectiveness of the transition to all-digital screening in the United States was evaluated by comparing biennial digital vs film mammography from ages 50 to 74 years (eg, the 2009U.S. Preventive Services Task Force guidelines). We also examined five alternative digital screening scenarios: 1) biennial screening from ages 40 to 74 years; 2) annual screening from ages 50 to 74 years; 3) annual screening from ages 40 to 74 years; 4) annual screening from ages 40 to 49 years followed by biennial screening from ages 50 to 74 years; and 5) annual screening from ages 40 to 74 years for those with dense breast tissue (Breast Imaging Reporting and Data System [BI-RADS] 3 or 4) and biennial screening otherwise (BI-RADS 1 or 2). The last scenario was included to help guide decision-making about new state and federal legislative efforts about breast density notification. All scenarios were also compared with a no screening scenario.
The models include model D (Dana-Farber Cancer Institute), model E (Erasmus University Medical Center), model G-E (Georgetown University Medical Center and Albert Einstein College of Medicine), model M (MD Anderson Cancer Center), and model W (University of Wisconsin and Harvard Medical School) and have been described elsewhere. Briefly, they begin with estimates of incidence without screening and treatment and then overlay screening use and improvements in survival associated with adjuvant treatment. Some model continuous-time tumor growth, whereas others consider progression through discrete preclinical and clinical disease states, and one makes no natural history assumptions (Supplementary Table 1, available online http://jnci.oxfordjournals.org/content/106/6/dju092/suppl/DC1). On the basis of mammography sensitivity (or thresholds of detection), screening can identify disease in the preclinical period possibly at an earlier stage or smaller size than might occur by clinical detection, resulting in a reduction in breast cancer mortality. Age, estrogen receptor status, and tumor size/stage–specific treatment have independent effects on mortality. Women can die of breast cancer or other causes. The models replicate US population breast cancer trends.
For this analysis, we used the cohort of women born in 1960. Outcomes were counted for their lifetimes, beginning at age 40, assuming they adhered to screening schedules and received recommended treatment based on age and tumor characteristics. All models used common inputs and assumptions (Table 1).
Breast Density. Women were assigned a density based on the distribution of BI-RADS breast density categories among ages 40 to 49 years observed in the BCSC (Table 1). Based on BCSC data, density was assumed to decrease by one BI-RADS category at age 50 years and remain at that level thereafter for 41% of women across all categories, reflecting perimenopausal reductions in breast density; the remainder maintained the same density after age 50 years. Breast cancer incidence was conditional on relative risks by BI-RADS categories at ages 40 to 49 years.
Mammography Performance. Mammography sensitivity, specificity, and cancer detection rates were estimated for film and digital mammography by breast density, age group (ages 40–49 or 50–74 years), and screening interval (first, annual, biennial) by fitting logistic regression models to data on nearly two million examinations performed between 2001 and 2008 in women aged 40 to 74 years with BI-RADS information in the BCSC (Table 2; Supplementary Tables 2 and 3, available online http://jnci.oxfordjournals.org/content/106/6/dju092/suppl/DC1). In the models, mammography performance changed according to age and breast density.
Health Effects. We estimated breast cancer mortality, mortality reductions, and life-years. Age- and sex-specific health utilities, adjusted for diagnosis and treatment, were used to estimate quality-adjusted life-years (QALYs) (Table 1). In our base case, we assumed no utility decrements associated with screening participation or experiencing false positives. In supplemental analysis, we included small reductions in utility from screening participation (0.006 for 1 week) and a positive screen (0.105 for 5 weeks).
Costs. Medicare reimbursement rates were used for costs of digital and film screens. Diagnostic costs were based on use patterns in the Group Health BCSC registry and average Medicare reimbursement rates within age and screening result strata (true positive or false positive) (Table 1; Supplementary Table 4, available online http://jnci.oxfordjournals.org/content/106/6/dju092/suppl/DC1). Treatment costs were based on published estimates. All costs were converted to 2012 US dollars.
Costs and health effects were discounted at 3% as recommended. The analysis was conducted using a federal payer perspective. Within models, all strategies were first compared with no screening and then ranked by total costs and compared with each other. If a strategy was more expensive and yielded fewer QALYs, it was considered "dominated." Incremental cost-effectiveness ratios between each strategy and the next most costly nondominated strategy were calculated as the difference in costs divided by the difference in QALYs. If the incremental ratio for one strategy was higher than the incremental ratio for the next more costly and effective strategy, it was considered "weakly dominated" and excluded. Final rankings were compared across models, and results per strategy and differences between strategies are presented as medians across models.
Comparative modeling is one way to test the impact of model structure and parameter uncertainty on results. One-way sensitivity analyses were also conducted in each model to explore the impact of varying key parameters, including reductions in digital prices, improvements in digital specificity, and effects of density on incidence.
Methods
We considered eight screening scenarios. First, the comparative effectiveness of the transition to all-digital screening in the United States was evaluated by comparing biennial digital vs film mammography from ages 50 to 74 years (eg, the 2009U.S. Preventive Services Task Force guidelines). We also examined five alternative digital screening scenarios: 1) biennial screening from ages 40 to 74 years; 2) annual screening from ages 50 to 74 years; 3) annual screening from ages 40 to 74 years; 4) annual screening from ages 40 to 49 years followed by biennial screening from ages 50 to 74 years; and 5) annual screening from ages 40 to 74 years for those with dense breast tissue (Breast Imaging Reporting and Data System [BI-RADS] 3 or 4) and biennial screening otherwise (BI-RADS 1 or 2). The last scenario was included to help guide decision-making about new state and federal legislative efforts about breast density notification. All scenarios were also compared with a no screening scenario.
Model Overview
The models include model D (Dana-Farber Cancer Institute), model E (Erasmus University Medical Center), model G-E (Georgetown University Medical Center and Albert Einstein College of Medicine), model M (MD Anderson Cancer Center), and model W (University of Wisconsin and Harvard Medical School) and have been described elsewhere. Briefly, they begin with estimates of incidence without screening and treatment and then overlay screening use and improvements in survival associated with adjuvant treatment. Some model continuous-time tumor growth, whereas others consider progression through discrete preclinical and clinical disease states, and one makes no natural history assumptions (Supplementary Table 1, available online http://jnci.oxfordjournals.org/content/106/6/dju092/suppl/DC1). On the basis of mammography sensitivity (or thresholds of detection), screening can identify disease in the preclinical period possibly at an earlier stage or smaller size than might occur by clinical detection, resulting in a reduction in breast cancer mortality. Age, estrogen receptor status, and tumor size/stage–specific treatment have independent effects on mortality. Women can die of breast cancer or other causes. The models replicate US population breast cancer trends.
For this analysis, we used the cohort of women born in 1960. Outcomes were counted for their lifetimes, beginning at age 40, assuming they adhered to screening schedules and received recommended treatment based on age and tumor characteristics. All models used common inputs and assumptions (Table 1).
Model Parameters
Breast Density. Women were assigned a density based on the distribution of BI-RADS breast density categories among ages 40 to 49 years observed in the BCSC (Table 1). Based on BCSC data, density was assumed to decrease by one BI-RADS category at age 50 years and remain at that level thereafter for 41% of women across all categories, reflecting perimenopausal reductions in breast density; the remainder maintained the same density after age 50 years. Breast cancer incidence was conditional on relative risks by BI-RADS categories at ages 40 to 49 years.
Mammography Performance. Mammography sensitivity, specificity, and cancer detection rates were estimated for film and digital mammography by breast density, age group (ages 40–49 or 50–74 years), and screening interval (first, annual, biennial) by fitting logistic regression models to data on nearly two million examinations performed between 2001 and 2008 in women aged 40 to 74 years with BI-RADS information in the BCSC (Table 2; Supplementary Tables 2 and 3, available online http://jnci.oxfordjournals.org/content/106/6/dju092/suppl/DC1). In the models, mammography performance changed according to age and breast density.
Health Effects. We estimated breast cancer mortality, mortality reductions, and life-years. Age- and sex-specific health utilities, adjusted for diagnosis and treatment, were used to estimate quality-adjusted life-years (QALYs) (Table 1). In our base case, we assumed no utility decrements associated with screening participation or experiencing false positives. In supplemental analysis, we included small reductions in utility from screening participation (0.006 for 1 week) and a positive screen (0.105 for 5 weeks).
Costs. Medicare reimbursement rates were used for costs of digital and film screens. Diagnostic costs were based on use patterns in the Group Health BCSC registry and average Medicare reimbursement rates within age and screening result strata (true positive or false positive) (Table 1; Supplementary Table 4, available online http://jnci.oxfordjournals.org/content/106/6/dju092/suppl/DC1). Treatment costs were based on published estimates. All costs were converted to 2012 US dollars.
Analysis
Costs and health effects were discounted at 3% as recommended. The analysis was conducted using a federal payer perspective. Within models, all strategies were first compared with no screening and then ranked by total costs and compared with each other. If a strategy was more expensive and yielded fewer QALYs, it was considered "dominated." Incremental cost-effectiveness ratios between each strategy and the next most costly nondominated strategy were calculated as the difference in costs divided by the difference in QALYs. If the incremental ratio for one strategy was higher than the incremental ratio for the next more costly and effective strategy, it was considered "weakly dominated" and excluded. Final rankings were compared across models, and results per strategy and differences between strategies are presented as medians across models.
Sensitivity Analyses
Comparative modeling is one way to test the impact of model structure and parameter uncertainty on results. One-way sensitivity analyses were also conducted in each model to explore the impact of varying key parameters, including reductions in digital prices, improvements in digital specificity, and effects of density on incidence.
Source...