Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Chapter 5: Additional Common Inputs for Analyzing Impact of Adjuvant Therapy and Mammography on U.S. Mortality

Chapter 5: Additional Common Inputs for Analyzing Impact of Adjuvant Therapy and Mammography on... Abstract In estimating the impact of mammography and adjuvant treatment on U.S. breast cancer mortality rates, several parameters were common to all the Cancer Intervention and Surveillance Modeling Network (CISNET) models participating in the breast cancer base case. Models either used the parameters directly as input or calibrated their models to reproduce the common set of parameters. This chapter describes the common input parameters that are not specifically discussed elsewhere in the monograph. Participants from the Cancer Intervention and Surveillance Modeling Network (CISNET) participated in a joint analysis that looked at the effect of mammography screening and the introduction of adjuvant therapy on breast cancer mortality. The purpose of this base case analysis was twofold: to provide an answer to an important breast cancer research question from several different independent models and to provide a point of comparison between models. Common inputs were used by all models when appropriate to facilitate comparisons of modeling results. These common inputs included adjuvant treatment dissemination and efficacy (1), mortality from causes other than breast cancer (2), and changes in underlying risk of developing breast cancer (3) discussed in previous chapters. This chapter reviews the remaining common inputs used by all the modeling groups, including dissemination of mammography screening, prevalence of breast cancer in 1975, and prescreening/preadjuvant therapy survival, stage distribution, and mortality rates. How these inputs were used varied with the different modeling approaches. For example, prevalence of cancer patients alive in 1975 was directly input into some models, whereas others used this information to calibrate model parameters so that the models could reproduce breast cancer prevalence in 1975. Table 1 lists the common input parameters used in the base case modeling along with a brief description of each. The goal was to derive inputs that are representative of the United States, but for many of the inputs (i.e., treatment dissemination, cohort risk, prescreening stage distribution), the closest approximation that could be derived came from the Surveillance, Epidemiology, and End Results (SEER) 9 registries (http://seer.cancer.gov). SEER has been shown to be somewhat representative of the United States (4,5). Table 1.  Common input parameters used in modeling the impact of mammography and adjuvant treatment on U.S. mortality Base case inputs  Brief description  Treatment dissemination*  Use of adjuvant chemotherapy and tamoxifen by age, stage, and ER status from 1975 to 2000  Treatment efficacy*  Survival benefit of adjuvant chemotherapy and tamoxifen by age and ER status  Other-cause mortality†  Risk of dying from causes other than breast cancer by age and calendar year  Cohort trend in risk‡  Background trend in the risk of breast cancer giving an estimate of what incidence would have been without screening  Mammography dissemination  Use of screening mammography by birth cohort from 1975 to 2000  Disease prevalence in 1975  No. of women alive in 1975 who had been previously diagnosed with breast cancer  Prescreening survival  Breast cancer specific survival estimated for patients diagnosed before mammography was commonly used for screening  Prescreening stage distribution  Stage of disease at diagnosis before mammography was commonly used for screening  Mortality  Breast cancer mortality rates for the United States and the SEER 9 regions  Base case inputs  Brief description  Treatment dissemination*  Use of adjuvant chemotherapy and tamoxifen by age, stage, and ER status from 1975 to 2000  Treatment efficacy*  Survival benefit of adjuvant chemotherapy and tamoxifen by age and ER status  Other-cause mortality†  Risk of dying from causes other than breast cancer by age and calendar year  Cohort trend in risk‡  Background trend in the risk of breast cancer giving an estimate of what incidence would have been without screening  Mammography dissemination  Use of screening mammography by birth cohort from 1975 to 2000  Disease prevalence in 1975  No. of women alive in 1975 who had been previously diagnosed with breast cancer  Prescreening survival  Breast cancer specific survival estimated for patients diagnosed before mammography was commonly used for screening  Prescreening stage distribution  Stage of disease at diagnosis before mammography was commonly used for screening  Mortality  Breast cancer mortality rates for the United States and the SEER 9 regions  * (1). † (2). ‡ (3). View Large Each section of this paper describes a set of input parameters used in modeling the base case question. Since various combinations of stage and size of disease were used by the CISNET models, several of the inputs were provided for different classification systems. We demonstrate inputs using historical stage, and additional information using other staging systems can be found as supplementary material when appropriate (http://jncicancerspectrum.oxfordjournals.org/jncimono/content/vol2006/issue36). DISSEMINATION OF MAMMOGRAPHY To estimate the impact of screening on population incidence and mortality rate, it is first necessary to quantify the screening practices in the United States over the years of interest. Cronin et al. modeled the dissemination for mammography in the U.S. population from 1975 until 2000 (6). Two distinct statistical modeling efforts were performed to estimate the time of a woman's first mammography exam and exams after the initial mammography. The two pieces were combined through a simulation program to generate screening exam histories for individual women. The simulation does not model diagnosis of cancer or death, so the users of the simulation program (posted on the public CISNET Web site, http://cisnet.cancer.gov) would truncate an individual's screening history at the time of diagnosis or death. A summary of data used in the analysis, statistical methods, and results are given below; detailed descriptions can be found in Cronin et al. (6). Cross-sectional survey data was used to estimate the cumulative distribution for the time to first mammography. Periodically the National Health Information Survey (NHIS) includes questions on the use of mammography to monitor cancer-screening behaviors. Estimates of the dissemination of an initial screening test by 5-year birth cohorts centered on calendar years ending in 0 and ending in 5 are made using cross-sectional estimates of the percentage of the population that reported ever having a mammogram from the 1987, 1990, 1992, 1993, 1994, 1998, and 2000 NHIS surveys (National Center for Health Statistics, National Health Interview Survey Web site http://www.cdc.gov/nchs/nhis.htm). Estimates of the proportion of women in a particular birth cohort having their first mammogram between two NHIS surveys is computed by subtracting the proportion reporting ever having a mammogram in the earlier survey from the proportion reported in the later survey. Since the observed data could be used to construct only a portion of the life history from 1987 to 2000, a dissemination of innovations model (7) was fitted to extrapolate the curve for the entire life history of age at first mammography for a birth cohort. The model assumes that some women never receive a mammogram and that no screening occurred before 1975 or before age 30. Fig. 1, A shows the estimated distribution for the time of first mammogram by birth cohort. Fig. 1. View largeDownload slide A) Cumulative distribution for the age of first mammography exam by 5-year birth cohort. B) Modeled cumulative distribution of time between subsequent mammography exams for women aged 50–59 years when the exam at the beginning of the interval is not a woman's first mammography. Models were stratified for the annual, biennial, and irregular screening groups. C) Percentage of the screened population that fall into each of the three defined groups (annual screener, biennial screener, irregular screener) conditioned on age. Distribution assumed constant over calendar years. *Figures 1, A, B, and C originally appeared in Cronin et al., “Modeling The Dissemination Of Mammography In The United States,” Cancer Causes Control 2005;16:701–12 as figures 1b, 3, and 4. Reproduced by permission of Springer Science and Business Media. Fig. 1. View largeDownload slide A) Cumulative distribution for the age of first mammography exam by 5-year birth cohort. B) Modeled cumulative distribution of time between subsequent mammography exams for women aged 50–59 years when the exam at the beginning of the interval is not a woman's first mammography. Models were stratified for the annual, biennial, and irregular screening groups. C) Percentage of the screened population that fall into each of the three defined groups (annual screener, biennial screener, irregular screener) conditioned on age. Distribution assumed constant over calendar years. *Figures 1, A, B, and C originally appeared in Cronin et al., “Modeling The Dissemination Of Mammography In The United States,” Cancer Causes Control 2005;16:701–12 as figures 1b, 3, and 4. Reproduced by permission of Springer Science and Business Media. Data on mammography usage from the Breast Cancer Surveillance Consortium (BCSC) were used to model repeat screening behavior. BCSC is an NCI-supported research initiative that collects population-based longitudinal data on mammography usage and performance in clinical practice through mammography registries that are linked to cancer outcomes (8) (http://breastscreening.cancer.gov). The BCSC collects information on screening mammograms from 1994 onward in seven geographically defined research sites representing approximately 5% of the U.S. population. This unique dataset captures multiple screening events for the same woman, thus allowing for the estimation of the period between subsequent mammogram exams. To model repeat screening behaviors, three general groups of screeners were defined a priori to represent regular annual screeners (defined as individuals with a mean gap time between screening exams of ≤1.5 years), biennial screeners (individuals with a mean gap time of 1.5–2.5 years), and irregular screeners (individuals with a mean gap time of >2.5 years). Individuals from the BCSC data were classified into one of the three groups listed above on the basis of their observed screening history. Stratified survival analyses with event times defined as the time between subsequent screening mammograms were performed using gamma frailty models (9) for each of the three defined groups to account for correlations between multiple intervals for one individual. Gap times between subsequent screening mammography exams were defined on the basis of either two observed exams recorded in the database or by personal recall for the time since last mammogram. Fig. 1, B shows the cumulative distribution of the time until next screening exam for each of the three groups (annual-, biennial-, and irregular- gap times). In addition to the survival curves from each group, the age-specific probabilities of a woman falling into each of the defined groups—annual, biennial, and irregular screener—were calculated and are shown in Fig. 1, C. To simulate an individual history of screening, the user must input an individual's date of birth. The simulation program performs the following steps: Generate date of first mammography on the basis of cumulative distribution of time to first screening exam conditioned on birth cohort. Generate a gamma frailty parameter that an individual will keep throughout her lifetime. This parameter represents the tendency to have longer or shorter gap times and induces the correlation of gap times of an individual over the course of her life. Generate a uniform random number between zero and one that is used to define the repeat mammography group (annual, biennial, or irregular time). The uniform number between 0 and 1 can be thought of as representing a person's tendency to get repeated screening exams relative to others in the population, with numbers close to 0 having the highest tendency and numbers close to 1 having the lowest tendency to return for later screening exams. The individual keeps the same random number for life, although they may change groups depending on their age. For example, in Fig. 1, C, a woman is assigned a number of .32 representing the 32nd percentile of the population, which would put her into the irregular screening group before age 40, biennial group between ages 40 and 49, the annual group between ages 50 and 79, and the irregular gap time group after age 70. This change of groups is consistent with most regular screening occurring (and associated highest percentage of the population in the annual and biennial groups) between the ages 50 and 69. At each mammography exam, generate a gap time to next screening exam conditioned on their frailty parameter and current group (annual, biennial, or irregular). More details on the modeling and the mammography dissemination program are available on the CISNET public Web site (http://cisnet.cancer.gov/interfaces). PREVALENCE OF BREAST CANCER IN 1975 For the base case input, we wanted to estimate complete prevalence, i.e., the proportion of women alive on July 1, 1975, with a previous diagnosis of breast cancer, no matter how long ago that diagnosis was. These women were at risk of dying from their breast cancer during the period of interest, 1975–2000. From the Connecticut Tumor Registry (CTR), we could estimate 35-year prevalence by using breast cancer incidence cases reported in 1940–1975; however, we cannot directly calculate complete prevalence. To obtain complete prevalence, we use a method that computes prevalence as a function of modeled incidence and survival (10). The method uses data from the CTR from 1940 to 1982 to avoid screening influencing the fitted trends. A polynomial model using age at diagnosis, year at diagnosis, and birth cohort is fitted to the logit of the breast cancer incidence rates. Survival is modeled using a cure model to extrapolate survival curves before 1940. Prevalence is derived from convolution of estimated/modeled cancer incidence and survival over time. Since prevalence estimates obtained by this convolution method are based on modeled incidence and survival projected into years before 1940, they represent total prevalence, i.e., the prevalence of all past diagnosis. Deaths from other causes are also taken into consideration. To adjust the prevalence estimate calculated from CTR to the SEER-9 regions, we calculated the prevalence at 1/1/1983 for people diagnosed between 1975 and 1982 (8-year limited duration prevalence) from both the CTR and SEER-9 registry using the prevalence section of the SEER*Stat software (http://seer.cancer.gov/seerstat). The age-specific ratio between the SEER-9 and CTR estimates on 1/1/1983 was used to adjust the age-specific 1975 Connecticut breast cancer prevalence proportion to SEER. This adjustment assumes that the difference between SEER and CTR in complete prevalence is the same as the one observed in 8-year prevalence and that the ratio in 1983 is the same as the ratio in 1975. Table 2 gives the estimated age-specific prevalence percent for breast cancer in the SEER 9 regions. Table 2.  Age-specific breast cancer incidence, mortality, and prevalence estimates for prescreening and preadjuvant treatment period*   Incidence: No.of cases per 100 000 women for SEER 9  Mortality: No. of breast cancer deaths per 100 000     Prevalence: % of women alive that have been diagnosed with breast cancer for SEER 9  Age, y    SEER 9  United States    0–24  0.3  0.04  0.06  0.00  25–29  8.5  1.2  1.6  0.02  30–34  25.9  5.7  5.6  0.07  35–39  57.7  13.4  13.3  0.27  40–44  108.6  23.7  24.4  0.64  45–49  171.7  43.2  43.4  1.10  50–54  197.1  62.0  59.1  1.53  55–59  221.0  75.8  74.2  1.99  60–64  260.2  88.5  84.4  2.29  65–69  284.3  99.1  92.9  2.56  70–74  306.2  108.2  104.4  2.90  75–79  332.9  125.2  117.9  3.06  80–84  342.7  130.9  132.8  3.27    Incidence: No.of cases per 100 000 women for SEER 9  Mortality: No. of breast cancer deaths per 100 000     Prevalence: % of women alive that have been diagnosed with breast cancer for SEER 9  Age, y    SEER 9  United States    0–24  0.3  0.04  0.06  0.00  25–29  8.5  1.2  1.6  0.02  30–34  25.9  5.7  5.6  0.07  35–39  57.7  13.4  13.3  0.27  40–44  108.6  23.7  24.4  0.64  45–49  171.7  43.2  43.4  1.10  50–54  197.1  62.0  59.1  1.53  55–59  221.0  75.8  74.2  1.99  60–64  260.2  88.5  84.4  2.29  65–69  284.3  99.1  92.9  2.56  70–74  306.2  108.2  104.4  2.90  75–79  332.9  125.2  117.9  3.06  80–84  342.7  130.9  132.8  3.27  * Incidence included cases diagnosed in the Surveillance, Epidemiology, and End Results (SEER) 9 region from 1975 to 1979. Mortality rates are for the SEER 9 region and the whole United States based on deaths occurring between 1973 and 1975. Prevalence is estimated for July 1, 1975. Incidence based on SEER Statistics Review, 1975–2002; available at: http://seer.cancer.gov/csr/1975_2002. View Large PRESCREENING AND PREADJUVANT TREATMENT ESTIMATES OF BREAST CANCER SURVIVAL Prescreening and preadjuvant treatment survival by stage at diagnosis was calculated from the SEER 9 data with cases that were diagnosed between 1975 and 1979 with 25 years of follow-up. The cause-specific SEER*Stat survival option was used with the cause of death defined as breast cancer. Estimates of cause-specific survival were calculated for 10-year age groups 30–39, 40–49, 50–59, 60–69, and 70–84. Different stage systems (11) to match the staging systems were used by the modelers: Dana-Farber used American Joint Committee on Cancer (AJCC) staging; University of Texas M. D. Anderson Cancer Center used AJCC with nodal status (positive or negative); Wisconsin used SEER historical stage; Stanford and Rochester used SEER historical stage and tumor size categories within stage; and Erasmus used size categories along with whether or not the tumor has metastasized. Between 1975 and 1979 the only stage variable recorded in the SEER database was historical stage. However, other stage distributions were approximated using available extent of disease information. Twenty-five-year life tables for each of the combination of age and stage can be found as Supplementary Data on the JNCI Web site (http://jncicancerspectrum.oxfordjournals.org/jncimono/content/vol2006/issue36). Fig. 2 shows cause-specific survival by age and historical stage for comparison. Fig. 2. View largeDownload slide Cause-specific survival for women diagnosed with breast cancer between 1975 and 1979 in the Surveillance, Epidemiology, and End Results (SEER) 9 regions by age and historical stage. Fig. 2. View largeDownload slide Cause-specific survival for women diagnosed with breast cancer between 1975 and 1979 in the Surveillance, Epidemiology, and End Results (SEER) 9 regions by age and historical stage. PRESCREENING AND PREADJUVANT TREATMENT STAGE DISTRIBUTION AT DIAGNOSIS The prescreening and preadjuvant treatment stage distribution by age for breast cancer cases was calculated from the SEER 9 database using cases diagnosed in 1975–1979. We assumed that unstaged cases would follow the same distribution as staged cases and they were excluded from the calculation. Fig. 3 shows the stage distribution for historical stages (in situ, local, regional, distant) by age at diagnosis. Stage distributions were also estimated by size of tumor and by AJCC staging with nodal status using an approximate translation between extent of disease information available in those years and the AJCC staging rules (Supplementary Data, available at http://jncicancerspectrum.oxfordjournals.org/jncimono/content/vol2006/issue36). Fig. 3. View largeDownload slide Historical stage distribution for breast cancer cases diagnosed between 1975 and 1979 in the Surveillance, Epidemiology, and End Results (SEER) 9 area by age at diagnosis. Fig. 3. View largeDownload slide Historical stage distribution for breast cancer cases diagnosed between 1975 and 1979 in the Surveillance, Epidemiology, and End Results (SEER) 9 area by age at diagnosis. BREAST CANCER MORTALITY RATES FOR THE UNITED STATES AND FOR SEER REGIONS Prescreening and preadjuvant treatment mortality rates were calculated for the United States and the SEER 9 regions using data from the National Center for Health Statistics (12). The rates are calculated using women who died of breast cancer between 1973 and 1975 on the basis of the cause of death listed on their death certificates. Table 2 lists age-specific mortality rates per 100 000 for both the United States and the SEER 9 regions. DISCUSSION The inputs described in this chapter, along with the dissemination and effectiveness of adjuvant therapy [discussed in chapter 2 (1)], and the risk of other mortality other than breast cancer [discussed in chapter 3 (2)], and changes in underlying risk of developing breast cancer [discussed in chapter 4 (3)], provided a common starting point for the CISNET models and facilitated model comparisons. Controlling various model inputs allows for a common context in which to make comparisons. For example, differences in the estimated mortality reduction due to mammography screening would reflect differences in benefit rather than differences in mammography usage if all models use the same screening patterns obtained from the dissemination of mammography model. The usefulness of the inputs described goes far beyond the base case analysis. It provides a basis for modeling other research questions of interest since the introduction of screening mammography and adjuvant therapy. They can also serve as a basis for projecting incidence and mortality trends into the future. References (1) Mariotto AB, Feuer EJ, Harlan LC, Abrams J. Dissemination of adjuvant multiagent chemotherapy and tamoxifen for breast cancer in the United States using estrogen receptor information: 1975–1999. J Natl Cancer Inst Monogr  2006; 36: 7–15. Google Scholar (2) Rosenberg MA. Competing risks to breast cancer mortality. J Natl Cancer Inst Monogr  2006; 36: 15–9. Google Scholar (3) Holford TR, Cronin KA, Mariotto AB, Feuer EJ. Changing patterns in breast cancer incidence trends. J Natl Cancer Inst Monogr  2006; 36: 19–25. Google Scholar (4) Frey CM, McMillen MM, Cowan CD, Horm JW, Kessler LG. Representativeness of the Surveillance, Epidemiology, and End Results program data: recent trends in cancer mortality rates. J Natl Cancer Inst  1992; 84: 872–7. Google Scholar (5) Pickle LW, Feuer EJ, Edwards BK. U.S. predicted cancer incidence, 1999: complete maps by county and state from spatial projection models. NCI Cancer Surveillance Monograph Series, Number 5. Bethesda (MD): National Cancer Institute; 2003. NIH Publication No. 03-5435. Google Scholar (6) Cronin KA, Yu B, Krapcho M, Miglioretti DL, Fay MP, Izmirlian G, et al. Modeling the dissemination of mammography in the United States. Cancer Causes Control  2005; 16: 701–12. Google Scholar (7) Mahajan V, Peterson RA. Models for innovation diffusion. Sage University Paper Series on Quantitative Applications in the Social Sciences 07-048. Newbury Park (CA): Sage; 1985. Google Scholar (8) Ballard-Barbash R, Taplin SH, Yankaskas BC. Breast Cancer Surveillance Consortium: a national mammography screening and outcomes database. AJR Am J Roentgenol  2004; 169: 1001–8. Google Scholar (9) Therneau TM, Grambsch PM. Modeling survival data: extending the Cox model. New York (NY): Springer; 2000. Google Scholar (10) Verdecchia A, De Angelis G, Capocaccia R. Estimation and projections of cancer prevalence from cancer registry data. Stat Med  2002; 21: 3511–26. Google Scholar (11) American Joint Committee on Cancer. Manual for staging of cancer. 3rd ed. Philadelphia (PA): J.B. Lippincott; 1988. Google Scholar (12) Surveillance, Epidemiology, and End Results (SEER) Program (http://www.seer.cancer.gov) SEER*Stat Database: Mortality—All COD, Total U.S. (1969–2001), National Cancer Institute, DCCPS, Surveillance Research Program, Cancer Statistics Branch, released December 2003. Underlying mortality data provided by NCHS (http://www.cdc.gov/nchs). Google Scholar © The Author 2006. Published by Oxford University Press. All rights reserved. For Permissions, please e-mail: journals.permissions@oxfordjournals.org. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png JNCI Monographs Oxford University Press

Chapter 5: Additional Common Inputs for Analyzing Impact of Adjuvant Therapy and Mammography on U.S. Mortality

Loading next page...
 
/lp/oxford-university-press/chapter-5-additional-common-inputs-for-analyzing-impact-of-adjuvant-hxTN2V0zml
Publisher
Oxford University Press
Copyright
© The Author 2006. Published by Oxford University Press. All rights reserved. For Permissions, please e-mail: journals.permissions@oxfordjournals.org.
ISSN
1052-6773
eISSN
1745-6614
DOI
10.1093/jncimonographs/lgj005
pmid
17032891
Publisher site
See Article on Publisher Site

Abstract

Abstract In estimating the impact of mammography and adjuvant treatment on U.S. breast cancer mortality rates, several parameters were common to all the Cancer Intervention and Surveillance Modeling Network (CISNET) models participating in the breast cancer base case. Models either used the parameters directly as input or calibrated their models to reproduce the common set of parameters. This chapter describes the common input parameters that are not specifically discussed elsewhere in the monograph. Participants from the Cancer Intervention and Surveillance Modeling Network (CISNET) participated in a joint analysis that looked at the effect of mammography screening and the introduction of adjuvant therapy on breast cancer mortality. The purpose of this base case analysis was twofold: to provide an answer to an important breast cancer research question from several different independent models and to provide a point of comparison between models. Common inputs were used by all models when appropriate to facilitate comparisons of modeling results. These common inputs included adjuvant treatment dissemination and efficacy (1), mortality from causes other than breast cancer (2), and changes in underlying risk of developing breast cancer (3) discussed in previous chapters. This chapter reviews the remaining common inputs used by all the modeling groups, including dissemination of mammography screening, prevalence of breast cancer in 1975, and prescreening/preadjuvant therapy survival, stage distribution, and mortality rates. How these inputs were used varied with the different modeling approaches. For example, prevalence of cancer patients alive in 1975 was directly input into some models, whereas others used this information to calibrate model parameters so that the models could reproduce breast cancer prevalence in 1975. Table 1 lists the common input parameters used in the base case modeling along with a brief description of each. The goal was to derive inputs that are representative of the United States, but for many of the inputs (i.e., treatment dissemination, cohort risk, prescreening stage distribution), the closest approximation that could be derived came from the Surveillance, Epidemiology, and End Results (SEER) 9 registries (http://seer.cancer.gov). SEER has been shown to be somewhat representative of the United States (4,5). Table 1.  Common input parameters used in modeling the impact of mammography and adjuvant treatment on U.S. mortality Base case inputs  Brief description  Treatment dissemination*  Use of adjuvant chemotherapy and tamoxifen by age, stage, and ER status from 1975 to 2000  Treatment efficacy*  Survival benefit of adjuvant chemotherapy and tamoxifen by age and ER status  Other-cause mortality†  Risk of dying from causes other than breast cancer by age and calendar year  Cohort trend in risk‡  Background trend in the risk of breast cancer giving an estimate of what incidence would have been without screening  Mammography dissemination  Use of screening mammography by birth cohort from 1975 to 2000  Disease prevalence in 1975  No. of women alive in 1975 who had been previously diagnosed with breast cancer  Prescreening survival  Breast cancer specific survival estimated for patients diagnosed before mammography was commonly used for screening  Prescreening stage distribution  Stage of disease at diagnosis before mammography was commonly used for screening  Mortality  Breast cancer mortality rates for the United States and the SEER 9 regions  Base case inputs  Brief description  Treatment dissemination*  Use of adjuvant chemotherapy and tamoxifen by age, stage, and ER status from 1975 to 2000  Treatment efficacy*  Survival benefit of adjuvant chemotherapy and tamoxifen by age and ER status  Other-cause mortality†  Risk of dying from causes other than breast cancer by age and calendar year  Cohort trend in risk‡  Background trend in the risk of breast cancer giving an estimate of what incidence would have been without screening  Mammography dissemination  Use of screening mammography by birth cohort from 1975 to 2000  Disease prevalence in 1975  No. of women alive in 1975 who had been previously diagnosed with breast cancer  Prescreening survival  Breast cancer specific survival estimated for patients diagnosed before mammography was commonly used for screening  Prescreening stage distribution  Stage of disease at diagnosis before mammography was commonly used for screening  Mortality  Breast cancer mortality rates for the United States and the SEER 9 regions  * (1). † (2). ‡ (3). View Large Each section of this paper describes a set of input parameters used in modeling the base case question. Since various combinations of stage and size of disease were used by the CISNET models, several of the inputs were provided for different classification systems. We demonstrate inputs using historical stage, and additional information using other staging systems can be found as supplementary material when appropriate (http://jncicancerspectrum.oxfordjournals.org/jncimono/content/vol2006/issue36). DISSEMINATION OF MAMMOGRAPHY To estimate the impact of screening on population incidence and mortality rate, it is first necessary to quantify the screening practices in the United States over the years of interest. Cronin et al. modeled the dissemination for mammography in the U.S. population from 1975 until 2000 (6). Two distinct statistical modeling efforts were performed to estimate the time of a woman's first mammography exam and exams after the initial mammography. The two pieces were combined through a simulation program to generate screening exam histories for individual women. The simulation does not model diagnosis of cancer or death, so the users of the simulation program (posted on the public CISNET Web site, http://cisnet.cancer.gov) would truncate an individual's screening history at the time of diagnosis or death. A summary of data used in the analysis, statistical methods, and results are given below; detailed descriptions can be found in Cronin et al. (6). Cross-sectional survey data was used to estimate the cumulative distribution for the time to first mammography. Periodically the National Health Information Survey (NHIS) includes questions on the use of mammography to monitor cancer-screening behaviors. Estimates of the dissemination of an initial screening test by 5-year birth cohorts centered on calendar years ending in 0 and ending in 5 are made using cross-sectional estimates of the percentage of the population that reported ever having a mammogram from the 1987, 1990, 1992, 1993, 1994, 1998, and 2000 NHIS surveys (National Center for Health Statistics, National Health Interview Survey Web site http://www.cdc.gov/nchs/nhis.htm). Estimates of the proportion of women in a particular birth cohort having their first mammogram between two NHIS surveys is computed by subtracting the proportion reporting ever having a mammogram in the earlier survey from the proportion reported in the later survey. Since the observed data could be used to construct only a portion of the life history from 1987 to 2000, a dissemination of innovations model (7) was fitted to extrapolate the curve for the entire life history of age at first mammography for a birth cohort. The model assumes that some women never receive a mammogram and that no screening occurred before 1975 or before age 30. Fig. 1, A shows the estimated distribution for the time of first mammogram by birth cohort. Fig. 1. View largeDownload slide A) Cumulative distribution for the age of first mammography exam by 5-year birth cohort. B) Modeled cumulative distribution of time between subsequent mammography exams for women aged 50–59 years when the exam at the beginning of the interval is not a woman's first mammography. Models were stratified for the annual, biennial, and irregular screening groups. C) Percentage of the screened population that fall into each of the three defined groups (annual screener, biennial screener, irregular screener) conditioned on age. Distribution assumed constant over calendar years. *Figures 1, A, B, and C originally appeared in Cronin et al., “Modeling The Dissemination Of Mammography In The United States,” Cancer Causes Control 2005;16:701–12 as figures 1b, 3, and 4. Reproduced by permission of Springer Science and Business Media. Fig. 1. View largeDownload slide A) Cumulative distribution for the age of first mammography exam by 5-year birth cohort. B) Modeled cumulative distribution of time between subsequent mammography exams for women aged 50–59 years when the exam at the beginning of the interval is not a woman's first mammography. Models were stratified for the annual, biennial, and irregular screening groups. C) Percentage of the screened population that fall into each of the three defined groups (annual screener, biennial screener, irregular screener) conditioned on age. Distribution assumed constant over calendar years. *Figures 1, A, B, and C originally appeared in Cronin et al., “Modeling The Dissemination Of Mammography In The United States,” Cancer Causes Control 2005;16:701–12 as figures 1b, 3, and 4. Reproduced by permission of Springer Science and Business Media. Data on mammography usage from the Breast Cancer Surveillance Consortium (BCSC) were used to model repeat screening behavior. BCSC is an NCI-supported research initiative that collects population-based longitudinal data on mammography usage and performance in clinical practice through mammography registries that are linked to cancer outcomes (8) (http://breastscreening.cancer.gov). The BCSC collects information on screening mammograms from 1994 onward in seven geographically defined research sites representing approximately 5% of the U.S. population. This unique dataset captures multiple screening events for the same woman, thus allowing for the estimation of the period between subsequent mammogram exams. To model repeat screening behaviors, three general groups of screeners were defined a priori to represent regular annual screeners (defined as individuals with a mean gap time between screening exams of ≤1.5 years), biennial screeners (individuals with a mean gap time of 1.5–2.5 years), and irregular screeners (individuals with a mean gap time of >2.5 years). Individuals from the BCSC data were classified into one of the three groups listed above on the basis of their observed screening history. Stratified survival analyses with event times defined as the time between subsequent screening mammograms were performed using gamma frailty models (9) for each of the three defined groups to account for correlations between multiple intervals for one individual. Gap times between subsequent screening mammography exams were defined on the basis of either two observed exams recorded in the database or by personal recall for the time since last mammogram. Fig. 1, B shows the cumulative distribution of the time until next screening exam for each of the three groups (annual-, biennial-, and irregular- gap times). In addition to the survival curves from each group, the age-specific probabilities of a woman falling into each of the defined groups—annual, biennial, and irregular screener—were calculated and are shown in Fig. 1, C. To simulate an individual history of screening, the user must input an individual's date of birth. The simulation program performs the following steps: Generate date of first mammography on the basis of cumulative distribution of time to first screening exam conditioned on birth cohort. Generate a gamma frailty parameter that an individual will keep throughout her lifetime. This parameter represents the tendency to have longer or shorter gap times and induces the correlation of gap times of an individual over the course of her life. Generate a uniform random number between zero and one that is used to define the repeat mammography group (annual, biennial, or irregular time). The uniform number between 0 and 1 can be thought of as representing a person's tendency to get repeated screening exams relative to others in the population, with numbers close to 0 having the highest tendency and numbers close to 1 having the lowest tendency to return for later screening exams. The individual keeps the same random number for life, although they may change groups depending on their age. For example, in Fig. 1, C, a woman is assigned a number of .32 representing the 32nd percentile of the population, which would put her into the irregular screening group before age 40, biennial group between ages 40 and 49, the annual group between ages 50 and 79, and the irregular gap time group after age 70. This change of groups is consistent with most regular screening occurring (and associated highest percentage of the population in the annual and biennial groups) between the ages 50 and 69. At each mammography exam, generate a gap time to next screening exam conditioned on their frailty parameter and current group (annual, biennial, or irregular). More details on the modeling and the mammography dissemination program are available on the CISNET public Web site (http://cisnet.cancer.gov/interfaces). PREVALENCE OF BREAST CANCER IN 1975 For the base case input, we wanted to estimate complete prevalence, i.e., the proportion of women alive on July 1, 1975, with a previous diagnosis of breast cancer, no matter how long ago that diagnosis was. These women were at risk of dying from their breast cancer during the period of interest, 1975–2000. From the Connecticut Tumor Registry (CTR), we could estimate 35-year prevalence by using breast cancer incidence cases reported in 1940–1975; however, we cannot directly calculate complete prevalence. To obtain complete prevalence, we use a method that computes prevalence as a function of modeled incidence and survival (10). The method uses data from the CTR from 1940 to 1982 to avoid screening influencing the fitted trends. A polynomial model using age at diagnosis, year at diagnosis, and birth cohort is fitted to the logit of the breast cancer incidence rates. Survival is modeled using a cure model to extrapolate survival curves before 1940. Prevalence is derived from convolution of estimated/modeled cancer incidence and survival over time. Since prevalence estimates obtained by this convolution method are based on modeled incidence and survival projected into years before 1940, they represent total prevalence, i.e., the prevalence of all past diagnosis. Deaths from other causes are also taken into consideration. To adjust the prevalence estimate calculated from CTR to the SEER-9 regions, we calculated the prevalence at 1/1/1983 for people diagnosed between 1975 and 1982 (8-year limited duration prevalence) from both the CTR and SEER-9 registry using the prevalence section of the SEER*Stat software (http://seer.cancer.gov/seerstat). The age-specific ratio between the SEER-9 and CTR estimates on 1/1/1983 was used to adjust the age-specific 1975 Connecticut breast cancer prevalence proportion to SEER. This adjustment assumes that the difference between SEER and CTR in complete prevalence is the same as the one observed in 8-year prevalence and that the ratio in 1983 is the same as the ratio in 1975. Table 2 gives the estimated age-specific prevalence percent for breast cancer in the SEER 9 regions. Table 2.  Age-specific breast cancer incidence, mortality, and prevalence estimates for prescreening and preadjuvant treatment period*   Incidence: No.of cases per 100 000 women for SEER 9  Mortality: No. of breast cancer deaths per 100 000     Prevalence: % of women alive that have been diagnosed with breast cancer for SEER 9  Age, y    SEER 9  United States    0–24  0.3  0.04  0.06  0.00  25–29  8.5  1.2  1.6  0.02  30–34  25.9  5.7  5.6  0.07  35–39  57.7  13.4  13.3  0.27  40–44  108.6  23.7  24.4  0.64  45–49  171.7  43.2  43.4  1.10  50–54  197.1  62.0  59.1  1.53  55–59  221.0  75.8  74.2  1.99  60–64  260.2  88.5  84.4  2.29  65–69  284.3  99.1  92.9  2.56  70–74  306.2  108.2  104.4  2.90  75–79  332.9  125.2  117.9  3.06  80–84  342.7  130.9  132.8  3.27    Incidence: No.of cases per 100 000 women for SEER 9  Mortality: No. of breast cancer deaths per 100 000     Prevalence: % of women alive that have been diagnosed with breast cancer for SEER 9  Age, y    SEER 9  United States    0–24  0.3  0.04  0.06  0.00  25–29  8.5  1.2  1.6  0.02  30–34  25.9  5.7  5.6  0.07  35–39  57.7  13.4  13.3  0.27  40–44  108.6  23.7  24.4  0.64  45–49  171.7  43.2  43.4  1.10  50–54  197.1  62.0  59.1  1.53  55–59  221.0  75.8  74.2  1.99  60–64  260.2  88.5  84.4  2.29  65–69  284.3  99.1  92.9  2.56  70–74  306.2  108.2  104.4  2.90  75–79  332.9  125.2  117.9  3.06  80–84  342.7  130.9  132.8  3.27  * Incidence included cases diagnosed in the Surveillance, Epidemiology, and End Results (SEER) 9 region from 1975 to 1979. Mortality rates are for the SEER 9 region and the whole United States based on deaths occurring between 1973 and 1975. Prevalence is estimated for July 1, 1975. Incidence based on SEER Statistics Review, 1975–2002; available at: http://seer.cancer.gov/csr/1975_2002. View Large PRESCREENING AND PREADJUVANT TREATMENT ESTIMATES OF BREAST CANCER SURVIVAL Prescreening and preadjuvant treatment survival by stage at diagnosis was calculated from the SEER 9 data with cases that were diagnosed between 1975 and 1979 with 25 years of follow-up. The cause-specific SEER*Stat survival option was used with the cause of death defined as breast cancer. Estimates of cause-specific survival were calculated for 10-year age groups 30–39, 40–49, 50–59, 60–69, and 70–84. Different stage systems (11) to match the staging systems were used by the modelers: Dana-Farber used American Joint Committee on Cancer (AJCC) staging; University of Texas M. D. Anderson Cancer Center used AJCC with nodal status (positive or negative); Wisconsin used SEER historical stage; Stanford and Rochester used SEER historical stage and tumor size categories within stage; and Erasmus used size categories along with whether or not the tumor has metastasized. Between 1975 and 1979 the only stage variable recorded in the SEER database was historical stage. However, other stage distributions were approximated using available extent of disease information. Twenty-five-year life tables for each of the combination of age and stage can be found as Supplementary Data on the JNCI Web site (http://jncicancerspectrum.oxfordjournals.org/jncimono/content/vol2006/issue36). Fig. 2 shows cause-specific survival by age and historical stage for comparison. Fig. 2. View largeDownload slide Cause-specific survival for women diagnosed with breast cancer between 1975 and 1979 in the Surveillance, Epidemiology, and End Results (SEER) 9 regions by age and historical stage. Fig. 2. View largeDownload slide Cause-specific survival for women diagnosed with breast cancer between 1975 and 1979 in the Surveillance, Epidemiology, and End Results (SEER) 9 regions by age and historical stage. PRESCREENING AND PREADJUVANT TREATMENT STAGE DISTRIBUTION AT DIAGNOSIS The prescreening and preadjuvant treatment stage distribution by age for breast cancer cases was calculated from the SEER 9 database using cases diagnosed in 1975–1979. We assumed that unstaged cases would follow the same distribution as staged cases and they were excluded from the calculation. Fig. 3 shows the stage distribution for historical stages (in situ, local, regional, distant) by age at diagnosis. Stage distributions were also estimated by size of tumor and by AJCC staging with nodal status using an approximate translation between extent of disease information available in those years and the AJCC staging rules (Supplementary Data, available at http://jncicancerspectrum.oxfordjournals.org/jncimono/content/vol2006/issue36). Fig. 3. View largeDownload slide Historical stage distribution for breast cancer cases diagnosed between 1975 and 1979 in the Surveillance, Epidemiology, and End Results (SEER) 9 area by age at diagnosis. Fig. 3. View largeDownload slide Historical stage distribution for breast cancer cases diagnosed between 1975 and 1979 in the Surveillance, Epidemiology, and End Results (SEER) 9 area by age at diagnosis. BREAST CANCER MORTALITY RATES FOR THE UNITED STATES AND FOR SEER REGIONS Prescreening and preadjuvant treatment mortality rates were calculated for the United States and the SEER 9 regions using data from the National Center for Health Statistics (12). The rates are calculated using women who died of breast cancer between 1973 and 1975 on the basis of the cause of death listed on their death certificates. Table 2 lists age-specific mortality rates per 100 000 for both the United States and the SEER 9 regions. DISCUSSION The inputs described in this chapter, along with the dissemination and effectiveness of adjuvant therapy [discussed in chapter 2 (1)], and the risk of other mortality other than breast cancer [discussed in chapter 3 (2)], and changes in underlying risk of developing breast cancer [discussed in chapter 4 (3)], provided a common starting point for the CISNET models and facilitated model comparisons. Controlling various model inputs allows for a common context in which to make comparisons. For example, differences in the estimated mortality reduction due to mammography screening would reflect differences in benefit rather than differences in mammography usage if all models use the same screening patterns obtained from the dissemination of mammography model. The usefulness of the inputs described goes far beyond the base case analysis. It provides a basis for modeling other research questions of interest since the introduction of screening mammography and adjuvant therapy. They can also serve as a basis for projecting incidence and mortality trends into the future. References (1) Mariotto AB, Feuer EJ, Harlan LC, Abrams J. Dissemination of adjuvant multiagent chemotherapy and tamoxifen for breast cancer in the United States using estrogen receptor information: 1975–1999. J Natl Cancer Inst Monogr  2006; 36: 7–15. Google Scholar (2) Rosenberg MA. Competing risks to breast cancer mortality. J Natl Cancer Inst Monogr  2006; 36: 15–9. Google Scholar (3) Holford TR, Cronin KA, Mariotto AB, Feuer EJ. Changing patterns in breast cancer incidence trends. J Natl Cancer Inst Monogr  2006; 36: 19–25. Google Scholar (4) Frey CM, McMillen MM, Cowan CD, Horm JW, Kessler LG. Representativeness of the Surveillance, Epidemiology, and End Results program data: recent trends in cancer mortality rates. J Natl Cancer Inst  1992; 84: 872–7. Google Scholar (5) Pickle LW, Feuer EJ, Edwards BK. U.S. predicted cancer incidence, 1999: complete maps by county and state from spatial projection models. NCI Cancer Surveillance Monograph Series, Number 5. Bethesda (MD): National Cancer Institute; 2003. NIH Publication No. 03-5435. Google Scholar (6) Cronin KA, Yu B, Krapcho M, Miglioretti DL, Fay MP, Izmirlian G, et al. Modeling the dissemination of mammography in the United States. Cancer Causes Control  2005; 16: 701–12. Google Scholar (7) Mahajan V, Peterson RA. Models for innovation diffusion. Sage University Paper Series on Quantitative Applications in the Social Sciences 07-048. Newbury Park (CA): Sage; 1985. Google Scholar (8) Ballard-Barbash R, Taplin SH, Yankaskas BC. Breast Cancer Surveillance Consortium: a national mammography screening and outcomes database. AJR Am J Roentgenol  2004; 169: 1001–8. Google Scholar (9) Therneau TM, Grambsch PM. Modeling survival data: extending the Cox model. New York (NY): Springer; 2000. Google Scholar (10) Verdecchia A, De Angelis G, Capocaccia R. Estimation and projections of cancer prevalence from cancer registry data. Stat Med  2002; 21: 3511–26. Google Scholar (11) American Joint Committee on Cancer. Manual for staging of cancer. 3rd ed. Philadelphia (PA): J.B. Lippincott; 1988. Google Scholar (12) Surveillance, Epidemiology, and End Results (SEER) Program (http://www.seer.cancer.gov) SEER*Stat Database: Mortality—All COD, Total U.S. (1969–2001), National Cancer Institute, DCCPS, Surveillance Research Program, Cancer Statistics Branch, released December 2003. Underlying mortality data provided by NCHS (http://www.cdc.gov/nchs). Google Scholar © The Author 2006. Published by Oxford University Press. All rights reserved. For Permissions, please e-mail: journals.permissions@oxfordjournals.org.

Journal

JNCI MonographsOxford University Press

Published: Oct 1, 2006

There are no references for this article.