Originally published as JCO Early Release 10.1200/JCO.2007.11.6079 on August 6 2007

This Article Abstract Full Text (PDF) Purchase Article View Shopping Cart Alert me when this article is cited Alert me if a correction is posted Services Email this article to a colleague Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Save to my personal folders Download to citation manager Rights & Permissions Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Anderson, W. F. Articles by Pfeiffer, R. M. Search for Related Content PubMed PubMed Citation Articles by Anderson, W. F. Articles by Pfeiffer, R. M.

Shifting Breast Cancer Trends in the United States

William F. Anderson, Anne S. Reiner, Rayna K. Matsuno, Ruth M. Pfeiffer

From the Biostatistics Branch, Department of Health and Human Services, National Institutes of Health, National Cancer Institute, Division of Cancer Epidemiology and Genetics, Bethesda, MD

Address reprint requests to William F. Anderson, MD, MPH, Biostatistics Branch, Department of Health and Human Services, National Institutes of Health, National Cancer Institute, Division of Cancer Epidemiology and Genetics, EPS, Room 8036, 6120 Executive Blvd, Bethesda, MD 20892-7244; e-mail: wanderso{at}mail.nih.gov

	ABSTRACT

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION AUTHORS' DISCLOSURES OF... AUTHOR CONTRIBUTIONS Appendix REFERENCES

 
Purpose: United States breast cancer incidence rates declined during the years 1999 to 2003, and then reached a plateau. These recent trends are impressive and may indicate an end to decades of increasing incidence.

Methods: To put emerging incidence trends into a broader context, we examined age incidence patterns (frequency and rates) during five decades. We used age density plots, two-component mixture models, and age-period-cohort (APC) models to analyze changes in the United States breast cancer population over time.

Results: The National Cancer Institute's Connecticut Historical Database and Surveillance, Epidemiology, and End Results program collected 600,000+ in situ and invasive female breast cancers during the years 1950 to 2003. Before widespread screening mammography in the early 1980s, breast cancer age-at-onset distributions were bimodal, with dominant peak frequency (or mode) near age 50 years and smaller mode near age 70 years. With widespread screening mammography, bimodal age distributions shifted to predominant older ages at diagnosis. From 2000 to 2003, the bimodal age distribution returned to dominant younger ages at onset, similar to patterns before mammography screening. APC models confirmed statistically significant calendar-period (screening) effects before and after 1983 to 1987.

Conclusion: Breast cancer in the general United States population has a bimodal age at onset distribution, with modal ages near 50 and 70 years. Amid a background of previously increasing and recently decreasing incidence rates, breast cancer populations shifted from younger to older ages at diagnosis, and then back again. These dynamic fluctuations between early-onset and late-onset breast cancer types probably reflect a complex interaction between age-related biologic, risk factor, and screening phenomena.

	INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION AUTHORS' DISCLOSURES OF... AUTHOR CONTRIBUTIONS Appendix REFERENCES

 
Female breast cancer incidence rates in the United States peaked in 1998 at 77.1 per 100,000 woman-years,¹ and then began to decline, primarily among older women.^1-4 Since 1998, overall breast cancer incidence rates have decreased 9.8%,¹ with a decrease of 12% among women age 50 to 69 years.⁴ These current trends may indicate an end to decades of increasing breast cancer incidence, spurring debate and research.⁵

The most recent breast cancer declines have been attributed to a reduction in hormone replacement therapy usage, after the Women Health Initiative (WHI) report in July 2002.^4,6-8 However, decreasing incidence before 2002 cannot be attributed to the WHI announcement; other birth-cohort (risk factor or exposure) and/or calendar-period (screening) effects must play a role. To develop additional etiologic clues for recent breast cancer declines as well as to provide a broader context for breast cancer incidence in the United States, we examined breast cancer incidence patterns (rates and age distributions) during five decades.

We used incidence data for the US female breast cancer population from the National Cancer Institute's Connecticut Historical Database (CHD) and the Surveillance, Epidemiology, and End Results (SEER) program, covering a calendar period from 1950 to 2003. We analyzed the age distributions at diagnosis with age-density plots and two-component mixture models. We also plotted breast cancer incidence rates using age-period-cohort (APC) models to assess birth-cohort (risk factor or exposure) and calendar-period (screening) effects.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION AUTHORS' DISCLOSURES OF... AUTHOR CONTRIBUTIONS Appendix REFERENCES

 
Data on in situ and invasive female breast cancer occurrences were obtained from the CHD (1950 to 1972) and the 13-Registry Database of the SEER (1973 to 2003).⁹ SEER's 13-Registry Database included tumor registries in Atlanta, Connecticut, Detroit, Hawaii, Iowa, New Mexico, San Francisco-Oakland, Seattle-Puget Sound, Utah, Los Angeles, San Jose-Monterey, rural Georgia, and the Alaskan Native Tumor Registry.

Tumor characteristics were categorized according to behavior (in situ or invasive), SEER historical stage A, and histologic grade. SEER's historical stage A described in situ disease as noninvasive breast cancer, local invasive disease as limited to the breast, invasive regional disease as limited to nearby lymph nodes or other organs, and distant disease as systemic metastases. In a previous study using the CHD,¹⁰ we demonstrated increasing incidence rates for tumors classified as in situ and local SEER stages versus decreasing temporal trends for tumors designated regional and distant SEER stages. On the basis of this prior experience, we also grouped SEER historic stage A into early (in situ + local) or late stages at diagnosis (regional + distant). Histologic grade was dichotomized into low grade (grade 1, well-differentiated tumors, and grade 2, moderately differentiated tumors) and high grade (grade 3, poorly differentiated, and grade 4, undifferentiated or anaplastic).

Statistical Analysis
Median ages at diagnosis were compared using the Kruskal-Wallis test (PROC NPAR1WAY, SAS version 9.1; SAS Institute, Cary, NC). We calculated age-adjusted breast cancer incidence rates (2000 US standard population), using case and population data from SEER*Stat 6.2.3.⁹ Relative risks were calculated as incidence rate ratios (IRRs) with 95% CIs for a given characteristic compared with a referent characteristic with an assigned IRR of 1.0.

Kernel density estimation in S-Plus version 7.06 (Insightful Corporation, Seattle, WA) was used to produce smoothed age distribution curves at diagnosis, as previously described.^11,12 In brief, the kernel smoother estimated the underlying probability density function for breast cancer incidence by age at diagnosis in single years. We graphed the corresponding age density plots for six 10-year calendar periods to capture breast cancer occurrences before widespread screening mammography (1950 to 1959 and 1960 to 1969), during the introduction of screening mammography (1970 to 1979 and 1980 to 1989), and after a saturation phase for widespread screening mammography (1990 to 1999 and 2000 to 2003).

Two-component mixture models characterized the age distributions overall and for each calendar period. The two-component mixture model considered the probability of a breast cancer occurrence being in an earlier or later group (or population) according to age at diagnosis. The probability density function of y = (x – 1)/, where x denotes the age at diagnosis, was g(y) = pf(y; ₀) + (1 – p) f(y; ₁). The probability density function f(y, ₀) reflects early ages at diagnosis, whereas the probability function f(y, ₁) corresponds to late ages at diagnosis. The mixing probability p represents the proportion of women in the early breast cancer group and was used to summarize the impact of mammography on the age distribution over time.

We used two different densities for the component densities f(y, ): normal probability densities f = (y, µ, ), with mean µ and standard deviation ; and semi-nonparametric densities that multiply the normal density with a polynomial component, allowing for skewness and heavier tails than the normal density. We chose a polynomial of degree 1, yielding f(y, ) = (y, µ, )(0 + 1)², properly standardized.¹³ Model fits for the different parameterizations were compared using the Akaike information criterion (AIC).¹⁴ In the tables we present the models and AIC values for the models with the best fit. Additional details regarding the estimation of the parameters (, p, ₀, ₁) are described elsewhere.^15,16,16a

Age-specific incidence rates were calculated by 14 5-year age groups (ages 20 to 24, 25 to 29, ..., 80 to 84, and 85+ years). To account for age, calendar period, and birth-cohort effects simultaneously, we analyzed age-specific incidence data for the years 1973 to 2002 with APC models, using Poisson regression (PROC GENMOD, SAS 9.1). We used 13 5-year age intervals (20 to 24, 25 to 29, ..., 80 to 84), six 5-year calendar periods of diagnosis (1973 to 1977, 1978 to 1982, ..., 1998 to 2002), and 18 5-year birth cohorts (referred to by the last year of birth: 1893, 1898, ..., 1978).

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION AUTHORS' DISCLOSURES OF... AUTHOR CONTRIBUTIONS Appendix REFERENCES

 
Descriptive Statistics
SEER's 13-Registry Database collected data for 597,167 female breast cancer patients (1973 to 2003), including in situ (n = 78,054), invasive (n = 499,861), and unknown stages (n = 19,252; listed in Table 1). The overall median age at diagnosis was 61 years. Median ages initially increased, and then decreased over time (Table 2); that is, from 60 years (1973 to 1979) to 63 years (1980 to 1989) to 62 years (1990 to 1999) to 60 years (2000 to 2003; P < .001). Overall incidence rates increased nearly 1.5-fold from 151.3 per 100,000 woman-years (1973 to 1979) to 225.1 per 100,000 woman-years (2000 to 2003).

View larger version (19K): [in this window] [in a new window] [PowerPoint Slide for Teaching]   Fig 1. Age-specific incidence trends according to early versus late Surveillance, Epidemiology, and End Results stage (A: 1973 to 1979, B: 1980 to 1989, C: 1990 to 1999, and D: 2000 to 2003) and low versus high tumor grade (E: 1973 to 1979, F: 1980 to 1989, G: 1990 to 1999, and H: 2000 to 2003). Age-specific relative risks were expressed as an incidence rate ratio, with a given rate compared to the overall rates.

Age Incidence Patterns
As did median ages at diagnosis (Table 2), density plots shifted from younger to older ages, and then back to younger ages at diagnosis (Fig 2). Before widespread screening mammography (1950 to 1959 and 1960 to 1969; Figs 2A and 2B), age distributions at diagnosis were bimodal with a dominant peak frequency (or mode) near age 50 years and a smaller mode near age 70 years. During the calendar period 1970 to 1979, the age distribution began to shift to older ages in both the CHD and the SEER 13-Registry Database (Fig 2C), obscuring the previous bimodal patterns in Connecticut. Bimodal age distributions began to re-emerge during the calendar period 1980 to 1989 (Fig 2D), becoming fully developed during 1990 to 1999 (Fig 2E). During 1980 to 1989 and 1990 to 1999, bimodal age distributions had a dominant late-onset mode and smaller early-onset mode. During 2000 to 2003, the bimodal age distribution returned to predominantly younger ages at onset, with a peak frequency near age 50 years (Fig 2F), similar to the density plots before mammography (Figs 2A and 2B).

View larger version (22K): [in this window] [in a new window] [PowerPoint Slide for Teaching]   Fig 2. Age distributions at diagnosis (in situ plus invasive breast cancer) for the Connecticut Historical Database and the SEER, Surveillance, Epidemiiology, and End Results (SEER) -13 Registry Database by six calendar periods (A: 1950 to 1959, B: 1960 to 1969, C: 1970 to 1979, D: 1980 to 1989, E: 1990 to 1999, and F: 2000 to 2003). Bimodal age distributions changed over time, shifting from younger to older ages at diagnosis, and then back again.

For all calendar periods (Table 2), the data fit semi-nonparametric densities better than normal probability densities; therefore, only semi-nonparametric models (single density and mixture) are presented in Table 2. During the calendar period 1973 to 1979, a single semi-nonparametric model with mean age of 49 years provided better fit than a two-component mixture model, consistent with a lack of separation between the bimodal peak frequencies in Figure 2C. That is, during 1973 to 1979, the AIC value for a single density (–243,859) was greater than the AIC value for a mixture (–244,973; ie, a smaller negative value). In contrast, during the remaining three time periods (1980 to 1989, 1990 to 1999, and 2000 to 2003), the data fit a two-component semi-nonparametric mixture model better than a single density as assessed by the AIC. For example, during the calendar-period 2000 to 2003, the AIC value for a mixture (–519,847) was greater than the AIC value for a single density (–522,813). From 1990 to 1999 to 2000 to 2003, the probability of being in the early-onset group increased significantly from 62% (95% CI, 59.5% to 65.2%) to 69% (95% CI, 66.0% to 72.6%), consistent with a shift toward earlier age distributions at diagnosis (Figs 2E and 2F). Notably, although the fraction of patients in the early-onset and late-onset breast cancer populations varied over time, modal ages remained near 50 and 70 years for all time periods.

Similar to median ages at diagnosis, density plots, and two-component mixture models, the age-specific incidence rate curves also drifted toward earlier age distributions at diagnosis (Fig 3). During 1973 to 1979, rates increased rapidly until age 50 years, paused, and then continued to increase steadily among older women. The pause or inflection in age-specific incidence rates around age 50 years has been termed Clemmesen's hook and has been attributed to menopause.^17,18 With each succeeding decade, a rising bend developed in the age-specific incidence rate curve among women ages 40 to 80 years. Rates then decreased markedly for the oldest women.

View larger version (14K): [in this window] [in a new window] [PowerPoint Slide for Teaching]   Fig 3. Age-specific incidence rates for the Surveillance, Epidemiology, and End Results 13 Registry Database by four calendar periods. For all time periods, age-specific rates increased rapidly until age 50 years, paused at the so-called Clemmesen's hook, and then rose at a slower pace. After Clemmesen's hook, a bulge developed in the rate curve for succeeding time periods.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION AUTHORS' DISCLOSURES OF... AUTHOR CONTRIBUTIONS Appendix REFERENCES

 
Breast cancer in the general US population has a bimodal age of onset distribution, with modal ages near 50 and 70 years. During five decades, these bimodal breast cancer populations shifted from younger to older ages at diagnosis, and then back to younger age groups. To explore further the dynamic fluctuations between early-onset and late-onset breast cancer types, we combined descriptive techniques (density plots and age incidence patterns) with structured biomathematical models (APC and mixture models).

Before widespread screening mammography in the early 1980s, changing reproductive risk factor patterns purportedly dominated breast cancer incidence trends,^20-22 promoting tumors of low malignant potential among postmenopausal women (Figs 2A to C).^23,24 After the implementation of screening mammography,^10,25-28 there was an additional increase of indolent tumors among older women (Fig 2D). Screening mammography attained the Healthy People 2010 Objective in the 2000 National Health Interview Survey, with 70.1% of women having biennial mammograms.²⁹ Almost simultaneously, breast cancer populations in SEER began to shift from later to earlier ages at onset (Fig 2E), probably because prevalent older screened breast cancer patients were removed from the general population. Recent declines in hormone replacement therapy usage after the July 2002 WHI announcement have likely accelerated this decreasing incidence trend among older women (Fig 2F).^2-4

Age-specific incidence rates also changed after screening mammography in US (Fig 3) as well as in Europe.^26-28 After widespread screening mammography in the early 1980s, age-specific rates among women age 40 to 80 years increased steadily from their baseline rate during 1973 to 1979, characterized by an increasing bend after Clemmesen's hook. The increasing bend in rates between ages 40 to 80 years has been attributed to a progressive increase in incidence from one generation (or birth cohort) to the next³⁰ (ie, a greater increase in breast cancer incidence among younger than older women). This age-specific shift in the US breast cancer population was also seen in APC models of breast cancer incidence. In addition, Holford et al²² recently presented APC models for in situ and invasive breast cancer in the CHD and SEER from 1940 to 2000, and found incidence rate trends like ours.

Median ages at diagnosis and two-component mixture models (Table 2) also drifted from younger to older ages, and then back again, similar to density plots (Fig 2). Before widespread screening mammography (Figs 2A and 2B), US breast cancer populations were bimodal, with a prominent early-onset peak frequency (or mode) near age 50 years along with a smaller late-onset mode at approximately 70 years. During the 1970s (Fig 2C; Table 2), the breast cancer population began to transition toward later ages at onset, obscuring the earlier separation between the bimodal peak frequencies. During this time period, a single density with a median age of 60 years fit the data better than a mixture model (Table 2). After widespread mammography screening during 1980 to 1989 and 1990 to 1999, there was a well-defined bimodal age distribution with prominent late-onset peak frequency near age 70 years. On reaching the mammography screening saturation point of 70.1%,²⁹ predominant early-onset age distributions returned during 2000 to 2003.

A major strength of this study was its large-scale population-based design. The SEER 13-Registry database covers 14% of the US population,⁹ with meticulous and consistent data collection and standards. SEER rates are considered to be nationally representative. In addition, we complemented our descriptive tools (density plots and age-specific incidence rate curves) with a more structured biomathematical approach (APC and mixture models).

Mixture models, however, rely on the assumption of two different subpopulations. To check the validity of this assumption, we compared the fit of a mixture model to that of a single density. We also tested models of increased complexity (semi-nonparametric densities v normal probability densities) to avoid identifiability issues due to overfitting. Indeed, bimodal breast cancer populations fit the data better than a single density for all calendar periods (except 1973 to 1979), even after allowing for skewness and heavy tails. Nonetheless, hypothesis-driven studies will be required to confirm definitively dual breast cancer populations within the general US population.

Another potential criticism is the inclusion of data for patients with in situ breast cancer, which might be affected by overdiagnosis.²⁵ However, in a sensitivity analysis, we excluded in situ breast cancers and observed similar trends for invasive breast cancer occurrences only (Appendix Fig A2, online only). Registry-based analyses also lack individual-level risk factor and screening information. A change in the age distribution pattern for a cancer population might be explained by a risk factor for which the distribution in the general population was changing differentially by age, such as obesity. However, an increasing prevalence of obesity would not explain the reversal of the age distribution patterns during 2000 to 2003.

In sum, amid long-term increasing and more recent decreasing breast cancer trends, the age structure for the breast cancer population in the United States demonstrated shifting bimodal age distributions. Although the fraction of patients in the early-onset and late-onset breast cancer populations varied over time, the modal ages remained unchanged near ages 50 and 70 years. We have observed similar modal ages according to different histopathologic subtypes, tumor characteristics, hormone receptor expression, and molecular signatures.^16,31

Shifting bimodal breast cancer populations support the notion that breast cancer is a heterogeneous disease with a complex interplay between age-specific, risk factor (exposure), and screening effects. Future studies that include the assessment of age-related biologic, risk factor, and screening phenomena are needed to examine further the conceptual framework for a bimodal breast cancer model. This effort is warranted given the implications for breast cancer intervention. If breast cancer is fundamentally heterogeneous, with two or more distinct breast cancer types, we need a stratified approach for breast cancer prevention and treatment.^32,33

	AUTHORS’ DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION AUTHORS' DISCLOSURES OF... AUTHOR CONTRIBUTIONS Appendix REFERENCES

 
The author(s) indicated no potential conflicts of interest.

	AUTHOR CONTRIBUTIONS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION AUTHORS' DISCLOSURES OF... AUTHOR CONTRIBUTIONS Appendix REFERENCES

 
Conception and design: William F. Anderson, Anne S. Reiner, Rayna K. Matsuno, Ruth M. Pfeiffer

Administrative support: William F. Anderson

Provision of study materials or patients: William F. Anderson

Collection and assembly of data: William F. Anderson, Anne S. Reiner, Rayna K. Matsuno, Ruth M. Pfeiffer

Data analysis and interpretation: William F. Anderson, Anne S. Reiner, Rayna K. Matsuno, Ruth M. Pfeiffer

Manuscript writing: William F. Anderson, Anne S. Reiner, Rayna K. Matsuno, Ruth M. Pfeiffer

Final approval of manuscript: William F. Anderson, Anne S. Reiner, Rayna K. Matsuno, Ruth M. Pfeiffer

	Appendix

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION AUTHORS' DISCLOSURES OF... AUTHOR CONTRIBUTIONS Appendix REFERENCES

 

View larger version (21K): [in this window] [in a new window] [PowerPoint Slide for Teaching]   Fig A1. Age-specific incidence trends according to in situ, local, regional, and distant Surveillance, Epidemiology, and End Results stages (A: 1973 to 1979, B: 1980 to 1989, C: 1990 to 1999, and D: 2000 to 2003). Age-specific relative risks were expressed as an incidence rate ratio, with a given rate compared to the overall rates.

View larger version (22K): [in this window] [in a new window] [PowerPoint Slide for Teaching]   Fig A2. Age distributions at diagnosis (invasive breast cancer) for the Connecticut Historical Database and the Surveillance Epidemiology, and End Results (SEER)-13 Registry Database by six calendar periods (A: 1950 to 1959, B: 1960 to 1969, C: 1970 to 1979, D: 1980 to 1989, E: 1990 to 1999, and F: 2000 to 2003). Bimodal age distributions changed over time, shifting from younger to older ages at diagnosis, and then back again.

	ACKNOWLEDGMENTS

 
We thank Susan S. Devesa, PhD, for her editorial comments.

	NOTES

 
published online ahead of print at www.jco.org on July 30, 2007.

Supported in part by the Intramural Research Program of the National Institutes of Health/National Cancer Institute.

Authors’ disclosures of potential conflicts of interest and author contributions are found at the end of this article.

	REFERENCES

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION AUTHORS' DISCLOSURES OF... AUTHOR CONTRIBUTIONS Appendix REFERENCES

 
1. Ries LAG, Melbert D, Krapcho M, et al: SEER Cancer Statistics Review, 1975-2004, National Cancer Institute Bethesda, MD, based on November 2006 SEER data submission, posted to the SEER Web site, 2007. http://seer.cancer.gov/

2. Howe HL, Wu X, Ries LA, et al: Annual report to the nation on the status of cancer, 1975-2003, featuring cancer among U.S. Hispanic/Latino populations. Cancer 107:1711-1742, 2006[CrossRef][Medline]

3. Clarke CA, Glaser SL, Uratsu CS, et al: Recent declines in hormone therapy utilization and breast cancer incidence: Clinical and population-based evidence. J Clin Oncol 24:e49-e50, 2006[Free Full Text]

4. Ravdin PM, Cronin KA, Howlader N, et al: The decrease in breast-cancer incidence in 2003 in the United States. N Engl J Med 356:1670-1674, 2007[Abstract/Free Full Text]

5. McNeil C: Breast cancer decline mirrors fall in hormone use, spurs both debate and research. J Natl Cancer Inst 99:266-267, 2007[Free Full Text]

6. Hersh AL, Stefanick ML, Stafford RS: National use of postmenopausal hormone therapy: Annual trends and response to recent evidence. JAMA 291:47-53, 2004[Abstract/Free Full Text]

7. Risks and benefits of estrogen plus progestin in healthy postmenopausal women: Principal results from the Women's Health Initiative randomized controlled trial. JAMA 288:321-333, 2002[Abstract/Free Full Text]

8. Glaser SL, Clarke CA, Gomez SL, et al: Cancer surveillance research: A vital subdiscipline of cancer epidemiology. Cancer Causes Control 16:1009-1019, 2005[CrossRef][Medline]

9. SEER: Surveillance, Epidemiology, and End Results (SEER) Program. SEER*Stat Database: Incidence-Connecticut Historical, Aug 1999 Sub (1935-1997), National Cancer Institute, DCCPS, Surveillance Research Program, Cancer Statistics Branch, 2006. http://seer.cancer.gov/

10. Anderson WF, Jatoi I, Devesa SS: Assessing the impact of screening mammography: Breast cancer incidence and mortality rates in Connecticut (1943-2002). Breast Cancer Res Treat 99:333-340, 2006[CrossRef][Medline]

11. Silverman BW: Density Estimation for Statistics and Data Analysis. London, United Kingdom, Chapman and Hall, 1986

12. Anderson WF, Chatterjee N, Ershler WB, et al: Estrogen receptor breast cancer phenotypes in the Surveillance, Epidemiology, and End results database. Breast Cancer Res Treat 76:27-36, 2002[CrossRef][Medline]

13. Gallant AR, Nychka DW: Semi-nonparametric maximum likelihood estimation. Econometrica 55:363-390, 1987[CrossRef]

14. Akaike H: Information theory and an extension of the maximum likelihood principle, in Petrov BN, Csaki F (eds): 2nd International Symposium on Information Theory. Budapest, Hungary, 1973, pp 267-281

15. Pfeiffer RM, Gail MH, Brown LM: Probability of Helicobacter pylori infection based on IgG levels and other covariates using a mixture model. J Epidemiol Biostat 5:267-275, 2000[Medline]

16. Anderson WF, Pfeiffer RM, Dores GM, et al: Comparison of age frequency distribution patterns for different histopathologic types of breast carcinoma. Cancer Epidemiol Biomarkers Prev 15:1899-1905, 2006[Abstract/Free Full Text]

16. Pfeiffer RM, Carroll RJ, Wheeler W, et al: Combining assays for estimating prevalence of human herpesvirus 8 infection using multivariate mixture models. Biostatistics (in press)

17. Clemmesen J: Carcinoma of the breast. Br J Radiol 21:583-590, 1948[Medline]

18. Lilienfeld AM, Johnson EA: The age distribution in female breast and genital cancers. Cancer 8:875-882, 1955[CrossRef][Medline]

19. Feuer EJ: Modeling the impact of adjuvant therapy and screening mammography on U.S. Breast cancer mortality between 1975 and 2000: Introduction to the problem. J Natl Cancer Inst Monogr 36:2-6, 2006[Free Full Text]

20. Chu KC, Tarone RE, Kessler LG, et al: Recent trends in U.S. breast cancer incidence, survival, and mortality rates. J Natl Cancer Inst 88:1571-1579, 1996[Abstract/Free Full Text]

21. Tarone RE: Breast cancer trends among young women in the United States. Epidemiology 17:588-590, 2006[CrossRef][Medline]

22. Holford TR, Cronin KA, Mariotto AB, et al: Changing patterns in breast cancer incidence trends. J Natl Cancer Inst Monogr 19-25, 2006[Abstract/Free Full Text]

23. Fox MS: On the diagnosis and treatment of breast cancer. JAMA 241:489-494, 1979[Abstract]

24. Adami H-O: Breast cancer incidence and mortality: Aspects on aetiology, time trends and curability. Acta Chir Scand 519:9-14, 1984 (suppl)

25. Ernster VL, Ballard-Barbash R, Barlow WE, et al: Detection of ductal carcinoma in situ in women undergoing screening mammography. J Natl Cancer Inst 94:1546-1554, 2002[Abstract/Free Full Text]

26. Hemminki K, Rawal R, Bermejo JL: Mammographic screening is dramatically changing age-incidence data for breast cancer. J Clin Oncol 22:4652-4653, 2004[Free Full Text]

27. Hemminki K, Bermejo JL: Effects of screening for breast cancer on its age-incidence relationships and familial risk. Int J Cancer 117:145-149, 2005[CrossRef][Medline]

28. Jonsson H, Johansson R, Lenner P: Increased incidence of invasive breast cancer after the introduction of service screening with mammography in Sweden. Int J Cancer 117:842-847, 2005[CrossRef][Medline]

29. Swan J, Breen N, Coates RJ, et al: Progress in cancer screening practices in the United States: Results from the 2000 National Health Interview Survey. Cancer 97:1528-1540, 2003[CrossRef][Medline]

30. Parkin DM, Bray FI, Devesa SS: Cancer burden in the year 2000: The global picture. Eur J Cancer 37:S4-S66, 2001[Medline]

31. Anderson WF, Jatoi I, Devesa SS: Distinct breast cancer incidence and prognostic patterns in the NCI's SEER program: Suggesting a possible link between etiology and outcome. Breast Cancer Res Treat 90:127-137, 2005[CrossRef][Medline]

32. Burstein HJ, Winer EP: HER2 or not HER2: That is the question. J Clin Oncol 23:3656-3659, 2005[Free Full Text]

33. Anderson WF, Matsuno RK: Breast cancer heterogeneity: A mixture of at least two main types? J Natl Cancer Inst 98:948-951, 2006[Free Full Text]

Submitted March 6, 2007; accepted June 4, 2007.

This article has been cited by other articles:

				A. N.A. Tosteson, N. K. Stout, D. G. Fryback, S. Acharyya, B. A. Herman, L. G. Hannah, E. D. Pisano, and for the DMIST Investigators Cost-Effectiveness of Digital Mammography Breast Cancer Screening Ann Intern Med, January 1, 2008; 148(1): 1 - 10. [Abstract] [Full Text] [PDF]

This Article Abstract Full Text (PDF) Purchase Article View Shopping Cart Alert me when this article is cited Alert me if a correction is posted Services Email this article to a colleague Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Save to my personal folders Download to citation manager Rights & Permissions Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Anderson, W. F. Articles by Pfeiffer, R. M. Search for Related Content PubMed PubMed Citation Articles by Anderson, W. F. Articles by Pfeiffer, R. M.