Originally published as JCO Early Release 10.1200/JCO.2008.17.3914 on December 1 2008

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 Google Scholar Google Scholar Articles by Ozanne, E. M. Articles by Belkora, J. Search for Related Content PubMed PubMed Citation Articles by Ozanne, E. M. Articles by Belkora, J. Social Bookmarking                            What's this?

Sensitivity to Input Variability of the Adjuvant! Online Breast Cancer Prognostic Model

Elissa M. Ozanne, Dejana Braithwaite, Karen Sepucha, Dan Moore, Laura Esserman, Jeffrey Belkora

From the Institute for Technology Assessment and Health Decision Research Unit, Massachusetts General Hospital, Harvard Medical School, Boston, MA; and the Departments of Biostatistics and Epidemiology and Surgery, University of California at San Francisco, San Francisco, CA

Corresponding author: Elissa M. Ozanne, PhD, Institute for Technology Assessment at Massachusetts General Hospital, 101 Merrimac St, 10th Fl, Boston, MA 02114; e-mail: elissa{at}mgh-ita.org

	ABSTRACT

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

 
Purpose Adjuvant! Online (www.adjuvantonline.org) is a software model that predicts the benefit of adjuvant therapy for women with early-stage breast cancer. The model has been validated, is widely consulted, and has been shown to influence patient choices in the clinical setting. Adjuvant! requires the clinician to input patient age, tumor size, grade, hormone receptor status, number of positive lymph nodes, and comorbidity level. Because comorbidity is strongly and independently associated with survival, this study tested the hypothesis that Adjuvant! predictions would be sensitive to comorbidity inputs.

Methods Investigators used single-variable deterministic sensitivity analysis to evaluate the effect of varying each input of the model independently for three representative case examples based on National Comprehensive Cancer Network guidelines (NCCN). The main outcome of interest was 10-year mortality prediction.

Results The analyses show that Adjuvant!'s 10-year mortality predictions are most sensitive to patients’ comorbidity levels and the extent of nodal involvement for the cases, particularly among older women. Comorbidity was the most influential input except in younger women, aged 40 years.

Conclusion The Adjuvant! model is sensitive to patient comorbidity, and impact on the model outputs are significant enough that they are likely to affect physician recommendations and patients’ treatment choices. For example, incorrect assessments of comorbidities could lead physicians to overtreat or undertreat a patient who is in a gray zone relative to the NCCN guidelines. These results point to the importance of accurately assessing comorbidities in patients with breast cancer when using Adjuvant! and highlight the need for a standardized process of comorbidity ascertainment.

	INTRODUCTION

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

 
Treatment decisions regarding adjuvant therapy for breast cancer are challenging. Physicians and patients must estimate the residual risk patients face after surgery in addition to the risks and benefits of adjuvant treatment options to make high-quality decisions about patient care. Recognizing these challenges, a handful of decision support tools have been developed. One such tool, Adjuvant! Online, is a computerized, Web-accessible risk-assessment model that predicts mortality and recurrence risk and the benefit of adjuvant therapy for women with early-stage breast cancer (www.adjuvantonline.org).^1-3 This model has been validated in a large, population-based study⁴ and is widely used. An estimated 75% of oncologists in the United States are reported to consult Adjuvant!,⁵ and it has also been shown that the model influences patient treatment choice in a clinical setting.⁶

Adjuvant! requires six inputs that are well established as powerful predictors of mortality and recurrence: patient age, tumor size, grade, hormone receptor status, number of positive lymph nodes, and comorbidity level. Adjuvant!'s use of widely accepted definitions for the first five inputs helps to minimize errors that may arise due to the subjective nature of user interpretation or judgment, which therefore minimizes variation in the assessment of these inputs. Although five inputs (all except comorbidity level) are typically documented clearly in the patient chart, second opinions or review by multidisciplinary teams may lead to changes in these inputs. In fact, a recent study found 45% of imaging and 29% of pathology reviews resulted in changes to the original chart documentation,⁷ suggesting possible inconsistency of user interpretation of even these well-defined inputs.

In contrast with the other inputs, the assessment of patient comorbidities, however, is often poorly integrated into routine care. This is due in part to the absence of clinically implemented comorbidity scoring systems.⁸ Therefore, the assessment of patient comorbidities adds an additional level of subjectivity, creating a potential barrier to reliability of this model, and possibly undermining the validity of Adjuvant! predictions when used in routine clinical settings. The comorbidity inputs for the Adjuvant! model include the following options: perfect health, minor problems, average for age, major problems (+10), major problems (+20), major problems (+30). The Adjuvant! documentation offers little guidance regarding these definitions of comorbidity and how patient medical histories should be translated into these input options. Rather, the online help files of Adjuvant! and an article by Ravdin et al³ include a short discussion on the comorbidity input, which states that the derivation of these input options stems from the Charlson Comorbidity Index,^9,10 and the suggestion that "major problems" is equivalent to adding 10 years to a woman's age (see Appendix, online only). The model is developed so that multiple major problems can add up to 30 years in age equivalence. Nevertheless, the help files do recognize that "adjustment for serious comorbidity is extremely complex and there are no simple sources for doing this" and "the method of adding years to the chronological age of the patient to adjust for comorbidity...is admittedly overly simplistic," requiring significant interpretation on the part of the clinician.

Although Adjuvant! has been validated against a registry of Canadian outcomes and found to be a good predictive model, such large-scale model validation does not account for local variation owing to user interpretation or judgment. Given the widespread use of this influential model, it is important to determine the extent to which variations in model inputs affect risk projections. Using clinical scenarios, this study examined how variation in comorbidity assessment may impact Adjuvant! predictions. We examined the potential clinical impact of changes in comorbidity assessments by conducting deterministic sensitivity analyses of the Adjuvant! model. We explored the value of increasing the accuracy of the assessment of a patient's comorbidity relative to other Adjuvant! inputs (ie, estrogen receptor [ER] status, extent of nodal involvement). Specifically, we evaluated the extent to which variations in inputs affected the 10-year baseline mortality and recurrence rate predictions and the prediction of adjuvant treatment benefit. We hypothesized that Adjuvant! would be highly sensitive to comorbidity assessment and that, in fact, Adjuvant! would be most sensitive to comorbidity assessment.

	METHODS

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

 
Our study examined how variation in input assessments impacts Adjuvant! predictions when used in routine clinical settings using deterministic sensitivity analysis, a technique that documents the amount of variation in a model's output associated with each input.¹¹ This one-way sensitivity analysis, often displayed using "tornado diagrams," allows us to compare the impact of changing one input at a time while holding all other inputs constant. This analysis determines the relative impact each input has on the model output. A base-case is established initially by using the base values for each input. One by one, each input is then varied between the minimum and maximum values, whereas all of the other input values are held constant. This is repeated for each input and the results are graphed in a tornado diagram, with the inputs presented in decreasing order of influence (determined by the magnitude of the swing from lowest to highest values). Although the model is described qualitatively by Ravdin et al,³ the true dynamics of the model remain unpublished, and therefore, this analysis should be considered a first-order approximation, as adjustments might need to be made for nonmonotonic relationships among variables or interaction terms in the model. The outcomes analyzed were 10-year risk of mortality from all causes, 10-year risk of recurrence, and 10-year prediction of treatment benefit.

Cases
We identified three situations described in the National Comprehensive Cancer Network (NCCN) Practice Guidelines in Oncology where adjuvant therapy recommendations were not clearly defined.¹² These are situations where physicians and patients may consult Adjuvant! and were chosen as representative patient scenarios for the base-cases. We translated each situation into the appropriate inputs for Adjuvant! (Table 1). In the last case (and for patients older than 70 years in the second case), the NCCN guidelines specifically mention that treatment should be tailored to individual patients considering patient comorbid conditions. To illustrate the impact comorbidity assessment has on treatment decisions, we performed deterministic sensitivity analysis for 10-year baseline mortality and recurrence rate predictions. We also generated expected mortality reduction for a second generation chemotherapy regime such as 6 cycles of fluorouracil, epirubicin, and cyclophosphamide for case 3, at each comorbidity level and at the base-case level for all other inputs. Version 8.0 of Adjuvant! was used in the analyses. We analyzed each case example in detail and then extrapolated the results to the general setting of adjuvant care for breast cancer.

	RESULTS

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

View larger version (17K): [in this window] [in a new window] [PowerPoint Slide for Teaching]   Fig 1. Adjuvant! predicted 10-year mortality ranges for different input values in case 1 (node-negative, estrogen receptor [ER] positive, grade 1, tumor < 1 cm. Decision: hormone therapy?). Bars show the minimum and maximum of predicted 10-year mortality for each input, holding the remaining factors constant. The inputs levels for comorbidity are delineated with gold vertical lines.

View larger version (16K): [in this window] [in a new window] [PowerPoint Slide for Teaching]   Fig 2. Adjuvant! predicted 10-year mortality ranges for different input values in case 2 (node-negative, estrogen receptor [ER] positive, grade 3, tumor 1 to 2 cm. Decision: hormone therapy alone v with chemotherapy?). Bars show the minimum and maximum of predicted 10-year mortality for each input, holding the remaining factors constant.

View larger version (11K): [in this window] [in a new window] [PowerPoint Slide for Teaching]   Fig 3. Adjuvant! predicted 10-year mortality ranges for different input values in case 3 (one to three positive nodes, estrogen receptor [ER] negative, grade 2, tumor < 1 cm. Decision: chemotherapy?). Bars show the minimum and maximum of predicted 10-year mortality for each input, holding the remaining factors constant.

The 10-year predicted recurrence rates are presented in Figure 4 for case 1. In these figures, the length of each bar represents the magnitude of the spread in the 10-year predicted recurrence rate between the lowest and highest values for each input. The base-case predicted recurrence rate without adjuvant therapy was 14.9% for the 40-year-old woman in case 1. The nodal involvement was by far the most influential input, followed by tumor size in Adjuvant!'s predictions of recurrence for all ages analyzed.

View larger version (14K): [in this window] [in a new window] [PowerPoint Slide for Teaching]   Fig 4. Adjuvant! predicted 10-year recurrence rates for different input values in case 1 (node-negative, estrogen receptor [ER] positive, grade 1, tumor < 1 cm. Decision: hormone therapy?). Bars show the minimum and maximum predicted 10-year recurrence rates for each input, holding the remaining factors constant.

	DISCUSSION

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

Despite Adjuvant!'s wide clinical use, its inputs, specifically comorbidity, have not been previously subjected to a sensitivity analysis. In commonly cited publications regarding Adjuvant!, the discussion of comorbidity assessment has been omitted entirely.^1-3,13 In addition, the large validation study of the Adjuvant! model⁴ used the default option of "minor problems" for all women in the study. When performing a population validation, the majority of women are likely to fall into this comorbidity category, allowing for successful validation. Although the average patient with breast cancer often has only average problems, this is not true for every patient, and further precision is needed when using this model on an individual patient level. Therefore, it is important to be able to accurately assess comorbidities for individual patients when using the Adjuvant! model. In addition to breast cancer, the Adjuvant! model is also available for lung and colon cancer. The importance of comorbidity is likely to be influential in these settings also, and possibly even more so for lung cancer predictions, given the high prevalence of comorbidities in patients with lung cancer.¹⁴

These results illustrate that changes in the comorbidity input result in significant variations in the model predictions, underscoring the importance of accurately assessing comorbidity in patients with breast cancer. Because there is no standard, widely adopted process to assess comorbidity for input into the model, clinicians may vary in their comorbidity assignment for patients. This is likely to lead to significantly different 10-year mortality assessments for local therapy and predicted benefit from adjuvant treatment. Consider a patient with breast cancer similar to case 3, a 70-year-old woman with an ER-negative, 1-cm, grade 2 tumor, with two lymph nodes involved. She presents with a history of high blood pressure and heart disease. She had a coronary bypass 5 years ago, but has no angina, is able to walk 2 miles daily, and her blood pressure is well controlled. Does this patient have "minor problems," "average for age," or "major problems (+10)," with a corresponding 44%, 48%, or 70% 10-year predicted mortality rate with local therapy alone? The possible range of absolute survival benefit from chemotherapy for this woman varies from 0.3% to 7.5%, depending on comorbidity level. A 0.3% absolute survival benefit is the same range as risk of death from complications of chemotherapy.¹⁵ Therefore, for women with major problems (+30), chemotherapy should not be used. In contrast, a 7.5% absolute survival benefit for women in perfect health exceeds the threshold for most women and would result in a decision to treat with chemotherapy.¹⁶ A clinician is not likely to mistake major problems for perfect health, but distinguishing between "minor problems," "average for age" and "major problems (+10)" is not straightforward. In this case, these slight misclassifications result in differences of 3.6% and 1.1%, respectively, in the 10-year prediction of treatment benefit, which are likely to influence treatment decisions.

Clearly, the model is influenced by comorbidity inputs, and yet little documentation or attention has been given to this input. Because of this apparent inattention, it is additionally concerning that large differences in the model predictions can be seen as a result of one incremental change in the comorbidity input. As seen in Figure 1 and Table A1, the source of the largest differences in model output is not predictable. For example, in case 1, the 40-year-old patient had the largest difference as a result of one incremental change between the major problems (+20) to major problems (+30) input levels (12.2% to 25.9%, a 13.7% difference in predicted mortality). In 60-year-olds, large differences occur at this same change (56.2% to 93.2%, a 37% difference in predicted mortality) and also between the major problems (+10) to major problems (+20) input levels (25.9% to 56.2%, a 30.3% difference in predicted mortality). However, for 75-year-olds, the largest difference is seen between the average health and major problems (+10) input levels (37.9% to 78.1%, a 40.2% difference in predicted mortality). This unpredictability of the model behavior increases the possibility for error in clinical use. It is possible that clinicians are less likely to use Adjuvant! with patients who have significant comorbidities and older patients who are more likely to have equipoise about indications for chemotherapy. Although these results are not unexpected in light of the model description given by Ravdin et al,³ it remains important for the implications of the model use to be understood by all clinicians who might use it.

Another possible error in the clinical use of Adjuvant! stems from the fact that the default comorbidity option for the program is "minor problems." As the life-expectancy of the general population increases, patients with breast cancer often present with one or more comorbid condition, such as heart disease, diabetes, hypertension, or arthritis at the time of diagnosis.^17,18 Importantly, epidemiologic evidence points to the growing burden of comorbidity among patients with breast cancer, and also shows that comorbidity is independently associated with all-cause survival among patients with breast cancer.¹⁹ With cancer death rates highest among people 65 years and older,²⁰ comorbidities are increasingly part of the setting of clinical decision making. To provide quality care and optimize the overall survival benefit among patients with breast cancer, we clearly require better estimates of the nature, severity, and impact of the different comorbidities to reduce any adverse effects of other health problems on cancer treatment of patients with breast cancer. Tools that provide such information should be integrated with models of breast cancer risk, including Adjuvant!, to enable comprehensive projections of survival.

Increasingly, biologic models and molecular tools will help us refine estimates of risk of recurrence and benefits of therapies. However, as Adjuvant! shows us, the impact of therapy must be projected against the background of overall health. In other words, patients must be alive to be at risk of disease or to realize the benefit of therapy. As models improve, it will be even more important to properly project the impact of comorbidities. Some tumor types may have the highest risk of recurrence in the first 5 years after diagnosis, and others 5 to 10 years out. In addition, the use of chemotherapy is likely to affect the recurrence risk in a patients’ first 5 years after treatment. Therefore, it may be useful in future models to provide both 5- and 10-year projections of risk and benefit.

Although our results are persuasive, they should be interpreted with the study limitations in mind. The analysis used was deterministic in nature was not able to account for possible interactions between inputs or nonmonotonic behavior of the Adjuvant! model. In addition, our analysis examined the difference in the model results based on changes across the full range of inputs, which is not likely to represent common errors in the clinical setting. To mitigate this limitation, we examined the effect of incremental changes in the inputs. Finally, although our case examples are thought to be representative of common cases in clinical care, we did not examine all scenarios that could possibly be used with the Adjuvant! model in our analyses.

In summary, our findings show that the Adjuvant! model is often more sensitive to comorbidity assignment than to other inputs. In addition, the changes in the comorbidity outputs are significant enough that they are likely to affect patients’ treatment choices. Our results demonstrate that even with a tumor where biology is uniformly accepted to confer poor prognosis, and the benefit of chemotherapy is thought to be high (eg, node-positive, ER-negative disease), for an older woman with comorbid conditions, this may not be the case. Therefore, assessing patients’ overall health status is key to quality care in this setting. These results point to the importance of accurately assessing comorbidity in patients with breast cancer when using Adjuvant! The ability to facilitate comorbidity assessments so they can consistently and reliably be integrated into the Adjuvant! model is likely to improve the clinical use of the model and the quality of patient decision making regarding adjuvant treatment for breast cancer.

	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: Elissa M. Ozanne, Dejana Braithwaite, Karen Sepucha, Dan Moore, Laura Esserman, Jeffrey Belkora

Collection and assembly of data: Elissa M. Ozanne, Dejana Braithwaite, Karen Sepucha, Dan Moore, Laura Esserman, Jeffrey Belkora

Data analysis and interpretation: Elissa M. Ozanne, Dejana Braithwaite, Karen Sepucha, Dan Moore, Laura Esserman, Jeffrey Belkora

Manuscript writing: Elissa M. Ozanne, Dejana Braithwaite, Karen Sepucha, Dan Moore, Laura Esserman, Jeffrey Belkora

Final approval of manuscript: Elissa M. Ozanne, Dejana Braithwaite, Karen Sepucha, Dan Moore, Laura Esserman, Jeffrey Belkora

	Appendix

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

 
The text below is found in the Adjuvant! help files by following this navigational path:

Online Resources >> Adjuvant! Online Documentation >> Adjuvant! for Breast Cancer online documentation >> Using Adjuvant! >> Entering Patient Information.

Comorbidity
This is an estimate of the general health of the individual for whom the estimates are being made. The default is "minor problems." The possible answers are "perfect health," "minor problems," "average for age," "major problems (+10)," "major problems (+20)," and "major problems (+30)."

The reason for such a broad range of estimates is that chronological age clearly does not define natural mortality rates if comorbidity is not taken into account. The age-specific mortality rates for average women in the United States population (Centers for Disease Control and Prevention, National Center for Health Statistics: Vital Statistics of the United States, vol II, section 6, 1989, p 7), given in Adjuvant! as "average for age," are clearly an overestimate for most women. Much of the mortality at any given age is driven by preexisting health problems. Therefore, most women have natural mortality rates that are going to be predictably somewhat better than that of the general population. The adjustment for the lack of comorbidity is based on the fact that actuaries have recognized (but not precisely defined) a population of "select" patients with no comorbidity who have lower short-term mortality rates (Medical Risks: Trends in Mortality by Age and Time Elapsed: A Reference Volume, Sponsored by the Association of Life Insurance Medical Directors of America and the Society of Actuaries, Praeger, NY, 1990; Transactions of Society of Actuaries, 1982 Reports of Mortality and Morbidity Experience, Society of Actuaries, 1982), which, in older women, may be only one third of the age-adjusted average initially, but come up to average within approximately 10 to 15 years. This "select" population has mortality rates that over a 10-year period will be approximatley one half the expected cumulative rate. This "select" population's natural mortality estimates are provided by selecting "perfect health." This group is based on nonbreast cancer–related mortality rates in the overviews of clinical trials (which also exclude most women with major comorbidities). These rates seem to lie between that expected for the average woman and women in a "select" population. Therefore Adjuvant! selects as a default "minor problems," which is the average of "perfect health" and "average for age."

Adjustment for serious comorbidity is extremely complex, and there are no simple sources for doing this. The impact of having a serious comorbidity is influenced by its severity, time since onset, and other comorbidities. The method of adding years to the chronological age of the patient to adjust for comorbidity is derived from an analysis of the effects of comorbidity of patients with breast cancer on non–breast cancer mortality^9,10 and is admittedly overly simplistic. History of myocardial infarction, congestive heart failure, diabetes, or vascular disease has been shown to increase the risk of mortality, in a way and on average, the equivalent of adding 10 years to the age. Such adjustments are arbitrary and imprecise, but particularly for some patients, may help in evaluating whether adjuvant therapy is worthwhile.

There are also subsets of patients for whom age-specific mortality rates are uncertain. For example, for patients with breast cancer who are BRCA1 positive, mortality rates not related to their initial breast cancer might be expected to be somewhat higher than in the general population because of their much higher risk of second cancers.

	NOTES

 
published online ahead of print at www.jco.org on December 1, 2008

Supported in part by Grant No. MRSG112037 from the American Cancer Society E.M.O.).

Presented in part at the 4th International Shared Decision Making Conference, May 30-June 1, 2007, Freiburg, Germany.

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. Ravdin PM: A computer based program to assist in adjuvant therapy decisions for individual breast cancer patients. Bull Cancer 82:561s-564s, 1995 (suppl 5)[Medline]

2. Ravdin PM: A computer program to assist in making breast cancer adjuvant therapy decisions. Semin Oncol 23:43-50, 1996 (suppl 2)[Medline]

3. Ravdin PM, Siminoff LA, Davis GJ, et al: Computer program to assist in making decisions about adjuvant therapy for women with early breast cancer. J Clin Oncol 19:980-991, 2001[Abstract/Free Full Text]

4. Olivotto IA, Bajdik CD, Ravdin PM, et al: Population-based validation of the prognostic model ADJUVANT! for early breast cancer. J Clin Oncol 23:2716-2725, 2005[Abstract/Free Full Text]

5. Love N: Management of breast cancer in the adjuvant and metastatic settings. Patterns of Care 2, 2005

6. Peele P, Siminoff L, Xu X, et al: Decreased use of adjuvant breast cancer therapy in a randomized controlled trial of a decision aid with individualized risk information. Med Decis Making 25:301-307, 2005[Abstract/Free Full Text]

7. Newman E, Guest A, Helvie M, et al: Changes in surgical management resulting from case review at a breast cancer multidisciplinary tumor board. Cancer 107:2346-2351, 2006[CrossRef][Medline]

8. Hall WH, Ramachandran R, Narayan S, et al: An electronic application for rapidly calculating Charlson comorbidity score. BMC Cancer 4:94, 2004[CrossRef][Medline]

9. Charlson M, Szatrowski TP, Peterson J, et al: Validation of a combined comorbidity index. J Clin Epidemiol 47:1245-1251, 1994[CrossRef][Medline]

10. Charlson ME, Pompei P, Ales KL, et al: A new method of classifying prognostic comorbidity in longitudinal studies: Development and validation. J Chronic Dis 40:373-383, 1987[CrossRef][Medline]

11. Clemen R, Reilly T: Making Hard Decisions. Pacific Grove, CA, Brooks/Cole, 2001

12. National Comprehensive Cancer Network: NCCN Breast Cancer Clinical Practice Guidelines in Oncology version 2.2007. http://www.nccn.org

13. Siminoff LA, Gordon NH, Silverman P, et al: A decision aid to assist in adjuvant therapy choices for breast cancer. Psychooncology 15:1001-1013, 2006[CrossRef][Medline]

14. Tammemagi CM, Neslund-Dudas C, Simoff M, et al: Impact of comorbidity on lung cancer survival. Int J Cancer 103:792-802, 2003[CrossRef][Medline]

15. Smith RE, Bryant J, DeCillis A, et al: Acute myeloid leukemia and myelodysplastic syndrome after doxorubicin-cyclophosphamide adjuvant therapy for operable breast cancer: The National Surgical Adjuvant Breast and Bowel Project Experience. J Clin Oncol 21:1195-1204, 2003[Abstract/Free Full Text]

16. Ravdin PM, Siminoff IA, Harvey JA: Survey of breast cancer patients concerning their knowledge and expectations of adjuvant therapy. J Clin Oncol 16:515-521, 1998[Abstract]

17. Yancik R, Havlik RJ, Wesley MN, et al: Cancer and comorbidity in older patients: A descriptive profile. Ann Epidemiol 6:399-412, 1996[CrossRef][Medline]

18. Satariano WA: Aging, comorbidity, and breast cancer survival: An epidemiologic view. Adv Exp Med Biol 330:1-11, 1993[Medline]

19. Tammemagi CM, Nerenz D, Neslund-Dudas C, et al: Comorbidity and survival disparities among black and white patients with breast cancer. JAMA 294:1765-1772, 2005[Abstract/Free Full Text]

20. American Cancer Society: Cancer Facts and Figures, 2008. American Cancer Society, 2008. Available at: http://www.cancer.org/docroot/STT/content/STT_1x_Cancer_Facts_and_Figures_2008.asp

Submitted April 7, 2008; accepted September 10, 2008.

CiteULike    Complore    Connotea    Del.icio.us    Digg    Facebook    Reddit    Technorati    Twitter    What's this?

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 Google Scholar Google Scholar Articles by Ozanne, E. M. Articles by Belkora, J. Search for Related Content PubMed PubMed Citation Articles by Ozanne, E. M. Articles by Belkora, J. Social Bookmarking                            What's this?