SEARCH

SEARCH BY CITATION

Keywords:

  • breast cancer;
  • competing mortality;
  • noncancer mortality;
  • competing risks;
  • racial disparities

Abstract

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. CONFLICT OF INTEREST DISCLOSURES
  8. REFERENCES

BACKGROUND:

Death in the absence of disease recurrence (competing mortality) is an important determinant of disease-free survival (DFS) in early breast cancer. The authors sought to identify predictors of this event using competing risks modeling.

METHODS:

A cohort study was made of 1231 consecutive women with stage I to II invasive breast cancer diagnosed between 1986 and 2004, treated with breast conservation therapy. Median follow-up was 82 months. The authors used a parametric competing risks regression model to analyze factors associated with the cumulative incidence of competing mortality. They generated a risk score from the model coefficient estimates and stratified patients according to low and high risk score for analysis.

RESULTS:

Ten-year DFS was 69.7% (95% confidence interval [CI], 66.2%-72.9%). The 10-year cumulative incidence of locoregional recurrence (LRR) was 4.4% (95% CI, 3.0%-5.8%), distant recurrence was 7.1% (95% CI, 5.4%-8.9%), and competing mortality was 18.7% (95% CI, 15.9%-21.6%). On multivariate analysis, competing mortality was associated with increasing age (hazard ratio [HR], 1.83 per 10 years; 95% CI, 1.58-2.12), black race (HR, 1.71; 95% CI, 1.17-2.51), and comorbid disease (HR, 1.93, 95% CI, 1.40-2.65). Ten-year cumulative incidences of competing mortality, locoregional recurrence, and distant recurrence for patients at low (n = 638) versus high (n = 593) risk of competing mortality were 7.2% versus 30.6% (P < .001), 4.4% versus 4.4% (P = .97), and 8.6% versus 5.6% (P = .12), respectively.

CONCLUSIONS:

Competing mortality is an important event influencing 10-year DFS in early breast cancer and is associated with increasing age, black race, and comorbid disease. Stratifying patients according to competing mortality risk may be useful in designing clinical trials. Cancer 2010. © 2010 American Cancer Society.

Disease-free survival (DFS) and overall survival (OS) are often used as primary endpoints in cancer clinical trials. Both represent composite endpoints that include cancer-specific (eg, cancer recurrence or mortality) and nonspecific (eg, noncancer mortality) outcomes as events.1 Some problems with composite endpoints arise in patients at high risk of noncancer mortality. These have implications for how outcome data are reported and analyzed and for design of clinical trials.

First, it is not possible to know the extent to which cancer-specific events contribute to a composite endpoint without analyzing noncancer (ie, competing) mortality separately. In such competing risks situations, analysis of cumulative incidence functions has been advocated as a way to decompose composite endpoints into incidences of constituent events.2 Second, because an individual who dies from causes unrelated to cancer is no longer at risk of cancer-specific events of interest, the specific effects of cancer treatments cannot be estimated as precisely in patients at high risk of competing mortality.3 Third, clinical trials of cancer therapies that use composite primary endpoints could be underpowered to show a benefit when they involve patients at high risk of competing mortality.4

Improved cancer therapies are likely to have minimal benefit in patients at high risk of noncancer mortality. It is possible that for such patients, more intensive management of comorbid illnesses or social programs aimed at behavior modification5, 6 would have a greater effect on DFS and OS than cancer therapies. Early breast cancer patients are at higher risk of noncancer mortality than breast cancer mortality.7, 8 Clinical trials of breast cancer therapies might be more efficiently designed if patients were stratified according to competing mortality risk, or if effective noncancer interventions were coimplemented for high-risk patients. Identifying factors associated with competing mortality is thus an important endeavor in breast and other cancers, to identify which patients stand the most to gain (on the margin) from improved cancer therapy.

Although many studies have analyzed factors associated with events such as locoregional recurrence,9-11 distant recurrence,9 and second malignancies,12 fewer have analyzed factors associated with competing mortality. Studies in breast cancer have shown that age, race, and comorbid disease are associated with noncancer mortality.7, 8, 13-15 However, these studies analyzed cancer death, rather than recurrence, as a competing risk for noncancer mortality (ie, competing as a first event). This approach is not ideal, because patients with recurrent disease may die of noncancer causes, yet still benefit from additional cancer treatment. Patients who die without disease recurrence, in contrast, would be expected to derive minimal benefit from additional cancer therapy. We therefore sought to model predictors of competing mortality in the context of the competing risk of recurrence.

Multivariate regression models of cumulative incidence functions are well suited for analyzing competing risks data.16, 17 We hypothesized that these models would be useful for identifying factors associated with competing mortality, and sought to develop methods to stratify patients with early breast cancer according to competing mortality risk.

MATERIALS AND METHODS

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. CONFLICT OF INTEREST DISCLOSURES
  8. REFERENCES

Patients

The study population included patients aged ≥40 years with American Joint Committee on Cancer pathologic stage I to II invasive breast carcinoma treated with lumpectomy and adjuvant radiation therapy. Patients treated with mastectomy or lumpectomy alone were excluded. Patients with bilateral breast cancer or a history of cancer diagnosis within the previous 10 years (other than a squamous cell or basal cell skin cancer) were also excluded. Our analysis was restricted to 1231 patients meeting the above criteria, treated between January 1986 and May 2004 at 1 of 3 centers: 1) the University of Illinois-Chicago, a university facility in west Chicago serving a socioeconomically disadvantaged population; 2) La Grange Memorial Hospital, a suburban satellite facility; and 3) the University of Chicago, a tertiary referral center in south Chicago. All data were collected prospectively on a protocol designed to analyze outcomes for breast cancer patients treated with lumpectomy and adjuvant radiation therapy. The data were not collected specifically for the purpose of this study, so we could not prospectively control which data were collected for this analysis. The protocol was approved by the institutional review board at each institution.

Treatment

Surgery according to the protocol consisted of partial mastectomy with axillary staging, consisting of either axillary lymph node dissection or sentinel lymph node biopsy followed by axillary dissection as appropriate. As part of the protocol, data were collected on radiation therapy, which typically consisted of daily whole breast photon radiation followed by an electron boost to the tumor bed. Similarly, data were collected on chemotherapy, which typically consisted of doxorubicin (Adriamycin) and cyclophosphamide, with or without a taxane; cyclophosphamide, doxorubicin Adriamycin, and fluorouracil; or cyclophosphamide, methotrexate, and fluorouracil. Adjuvant therapy with tamoxifen was recorded, but the use of aromatase inhibitors was not standard for most of the study period, and data on their use were not available.

Outcomes Evaluation

As part of the protocol, follow-up was scheduled at 3- to 6-month intervals for the first 2 years, 6-month intervals for years 3 to 5, and on a yearly basis thereafter. Unilateral mammograms were recommended every 6 months on the affected breast, and bilateral mammograms were recommended yearly. Chest x-ray and liver function tests were recommended yearly in the absence of symptoms or findings worrisome for metastatic spread, and more extensive testing was recommended as appropriate to the clinical situation. We used information from these assessments to identify DFS events. Follow-up ended May 31, 2007.

To ensure completeness of data for patients receiving portions of their care at outside facilities, relevant outside records were obtained at the time of initial consultation, and outcome data were linked to the University of Chicago cancer registry database. This registry tracks patient outcomes under an agreement with area hospitals and incorporates data from the Social Security Death Index and Illinois Department of Public Health websites. The registry determines cause of death from multiple sources, including hospital records and death certificate information.

Definitions

Race was self-described by the patient at the time of consultation, and patients were categorized as black versus nonblack, based on previous studies associating black or African American race with competing mortality.7, 15 Tumors were recorded as “mammographically detected” if they were detected by screening mammogram and were not associated with clinical signs or symptoms. Patients who were either estrogen receptor or progesterone receptor positive by immunohistochemistry were classified as hormone receptor positive. Comorbid disease data was elicited from the patients' past medical and medication history from the surgical, medical, and/or radiation oncologists' initial consultation. Patients were classified as low or high socioeconomic status (SES) by estimating the median household income for each patient using 2000 census tract data linked to the patient's Zip code, with “low” SES indicating a household income below the median.

Comorbid disease was classified as a binary variable based on whether the consulting physicians documented a history of any of the following: diabetes mellitus (DM), chronic obstructive pulmonary disease (COPD), hypertension, cardiovascular disease, or cerebrovascular accident. These diseases were specifically analyzed because of their relatively high prevalence and expected correlation with noncancer mortality.15 The number of comorbid diseases and dummy variables for each comorbidity were also analyzed in different model specifications, but for the final model all comorbid diseases were combined into 1 binary term.

Statistical Analysis

For evaluation of univariate covariate effects on cumulative probability of cause-specific events (locoregional recurrence or distant recurrence and competing mortality), and variable selection for prediction, a semiparametric method of the proportional hazards model for subdistributions of the cause-specific events was used.16 Covariates included age and tumor size as continuous variables and lymph node positive status, hormone receptor positive status, tumor grade, whether the patient received chemotherapy, whether the patient received tamoxifen, margin status, race, SES, method of detection (clinical vs mammographic), and presence of comorbid disease as dichotomous variables. Regression models were tested for interactions between race and other covariates. A parametric competing risks regression model17 was built with 95% pointwise confidence intervals to predict the cumulative probability of competing mortality events, adjusted for age, race, and comorbidity. All covariates with P values <.01 from the univariate analysis were re-evaluated in a multivariate model and screened to be included in the final model.

DFS was estimated using the Kaplan-Meier method.18 Time to a DFS event was defined as time from diagnosis to locoregional recurrence or distant recurrence, or death from any cause, whichever occurred first. Patients not experiencing a DFS event were censored at their latest follow-up date. Data regarding second malignancies were not collected. The Cox proportional hazards regression model19 was used to analyze factors associated with DFS. The proportional hazards assumption was tested by analyzing scaled Schoenfeld residuals.20 Missing values were imputed for both receptor positive status (23.6% of the sample) and tumor grade (22.4%) using the Markov Chain Monte Carlo method21 (MI procedure, SAS Institute, Cary, NC) with a single iteration (ie, single imputation). For model checking, internal validation procedures were adopted to evaluate accuracy of the prediction model. To evaluate the model's discrimination ability, the rank correlation (concordance index) among pairs of competing mortality event times and corresponding predicted probabilities of not experiencing events was measured.22 To evaluate potential bias in prediction, the calibration plot between observed and predicted probabilities of not experiencing competing mortality events was displayed. Each point in the calibration plot marks the Kaplan-Meier estimate of the events among a subset of patients versus the mean value of the predicted probabilities in that particular subset. Subsets of patients are consecutively selected by sliding a window with a fixed width after ordering the predicted probabilities.23 We also performed split-sample validation by dividing the sample into 75% training set and a 25% test set by random subsampling (without replacement). A competing mortality risk model was developed on the training set and tested for event stratification in the test set. Analyses were conducted with SAS and R software.

Competing Mortality Risk Score

The competing mortality risk score derived from the parametric regression model was computed as:

  • equation image

with age in years, and race and comorbidity as dichotomous variables (1 if condition present, 0 otherwise). The risk score was constructed using the coefficient estimates from the regression model as weights for each covariate, then transforming to the 0-100 scale by first subtracting the minimum value, then dividing by the maximum value, then multiplying by 100. As shown by the magnitude of the coefficients in this model, the effect of black race (0.56) was comparable to having a major comorbid cardiopulmonary disease, or to being approximately 9 years older. The median and mean risk scores were 38.0 and 39.4, respectively. Patients were divided into low and high competing mortality risk categories, according to whether the risk score was lower or higher than the mean.

RESULTS

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. CONFLICT OF INTEREST DISCLOSURES
  8. REFERENCES

Patient, Tumor, and Treatment Characteristics

The mean tumor size was 1.55 cm (range, 0-7 cm) (Table 1). Sixty-nine percent of patients had stage I disease, 49% had mammographically detected tumors, and 18% had lymph node-positive disease (Table 1). Black patients were, on average, more likely to have low SES and comorbid illnesses, and to have larger tumors that were more often clinically detected. They were also more likely to present with stage II disease and have hormone receptor-negative tumors.

Table 1. Patient and Tumor Characteristics
CharacteristicAll, No. (%)Nonblack, No. (%)Black, No. (%)P
  1. SES indicates socioeconomic status.

Total1231816415 
Age, y [range]60 [40-91]60 [40-91]61 [40-89].038
Low SES610 (49.6)249 (30.5)361 (87.0)<.001
Comorbid disease456 (37.0)229 (28.1)227 (54.7)<.001
 Hypertension381 (31.0)180 (22.1)201 (48.4)<.001
 Diabetes mellitus102 (8.3)31 (3.8)71 (17.1)<.001
 Cardiovascular disease80 (6.5)45 (5.5)35 (8.4).050
 Cerebrovascular disease14 (1.1)4 (0.5)10 (2.4).007
Chronic obstructive pulmonary disease23 (1.9)16 (2.0)7 (1.7).83
Tumor size, cm [range]1.55 [0-7.0]1.49 [0-6.7]1.67 [0.2-7.0].002
Mammographically detected607 (49.3)424 (52.0)183 (44.1).009
Stage   <.001
 I844 (68.6)229 (28.1)158 (38.1) 
 II387 (31.4)587 (71.9)257 (61.9) 
T classification   .041
 T1977 (79.4)664 (81.4)313 (75.4) 
 T2247 (20.1)148 (18.1)99 (23.8) 
 T37 (0.6)4 (0.5)3 (0.8) 
Lymph node positive222 (18)127 (15.6)94 (22.7).002
Hormone receptor status   <.001
 Negative216 (17.6)132 (16.2)84 (20.2) 
 Positive724 (58.8)548 (67.2)176 (42.4) 
 Unknown291 (23.6)136 (16.7)155 (37.4) 
Grade   .10
 1147 (11.9)103 (12.6)43 (10.4) 
 2463 (37.6)321 (39.3)142 (34.2) 
 3346 (28.1)220 (27.0)126 (30.4) 
 Unknown275 (22.4)172 (21.1)104 (25.1) 
Chemotherapy405 (32.9)262 (32.1)143 (34.5).41
Tamoxifen
 Hormone receptor positive601 (92.7)451 (95.6)150 (85.2)<.001
 Hormone receptor negative/unknown346 (59.3)216 (62.8)130 (54.4).31
Margin positive68 (5.5)47 (5.8)21 (5.1).61

For patients who underwent an axillary dissection, the median number of lymph nodes removed was 13. The median whole breast irradiation dose was 4600 centigrays (cGy), and median boost dose was 1400 cGy. Adjuvant chemotherapy was given to 32.9% of patients (22.8% of lymph node-negative patients and 80.4% of lymph node-positive patients) (Table 1). Of the patients who received chemotherapy, 73.6% received anthracycline-based chemotherapy. For hormonal therapy, 52.3% of women received tamoxifen (hormone receptor positive, 63.2; receptor negative, 16.6%).

Outcomes

The median follow-up for surviving patients was 82 months (range, 3-233). Ten-year DFS was 69.7% (95% confidence interval [CI], 66.2%-72.9%). The 10-year cumulative incidence of locoregional recurrence, distant recurrence, and competing mortality were 4.4% (95% CI, 3.0%-5.8%), 7.1% (95% CI, 5.4%-8.9%), and 18.7% (95% CI, 15.9%-21.6%), respectively (Fig. 1).

thumbnail image

Figure 1. Cumulative incidence curves are shown of individual events that are additive to the complement of disease-free survival (1-DFS). Red indicates locoregional recurrence (LRR); blue, distant recurrence (DR); green, competing mortality (CM); black: 1-DFS.

Download figure to PowerPoint

Regression Analysis

On multivariate Cox regression analysis, poorer DFS was significantly associated with increasing age (hazard ratio [HR], 1.38; 95% CI, 1.24-1.55), black race (HR, 1.48; 95% CI, 1.16-1.88), history of comorbid disease (HR, 1.56; 95% CI, 1.22-2.00), lymph node-positive status (HR, 1.54; 95% CI, 1.17-2.04), tumor size (HR, 1.23; 95% CI, 1.10-1.37), and mammographic detection (HR, 0.74; 95% CI, 0.57-0.95). For DFS, the unadjusted HR for black race was 1.90 (95% CI, 1.51-2.40). Adjusted for age alone, the HR was 1.86 (95% CI, 1.48-2.35), and adjusted for age and tumor-related factors (nodal status, tumor size, hormone receptor status, and detection method), it was 1.62 (95% CI, 1.28-2.06).

On univariate analysis using semiparametric competing risks regression, competing mortality was associated with increasing age, SES, black race, comorbid disease, and chemotherapy use (Table 2). When the regression coefficient estimates were adjusted using the multivariate model, the association between competing mortality and age, race, and comorbid disease remained statistically significant (Table 3). Interaction terms between race and age or comorbid disease were not statistically significant (P > .05). When adjusted for tumor-related factors, the (subdistribution) hazard ratios for age and comorbid disease were affected only minimally, and the estimate for black race (HR, 1.59; 95% CI, 1.18-2.14) differed slightly from the estimate in Table 3 (HR, 1.71; 95% CI, 1.17-2.51).

Table 2. Univariate Analysis of Factors Associated With Competing Mortality
CharacteristicHazard Ratioa (95% CI)P
  • CI indicates confidence interval; SES, socioeconomic status.

  • a

    Hazard ratio of the subdistribution.

Age, per 10 years1.96 (1.70-2.25)<.001
Low SES1.65 (1.23-2.21)<.001
Comorbid disease3.01 (2.22-4.09)<.001
Black race2.11 (1.58-2.82)<.001
Pathologic tumor size, cm1.10 (0.95-1.26).20
Mammographically detected0.82 (0.61-1.09).17
Lymph node positive1.41 (1.00-2.00).05
Hormone receptor positive0.97 (0.70-1.36).88
Grade 30.90 (0.67-1.22).50
Tamoxifen0.81 (0.60-1.08).15
Chemotherapy0.61 (0.43-0.86).006
Final margin negative0.78 (0.41-1.48).44
Table 3. Multivariate Analysis of Factors Associated With Competing Mortality
CharacteristicHazard Ratioa (95% CI)P
  • CI indicates confidence interval; SES, socioeconomic status.

  • a

    Hazard ratio of the subdistribution.

Age, per 10 years1.83 (1.58-2.12)<.001
Low SES1.00 (0.68-1.46).99
Comorbid disease1.93 (1.40-2.65)<.001
Black race1.71 (1.17-2.51).006
Chemotherapy1.20 (0.80-1.80).37

When the effects of comorbidities were analyzed separately (each adjusted for age and race), COPD (HR, 2.94; 95% CI, 1.71-5.05), DM (HR, 1.92; 95% CI, 1.30-2.83), and hypertension (HR, 1.79; 95% CI, 1.32-2.44) were associated with a statistically significant increased incidence of competing mortality, whereas cardiovascular (HR, 1.16; 95% CI, 0.72-1.88) and cerebrovascular disease (HR, 1.18; 95% CI, 0.42-3.31) were associated with a nonsignificant increased incidence of competing mortality. Small patient numbers in subgroups comprising the cardiovascular disease group precluded meaningful conclusions about specific diseases in this category. Age- and race-adjusted models simultaneously controlling for each condition (with dummy variables) indicated that the effects of COPD (P < .001), DM (P = .012), and hypertension (P = .006) were independent.

Among women younger than 60 years with comorbid disease, the 5-year cumulative incidence of competing mortality was comparable to recurrence (Table 4). In patients younger than 60 years with comorbid disease (not stratified by race), the 5-year cumulative incidence of locoregional recurrence, distant recurrence, and competing mortality was 4.7% (95% CI, 1.3%-8.1%), 5.5% (95% CI, 1.8%-9.2%), and 9.6% (95% CI, 4.8%-14.5%), respectively. Racial disparities in competing mortality were especially apparent in patients older than 60 years (Table 4), with black patients having a 5-year cumulative incidence of competing mortality of 15.3% (95% CI, 10.3%-20.3%), compared with 6.8% (95% CI, 4.2%-9.5%) in nonblack patients (P < .001).

Table 4. Five-Year Cumulative Incidences of Competing Mortality, Locoregional Recurrence, and Distant Recurrence, by Subgroup
OutcomeAge <60 Years
No ComorbidityComorbidity
Nonblack, n = 341Black, n = 107Nonblack, n = 77Black, n = 81
  1. Estimates are shown as percentages with 95% confidence intervals in parentheses.

Competing mortality1.8 (0.2-3.4)2.6 (0.0-6.1)8.4 (2.0-14.9)10.8 (3.7-17.9)
Locoregional recurrence3.4 (1.3-5.4)1.1 (0.0-3.3)1.4 (0.0-4.0)7.9 (1.8-13.9)
Distant recurrence6.4 (3.6-9.2)5.3 (0.7-9.9)5.7 (0.3-11.0)5.3 (0.2-10.3)
 Age ≥60 Years
No ComorbidityComorbidity
Nonblack, n = 246Black, n = 81Nonblack, n = 152Black, n = 146
Competing mortality5.7 (2.5-8.9)9.9 (2.9-16.9)9.0 (4.3-13.7)17.9 (11.4-24.4)
Locoregional recurrence1.0 (0.0-2.3)2.8 (0.0-6.5)1.4 (0.0-3.3)5.1 (1.4-8.8)
Distant recurrence1.3 (0.0-2.7)4.0 (0.0-8.5)1.4 (0.0-3.3)3.7 (0.5-6.9)

We used multivariate competing risks regression models to analyze the cumulative incidences of locoregional recurrence and distant recurrence (data not shown). Having a positive margin (P < .001) was the lone factor associated with an increased cumulative incidence of locoregional recurrence, whereas greater tumor size (P = .009) and clinical detection (P = .039) were associated with an increased cumulative incidence of distant recurrence.

Competing Mortality Risk Score

Patients with a low (n = 638) versus high (n = 593) competing mortality risk score had 10-year cumulative incidences of locoregional recurrence, distant recurrence, and competing mortality of 4.4% versus 4.4% (P = .97), 8.6% versus 5.6% (P = .12), and 7.2% versus 30.6% (P < .001), respectively (Fig. 2). Ten-year DFS was 79.8% versus 59.4% (P < .001) for patients with low versus high competing mortality risk scores, respectively. Patients in the highest quartile of the risk score (n = 309) had 10-year cumulative incidences of locoregional recurrence, distant recurrence, and competing mortality of 4.9% (95% CI, 2.1-7.7%), 5.7% (95% CI, 2.7-8.6%) and 40.8% (95% CI, 33.9-47.7%), respectively.

thumbnail image

Figure 2. Cumulative incidence curves are shown for low (Top) versus high (Bottom) competing mortality (CM) risk score. Red indicates locoregional recurrence (LRR); blue, distant recurrence (DR); green, CM; black, complement of disease-free survival (1-DFS).

Download figure to PowerPoint

Even among women younger than 60 years with high-risk pathologic features (either estrogen receptor/progesterone receptor-negative or lymph node-positive disease), we observed a higher incidence of competing mortality than disease recurrence in patients with a high competing mortality risk score (n = 45). At 5 years, the cumulative incidences of competing mortality and recurrence (locoregional recurrence or distant recurrence) were 18.0% (95% CI, 5.7%-30.3%) and 14.4% (95% CI, 3.8%-25.0%), respectively. By 10 years, the incidences of competing mortality and recurrence were 31.8% (95% CI, 15.5%-48.1%) versus 16.8% (95% CI, 5.4%-28.2%).

Model Validation

The concordance index under the prediction model developed on the whole sample was 74.2%. The plot comparing 10-year observed and predicted probabilities of not experiencing competing mortality showed a straight line with the slope of 1, confirming that the prediction model was unbiased. Coefficient estimates for age, black race, and comorbid disease in the training sample were 0.068 (P < .001), 0.50 (P = .005), and 0.66 (P < .001), respectively, which were comparable to the estimates based on the whole sample. The training model performed well on the test sample in terms of stratifying patients into low- and high-risk groups, with a 10-year cumulative incidence of competing mortality of 7.0% versus 34.3% (P < .001), respectively.

DISCUSSION

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. CONFLICT OF INTEREST DISCLOSURES
  8. REFERENCES

In this cohort, competing mortality was the predominant event influencing DFS and was correlated with age, black race, and comorbid disease. Comorbid disease is a potentially modifiable risk factor. Directing more intensive medical care to high-risk patients, particularly those with major cardiopulmonary diseases, is 1 strategy to narrow outcomes disparities between African Americans and whites. Others have reported that African American and low-SES breast cancer patients receive poorer healthcare maintenance.24 Intensifying efforts to ensure these patients receive adequate primary or preventive health maintenance efforts is 1 approach to improving outcomes. The degree to which medical management and health maintenance can be optimized in this population is not well known, however.

Age, comorbidity, and race are factors known to be associated with noncancer mortality.7, 8, 13-15 By using Surveillance, Epidemiology, and End Results data, Schairer et al7 found that in localized breast cancer, noncancer mortality exceeds breast cancer mortality for patients older than 50 years, and that, for patients younger than 70 years, the probability of noncancer mortality is higher in black than white patients. Tammemagi et al15 found that carefully accounting for comorbid illnesses, particularly DM and hypertension, explained much of the racial disparity in noncancer mortality. However, we found that, even adjusting for age and comorbid disease, black race was independently associated with increased competing mortality. Furthermore, this was invariant with respect to how comorbid disease was coded in our analysis. There were several differences, however, between Tammemagi et al and this study, which included only stage I to II patients, a higher proportion of black patients, and a different endpoint (competing mortality as a first event), statistical model (competing risks regression), and index of comorbid disease (binary). A limitation of this study was that we were unable to test a validated index of comorbidity. Although we accounted for DM, hypertension, and other cardiovascular diseases, more detailed indices might improve this model.

There are several advantages to the approach reported here for modeling cumulative incidence functions. These functions pertain to the probability of observing a specific event of interest in the presence of competing events,25 and provide a way to decompose a composite endpoint into its constituent events, more clearly characterizing the effects of risk factors (or treatments) on observed outcomes.26 With this approach, recurrence probabilities can be presented in the context of the probabilities of other competing events, which is clinically relevant in the follow-up and management of early breast cancer. Patients who ultimately die of noncancer-related causes may still be at high risk of recurrence and benefit from additional cancer therapy, inasmuch as preventing recurrence is a goal of treatment. In contrast, patients more likely to die than to have recurrence benefit less, on the margin, from additional cancer therapy. The results of this study might be particularly useful for the design of studies focusing on elderly or socioeconomically disadvantaged populations. On the basis of our results, the use of cause-specific endpoints, such as breast cancer mortality or distant recurrence, might be preferable in designing clinical trials in such populations.

A strength of this study is that our cohort represents a racially and socioeconomically heterogeneous patient population receiving relatively homogeneous primary treatment (lumpectomy and radiotherapy). A limitation is that this was a retrospective analysis; therefore, validation of the risk score in a population-based cohort would be desirable to assure its accuracy and utility. Most black women in our cohort were classified as having low SES, and it was not possible to separate completely the effects attributable to low SES from those attributable to race. Although the observed associations between competing mortality and age, race, and comorbidity were consistent with other studies, our findings are subject to possible biases because of referral patterns or temporal trends. Misclassification of outcomes is a potential limitation, because mortality classified as noncancer-related may include deaths directly or indirectly related to the presence of breast cancer.27, 28 However, we did not observe strong correlations between competing mortality and tumor-specific characteristics, which we would have expected if misclassification were common.

Although we did not find significant interactions between race and comorbid disease, we observed that black patients, particularly those older than 60 years, appeared to be at higher risk of competing mortality than nonblack patients even within comorbidity subgroups (Table 4). Poorer control or management of major comorbid illnesses in black patients, both before and after the diagnosis of breast cancer, may account for some disparity that our regression analysis could not capture. Quantitative data with baseline and follow-up blood pressure, glucose measurements, and other physiologic indicators could be useful to investigate whether undiagnosed comorbidities partially account for these findings.

In summary, competing mortality is an important event influencing long-term DFS for patients with early breast cancer. Stratifying patients according to competing mortality risk may be useful in designing clinical trials. Furthermore, efforts to reduce racial disparities in outcomes could be aided by more intensively addressing the high risk of competing mortality in black patients.

Acknowledgements

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. CONFLICT OF INTEREST DISCLOSURES
  8. REFERENCES

We thank Amber Meriwether, BA, for assistance with data collection and processing

REFERENCES

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. CONFLICT OF INTEREST DISCLOSURES
  8. REFERENCES