The Effect of Comorbidity on 3-Year Survival of Women with Primary Breast Cancer

Methods

The information we report derives from a study begun in 1984 in the Detroit metropolitan area to assess the functional and psychosocial status of women after the diagnosis and treatment of breast cancer [3]. Women ages 40 to 84 years who were recently found to have histologically confirmed, invasive breast cancer were identified through the Metropolitan Detroit Cancer Surveillance System (MDCSS) in the Division of Epidemiology at the Michigan Cancer Foundation. The MDCSS, 1 of 9 participants in the National Cancer Institute's Surveillance, Epidemiology, and End Results (SEER) program, is a population-based cancer surveillance system that monitors the 4 million residents of the Detroit metropolitan area (Wayne, Oakland, and Macomb counties). Medical and demographic information on all patients with cancer, except those with nonmelanoma skin cancer, is obtained from 66 tricounty hospitals, 8 private laboratories, 15 radiation therapy clinics, and other facilities (such as hospices). Information is also obtained from 4 hospitals outside the metropolitan area that treat patients who have cancer and who live in the Detroit area. As a measure of the completeness of case identification, only 1.2% of all cases of cancer occurring among metropolitan residents are identified solely through death certificates [6].

Case ascertainment for the study used the established MDCSS rapid-reporting system, which allows identification of cases within 4 weeks of diagnosis. This system facilitates earlier interview (2 to 4 months after diagnosis).

The names of women identified in two case series through rapid reporting were matched to MDCSS files, and women with a previous diagnosis of breast cancer were excluded from the analysis.

In the first case series, 571 eligible women ages 55 to 84 years were identified in a 7-month period between 1984 and 1985. Of these, 463 (81.1%) were successfully interviewed. In the second case series, 548 of 620 women (88.4%) were interviewed during a 7-month period between 1987 and 1988. The second cohort was selected to broaden the age range from 40 to 84 years and to increase the representation of women ages 75 to 84 years. No substantial differences were found between the two cohorts in treatment patterns or in the association between other variables and survival. Therefore, the two cohorts were combined for all analyses. Of the 1011 patients in the two cohorts, 829 were white and 170 were black.

Although the response rate was relatively high (84.9%), women with remote-stage disease were less likely to be interviewed (70.5%) than were women with local (84.3%) or regional disease (85.6%) (P < 0.01). In addition, women ages 40 to 54 years were more likely to be interviewed (92.7%) than were older women (55 to 64 years, 81%; 65 to 74 years, 80.5%; 75 to 84 years, 80%) (P < 0.01). No difference was found, however, in the response rates for black and white women (P > 0.2).

Trained interviewers conducted the personal interviews 2 to 4 months after diagnosis. The questionnaire, which required 45 to 60 minutes to complete, included questions on physical functioning, health practices, and social and economic resources. Most interviews were conducted in the patient's home; 5% were done by telephone.

Clinical, Social, and Behavioral Factors

Age, stage of breast cancer, and treatment status were included in each analysis. For the purpose of analysis, age at the time of diagnosis was recorded, and four age categories were created (40 to 54 years, 55 to 64 years, 65 to 74 years, 75 to 84 years). Stage of breast cancer at diagnosis was determined from MDCSS files and was coded as a three-level variable (local, regional, or remote) according to criteria established by the National Cancer Institute's SEER program [7]. Results from the SEER program indicate that the 5-year relative survival rates for women with local, regional, and remote-stage breast cancer are 92%, 71%, and 18%, respectively [1].

Information on type of surgery (no surgery, partial mastectomy, or modified radical mastectomy) and type of adjuvant therapy (radiation or chemotherapy) was obtained from MDCSS files. In addition, each physician on record was contacted in a special survey to supplement the information from MDCSS files, especially for chemotherapy and hormonal therapies administered on an outpatient basis. Ninety percent of physicians provided complete treatment data. Treatment-related information from the MDCSS and the special survey was combined to create a five-level treatment variable (no surgery, partial mastectomy only, partial mastectomy with adjuvant therapy, modified radical mastectomy only, and modified radical mastectomy with adjuvant therapy). For this variable, radiation, chemotherapy, and hormonal therapies were grouped as “adjuvant therapy”. In all analyses, the five-level treatment variable was represented by four indicator variables, that is, as a nonordered categorical variable.

Information on several potential confounding variables was also derived from the interview. Race was coded as either black or white. Body mass index (BMI) (weight in kilograms divided by height in meters squared) was calculated from heights and weights reported by patients at the interview conducted 3 months after diagnosis. For the analysis, BMI was divided into three categories (<22 kg/m², 22 to 27.29 kg/m², >27.29 kg/m²). Financial adequacy was determined from the patient's self-reported current financial resources (from all sources) and their adequacy in meeting her daily needs. Education was defined by the highest number of completed years of schooling. History of alcohol use was represented by a summary measure derived from the patient's report of alcohol consumption (frequency, type, and amount) in the previous year [8]. For analysis, the alcohol summary measure was divided into four categories (0, 0.01 to 15.99, 16 to 44.99, and >45). Smoking history was represented as a two-level binary variable (currently smoking compared with not currently smoking). Finally, a variable was constructed to represent the period of entry into the study (1984 to 1985 or 1987 to 1988).

Vital Status

The MDCSS, through its contract with the SEER program, is required to conduct annual surveillance of the vital status of all patients with cancer who are listed in the registry. This is done through a biannual match of the MDCSS files against Michigan State vital statistics files and through annual contact with the patients' physicians. Annual vital status surveillance is complete for 87% of the patients with cancer in the MDCSS.

Metropolitan Detroit Cancer Surveillance System Quality Control Procedures

Accuracy and consistency in staging and ascertainment of treatment data for the MDCSS are ensured in three ways: 1) Staff are trained to conduct case finding and abstracting, including staging and treatment, in accordance with guidelines and training material developed by the SEER staff; 2) case finding and abstracting are manually reviewed and evaluated by senior staff members, and computerized editing procedures are used to review cancer staging and treatment; and 3) comprehensive annual quality control reviews are done by MDCSS senior abstracting staff and by SEER staff.

Assessment of Comorbidity

Medical abstractors from the MDCSS reviewed the medical records of all patients with cancer to code concurrent health conditions on a special abstract form designed to record eight specific chronic conditions (hypertension, myocardial infarction, other types of heart disease, diabetes, arthritis, respiratory disease, stroke, and other forms of cancer). In addition, space was available to record a history of other concurrent conditions. Medical records were reviewed for all hospitalizations of each patient during a 6-month period beginning at the time breast cancer was diagnosed. The records included the face sheet, the findings of history and physical examination, and the discharge summary. For each health condition, it was noted whether the diagnosis was made before, concurrent with, or subsequent to the diagnosis of the breast cancer. Two senior staff members (field administrators with more than 10 years of abstracting experience) reabstracted the records and compared them with those originally prepared by the medical abstractors. Inconsistencies, found in fewer than 5% of the cases, were reviewed and jointly resolved by senior staff and the abstractor.

Development of the Comorbidity Index

The most common comorbid conditions were high blood pressure, heart disease (other than myocardial infarction), arthritis, thyroid disorders, gallbladder disorders, and diabetes. (See Appendix Table 1 for the full set of 18 comorbid conditions.) In only a few instances was the time of diagnosis of the comorbid conditions either unknown (2.9%) or subsequent to the diagnosis of the breast cancer (3.9%). Therefore, all conditions were included, regardless of the timing of the diagnosis. Of the 18 conditions, 7 were found to be independently associated with all-cause mortality, breast cancer mortality, or mortality from other causes after adjustments were made for age, stage of breast cancer treatment, and the other comorbid conditions (P < 0.05). These 7 conditions included myocardial infarction, other types of heart disease, diabetes, other forms of cancer, and respiratory, gallbladder, and liver conditions. “Other forms of cancer” did not include metastases of breast cancer. Gallbladder and liver conditions were obtained from the abstract in the section listing “other concurrent conditions”.

A comorbidity index was developed based on the total number of 7 conditions present. We also generated a second index based on the remaining 11 conditions to determine whether these conditions were associated collectively with increased mortality. No such association was found (P > 0.2). Therefore, the comorbidity index based on the 7 conditions was used in all analyses.

Statistical Analysis

Descriptive results were based on 999 women with primary breast cancer. Survival analyses were restricted to the 936 women for whom complete information was available on stage of breast cancer, type of treatment, and comorbidity. Two-level variables were treated as simple indicator variables. To allow for the possibility of nonlinear relations, variables with three or more levels were considered categorical variables.

The Cox proportional-hazards regression model was used to test for the independence of the association between comorbidity and all-cause mortality, mortality from breast cancer, and mortality from causes other than breast cancer [9]. For analyses involving death from breast cancer, patients who died from other causes were removed from the cohort at the time of death. For analyses involving death from causes other than breast cancer, the patients dying from breast cancer were removed from the cohort at the time of death. The time variable for the Cox models was the time from diagnosis to the date of death or to the last day the patient was observed alive, within 3 years after diagnosis. Age and treatment were included as possible confounding variables. Race, BMI, financial adequacy, education, history of alcohol use, history of cigarette smoking, histologic type, tumor size, and period of entry into the study were also evaluated as possible confounding variables.

Results

Sample Characteristics

The demographic and behavioral characteristics of the study sample, by age, are shown in Table 1. When compared with younger patients, older patients had a lower BMI (P = 0.06), had fewer years of education (P < 0.001), and were less likely to currently smoke cigarettes (P < 0.001) and consume alcohol (P < 0.001). No statistical difference was found in the age distribution of black and white patients (P > 0.2), nor was an association found between patient age and financial adequacy (P > 0.2).

No statistical association was found between age and stage of disease (P = 0.10) (Table 2). However, an association was found between age and type of therapy for breast cancer (P < 0.001), with older patients being less likely than younger patients to receive surgical treatment (P = 0.02). Among those receiving a partial mastectomy, older women were less likely than younger women to receive adjuvant therapy (P < 0.001).

Comorbidity and Cause of Death

One hundred forty-five patients died in the 3 years after diagnosis (mean follow-up, 2.45 years). In 92 cases (63.4%), breast cancer was recorded as the underlying cause of death on the death certificate. Mortality in women ages 40 to 54 years at the time of diagnosis was nearly six times more likely to be attributable to breast cancer than to some other cause. In contrast, mortality in women ages 75 to 84 years at diagnosis was attributable to breast cancer in fewer than 50% of cases.

Table 3 shows the association of comorbidity with age, stage of disease, and treatment. A total of 288 (30.8%) of the patients had one comorbid condition, 124 (13.2%) had two, and 41 [4.4%] had three or more. Appendix Table 2 lists the frequency of each combination of comorbid conditions, in order of prevalence.) The level of comorbidity increased steadily with increasing age (P < 0.001). No association was found between comorbidity and stage (P = 0.15), but an overall association between comorbidity and type of treatment was observed (P < 0.001). Although the results shown in Table 3 suggest that increasing comorbidity was associated with either no surgery and partial mastectomy without adjuvant therapy, these associations were not statistically significant (P > 0.2).

The number of comorbid conditions was strongly associated with an increased risk for death after adjustments were made for age, stage of breast cancer, and type of treatment (Table 4). The adjusted all-cause, 3-year mortality rate was 188.4 per 1000 person-years of follow-up in patients with three or more comorbid conditions and 47.7 in women with no such conditions. This association between severe comorbid conditions and death can be attributed almost entirely to an increased risk for death from causes other than breast cancer (Table 4).

Adjusting for race, education, or financial adequacy did not alter the association between comorbidity and mortality. Moreover, neither BMI, cigarette smoking, alcohol consumption, histologic type, tumor size, nor period of entry into the study affected this association.

The effect of racial differences on mortality remained significant after adjustments were made for comorbidity (risk ratio, 1.63; P = 0.03). However, adjusting for comorbidity did reduce the risk for death associated with inadequate financial resources from 1.48 (P = 0.05) to 1.29 (P > 0.2).

Stage of Disease and Cause of Death

A strong association between stage of breast cancer and survival was observed after adjustments were made for age, treatment, and level of comorbidity (Table 5). The all-cause mortality rate for patients with remote disease was more than 7 times the rate for patients with localized disease. Although the association between stage of breast cancer and mortality was found in patients who died of breast cancer and in those who died of other causes, it was much stronger in the former group. The likelihood of death from breast cancer increased greatly with each stage of disease (Table 5). Women with localized disease were 60% more likely to die of breast cancer than of some other cause, whereas those with remote-stage cancer were more than 4 times more likely to die of breast cancer.

Comorbidity, Stage of Disease, and Survival

An interaction was found between comorbidity and stage of disease at diagnosis after adjustments were made for age and treatment (P = 0.02) (Figure 1). In general, patients with remote-stage breast cancer had a lower probability of survival than did patients with local or regional disease. However, among patients with three or more comorbid conditions, stage of disease seemed to have little additional effect on survival. Conversely, the effect of comorbidity on survival varied depending on the stage of breast cancer at diagnosis. For example, the age-adjusted, all-cause mortality rate among patients with localized disease ranged from 20.2 to 245.3 per 1000 person-years of follow-up as the level of comorbidity increased. Among patients with regional breast cancer, the age-adjusted mortality rate ranged from 67.8 to 204.3. In contrast, patients with remote-stage disease had a low probability of survival at each level of comorbidity. This interaction between stage of breast cancer and comorbidity was not affected by the inclusion of tumor size and histologic type in the model.

Discussion

Our results indicate that comorbidity in patients with breast cancer is strongly related to the risk for death from causes other than breast cancer, confirming findings from a previous study by Charlson and colleagues [2]. Although the presence of particular comorbid conditions may influence the reporting of an underlying cause of death other than breast cancer, given the magnitude of the association between comorbidity and death, it is very unlikely that this is the sole explanation. It is possible that comorbid conditions do not affect the course of the breast cancer. Rather, breast cancer (or perhaps treatment for breast cancer) may, in some instances, accelerate the course of other pathologic conditions. As a result, the risk for death from those conditions may be greater.

Our results also show an interaction between comorbidity and stage of disease, which may have implications for breast cancer screening. Screening is done to detect the disease at an early stage in order to increase the chances of the patient's survival. The results of our study seem to suggest that diagnosing breast cancer at an early stage may not confer the same advantage for women with severe comorbidity as it does for women with fewer or no concurrent medical conditions (Figure 1). If confirmed in other studies, measures of comorbidity could be used to evaluate the efficacy, benefits, and costs of screening, supporting recommendations made by the American Geriatrics Society and others [10-14].

Although our measure of comorbidity helped to explain the association between financial adequacy and survival, it did not explain racial differences in survival. The elevated risk for death found among black women may be attributable to factors other than comorbidity and stage of disease [15].

The effect of comorbidity on mortality rates and the relation between comorbidity and stage of breast cancer were robust. However, our study had several limitations. First, our measure of comorbidity was generated from an analysis of a single population of patients with breast cancer. The prognostic significance of this measure must be tested in other populations. Second, our measure did not account for time intervals between primary comorbid conditions, past or current treatments, the significance of specific combinations of conditions, or the relative severity of conditions. Each of these factors could have important implications for prognosis. For example, in our study and in others [2], the severity of a particular condition was defined in terms of the risk for death for persons with that condition. Other criteria, such as measures of function used by Greenfield and colleagues [16] in their study of comorbidity and breast cancer treatment, should also be considered. More sophisticated measures of comorbidity may contribute to our understanding of racial and socioeconomic differences in survival. Third, the follow-up period (3 years) was relatively short. Although the 3 years after diagnosis is a period of high mortality risk for patients with breast cancer, the long-term effects of comorbidity and stage must also be examined.

Care must be taken in considering the policy implications of our findings. To recommend special screening guidelines for women with severe comorbidity would be premature. Studies must be done in other populations and for longer periods, and investigators must consider the effects of competing causes of death. The most definitive results will come from clinical trials. As more refined measures of comorbidity are developed, it may be possible to conduct clinical trials in which women are stratified by level of comorbidity. Within each stratum, women could be randomly assigned to different screening protocols, perhaps each with different time intervals between mammograms. In accord with recent recommendations, investigators conducting clinical trials should include quality-of-life end points [17, 18]. The results of such studies could establish a basis for designing more effective screening procedures for breast cancer.