Background
In 1990 the National Institutes of Health Consensus Development Conference [
1] concluded that breast conserving surgery (BCS) followed by radiation is an appropriate method of primary treatment for the majority of women with early stage breast cancer (AJCC stages [
2] I, IIA, and IIB). Numerous clinical studies have shown that survival after BCS followed by radiation therapy is equivalent to survival following mastectomy for women in these stages [
3‐
5]. Some absolute indicators for mastectomy remain [
6,
7], in particular widespread malignant-type microcalcifications, previous radiotherapy, and a relation of tumor size to breast size that would not allow a cosmetically satisfactory result. Although there has been an increase in the use of BCS since the early 1990s, an apparent under-utilization of BCS among women for whom such treatment was not contraindicated has been documented [
8]. Geographic location, type of hospital and health plan, and personal preferences have been investigated as possible factors to explain the continued use of mastectomy [
9‐
12]. Comorbidity [
13] has been found to be an important predictor among older women.
The Surveillance, Epidemiology and End Results (SEER) cancer registries do not collect sufficient detail to examine some of these issues and medical records reviews are time consuming. Insurance claims data are a cost-effective alternative; they have already been collected and put into an electronic format by the insurance carriers. They cover large segments of the population, allow follow-up, use standardized codes [
14,
15], and do not rely on subject recall. In cancer research, insurance claims data have been used to estimate the effectiveness of cervical cancer screening [
16], to assess the role of screening practices in the incidence of prostate cancer [
17], and to estimate mammography participation [
18]. Medicare enrollees in Seattle and San Francisco who received care from a Health Maintenance Organization (HMO) were found to receive more BCS than women in Fee-For-Service (FFS) plans [
19]. Whereas a Medicare/SEER registry linked data set has been used extensively to examine treatment and screening issues in individuals 65 years and older [
20], linking of insurance claims data from private health plans in younger populations has been more difficult because of the large number of health plans in most geographic areas. However, Hawaii provides unique opportunities for insurance claims research because the majority of medical care is received within the state and more than 90% of the population [
21] are covered by a limited number of health plans. As a pilot project, we were able to link data from the Hawaii Tumor Registry and from a health plan in Hawaii. The objectives of this analysis were to describe breast cancer treatment and comorbid conditions using the insurance claims data and to examine possible determinants of BCS
vs. mastectomy for breast cancer patients with stage I and II disease.
Results
Based on the TNM stages, we identified 722 women who had stage I, IIA, or IIB breast cancer (66.4% of the 1,088 breast cancer cases in the linked data set). These 722 cases represented 32.8% (722 of 2,203 cases) of all early stage breast cancer cases recorded in the HTR during the study period. Approximately two-thirds of the women in the study population (Table
1) were diagnosed in stage I of the disease and only 24% and 12% in stage IIA and IIB, respectively. Approximately 30% of the study subjects were aged less than 50 years, close to half were 50 to 64 years old, and 21% were 65 years and older. Three out of four women resided on the island of Oahu. Overall, 52.8% of the women had received BCS. Of the cases diagnosed at stage I, 57% of the women had received BCS and 47% a mastectomy. This decreased to 50% for stage IIA and to 34% for stage IIB. Among women, 65 years and older, 56% had BCS, compared to 50% of those under 50 years of age and 53% of the 50–64 year old women. While 56% of women on Oahu underwent BCS, only 43% of the women residing on the outer islands received BCS. We observed no statistically significant differences between the BCS and mastectomy group in terms of ethnicity, comorbidity count, menopausal status, marital status, and insurance plan. We found very high agreement between HTR data and claims data in identifying BCS [κ = 0.91 (95% CI 0.88, 0.94)], only 32 women were misclassified. For the majority of women who received BCS (92.1%), the lumpectomy was followed by radiation therapy. Additional chemotherapy was given to 34.1% and 42.5% of the women who had a lumpectomy or a mastectomy, respectively.
Table 1
Determinants of Breast Conserving Surgery Among 722 Cases of Early Stage Breast Cancer
Ethnicity | Caucasian | 71 | 66 | 1 | 1 |
| Japanese | 170 | 147 | 1.08 (0.72–1.61) | 0.85 (0.55–1.31) |
| Hawaiian | 48 | 40 | 1.12 (0.65–1.91) | 1.24 (0.71–2.17) |
| Filipino | 35 | 47 | 0.69 (0.39–1.20) | 0.65 (0.37–1.15) |
| Chinese | 32 | 25 | 1.19 (0.64–2.22) | 0.97 (0.50–1.86) |
| Other | 24 | 16 | 1.45 (0.71–2.96) | 1.33 (0.64–2.77) |
Residence | Oahu | 299 | 232 |
1.71 (1.23–2.39)
|
1.67 (1.17–2.38)
|
| Other | 82 | 109 | 1 | 1 |
Age at diagnosis | <50 yrs | 110 | 109 | Continuous (per 10 years) | |
| 50–64 yrs | 186 | 166 |
1.01 (1.001–1.028)
| 1.01 (0.99–1.04) |
| 65+ yrs | 85 | 66 | | |
Menopausal status | Pre Post | 110 271 | 109 232 | 1 1.16 (0.84–1.59) | 1 0.75 (0.46–1.23) |
Marital status | Married | 262 | 244 | 0.88 (0.64–1.21) | 0.94 (0.68–1.32) |
| Unmarried | 119 | 97 | 1 | 1 |
Health plan | FFS | 330 | 300 | 0.89 (0.57–1.37) | 0.87 (0.55–1.38) |
| HMO | 51 | 41 | 1 | 1 |
TNM stage | I | 266 | 200 | 1 | 1 |
| IIA | 86 | 85 | 0.76 (0.54–1.08) | 0.99 (0.66–1.51) |
| IIB | 29 | 56 |
0.39 (0.24–0.63)
| 0.61 (0.32–1.19) |
Tumor size | <1 cm | 113 | 83 | Continuous | (per 1 cm) |
| 1–1.9 cm | 177 | 139 |
0.97 (0.96–0.99)
| 0.98 (0.96–1.00) |
| 2 + cm | 91 | 119 | | |
Tumor grade | Well diff. | 77 | 49 |
1.51 (1.02–2.24)
| 1.25 (0.83–1.89) |
| Other | 304 | 292 | 1 | 1 |
Comorbid conditions | 0 | 230 | 201 | Continuous (per score) | |
| 1 | 109 | 102 | 0.97 (0.78–1.19) | 0.91 (0.72–1.14) |
| 2+ | 42 | 38 | | |
In the univariate models (Table
1), we found that tumor size and grade, island of residence, age, and stage at diagnosis were predictors of breast cancer treatment. Women residing on Oahu were considerably more likely to have BCS than women living on all other islands in the state. For each one-centimeter increase in tumor size, there was a 3% lesser chance of undergoing BCS. Since size of tumor correlates to breast cancer stage, this also decreased the likelihood for women in stage IIA and IIB to receive BCS. Women with well-differentiated tumor grades were 50% more likely to undergo BCS as compared to women with all other grades. We found that for each ten-year increment in age, the chances of having BCS increased by 1%. We also observed that average tumor size was inversely related to age. Mean tumor sizes (in cm) with standard deviations were 1.82 ± 1.1, 1.59 ± 1.2, 1.39 ± 0.94 for women younger than 50 years, 50 to 64 years, and 65 years and older, respectively. Therefore, the smaller tumor sizes may account for the greater likelihood of BCS among older women.
In a combined model with all independent variables in a logistic regression, island of residence remained the only significant predictor of BCS in this population. Women living on Oahu were 67% more likely to have BCS than women on the outer islands. Although all other variables lost their statistical significance, associations for age, TNM stage, tumor size, and tumor grade remained similar in magnitude as in the univariate models. Although none of the ethnicity variables was significant, it appeared that Filipino women were less likely to receive BCS than women from all other groups. Residence on outer islands did not explain this observation. Women with a TNM stage of IIB were still 40% less likely to receive BCS than women diagnosed at stage I, but the relation lost its statistical significance due to the small number of cases. Menopausal or marital status, type of health plan, or the number of comorbidities were not related to the type of surgery.
Discussion
The place of residence at the time of diagnosis was the most important predictor for receiving BCS in this dataset of health plan members. Women with early stage breast cancer living on Oahu were 70% more likely to have BCS than women living on outer islands. Age at diagnosis and tumor size were also related to breast cancer treatment although they were not statistically significant in the combined model. Contrary to our expectation and a previous publication [
13], the number of comorbid conditions did not effect treatment in this analysis. The importance of geographic location can be explained by the availability of radiation facilities on the different Hawaiian Islands. Whereas on Oahu treatment facilities can be reached within one hour, distances on the outer islands are much farther. Only the Island of Hawaii and Maui have a radiation facility, but they are hard to reach from many parts of the islands. Women residing on Kauai, Molokai, and Lanai have to fly to Honolulu daily to receive the course of treatment or they have to remain there temporarily. Because this may be not be economically feasible for some women, as well as physically and psychologically challenging, it is possible that many women choose to undergo a mastectomy instead. Alternatively, patients and physicians in rural areas may favor mastectomy because of differences in education or cultural attitudes toward medical advances. We do not have information to explain why a small proportion of women who underwent BCS (6.9%) did not receive radiation. However, compared to other reports [
13,
30,
31], the proportion of women who received radiation is very high. An interview study with cancer patients [
32] showed that some patients decide against their physicians' recommendation for radiation because they fear the treatment or hold certain beliefs about cancer. For other patients, radiation may have been contraindicated due to comorbid conditions.
In 1994, a national BCS rate of 42.6% was reported [
33] with a rate of 46.7% for the Pacific region (Hawaii, California, Alaska, Washington, and Oregon). Our results for 1995 to 1998 indicate a slightly higher rate for Hawaii (52.8%). Geographic differences in BCS use have repeatedly been described in the literature. In western Washington [
34], women residing outside the Seattle area were less likely to have BCS than women residing inside the county, in particular if radiation therapy facilities were not available in their county. Comparing rates across the country showed that BCS was much more common in the Northeast than in the South [
33]. Low BCS rates were also found in rural areas, such as North and South Carolina [
10], Minnesota [
35], and the Southwest [
36]. An analysis of 1991–1992 SEER data [
37] demonstrated a strong association between BCS and distance to the next radiotherapy facility. The likelihood of undergoing BCS was 50% lower among women living more than 15 miles from the next radiotherapy facility.
In contrast to our results, a number of studies have shown that the likelihood of BCS decreases with increasing age [
11,
30,
33], possibly due to less concern about cosmetic results. Because of the relatively small proportion (21%) of women 65 years and older and the fact that the women 65 years and older in our study had a health plan other than Medicare, they were not directly comparable to the populations in the published literature. Therefore, our ability to observe an effect among older women was limited. In particular, we were not able to examine the treatment of women 80 years and older who have been described as receiving more BCS than women age 65 to 79 years [
13]. Our findings of lower BCS in smaller and well differentiated tumors agree with previous studies [
33], but the absence of an effect of comorbid conditions disagrees with a study in older women [
13] that showed a higher rate of BCS in women with comorbidities. Type of health plan did not predict type of treatment in our study probably because the majority of women in this data set had a FFS plan. In previous analyses with a larger variety of health plans, HMO members were considerably more likely to receive BCS in San Francisco and in Seattle [
19]. Medicare and Medicaid patients received more BCS in a North Carolina hospital [
7]. However, a study using administrative data [
38] described lower BCS use among Medicaid patients and treatment was similar by health plan [
39] in Northern California.
Our study had several limitations, including the possibility of incomplete claims histories if patients changed health plans during the course of treatment, but the high agreement with HTR information indicates that the validity of insurance claims information was high. Continuing efforts by the health plan through feedback loops has improved diagnostic coding [
40] in Hawaii, but as documented for Medicare claims, problems remain [
41]. As discussed above, our study sample of breast cancer patients does not represent the population of the state, excluding in particular Medicare and Medicaid recipients, members of a typical HMO, and the relatively small number of uninsured patients. The exclusion of Medicare claims lead to a younger population, whereas the lack of Medicaid claims biased the cases toward a higher socioeconomic status. On the other hand, the study population represented the population quite well in terms of ethnicity (Hawaii Department of Business 1998 1705 /id). Because our data set included 42.4% of all breast cancer cases under 65 years diagnosed during the study period, the results represents this population much better than the population of older women. Also, we were unable to measure some factors that have been found to be important in other reports, in particular, influence of physician age [
11], physician specialty [
42], type of hospital (teaching hospitals and hospitals with radiation facilities, private, county and public) [
36,
38,
39,
43], socio-economic status and education [
42], tumor-breast ratio and the expected cosmetic results [
7,
9], physician/patient interactions before surgery [
42], and psychological factors, such as fear of radiation or cancer [
9]. The strengths of using of insurance claims data for this analysis were twofold. First, we had more detailed treatment information available than collected by the tumor registry. Second, the diagnostic codes in the claims data allowed us to estimate the number of comorbid conditions before the cancer diagnosis.
The results of this pilot study combining tumor registry and insurance claims data raises an important issue for cancer practice in the State of Hawaii and elsewhere. Health care providers and insurance plans need to work with other agencies to develop viable solutions to facilitate access to radiation facilities for women residing in remote locations, such as islands with no radiation facility. Attitudes of patients and physicians living in rural areas may be important in the choice of treatment, but could not be investigated in this project. Future research that includes interviews with patients and physicians may investigate attitudes toward certain types of treatment and identify additional barriers to BCS, such as psychological problems. In order to understand how treatment decisions were made in practice, qualitative information from physicians and breast cancer patients should be collected and examined in detail. Further validation of treatment and comorbidity from insurance claims data would strengthen this type of research in the future. The inclusion of a larger proportion of breast cancer cases is needed to establish the validity of our findings for the population at-large.