Breast cancer treatment

Despite an increasing trend in the use of breast conserving surgery (B CS) to treat early-stage breast cancer (Silliman et al. 1997; Guadagnoli et al. 1998), substantial variability exists in use of BCS among older women (Nattinger et al. 1996; Wennberg and Cooper, 1996), with the oldest women receiving less BCS, and when treated by BCS, receiving radiotherapy less often than others (Mandelblatt et al. 2000; Busch et al. 1996; Bahlard-Barbash et al. 1996). Numerous studies have examined the roles of factors such as underlying health (Silliman et al. 1997; Albain et al. 1996), age, or socioeconomic biases (Lazovich et al. 1991; Albain et al. 1996; Michalski and Nattinger 1997), physicians' attitudes toward treatment, and patient involvement in treatment decisions (Silliman et al. 1989; Liberati et a). 1987; Liberati et al. 1991), geographic variations or barriers in access to services (Farrow, Hunt, and Samet 1992; Nattinger et al. 1992; Nattinger et al. 1996; Albain et al. 1996; Osteen et al. 1994; Hand et al . 1991), and different care delivery systems (Riley et al. 1999; Potosky et al. 1997).

Only one study (Hadiley, Mitchell, and Mandelblatt 2001) has investigated whether variations in Medicare's fees for BCS and mastectomy (MST) influence the surgical treatment received by elderly Medicare beneficiaries who had breast surgery. Analyzing small-area data on the percentage of elderly Medicare breast surgery patients receiving BCS in 1994, that study found that a 10 percent higher fee for BCS was associated with a 7-10 percent increase in the percentage of beneficiaries receiving BOS in an area, while a 10 percent lower MST fee increased the BCS percentage by 2-3 percent. While suggestive of a fee effect, these findings may have been influenced by several potential limitations. The results may reflect an ecological fallacy because the analysis was conducted at the area level--the same results may not hold for individual patients. The Medicare claims data used in the analysis were not limited to confirmed cases of newly diagnosed localized breast cancer. Thus, it was not possible to exclude cases wh ere minimally invasive surgery was used to rule out a cancer diagnosis from those where the procedure was used as a treatment. Nor was it possible to distinguish women who received breast conserving surgery only (BCSO) from those who received breast conserving surgery plus radiation therapy (BCSRT). If there are differences in the factors that determine the receipt of either of these treatments relative to mastectomy, then the inability to distinguish between them may have biased the earlier result. In particular, it is not clear that the potential effects of the MST and BCS fees should be the same in considering these two treatments relative to mastectomy.

This research brief extends that research, offering several improvements in methodology. We analyze data for individual elderly Medicare beneficiaries with confirmed cases of newly diagnosed early-stage breast cancer. Using data on individual patients avoids the ecological fallacy. In addition, we treat BCSO and BCSRT as distinct surgical outcomes relative to MST, which allows fee effects to differ among the three possible treatment choices. We also use more refined measures of Medicare fees than the prior study. Adjusting average fees across areas for the use of modifiers and variations in specific procedure codes provides a more accurate measure of the true variation in Medicare fees for these procedures.

METHODS (1)

Patient Population

The sample for this analysis was a subset of 1,787 elderly Medicare patients who were treated for early-stage (I, IIa, and IIb) breast cancer in 1994 and were part of the Breast Cancer OPTIONS (Outcomes and Preferences for Treatment in Older Women, Nationwide Study) Project. The OPTIONS sample was designed to be representative of all elderly beneficiaries with newly diagnosed, early-stage breast cancer in Medicare's fee-for-service program between 1992 and 1994. Approximately 3,800 cases were sampled over the three years. This analysis is limited to women treated in 1994 because that is the year for which the Medicare fee variables were constructed.

The sample was drawn from Medicare's database containing all (inpatient and outpatient) claims for a 5-percent random sample of all beneficiaries. Potentially eligible cases were extracted from claims with either a breast cancer diagnosis or a breast surgery procedure code. Following earlier studies (Nattinger et al. 1992; Nattinger et al. 1996), we excluded women whose claims indicated a history of prior breast cancer, carcinoma in-situ without invasive disease, codes for metastatic disease, bilateral breast procedures, and diagnoses of breast cancer without a surgical procedure code. We also excluded women for whom breast surgery was not the primary procedure code or for whom breast cancer was not the primary diagnosis. Women younger than 67 years old were deleted in order to allow for up to two years of prior Medicare claims to evaluate the effects of preexisting health status on treatment received. Finally, we excluded women whose claims were missing a physician identifier or where the physician provider number could not be matched in the Health Care Financing Administration's (HCFA) provider database.

To confirm the cancer diagnosis and determine cancer stage, which are not available from claims, we surveyed the physicians identified from the claims. Physicians were contacted in random order until the target sample size was attained. Of those contacted, 80.7 percent provided information, while 10.6 percent were unable to supply the information requested, and only 8.7 percent refused.

Based on physicians' reviews of their medical records, we deleted women who were ineligible (no cancer, late-stage disease, secondary or recurring cancer), whose eligibility could not be determined because of missing or incomplete records, or whose Medicare beneficiary numbers could not be matched to other Medicare data. We subsequently omitted approximately 2 percent of eligible women with some missing data, rather than imputing missing values, leaving a final analytic sample included 1,787 unweighted cases who were treated in 1994.

Data Sources

Most of the variables used in the analysis were constructed from Medicare data sources: the national claims history file, which contains all inpatient and outpatient claims; the beneficiary enrollment file, which contains location and basic demographic information; and the physician provider file. Disease stage was obtained from the physician survey to verify patients' eligibility. We used the 1990 census file of population characteristics by zip code to create proxy measures of socioeconomic status. Hospital information was obtained from the American Hospital Association's Annual Survey of Hospitals.

Model Specification

The empirical specification is motivated by an economic model of the demand for and supply of treatments for localized breast cancer. A woman's demand for a particular treatment is assumed to depend on out-of-pocket costs (money price), inconvenience costs (time and difficulty of travel, needing help at home), preferences, and clinical factors (disease stage and comorbidity).

The surgeon's likelihood of recommending one treatment over another (supply) is hypothesized to depend on the expected fee for each treatment, the costs of providing alternative treatments, and physicians' preferences. This model suggests that variations in either fee, holding the other constant, may affect the treatment recommendation when clinical factors are held constant. If provider costs are higher for one treatment than for another, then there may be a tendency to offer the lower cost treatment. Last, the surgeon's experience may reflect differences in preferences for alternative treatments il for example, older surgeons were slower to adopt the 1991 National Institutes of Health consensus guideline on the equivalency of BCS and MST in treating localized breast cancer (National Institutes of Health 1991).

Dependent Variable (Treatment Received)