Regional differences in recommended cancer treatment for the elderly

We chose to study treatments for cancers in which we had prior research experience and were therefore most familiar with recommended guidelines. Previous literature was reviewed to identify 7 recommended courses of treatment for colorectal, pancreatic, and prostate cancer that could be readily measured with claims data. These recommended courses include 1) removal and pathological examination of at least 12 lymph nodes during resected colon cancer, 2) adjuvant chemotherapy within 4 months of diagnosis for patients under age 80 with regional colon cancer, 3) radiation therapy within 6 months of diagnosis for patients under the age of 80 with regional spread of rectal cancer who received surgical resection, and 4) postoperative adjuvant chemotherapy for patients under age 80 years with regional rectal cancer [27]. Relatively low rates of adherence to these recommended treatments have been documented in the U.S. previously [28]. 5) Pancreatic resection is recommended for patients with localized disease, or for patients with regional pancreatic cancer that is not locally advanced [29]. Low rates of pancreatic resection have also been documented in the U.S. [15]. All of these treatments were recommended courses of care during the study’s sample period.

Appropriate treatment for men with non-metastatic prostate cancer depends on disease characteristics; therefore, we distinguish between cancers with a “favorable” versus an “unfavorable” prognosis. Patients with non-metastatic prostate cancer can be treated with external beam radiation therapy, brachytherapy, prostatectomy, cryosurgery, androgen deprivation therapy [16]. However, there is no compelling data from published randomized trials that men age 65 and older with relatively low risk prostate cancer derive a survival or quality of life benefit from cancer-directed treatment [30, 31]. In fact, randomized trials demonstrate that in older men with low-risk prostate cancer, observation yields similar survival and decreased morbidity compared with up-front treatment [31, 32]. Most men with a diagnosis of low-risk prostate cancer in the United States received up-front treatment with prostatectomy or radiotherapy and are thus exposed to the risk of treatment-induced urinary dysfunction, rectal bleeding, and impotence. We define appropriate treatments for prostate cancer as 6) receipt of cancer-directed treatment within one year of diagnosis for those patients with an unfavorable prognosis, and 7) no cancer-directed treatment within one year (observation) for patients with a favorable prognosis. Relatively high rates of cancer-directed treatment have been documented for prostate cancer patients with both favorable and unfavorable prognoses [16].

We analyzed data from the Texas Cancer Registry-Medicare linked database. Medicare provides health insurance for U.S. residents who are 65 or older. The linkage was performed under the guidance of TCR, the National Cancer Institute, and the Center for Medicare and Medicaid Services. The data used in this study includes patients diagnosed with the study cancers between 2004 and 2007 and their Medicare claims through 2008. Patients in the sample met the following criteria: diagnosed and reside in Texas, age 65 years or older, first diagnosed in 2004 to 2007 and not at time of death, first primary cancer and no second primary cancer within 12 months, histology confirmation, continuously enrolled in Medicare Parts A and B pre- and post-diagnosis, and not a member of an HMO 12 months pre- and post-diagnosis. Analysis was limited to appropriate subpopulations when recommended courses of treatment designated specific cancer stages for care. Only men were included in the analysis of prostate cancer treatment.

Patient-level variables evaluated included gender, age at diagnosis, race/origin, Charlson comorbidity score using a NCI defined algorithm, [33] tumor size, cancer stage, and year of diagnosis. Age was categorized into the intervals 65–70, 70–75, 76–80, and 81+. Analysis for colon and rectal treatments was limited to patients 80 and under, based on treatment guidelines. We also controlled for certain census tract-level indicators of patient socioeconomic status. Median income in the census tract of residence from the 2000 U.S. Census was included in the TCR-Medicare linked database. Median incomes were classified into quartiles ($ <31,000; $31,000–< $39,000; $39,000–<$53,000; and $53,000+). These variables included urban/rural location, percent of individuals who do not speak English, percent of individuals who have completed at least some college, and median income. We determined the hospital referral region (HRR) that each patient resided in. HRRs represent regional health care markets for tertiary care [34]. The U.S. is divided into 306 HRRs, and 22 are in Texas.

The TCR data and Medicare billing claims were used to determine the treatments within the specific time frame of cancer diagnosis; the methods to identify the treatments have been previously described in the literature [16, 28, 29]. Patients were identified from the TCR Patient Entitlement and Diagnosis Summary file using major site groups that SEER has defined based on the primary site and ICD-O-3 morphology. The TCR contained the cancer disease stage at diagnosis for each patient.

We also classified the prostate cancer patients as favorable risk (stage T1 or T2 tumor and low histologic grade) versus unfavorable risk (T3 or T4 tumor or intermediate/ high histologic grade) [16]. PSA information is not currently captured by TCR and so could not be incorporated into patient risk stratification.

We counted the number of specialists per 1,000 Medicare elderly patients who had at least one Medicare claim with an accompanying diagnosis of neoplasm (ICD9 diagnosis codes 140–239) in each Hospital Service Area (HSA). HSAs are local health care markets for hospital care. An HSA is a collection of ZIP codes whose residents receive most of their hospitalizations from the hospitals in that area [34]. There are 3,436 HSAs in the U.S. and 208 in Texas. The Dartmouth Atlas, which defined HRRs and HSAs, aggregates the HSAs to define HRRs. Specialist types were defined based on the treatment guideline being examined. For example, for surgical removal of lymph nodes during colorectal resection, we counted the number of surgeons per HSA. But for radiation therapy for a diagnosis of regional rectal cancer, the number of radiation oncologists was counted. Specialists associated with each cancer are listed in Additional file 1: Table S1.

We first report the mean treatment rates for each cancer type in our study across all patients in Texas. We then report the treatment rate in the median HRR in the state for each recommended course of care, as well as the lowest and highest treatment rates by HRR. We list the name of the HRR associated with each reported rate, so that we can look for similarities and differences in cancer care across regions in Texas.

In regression analyses for each treatment, the dependent variable was set equal to 1 if the patient received the recommended treatment given their diagnosis and 0 otherwise. We first estimated logistic regressions that included only the HRRs as explanatory variables to test for variation in appropriate treatment across parts of Texas. For each treatment, we set the HRR with the median treatment rate for the sample as the excluded category in the regressions. We then tested for differences in treatment rates across HRRs, adjusting for patient characteristics, census tract-level characteristics, and the supply of cancer specialists in the HSA. Cancer stage (local, regional, or distant versus in situ) was included in the colon cancer regression. For the other treatments, cancer stage was used to define the cancdidate treatment population and was therefore not a regressor. We include the supply specialists when measuring the determinants of favorable treatment for prostate cancer, because we seek to test whether the overly high rate of cancer-directed treatment for this patient group may be an example of “supply-sensitive care.” This term refers to clinical activities for which the frequency of use is related to the capacity of the local healthcare system, but which do not yield better health outcomes [11, 35]. All regression analyses were conducted using Stata 11.2, and standard errors were computed using the cluster option to account for correlation in unobservables across HRRs [36]. Means of all of the explanatory variables included in the regressions are listed by treatment status in Additional file 1: Table S2.