Variation in geographic access to chemotherapy by definitions of providers and service locations: a population-based observational study

Patients who live in areas where chemotherapy providers are scarce must travel extended distances to receive treatment [1–3]. Research suggests that some patients who may benefit from chemotherapy remain untreated because of where they live [1, 2, 4–6]. Thus measures of geographic access to chemotherapy have been constructed to better understand these potential barriers to treatment. The two most common geographic access to care measures include local area density and distance to the nearest chemotherapy provider [1, 7–9]. To construct these measures, researchers must accurately identify chemotherapy providers and their service locations. Failure to account for all types of service locations can bias estimates of geographic access to chemotherapy. For example, in a recent study, the median driving time to the nearest chemotherapy provider decreased from 51.6 to 19.2 min when visiting consulting clinics were included [10].

Measures of geographic access to chemotherapy are often created using Medicare claims because they offer detailed information regarding treatments, providers and patients [11]. Unfortunately, even with this information, creating measures of geographic access to chemotherapy is not completely straightforward, as noted in the literature [12, 13]. Baldwin et al. state that researchers may need to distinguish between those who claim to be oncologists (through specialty codes) and those who do the work of an oncologist (through billing codes) when identifying chemotherapy providers [12]. They also acknowledge that identifying service locations is not trivial, as providers can and do practice at more than one location [12]. Should researchers assume that chemotherapy can only be provided at locations where it has been in the past, i.e. “proven” service locations? Or is reasonable to assume that chemotherapy may be available at each practice location, i.e. “possible” service locations?

Although the challenges of identifying providers and service locations have been previously noted, no research has quantified the effect of these various definitions on measures of geographic access to care. As there exists no standard protocol in how to define providers and locations, the definitions vary across studies. Some have used specialty codes to identify providers [14–16]. One paper used billing codes [17] and another used both billing and specialty codes [5]. Another strategy has been to supplement Medicare data with outside sources, such as the American Medical Association (AMA) physician masterfile or the Area Health Resource Files [1, 2, 4, 18–20]. Although doing so has been shown to capture more oncologists [12, 13], our objective was slightly different. We were interested in how various methods of identifying chemotherapy providers and locations impacted the two most commonly used measures of geographic access to chemotherapy: local area density and distance to nearest provider.

Among the 216,438 unique providers with usable ZIP code information who cared for our study population in any capacity, 4,424 reported an oncology specialty code and 6,253 billed Medicare for a chemotherapy-related service. These 4,424 providers, as defined by specialty codes, provided chemotherapy in 5,076 locations (proven service locations) and practiced in a total of 5,157 locations (possible service locations). The 6,253 providers, as defined by chemotherapy billing codes, provided chemotherapy in 6,722 locations and practiced in 8,101 locations. Thus, on average, providers administered chemotherapy in 1.08-1.30 locations, depending on how provider and service locations were defined.

Geographic variation in access to chemotherapy

Geographic access to chemotherapy (defined here as any provider practicing in their proven locations within 20 min of a beneficiary) was mapped for three representative SEER registries in Fig. 2. Panel A identified chemotherapy providers through specialty codes and panel B through chemotherapy billing codes. Brownish-grey areas signify that no beneficiary was diagnosed with lung-cancer in that ZIP code during the study period. Regardless of which method was used to identify providers, fewer Iowa beneficiaries were within a 20-min driving radius of a chemotherapy provider than beneficiaries residing in New Jersey or Connecticut. Similarly, Iowa beneficiaries were less likely to reside in the same ZIP code as a chemotherapy provider than beneficiaries residing in Connecticut or New Jersey: 17.9 % when identified by specialty code and proven locations in Iowa versus 23.5 % in Connecticut and 26.9 % in New Jersey (Table 3). When identified by chemotherapy billing codes, 35.0 % of Iowa beneficiaries resided in the same ZIP code as a provider, a substantial and statistically significant increase over identification by specialty code (p < 0.0001). Indeed, this rate was similar to the rate in Connecticut (36.0 %), although still less than New Jersey (43.5 %).

Access to treatments for cancer, whether geographic or financial, continues to be of interest to policy makers and researchers [1, 10, 27, 28]. However, previous papers estimating geographic access to chemotherapy using Medicare claims have not assessed whether their results are sensitive to how chemotherapy providers and service locations are identified. We found that the method of identifying who, but not where, chemotherapy is provided affected our two measures of geographic access to chemotherapy.

In our analysis, a number of specialties in addition to hematology/oncology and medical oncology billed Medicare for chemotherapy-related services, with 42.8 % of the providers associated with chemotherapy claims reporting a specialty other than hematology/oncology or medical oncology. Even so, physicians who reported a specialty of hematology/oncology or medical oncology provided the vast majority of the chemotherapy services (91.0 % of the claims).

When we identified chemotherapy providers through chemotherapy billing codes rather than specialty codes, we found that average driving-time distances for beneficiaries decreased by almost a quarter and the average number of providers within 20 min increased two-fold. An additional 50 % of beneficiaries resided in the same ZIP code as a chemotherapy provider when providers were identified using billing codes instead of specialty codes (35.7 % versus 23.8 %). This suggests that a number of oncologists might be reporting a specialty other than hematology/oncology and medical oncology. There are a number of reasons why this may be the case. First, all oncologists are trained in internal medicine before they specialize in oncology. Also, specialty certification is independent of maintaining one’s license, the former not mandatory for practice. Finally, accuracy in specialty code reporting is not required for reimbursement. Any/all of these factors could cause underreporting of oncology specialty codes.

The increase in geographic access to chemotherapy when providers are identified through chemotherapy billing codes instead of specialty codes is consistent with the literature showing underreporting of oncology specialty on Medicare claims [12, 13]. This does not necessarily imply that all of the providers associated with chemotherapy claims were indeed oncologists. It may also be the case that other specialties truly are providing chemotherapy. One could concoct a scenario where chemotherapy was provided by a non-oncologist under extenuating circumstances, but to which future provision of chemotherapy from that same provider is unlikely. Thus there is an additional dimension which has yet to be explored in the literature - whether past provision of chemotherapy is indicative of future access.

Although our measures of geographic access to chemotherapy varied substantially by how we identified chemotherapy providers, there was no statistical difference across service location definitions. This was due to the fact that chemotherapy providers in this sample in general provided chemotherapy at each location in which they practiced.

There are limitations to our study. First, we only utilize Medicare claims. Inclusion of additional data would affect the magnitude of our estimates, as others have found [13]. Unfortunately, NCI no longer releases unencrypted Unique Physician Identification Number (UPIN) or National Provider Identifier (NPI), rendering these linkages impossible. Our contribution to the literature, however, is not providing “the” estimate of geographic access to chemotherapy but instead assessing how definitions of “who” and “where” affect these estimates, something that others have noted might be an issue but never quantified [12]. In addition, we only include Carrier claims, which comprise about 75 % of chemotherapy claims. However the outpatient files do not include specialty information.

Another limitation is that we only include lung cancer patients and their providers. However, unless lung cancer patients are systematically avoiding some chemotherapy providers by how the providers report their specialty, our estimates of geographic access to care across definitions of providers and locations will not be biased. If some oncologists are sufficiently specialized (e.g. at an NCI-designated comprehensive cancer center) such that they never treat lung cancer patients, then they would not be captured in our analysis. However, these sorts of locations would also likely have oncologists that do specialize in lung cancer. Although absolute measures of access would not be accurate, relative measures across the various definitions, our outcome of interest, would still be valid. We also only include a single year of data. In doing so we may not capture physicians and their service locations if they bill Medicare less than once a year. Finally, our distances are calculated between ZIP code centroids. We acknowledge that distances between centroids are not as accurate as mailing addresses, and that driving times within a centroid is not zero. In addition, for ZIP codes that cover a large geographic area, the next closest ZIP code may be more than 20 min away. If there are no chemotherapy providers in the same ZIP code as the patient, that area would be considered as not having access to chemotherapy as we have defined it. ZIP codes can also change over time, increasing the complexity of geographic analyses.

Improving geographic access to care and reducing health disparities have been of interest to healthcare providers and policy makers for some time and will continue to be so. Oncology services in particular are estimated to experience shortages in the coming years due to an aging population and insufficient growth in the physician workforce [29]. The effect is likely heterogeneous, as demographic compositions vary across areas and over time due to both aging in place and retirement migration. To the extent that measures of geographic access to care affect health policy, it is imperative that these measures be accurate. Our study suggests that measures of access to chemotherapy are more sensitive to how one identifies chemotherapy providers than service locations. Given that others have shown underreporting in specialty codes, perhaps both methods of identifying providers ought to be used to bound estimates of measures of geographic access to care.

HCPCS, Healthcare Common Procedure Coding System; ICD9-CM DX, International Classification of Diseases – 9^th edition, Clinical Modification, diagnosis code; NCI, National Cancer Institute; NPI, National Provider Identifier; PEDSF, Patient Entitlement and Diagnosis Summary File; SD, standard deviation; SEER, Surveillance, Epidemiology and End Results; UPIN, Unique Physician Identification Number; ZIP, Zone Improvement Plan