Medicare Claim Processors’ Reimbursement and G-CSF Choice Among Non-Hodgkin Lymphoma Patients | Page 2
Published Online: August 21, 2013
Xiaoyun Pan, PhD and Usha Sambamoorthi, PhD
Key Variable: Medicare Claim Processors. This variable was derived using information on patients’ county of residence within SEER programs and Medicare claim processors recorded in Medicare Part B claims. For each county, we used the most frequently billed Medicare claim processor. When 2 or more Medicare claim processors served a particular county we had a combination code. For example, the highest percentage of physician visit claims in Johnson County, Iowa, were submitted to the Medicare claim processor Iowa Wellmark Inc, and the highest percentage of outpatient claims in Johnson county were submitted to Nebraska Blue Cross. Therefore, for Johnson County, we combined both claim processors: Iowa Wellmark and Nebraska Blue Cross. Because claim processors serving a county might change across years (eg, some claim processors were terminated in a particular year), the claim processors were assigned annually to each county. After grouping of claim processors by each county, there were 108 Medicare claim processor groups in our study population.
Determination of Reimbursement Policies. Reimbursement policies were identifi ed in 2 steps using the average physician reimbursement amount for the first cycle of chemotherapy and the number of chemotherapy administration codes. In the fi rst step, for each patient, we defined the first cycle of chemotherapy as a fixed period of 21 days following the initiation of chemotherapy. Wechose 21 days as the cutoff period because most chemotherapy regimens last for 21 days.22 Chemotherapy was identified from the outpatient and physician visit claims. Healthcare Common Procedures Classifi cation System (HCPCS) codes (J8999-J9999, Q0083, Q0084, Q0085, J7150, 964XX, and 965XX) and revenue center codes (0331, 0332, and 0335) were used to identify chemotherapy regimens and administration procedures. Only for year 2005, the following codes were also used to identify chemotherapy claims: 51720, G0355, G0356, G0357, G0358, G0359, G0360, G0361, G0362, and G0363.23 In this step, total physician reimbursement amount derived from outpatient and physician visit claims during the first cycle of chemotherapy was used to define reimbursement policies. As our study period covered multiple years, total physician reimbursement amount was converted to 2008 dollars based on the Consumer Price Index for medical services published by the bureau of labor statistics.24 Medicare payments also include the Geographical Practice Cost Index (GPCI) component and refl ect costof-living adjustments based on geographical region. To standardize the different price levels for healthcare inputs across location and time, we divided the reimbursement amount by the GPCI from Federal Register. We estimated average reimbursement amount for each Medicare claim processor after controlling for clinical and non-clinical factors using regression techniques.
In the second step, the average estimated reimbursement amount per claim processor computed from the regression was grouped into top and bottom deciles. Medicare claim processors in these deciles were further examined for chemotherapy administration codes to determine single bundled payment and separate payments for services and drugs.
Prescription of G-CSF. We constructed an indicator variable representing presence or absence of any G-CSF prescription during the fi rst cycle of chemotherapy (ie, within 21 days following the initiation of chemotherapy). G-CSF drug was identified from outpatient and physician visits based on HCPCS codes (J1440, J1441, Q0453, S0135, and J2505).
Other Independent Variables. Demographic variables were: patient’s age at diagnosis (66-70, 71-75, 76-80, 81-84, >85 years), gender (female/male), race (white, African American, other). Patients’ clinical characteristics were: histology type, stage of cancer, lymph node involvement, prior-diagnosis inpatient and outpatient comorbidities, and grade. We used both physician and inpatient claims to identify comorbid conditions 1 year prior to NHL diagnosis. Using methodology developed by Klabunde et al,25 comorbidity index was computed using non-cancer Charlson condition26 from inpatient claims and physician or outpatient claims. Our comorbidity index may be limited in scope because it did not include all co-occurring conditions present in an individual. Area-level socioeconomic variables such as income level, education level, and percentage of white were used as independent variables as well.
Type of chemotherapy (anthracycline-based chemotherapy [ABC] with rituximab, ABC without rituximab, rituximab with other non-ABC chemotherapy, other non-ABC chemotherapy only, and rituximab only) was used as an independent variable because reimbursement amounts may vary by type of chemotherapy. Site of visit (outpatient vs office-based physician visits) was also distinguished and used as one of the independent variables. Year of first chemotherapy treatment was included as a time trend variable to account for differences in treatment patterns over time.
Statistical Techniques. Reimbursement policies were measured using estimates from an ordinary least square regression (OLS) model. The dependent variable was total physician reimbursement amount derived from outpatient and physician visit claims during the first cycle of chemotherapy. Indicator variables for each of the 108 Medicare claim process or groups were the key explanatory variables in the model. The reference group for Medicare claim processor variables was the claim processor group with the lowest estimated chemotherapy reimbursement amount. In addition to Medicare claim processor variables, this regression also included patient-level demographic and clinical factors and arealevel socioeconomic status listed in the measures section. Adjusted differences in average total reimbursement amounts by Medicare claim processor groups were tested by Chow F-statistics. The average estimated chemotherapy reimbursement amount by claim processor group was computed using the regression coefficients and intercept.
To evaluate the association between G-CSF prescription and average estimated reimbursement amounts, we conducted logistic regressions. In the first model, we included only patient-level demographic, clinical factors, and area socioeconomic status to describe the association between these variables and G-CSF prescription. In the second model, we additionally included policy-level variable (ie, average chemotherapy reimbursement levels by claim processor groups estimated from OLS regression).
PDF is available on the last page.