Ethnic Disparities in Imaging Utilization at Diagnosis of Non-Small Cell Lung Cancer

Rustain L. Morgan, MD, MS; Sana D. Karam, MD, PhD; Cathy J. Bradley, PhD

Disclosures

J Natl Cancer Inst. 2020;112(12):1204-1212. 

In This Article

Methods

Data

We conducted a retrospective cohort observational study using patients from the Surveillance, Epidemiology, and End Results (SEER)-Medicare linked database. SEER, a program of the National Cancer Institute (NCI), encompasses person-level information on cancer survival and incidence from 18 population-based tumor registries that cover approximately 28% of the United States.[1] Medicare claims and census information were linked to the SEER tumor registry data, which have additional information on patient demographic and tumor characteristics. By linking SEER data to Medicare claims, dates of service, payments, procedures, and diagnosis codes are also captured.

Cohort Selection

We selected patients aged 66 years or older at diagnosis whose first and only primary tumor was lung cancer (International Classification of Diseases [ICD]-O-3 site codes: C34.0, C34.1, C34.2, C34.3, C34.8, and C34.9) diagnosed between 2007 and 2015. We limited the study to patients with the following non-small cell histology types: adenocarcinoma, squamous cell carcinoma, large cell carcinoma, and other non-small cell carcinoma (n = 137 882). Patients who were diagnosed by autopsy, had unknown diagnosis dates, or survived less than 2 months were excluded, leaving 107 548 patients. We further limited the sample to patients continuously enrolled in fee-for-service Medicare Parts A and B for 12 months before through 12 months after the month of diagnosis (or until death if it occurred within 12 months) to ensure complete claims history. Additionally, we required that patients were enrolled in Medicare Part D prescription drug plans from the month of diagnosis through the following 12 months to capture oral chemotherapy treatment. We further limited the sample to patients who underwent PET with CT imaging during the diagnostic period (n = 36 469) (Figure 1). We used ICD, 9th and 10th revision, Clinical Modification codes, Current Procedural Terminology codes, and Healthcare Common Procedure Coding System codes to identify the diagnoses and procedures. The study was approved by the University of Colorado Multiple Institutional Review Board.

Figure 1.

Sample derivation. Newly diagnosed non-small cell lung cancer (NSCLC) in patients 66 years and older who survived at least 2 months and continuously participated in Medicare fee-for-service for 12 months before and following diagnosis. CT = computerized tomography imaging; PET = positron emission tomography imaging.

Outcomes

The primary outcome was imaging modality at diagnosis, and the secondary outcome was cancer-specific survival (CSS) differences between imaging modality. We identified imaging modality using Healthcare Common Procedure Coding System and Current Procedural Terminology procedure codes reported during the diagnostic period, defined as 3 months before diagnosis through the earlier of two potential dates: 2 months after the month of diagnosis or 30 days after the initiation of therapy. Patients were categorized as having 1) any PET imaging with or without CT imaging, or 2) CT imaging alone. CSS and overall survival (OS) were estimated 12 months after diagnosis. CSS time was determined using SEER dates of death, which include cause of death and are reported through December 2015. OS time was determined using Medicare dates of death reported through December 2016. Patients surviving longer than 12 months were censored at 12 months.

Statistical Analysis

Chi-square tests were performed to assess the univariate association of categorical variables and imaging modalities at diagnosis. Logistic regression was used to evaluate the adjusted impact of demographic, socioeconomic, and facility characteristics on choice of diagnostic imaging modality. A generalized estimating equation model was used to obtain standard errors that accounted for clustering by the facility that provided initial treatment.

Survival curves were generated using the Kaplan-Meier method, and Cox proportional hazards regression was used to estimate adjusted hazard ratios (HRs) for 12-month CSS and OS. In multivariable survival analysis, we used a robust sandwich covariance matrix estimate to account for clustering by facility.[12]

We examined a series of Cox proportional hazards models. In the base model, we adjusted for year of diagnosis, patient sex (male or female), age at diagnosis (66–69 years, 70–74 years, and 75 years or older), type of NSCLC (squamous or other), race or ethnicity (white, black, Hispanic, other), marital status (married or partnered, or single), SEER registry site, census tract Rural Urban Commuting Area Codes (urban commuting area or not an urban commuting area), census tract poverty level (less than study sample median or greater than or equal to study sample median level), census tract-level percentage high school education or less (less than study sample median level or greater than or equal to study sample median level), additional imaging (Magnetic Resonance Imaging [MRI] brain yes or no), and whether first cancer-directed therapy was provided by an NCI-designated center, teaching hospital, or community hospital (Supplementary Table 1, available online). We used Medicare claims from the year before diagnosis to estimate the Charlson Comorbidity Index according to the NCI's adaptation of the algorithm described by Klabunde et al.[13]

In subsequent Cox proportional hazards models, we added the following variables in a stepwise fashion: imaging modality, derived American Joint Committee on Cancer Stage Group 6th edition, and initial treatment. Initial treatment was defined as treatment initiated within 6 months of diagnosis, categorized as no treatment, chemotherapy, radiation therapy, surgery, as well as combinations of these treatments. We used Schoenfeld residuals to validate the proportional hazard assumption in the Cox model. All statistical analyses were performed with SAS 9.4 (SAS Institute, Cary, NC), were two-sided and were evaluated at a statistical significance level of P less than .05.

Comments

3090D553-9492-4563-8681-AD288FA52ACE

processing....