Lack of Adherence to Guideline-Based Imaging Before Subsequent Radiation in Patients with Non–Small Cell Lung Cancer: Impact on Patient Outcomes

Data Sources

This was a retrospective study of patients with NSCLC in the Surveillance, Epidemiology, and End Results (SEER) Medicare-linked database. We compared the outcomes of patients who underwent RT for oligoprogressive or recurrent disease based on whether their most recent imaging was CT or ¹⁸F-FDG PET/CT. SEER, a program of the National Cancer Institute, includes information on patient demographics, tumor characteristics at diagnosis, and treatment information from 18 population-based tumor registries that cover approximately 28% of the U.S. population. Medicare claims provide information on the health services received and on the facility where they were received, and census information is linked to the SEER data for additional geographically based sociodemographic factors (23,24).

Cohort Selection

We selected patients aged 66 y or older at diagnosis, whose first primary tumor was NSCLC, diagnosed from 2007 through 2015. Patients were included at this age so there would be at least 1 y of data to identify patients who had cancer before enrolling in Medicare. We excluded patients who were diagnosed via death certificate or at autopsy, as well as patients who had an unknown diagnosis date. We required continuous enrollment in fee-for-service Medicare parts A and B from 12 mo before through 12 mo after the month of diagnosis (or until death if it occurred within 12 mo) to ensure a complete claims history for comorbidities before diagnosis and for services received after diagnosis. We also excluded patients who had no paid claims in the 12 mo after diagnosis, patients who were stage 0, patients who were missing the census tract, and patients who had an unknown stage, rural measure, or race.

To best identify patients with disease requiring subsequent RT, we required a specific sequence of events to have occurred after diagnosis. From the cohort, we identified anyone who received initial treatment within 6 mo of diagnosis and had subsequent RT within 36 mo of completion of the initial treatment. A gap of 90 d or more was required to distinguish the initial treatment and subsequent treatment. As we were interested in outcomes by modality of imaging for oligoprogressive disease, we further limited our sample to patients with imaging occurring within 60 d before the subsequent RT. Our final sample with qualifying treatment and imaging included 5,017 patients. All inclusion and exclusion criteria are outlined in Supplemental Figure 1 (supplemental materials are available at http://jnm.snmjournals.org). Per NCCN guidelines regarding treatment, this sample could include patients with local or metastatic recurrent disease or patients with progression at a limited number of sites (oligoprogressive disease).

We used the International Classification of Diseases, ninth and 10th revisions; clinical modification codes; current procedural terminology codes; and Healthcare Common Procedure Coding System codes to identify the diagnoses and procedures.

Treatment Identification

Initial treatment was defined as chemotherapy, RT, or surgery for NSCLC within 6 mo after diagnosis. We categorized initial treatment as surgery alone, RT alone, chemotherapy alone, surgery and RT, surgery and chemotherapy, chemoradiotherapy, or surgery and chemoradiotherapy. We required at least 90 d to define separate treatment periods; subsequent RT after the initial treatment period was defined as having occurred at least 90 d after initial treatment. We also required this subsequent RT to have occurred within 36 mo of initial treatment. Codes used to identify initial treatment and subsequent RT can be found in Supplemental Table 1.

Imaging Identification

Imaging procedures of interest were defined as imaging occurring after initial treatment and within 60 d before subsequent RT. These were categorized into 2 groups: CT or CT angiography (CTA) of the abdomen, chest, and pelvis, and any ¹⁸F-FDG PET either with or without CT/CTA. Initial imaging at the time of diagnosis was also assessed. The imaging codes used in this study can be found in Supplemental Table 1.

Statistical Analysis

For comparing baseline characteristics of demographic and diagnostic information, χ² testing was performed to test statistically significant differences in each categoric variable by imaging modality. The categories tested were year of diagnosis, age at diagnosis (66–69, 70–74, or ≥75 y), race (White, Black, Hispanic, or other/unknown), sex (male or female), marital status (married/partner or single), region, residence (in an urban area or not), socioeconomic status (poverty level and percentage of individuals with a high school education or less), comorbidities before diagnosis, facility type (National Cancer Institute–designated center, teaching hospital, or other/no hospital), initial derived American Joint Committee on Cancer Stage Group (sixth edition), initial imaging results, histology subtype, and type of initial treatment. The reason for analyzing race is that multiple prior studies have shown that racial and ethnic minorities are not receiving the recommended imaging for their disease. We used Medicare claims from the year before diagnosis to estimate the Charlson comorbidity index according to the National Cancer Institute’s adaptation of the algorithm of Klabundle et al. (25).

Logistic regression was performed to identify predictors of imaging type, adjusting for the covariates included in the univariate table, and odds ratios (ORs) between groups were compared.

Overall survival and cancer-specific survival between imaging modalities was estimated at 36 mo after diagnosis. Overall survival was determined using Medicare-reported dates of death, which are reported through December 2018. However, cancer-specific survival was determined using SEER-reported dates of death, which include cause of death and are through December 2016. Patients living longer than 36 mo were censored at 36 mo.

The Kaplan–Meier method and unadjusted Cox proportional hazards were used for univariate survival analyses. Multivariate cancer-specific survival analyses were completed using Cox proportional-hazards models, adjusting for covariates. The proportional-hazards assumption was evaluated using Schoenfeld residuals, and violations were addressed using time-dependent interaction terms in multivariable models. A forward-stepwise analysis was completed. The base model included demographic characteristics; the stratified model added stage, initial treatment, and facility type to the base model; and the full model included all previous covariates and the imaging modality.

All statistical analyses were performed with SAS, version 9.4 (SAS Institute), and evaluated at a critical α of 0.05. The study was reviewed and approved by the University of Colorado Institutional Review Board. It was conducted under a data use agreement with the National Cancer Institute.