Data Source
The source of information for this study was the Puerto Rico Central Cancer Registry-Health Insurance Linkage Database (PRCCR-HILD). As one of the main surveillance systems in the island, the PRCCR monitors all cancer diagnoses obtained from health facilities in Puerto Rico that diagnose and treat cancer patients, such as hospitals, outpatient clinics, pathology laboratories and RT/chemotherapy sites. Demographic characteristics, date of cancer diagnosis, anatomic cancer site, histology type, method of diagnosis, stage of disease at diagnosis, therapy and follow-up status, and cause of death are included in the type of information collected. Data in the PRCCR files are linked to the insurance claims files from Medicare, Medicaid and private insurers in order to constitute the PRCCR-HILD. The process of linking claims from the health insurance databases is performed using a deterministic match similar to the one used by SEER-Medicare [12]. All data is de-identified so that the health information is not linked to individual patients.
The PRCCR has been part of the United States National Program of Cancer Registries (NPCR) since 1997. The Surveillance, Epidemiology, and End Results (SEER) Program and the North American Association of Central Cancer Registries (NAACCR) standards are used for coding data. In the most recent NPCR audit, the PRCCR complied with all the criteria, including the completeness of case ascertainment (>99.0%), comparable to the United States median (99.95%) [13].
Study Cohort
This study included patients aged 21 years or older in the PRCCR-HILD with a primary diagnosis and histologic confirmation of metastatic NSCLC (defined as stage III-B with malignant effusion (wet IIIB) or stage IV NSCLC [14] between January 2009 and December 2015. Only residents of Puerto Rico at diagnosis were included. NSCLC was required to be pathologically confirmed and diagnosed prior to death as the primary cancer diagnosis. To ensure we had adequate claims information for our analyses, patients with incomplete claims data regarding cancer diagnosis (cancer type and cancer stage), with multiple primary diagnoses, with missing diagnosis date, without claims data after the primary diagnosis, without enrollment data during the first year after diagnosis, or with a death certificate within 15 days from diagnosis were excluded (Figure 1).
Study Variables
The variables of interest for this study were identified via data in the PRCCR-HILD as well as insurance claims codes for the Current Procedural Terminology (CPT), Healthcare Common Procedure Coding System (HCPCS), and the International Classification of Diseases for Oncology 3th (ICD-O-3). The patient’s diagnosis was identified using the ICD-O-3 codes for metastatic NSCLC site codes C340-C349, excluding small cell lung cancer (histology codes: 80413, 80423, 80433, 80443, and 80453), lymphomas, and sarcomas.
Our primary outcome variable, use of palliative RT, was defined by a combination of patient cohort (metastatic NSCLC primary diagnosis) and patient claims using HCPCS, CPT, and ICD codes for radiation courses occurring within one year of diagnosis. Codes used for identifying receipt of palliative RT included stereotactic RT, conventional radiation treatment deliver, conventional IMRT delivery, high-energy neutron radiation, and proton treatment delivery. Previous work with surveillance registry has also considered that the RT for patients with advanced NSCLC is being provided with palliative intent as a standard of care [15].
The independent variables were categorized in order to assess for nonlinear trends. Demographic variables included age, sex, and marital status. Health services variables included health insurance type, density of RT centers, and geographic location. Density of RT centers was defined as the proportion of RT center per region of patient residency, as defined by Health Department Coalitions by Health Region [16,17]. This variable was categorized according to the distribution of the study sample: low (0-2 centers), medium (3-6 centers), high (> 7 centers). Geographic location referred to whether or not the patient’s municipality of residency belonged to a metropolitan area, as defined by the Office of Management and Budget (OMB) [18,19]. Clinical variables included surgery, chemotherapy, year of diagnosis (2009-2015), symptoms and comorbidity. Surgery and chemotherapy receipt were identified at any time within one year of diagnosis. Common pain symptoms due to cancer or its treatment were identified in the claims data and included at least one of the following: malaise and fatigue, generalized pain, dyspnea and respiratory abnormalities, dysphagia, acute pain, cough, hemoptysis, chest pain, brachial plexopathy, or superior cava syndrome. Comorbidity was assessed during one year prior diagnosis using the modified Charlson’s comorbidity index described by Klabunde et al. [20,21].
Statistics
First, descriptive statistics were used to examine the characteristics of the study population. Then, prevalence measures were used to determine the palliative RT use among the study participants. Univariate associations with palliative RT were determined with two-tailed Pearson chi-squared tests. Univariate logistic regression models were performed in order to determine the unadjusted magnitude of the associations. Then, all covariates, regardless their statistical significance in the univariate models, were included in a multivariate logistic regression model to assess covariates independently associated with the use of palliative RT. Exploratory analysis was conducted to identify interactions between the independent variables. These would be accounted for in the multivariate model if considered appropriate. Statistical significance was set at the P = 0.05 level. Results were presented as odds ratio (OR) with 95% confidence intervals and P values. Statistical analyses were performed using STATA software (Stata Corp., LP. College Station, TX).