Data Source
This retrospective analysis of the US Surveillance, Epidemiology, and End Results (SEER) program–Medicare linked database includes cancer registries on newly diagnosed cancer cases linked to administrative healthcare data (e.g., claims, eligibility for medical and pharmacy benefits, and demographic data). SEER is the largest population-based cancer registry, covering approximately 34% of the US population from 19 geographically disparate regions that are representative of the demographics of the national population [15]. Permission to use the SEER–Medicare data was obtained from the National Cancer Institute.
Individual patient-level data in the database comprise (1) primary and secondary diagnoses (in International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM] or ICD-10-CM formats); (2) procedures (in ICD-9-CM, ICD-10 CM, Healthcare Common Procedure Coding System [HCPCS], or Current Procedural Terminology, fourth edition formats); (3) setting of care (e.g., inpatient or emergency department) and date[s]) of service; (4) dispensed medications from outpatient pharmacies; and (5) total reimbursed amounts for medical care and prescription pharmacotherapies. The data were de-identified, compliant with the Health Insurance Portability and Accountability Act of 1996 and encompassed the study period of 1 January 2003, to 31 December 2014.
Patient Population
The study population included patients with resected stage IIB, IIC, or IIIA melanoma (AJCC, seventh edition) with pathologically confirmed diagnoses between 1 January 2010, and 31 December 2013, and with evidence of resection within 4 months following diagnosis. The date of resection was designated as the index date.
Patients were excluded if they had any of the following: (1) < 12 months of enrollment in Medicare part A or part B before and after the index date; (2) an age of < 18 years at the index date; (3) evidence of resection in the preindex period; (4) ocular/uveal melanoma or any other nonmelanoma malignancies (except nonmelanoma skin cancers and in situ cancers) in the preindex period; or (5) a record of enrollment in a health maintenance organization after the index date. The detailed sample–selection steps are outlined in Fig. 1.
The preindex period was from 2003 to the day before the index date. Patients were followed from the index date until death, disenrollment from the health plan, or end of the study period, whichever was the earliest. All available information from the SEER registry and patients’ claims were compiled for the preindex period through the end of follow-up.
Ethical approval was not required for this study as data were de-identified and did not contain any identifiable patient information.
Outcomes
Patient demographics and clinical characteristics were based on information collected during the preindex baseline period, which comprised age, sex, geographic location, race, insurance type, tumor thickness, lymph node involvement, Charlson Comorbidity Index (CCI) [16], and selected comorbidities.
Outcomes of interest consisted of locoregional and distant recurrence and healthcare costs following recurrence. Since recurrence was not captured in the SEER–Medicare database, a treatment-based algorithm was used to assess recurrence, which was identified by the presence of metastases or secondary cancer (including neoplasms of skin), end-of-life care (admission to hospice care, regardless of relationship to melanoma), evidence of death (as recorded in the SEER registry or Medicare data for any cause), or cancer treatment (chemotherapy, radiotherapy, or surgery) in the post-index period following a 3-month treatment-free interval after the primary treatment [17]. The healthcare costs following recurrence were not melanoma specific, but rather were inclusive of all-cause costs designed to capture the overall real-world economic burden to the healthcare system. The total healthcare cost following recurrence incorporated the cost of hospitalization, emergency room visits, office visits, skilled nursing facilities, home health aides, hospice, and prescription medications. In all instances, reimbursed amounts (i.e., insurance reimbursement) plus patient liability (e.g., co-pay, co-insurance) served as a proxy for costs. Costs were assessed in two ways: annualized mean healthcare costs were assessed in the year following recurrence, or cumulative costs per-patient per-month (PPPM) were calculated before and after recurrence. Healthcare costs were adjusted and inflated to 2018 US dollars (USD) using the consumer price index medical component [18].
Sensitivity analyses, which excluded end-of-life care and evidence of death from the original definition of recurrence, were conducted to assess 1- and 2-year recurrence rates. For these analyses, recurrence was identified by the presence of metastases, secondary cancer, or cancer treatment in the post-index period following a 3-month treatment-free interval after the primary treatment. Patients with stage IIB and stage IIC disease were grouped together to describe baseline characteristics and cost analyses.
Statistical Analyses
Descriptive statistics were used to analyze baseline demographics and clinical characteristics. The proportion of patients who experienced recurrence during follow-up was assessed. Kaplan–Meier analyses were used to estimate RFS and OS. Costs were annualized and adjusted to reflect 2018 USD and reported as mean costs PPPM. Cumulative costs were calculated from 12 months before the first recurrence to death, patient disenrollment, or end of follow-up, whichever was earliest. For those who had a second recurrence, costs were calculated only until the second recurrence to reflect the true costs of the first recurrence only. Total healthcare costs included costs by types of medical encounter and prescription medications. Multivariate logistic regression models were conducted to estimate risk factors for recurrence. Covariates used in the models were age, sex, race, marital status, location, histologic subtype, tumor stage, ulceration, node stage, CCI, receipt of adjuvant therapy, and days to resection. All analyses were conducted using SAS version 9.4 software (SAS Institute Inc., Cary, NC, USA) and were assessed in the aggregate population as well as in subgroups.