This population-based retrospective cohort study included patients diagnosed with PDAC at any time prior to death in Ontario, Canada. We captured deaths in all Ontarians aged 18 or over between April 1, 2010 and March 31, 2018 and used administrative and clinical databases linked with patient-level identifiers housed at ICES (formerly the Institute of Clinical and Evaluative Sciences). These databases have been used to conduct cohort studies examining end-of-life outcomes for patients with cancer and terminal non-cancer illnesses(16–19). This study received ethical approval from the Ottawa Health Science Network Research Ethics Board (OHSN-REB). Informed consent was waived due to the retrospective nature of the study by the OHSN-REB.
All methods were carried out in accordance with Declaration of Helsinki. We reported this study according to the Reporting of studies Conducted using Observational Routinely-collected health Data (RECORD) statement(20).
2.1 Data Sources
We identified patients with PDAC through the Ontario Cancer Registry (OCR), which contains information on incident cancers diagnosed since 1964 and is approximately 95% complete(21, 22), and used encrypted provincial health insurance numbers to link databases via ICES (Supplementary table 1). We used the Registered Persons Database(23) and Vital Statistics registry(24) to obtain demographic and vital status data respectively. To obtain information about health services utilization, we used the Ontario Health Insurance Plan (OHIP) claims database(25) (healthcare provider billings), the Canadian Institute for Health Information (CIHI) Discharge Abstract Database (DAD)(26) and Same Day Surgery Database (SDS)(27), Cancer Activity Level Reporting (ALR) database (radiation, chemotherapy, and oncology appointment visits)(28), National Ambulatory Care Reporting System(29), National Rehabilitation Reporting System (NRS)(30), and Continuing Care Reporting System(31). We obtained information on homecare through the InterRAI Reporting System(32, 33) (homecare assessments) and the Home Care Database (HCD)(34) (services provided or coordinated by local health networks in Ontario). All database codes used in this study are available in Supplementary table 2.
2.2. Cohort
We included patients who died from April 1, 2010 to December 31, 2017 and had a diagnosis of PDAC prior to death based on ICD-10 (International Statistical Classification of Diseases and Related Health Problems 10th revision) codes(35) (Supplementary table 2). We excluded patients who were not eligible for Ontario health insurance coverage at their PDAC diagnosis date, or who lost eligibility before their death or the end of the study period. These patients would not have been captured in administrative databases, which are linked through health insurance numbers. We also excluded patients aged under 18 years or greater than 105 years, whose age was not known, and those diagnosed with possible neuroendocrine tumors.
2.3 Index cancer treatment
The primary exposure was index cancer treatment, with five a priori defined treatment groups: no cancer-directed therapy, radiation, chemotherapy, surgery only, surgery and chemotherapy.
The surgery and chemotherapy group was defined as having undergone surgical intervention for pancreatic cancer with pancreatectomy and the receipt of at least one dose of chemotherapy within 120 days of index cancer diagnosis. The surgery only group was defined as having undergone surgical intervention for pancreatic cancer with pancreatectomy but no chemotherapy in the initial 120 days after diagnosis. The chemotherapy group was defined as having received at least one dose of chemotherapy, with no pancreatectomy, within 120 days of diagnosis. This included patients who underwent initial chemoradiation treatment. The radiation group was defined as having received at least one dose of radiation therapy with no pancreatectomy or chemotherapy in the initial 120 days after diagnosis. The no cancer-directed therapy group was defined as having received none of radiation, chemotherapy, or pancreatectomy within 120 days of diagnosis. Surgery, radiation, and chemotherapy treatments were identified using OHIP claims and ALR codes (Supplementary Table 2) (14, 36–38). As we were interested in index treatment decisions, we did not group patients based on treatments they received after 120 days from diagnosis.
2.4 Covariates
We collected data on age, sex, rural location of residence, neighborhood income quintile, comorbidities, location of primary pancreatic tumor, cancer stage, and index cancer treatment. We used the Charlson Comorbidity Index (CCI) to characterize comorbidities(39), with patients categorized as having a CCI score of 0, 1, or ≥2. Comorbidities included 18 chronic conditions (acute myocardial infarction, arrhythmia, asthma, cancer, congestive heart failure, chronic obstructive pulmonary disease, coronary artery disease, dementia, diabetes, hypertension, inflammatory bowel disease, non-psychotic mood and anxiety disorders, other mental health illnesses, osteoarthritis, osteoporosis, renal disease, rheumatoid arthritis, and stroke) with high prevalence and economic burden in Ontario.(40–42)
OCR data were used for tumor location and cancer staging(43–45) to derive the “best stage” grouping consistent with the American Joint Committee on Cancer staging manual(46). Residence was defined as rural (community size greater than 10,000 persons) or urban using postal codes in the RPDB, and income quintile as the median income of a patient’s postal code using Canadian census data(47, 48). Cohort identification, demographic variables, and covariate codes are available in Supplementary Table 3.
2.5 Outcomes
The primary outcomes were survival, healthcare encounters per 30 days in the last six months of life, and palliative care visits per 30 days within the last six months of life. We defined healthcare encounters as primary care visits, emergency department visits, and hospital admissions (including intensive care admissions). Palliative care visits included outpatient visits, homecare visits, and palliative inpatient admissions(19). We captured inpatient and outpatient palliative care treatments based on a set of CIHI DAD, HCD, and OHIP claims codes(49–51). Secondary outcomes included location of death, hospitalization within the last 30 days of life, and receipt of chemotherapy within the last 30 days of life. We categorized location of death as institutional (emergency department, acute care ward, intensive care unit, complex continuing care and rehabilitation facility, long-term care) and community (home or hospice)(52). Hospitalization and chemotherapy within the last the last 30 days of life were selected as measures of aggressive EOL care, in keeping with prior studies(53–55). All database codes for outcomes are available in Supplementary Table 4.
2.6 Data analysis
We calculated descriptive statistics for all patient at the time of their diagnosis. Variables and outcomes were stratified by index cancer treatment. We estimated the association between the exposure variables and outcomes using multivariable models. Survival was modelled with Cox regression, rates of end-of-life healthcare encounters and palliative care visits per 30 days were modelled using linear regression, and binary outcomes (location of death as institution vs. community, any hospitalization within the last 30 days of life, and any receipt of chemotherapy within the last 30 days of life) were modelled using logistic regression.
We adjusted all models for age, sex, rurality, neighbourhood income, comorbidity, location of the tumor in the pancreas and cancer stage. We performed one sensitivity analysis planned a priori by excluding patients without cancer stage data because we theorized that these patients were less likely to have undergone cancer-directed therapy. Hazard ratio, adjusted mean difference, and odds ratios were reported for Cox regression, linear regression, and logistic regression respectively, all with 95% confidence intervals. A P-value of < 0.05 was considered statistically significant for all analyses.