Health administrative data are defined as records passively collected in the course of patient care within a health care system [1], and are frequently linked to government-level records (i.e., census data, birth records) and disease-based registries (e.g., cancer registries). Population-level linked data are useful for large-scale epidemiologic and outcomes research [2], [3], [4], and have been utilized in many cancer-focused research studies including studies of outcomes [3], [4], [5], [6], [7], [8], [9], [10], [11], short- and long-term toxicities [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], surveillance [2], [22], healthcare utilization [12], [23], [24], [25], [26], quality improvement [5], [27], [28], [29], comparative effectiveness [30], [31], [32], [33], and health economics [24], [34], [35], [36], [37].

Unfortunately, the specifics of chemotherapy and radiation are often poorly captured by administrative data [1], [2], [40], [41], and cancer diagnosis and demographic information alone are insufficient for deducing what treatments were received [38], [39]. This gap is particularly relevant for those with cancer diagnoses early in life (i.e., pediatric, or adolescent, and young adult [AYA] populations) where long-term side effects of specific cancer therapies are anticipated to manifest in survivors [42]. To appropriately design surveillance strategies and prognosticate risk at the population level, accurate capture of cancer treatment details in administrative databases is desired. Several validation studies have assessed the capacity of cancer registry-type data [43], [44], [45], [46], [47], [48], [49] and insurance/physician claims-based data [48], [50], [51], [52] to identify cancer treatment details in the United States (US) and elsewhere. These validation studies have generally been unable to compare the utility of different types of administrative data within the same jurisdiction.

In Ontario, Canada, several levels of administrative data are available. In addition to general administrative data comprising physician billing claims and hospitalization records which are available in many jurisdictions, and two cancer registries (Ontario Cancer Registry [OCR], capturing all patients diagnosed in the province, and Pediatric Oncology Group of Ontario Networked Information System [POGONIS], for those treated at pediatric cancer centres), Ontario has an administrative database specific to cancer-related care. This database solely captures entries on an ‘activity-level’ basis and is used to demonstrate demand for services and inform policy. Unlike in POGONIS, the OCR does not capture treatment information; as such, most individual’s cancer-related treatment data is fragmented across each contact they have with the healthcare system and does not exist elsewhere in aggregate form. As well, mandated reporting is only in place for regional cancer centres; thus, data capture may not be comprehensive. We hypothesized that algorithms using cancer-specific administrative data would have superior validity in providing important treatment details. Our objective was to evaluate the validity of several algorithms using either general, cancer-specific, or both kinds of administrative data in identifying cancer treatment characteristics including treatment modality, agent, and dose, compared to chart-abstracted treatment data from a cohort of AYA cancer patients (IMPACT cohort) as a reference-standard.