OBJECTIVES: Increasing focus has been placed on real-world data (RWD) to address applicability of clinical trial results to a real-world population. However, in 2018, ASCO identified several gaps that affect the utility of RWD for comparative analyses in oncology patients. We describe over-arching gaps and propose practical methodological handling strategies.

METHODS: Gaps common to oncology RWD are: 1) patient risk-stratification based on variables not linked to diagnoses (ICD)-9/-10 coding (ie, cytogenetics/disease stage) [Case 1, below], and 2) identification of treatments, including determination of lines of therapy (LOT), as many agents are given in combination regimens [Case 2]. Besides these gaps, inherent bias within non-randomized studies further compound comparison validity and RWD utility. Utilizing proxies and employing a patient-level match for confounding variables ensures appropriate comparisons between oncology patient groups [Case 3].

RESULTS: Case 1, risk stratification and non-coded variables: In EHR data, identification of a “high-risk” population of patients (eg, a rare hematologic malignancy) is challenging as risk-stratification is based on unavailable data within an EHR. To overcome this, risk is calculated as a composite score of existing data supplemented with population-based estimates. Case 2, LOT identification: In EHR/claims data, LOT determination is critical for outcomes evaluations but difficult to ascertain. In an example of myeloma patients, pre-validated algorithmic approach was undertaken based on national treatment/monitoring guidelines and clinical input. Case 3, comparability of study groups in oncology: Outcomes between cohorts are influenced by patient-level variables impacting treatment choice and monitoring. Utilizing proxies and employing a patient-level match for confounding variables, two matching methods were compared. While propensity score matching retained more patients, a direct match more conservatively ensured confounding covariates are matched across cohorts.

CONCLUSIONS: Challenges associated with comparative analyses in retrospective oncology studies can be overcome by leveraging pragmatic algorithms collaboratively derived from variables available in RWD.