Real-World Data Integration for Causal Inference: Benefits, Costs, and Case Studies

OBJECTIVES: Causal inference from Real-World Data (RWD) is growing in importance, driven by the need for rapidly delivered and generalizable evidence to inform regulatory, payer, and patient/provider decision-making. Integrating multiple sources of RWD for the same patient (e.g., claims and electronic health records) can provide deeper insights into the patient's health journey and facilitate causal research. We summarize considerations for RWD integration and present case studies.

METHODS: A consensus-based decision-making process was used to develop an overview of the benefits and costs of using integrated RWD for causal research. We summarized common biases in observational study designs and how integrated RWD could affect these biases and the resulting causal effect estimates. Several case studies were chosen to illuminate the trade-offs associated with using integrated RWD for causal research.

RESULTS: We grouped benefits and costs associated with the use of integrated RWD into five domains: collaboration (e.g., efforts to align stakeholders), IT security and regulatory compliance (e.g., preserving health information privacy), data interoperability (e.g., ensuring adequate linkage), data quality (e.g., reliability of measures), and data availability (e.g., sample size impacts). Causal inference from RWD studies is subject to confounding, measurement, selection, and time-related biases; we discuss how each bias may be affected through the use of integrated RWD (e.g., integrated RWD may reduce bias from unmeasured confounders by allowing for the inclusion of additional covariates). Case studies illustrate practical applications (e.g., using an external control arm to enable causal inference in a study of patients with multiple myeloma).

CONCLUSIONS: Using integrated RWD can lower methodological and resource barriers to comparative effectiveness and safety assessments. However, integrating data requires trade-offs regarding variable consistency, available sample size, and selection bias. Taking these into account will enhance the quality and, thus, the impact of evidence from observational research.