Presentation (CTI)

OBJECTIVES: To propose and apply a structured, reproducible framework for identifying baseline confounders in localized or locally advanced high-risk prostate cancer, with broader applicability to comparative effectiveness research settings requiring adjustment for cross-study baseline differences, including indirect treatment comparisons and external control arm approaches.
METHODS: A multi-stage framework was developed, grounded in DAG-informed causal inference principles and integrating four sequential phases: systematic literature review, structured expert elicitation using a modified Delphi approach, quantitative consensus scoring, and causal filtering against pre-specified eligibility criteria. The first three phases are designed to be agnostic to the downstream comparative setting, yielding a generalizable candidate confounder set applicable across study contexts. The fourth phase operationalizes the framework for a defined target population, excluding variables that are not measurable at baseline or are post-exposure, and is therefore specific to the comparative setting of interest.
RESULTS: Applying this framework to localized or locally advanced high-risk prostate cancer reduced a heterogeneous pool of 119 candidate variables to 8 confounders: ISUP grade group/Gleason score, T-status, PSA level, lymph node involvement, core status, very high-risk classification, number of risk factors, and presence of comorbidities, capturing both tumor biology and baseline health status dimensions that plausibly influence treatment allocation and outcomes. Numerous candidate confounders were excluded due to insufficient expert consensus, limited causal relevance, or failure to satisfy pre-specified baseline eligibility criteria.
CONCLUSIONS: This study presents a structured, causally grounded approach to confounder identification, advancing the process from an implicit and variable practice toward a transparent and reproducible component of comparative study design. The framework yields a defensible, evidence-informed confounder set, strengthening the causal validity of comparative effectiveness analyses within health technology assessment contexts. The multi-stage design is resource-intensive, reflecting an inherent trade-off between methodological rigor and practical feasibility that should be considered when planning future applications.