A Plasmode Simulation-Based Bias Analysis for Residual Confounding by Unmeasured Variables Leveraging...

Area of Science:

• Health Informatics
• Epidemiology
• Real-World Evidence (RWE)

Background:

• Non-randomized studies often face bias due to unmeasured confounding variables.
• Existing methods may not fully capture the complexities of real-world healthcare data.
• Accurate bias assessment is crucial for reliable causal inference in observational research.

Purpose of the Study:

• To develop and validate a quantitative bias analysis approach for non-randomized studies.
• To simulate residual confounding using realistic assumptions and complex healthcare data.
• To assess the impact of unmeasured confounders in claims-only analyses using linked EHR data.

Main Methods:

• A plasmode simulation approach was employed, generating 500 cohorts from linked claims and EHR data.
• Simulated outcomes included neuropsychiatric hospitalizations and major adverse cardiovascular events (MACE).
• Residual confounding was modeled using EHR-measured variables (e.g., suicidal ideation, BMI, BP) not present in claims data.

Main Results:

• Simulations showed minimal standardized mean differences for key unmeasured confounders in the unadjusted sample.
• Adjustment using claims-based variables resulted in near-zero relative bias for both outcomes.
• EHR-measured confounders, when unmeasured in claims, were unlikely to cause substantial residual confounding in realistic simulations.

Conclusions:

• The proposed plasmode simulation method offers a robust way to quantify bias in non-randomized studies.
• This approach effectively addresses the challenge of unmeasured confounding in real-world evidence research.
• The findings support the reliability of claims-only analyses under specific conditions when bias is quantified.