OHDSI Center Events

CBER BEST Seminar Series: Negative controls and p-value calibration in RWE generation

Anyone can register and join for free.

Topic

Negative controls and p-value calibration in RWE generation

Speaker

Dr. Martijn Schuemie
Research Fellow, Janssen Research & Development

Description

Despite our best efforts to adjust for confounding, selection bias, and measurement error through clever epidemiological design, we cannot guarantee our effect estimates produced from real-world data are free from bias. Negative controls, exposure-outcome pairs where we believe there is no causal relationship, can help understand how accurate our study design and data are. By comparing the estimates we produce for our negative controls to the truth (of no causal) effect, we can infer a probability distribution for systematic error in our study. By integrating this expression of uncertainty about systematic error into our statistics, such as p-values, confidence intervals, and critical values when performing sequential testing, we can demonstrably achieve operating characteristics closer to nominal; Our p-values will lead to type-1 errors closer to our alpha, and our confidence intervals will show coverage closer to alpha. OHDSI has developed tools that facilitate the selection of negative controls, include those controls in our study designs, and perform empirical calibration, making these best practices available to everyone.