Assessing the prior event rate ratio method via probabilistic bias analysis on a Bayesian network

Area of Science:

• Epidemiology
• Biostatistics
• Health Services Research

Background:

• Unmeasured confounding is a significant challenge in real-world data observational studies.
• Prior event rate ratio (PERR) adjustment is a statistical method used to address unmeasured confounding.
• The PERR method's performance can vary, sometimes exacerbating bias instead of mitigating it.

Purpose of the Study:

• To investigate the robustness of the Prior Event Rate Ratio (PERR) adjustment method.
• To better understand the circumstances under which PERR adjustment may increase bias.
• To provide guidance on the appropriate use of PERR versus probabilistic bias analysis (PBA).

Main Methods:

• Utilized a Bayesian network to represent a generalized observational study with unmeasured confounding.
• Employed a Bayesian networks framework for probabilistic bias analysis (PBA), differing from previous Monte Carlo simulation methods.
• Applied the methodology to a real-world study evaluating the effectiveness of a high-dose influenza vaccine.

Main Results:

• The Bayesian network approach enabled comprehensive analysis of PERR performance across numerous parameter combinations.
• Analysis of a high-dose influenza vaccine study indicated that PERR adjustment was more likely to underestimate relative effectiveness.
• The PERR-adjusted relative effectiveness for the high-dose influenza vaccine was found to be potentially biased towards underestimation.

Conclusions:

• Prior Event Rate Ratio (PERR) adjustment is a valuable tool but not universally effective against unmeasured confounding.
• Probabilistic bias analysis (PBA) may be a more reliable approach in specific scenarios where PERR performance is uncertain.
• Developed general guidelines to assist researchers in choosing between PERR adjustment and PBA for their studies.