Three handy tips and a practical guide to improve your propensity score models.

Area of Science:

• Observational studies
• Real-world data analysis
• Biostatistics

Background:

• Real-world data (RWD) offers valuable insights into treatment safety and efficacy.
• Observational studies face confounding by indication, where patient characteristics influence treatment and outcomes.
• Propensity scores (PS) are commonly used to adjust for this bias in RWD studies.

Purpose of the Study:

• To highlight common methodological flaws in propensity score (PS) application.
• To provide a stepwise guide for estimating and using PS to avoid spurious conclusions.
• To improve the reliability of real-world evidence (RWE) derived from observational studies.

Main Methods:

• Discussion of common propensity score (PS) pitfalls, including perfect prediction and lack of generalizability.
• Stepwise description of propensity score (PS) estimation and utilization.
• Focus on methodological rigor in applying PS to observational data.

Main Results:

• Identified critical flaws in propensity score (PS) methodology that can invalidate study findings.
• Demonstrated how oversimplification in PS models can lead to inaccurate results.
• Highlighted the importance of careful variable selection and model validation in PS analysis.

Conclusions:

• Awareness and avoidance of common methodological flaws are crucial for valid propensity score (PS) utilization.
• Proper application of propensity scores (PS) enhances the credibility of real-world evidence (RWE).
• This viewpoint aims to guide researchers towards more robust and reliable analyses of real-world data (RWD).