K-Sample comparisons using propensity analysis.

Area of Science:

• Biostatistics
• Observational Studies
• Survival Analysis

Background:

• Observational studies often have non-comparable treatment groups due to imbalanced baseline characteristics.
• Propensity analysis methods like multinomial logistic regression are used to address this imbalance.
• Inverse Probability Weighting (IPW) is a common technique for comparing groups using estimated weights.

Purpose of the Study:

• To propose decision tree methods as a robust alternative for propensity analysis in observational studies.
• To introduce novel Inverse Probability Weighting (IPW) rank statistics (Dunnett-type and ANOVA-type tests) for comparing three or more treatment groups.
• To evaluate the performance of these new methods using simulations and a real data example.

Main Methods:

• Utilizing decision tree methods for propensity score estimation.
• Applying Inverse Probability Weighting (IPW) with proposed Dunnett-type and ANOVA-type rank statistics.
• Conducting simulations to assess finite sample performance of the weighted rank statistics.

Main Results:

• Decision tree methods offer a simple and robust alternative for propensity analysis.
• The proposed IPW rank statistics demonstrate effective performance in simulations for comparing multiple treatment groups.
• The IPW method facilitates unbiased estimation of population parameters for each treatment group.

Conclusions:

• The proposed decision tree-based propensity analysis and IPW rank statistics are effective for K-group comparisons on survival endpoints.
• These methods provide a valuable alternative for analyzing complex observational data.
• The presented techniques can be extended to other outcome types beyond survival data.