Analyzing medical costs with time-dependent treatment: The nested g-formula

Area of Science:

• Health Economics
• Causal Inference
• Biostatistics

Background:

• Rising healthcare expenditures necessitate advanced cost outcome analysis.
• Existing methods like inverse probability weighted regression models address cost censoring for intent-to-treat effects.
• Comparing costs across time-varying treatment regimes presents unique analytical challenges.

Purpose of the Study:

• To introduce and describe a nested g-computation procedure for comparing mean costs between time-varying treatment regimes.
• To highlight the advantages and limitations of this novel approach compared to existing regression-based models.
• To demonstrate the utility of the nested g-formula in informing public policy through a simulated cancer treatment cost analysis.

Main Methods:

• Description of a nested g-computation procedure for analyzing cost outcomes.
• Comparison of the nested g-formula with inverse probability weighted regression models.
• Application to a simulated dataset motivated by cancer treatment costs.

Main Results:

• Simulations reveal that intent-to-treat effects and joint causal effects from the nested g-formula can yield substantially different cost conclusions.
• The choice of inference framework (intent-to-treat vs. joint causal effects) is critical and must be prespecified.
• The nested g-formula provides a valuable, complementary perspective to existing cost outcome analysis methods.

Conclusions:

• The nested g-computation procedure is a valuable tool for analyzing cost outcomes in time-varying treatment scenarios.
• Prespecification of the target of inference is essential when choosing between different causal inference frameworks for cost analysis.
• This method can aid public policy decisions by providing nuanced insights into treatment-related costs.