Evaluating Public Health Interventions: 8. Causal Inference for Time-Invariant Interventions

Area of Science:

• Public Health
• Epidemiology
• Biostatistics

Background:

• Confounding is a major challenge in public health evaluations.
• Accurate causal inference requires robust methods to control for confounding.
• Time-invariant interventions necessitate specific approaches for causal analysis.

Purpose of the Study:

• To compare classical and newer methods for controlling confounding in public health.
• To estimate the causal effect of gender on mortality in an HIV treatment program.
• To evaluate the performance of various statistical methods for causal inference.

Main Methods:

• Overview of multivariable modeling, propensity score methods, inverse-probability weighting, doubly robust methods, and targeted maximum likelihood estimation.
• Causal effect estimation of gender on all-cause mortality in a Tanzanian HIV program (2004-2012).
• Comparison of bias reduction and efficiency across different causal inference techniques.

Main Results:

• Significant confounding was observed in the analysis.
• All methods consistently showed approximately 20% increased mortality in men.
• No method demonstrated a clear advantage over multivariable modeling for bias reduction or efficiency.

Conclusions:

• Classical multivariable modeling offers the greatest statistical efficiency for causal estimates when data is sufficient.
• The choice of method does not overcome limitations of unmeasured or poorly measured confounders.
• Effective causal inference relies on accurate measurement of all relevant confounding factors.