Single proxy control.

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Observational studies often suffer from unobserved confounding, biasing causal effect estimates.
• Negative control outcomes (NCOs) are valuable tools for detecting and correcting such biases.
• The Control Outcome Calibration Approach (COCA) offers a formal method for NCO-based confounding control.

Purpose of the Study:

• To establish nonparametric identification for the Control Outcome Calibration Approach (COCA).
• To relax the rank-preservation assumption in COCA, allowing for unrestricted effect heterogeneity.
• To propose novel estimation strategies for the enhanced COCA method.

Main Methods:

• Developed a nonparametric identification strategy for COCA using NCOs.
• Introduced three estimation approaches: extended propensity score, outcome bridge function, and doubly-robust methods.
• Applied the methods to evaluate the causal impact of Zika virus on birth rates in Brazil.

Main Results:

• Achieved nonparametric identification of average causal effects for the treated under COCA without rank preservation.
• Demonstrated that the enhanced COCA provides single-proxy confounding control.
• The proposed methods were successfully illustrated in a real-world epidemiological application.

Conclusions:

• The enhanced COCA method offers a robust and flexible approach to address residual confounding in observational studies.
• This work advances causal inference by enabling more accurate estimation of treatment effects in the presence of unobserved confounders.
• The proposed methods have broad applicability in various fields, including epidemiology and public health research.