Exposure definition sensitivity unmasks hidden confounding in crystalloid target trial emulation

Area of Science:

• Health research methodology
• Observational data analysis
• Causal inference

Background:

• Target trial emulation (TTE) is a common method for estimating causal effects from observational data.
• Standard diagnostic checks are often used to assess the validity of TTE studies.
• The reliability of TTE when faced with complex real-world data, such as treatment switching, remains a concern.

Purpose of the Study:

• To evaluate if standard diagnostic methods are sufficient for ensuring the validity of TTE studies.
• To investigate the impact of treatment switching on causal effect estimates in TTE.
• To identify potential improvements for TTE methodology.

Main Methods:

• Emulation of a randomized trial comparing crystalloid versus 0.9% saline using the MIMIC-IV database (n=42,883).
• External validation using the eICU-CRD database.
• Comparison of two exposure definitions: initial assignment and 48-hour dominant strategy.
• Assessment of propensity score diagnostics and negative control outcomes.

Main Results:

• Two exposure definitions yielded opposite estimates for major adverse kidney events within 30 days (MAKE-30) (OR: 0.49 vs. OR: 2.51), despite satisfactory propensity score diagnostics (C-statistic: 0.787; SMD: 0.048).
• High treatment switching (89.4%) in MIMIC-IV, reflecting local fluid management workflows, contrasted with eICU-CRD (1.2%).
• Negative control outcomes suggested residual confounding influenced by length of stay, which was not detected by standard TTE diagnostics.

Conclusions:

• Standard TTE diagnostics are insufficient to detect confounding from time-varying institutional practices and treatment switching.
• Operationalization sensitivity, characterization of treatment switching, and directed acyclic graph (DAG)-guided assessments should be incorporated into routine TTE.
• Improved methods are necessary for robust causal inference from observational data using TTE.