Target parameters and bias in non-causal change-score analyses with measurement errors

Area of Science:

• Statistics
• Biostatistics
• Econometrics

Background:

• Change-score analyses often involve debate regarding baseline score adjustment.
• Previous bias analyses under measurement error used simplified models and overlooked relevant target parameters.

Purpose of the Study:

• To clarify the relevance of baseline score adjustment in change-score analyses for prediction.
• To derive bias expressions for adjusted and unadjusted analyses under measurement error using a more realistic model.
• To provide guidance on when adjustment is warranted in prediction-focused change-score studies.

Main Methods:

• Considered two relevant target parameters: one adjusted for baseline and one unadjusted.
• Analyzed scenarios with and without measurement error.
• Derived analytic expressions for bias in adjusted and unadjusted analyses under a realistic measurement error model.

Main Results:

• Identified two distinct, potentially relevant target parameters for prediction in change-score analyses.
• Quantified bias associated with adjusting or not adjusting for mis-measured baseline scores.
• The choice of adjustment depends on the specific prediction goal and presence of measurement error.

Conclusions:

• Baseline score adjustment in change-score analyses is relevant for prediction, not just causal inference.
• A more realistic model for measurement error reveals nuanced bias considerations.
• Guidance is provided for optimal adjustment strategies based on study aims and data characteristics.