Change analysis for intermediate disease markers in nutritional epidemiology: a causal inference perspective

Area of Science:

• Epidemiology
• Biostatistics
• Nutritional Science

Background:

• Prospective studies commonly use change-score, concurrent change-change, and lagged change-change analyses to estimate diet's effect on disease markers.
• Concurrent change-change analysis is empirically suggested as the most robust, consistent, and biologically plausible approach.
• A causal inference perspective comparison of these methods, including their underlying causal models and interpretation, is lacking.

Purpose of the Study:

• To elucidate and compare the causal models, estimands, and interpretations of change-score, concurrent change-change, and lagged change-change analyses.
• To intuitively illustrate these approaches using directed acyclic graphs (DAGs).
• To clarify the strengths and limitations of concurrent change-change analysis through simulations.

Main Methods:

• Causal models and DAGs were employed to theoretically clarify the causal estimand and interpretation of each approach.
• Monte Carlo simulations were conducted to assess the performance of different approaches under varying degrees of time-invariant heterogeneity.
• The performance of concurrent change-change analysis was evaluated when its causal identification assumptions were violated.

Main Results:

• Concurrent change-change analysis estimates the contemporaneous effect of exposure on outcome, which is more relevant for diet-biomarker associations in prospective studies.
• This method provides unbiased estimates even with significant unobserved time-invariant confounding, unlike change-score and lagged change-change analyses.
• Estimation bias in concurrent change-change analysis increases linearly with the severity of violated causal identification assumptions.

Conclusions:

• Concurrent change-change analysis appears superior for studying diet and intermediate biomarkers in prospective studies due to its plausible estimand and ability to circumvent bias from unobserved heterogeneity.
• Application of this method necessitates careful examination of its underlying identification assumptions.
• This approach offers a promising avenue for more accurate dietary impact assessments in epidemiological research.