Background Within lifecourse epidemiology, there is substantial interest in relationships between exposures (X) that are measured longitudinally (e.g. at time 0, 1, 2; hence X₀, X₁, X₂) and outcomes (Y) that are measured once, cross-sectionally (e.g. Y₂ or Y_2+t). Methods used to analyse these relationships often involve conditioning on the outcome and suffer regression to the mean (RTM), making causal inference challenging. The aim of our study was to demonstrate this and to quantify the potential bias that can ensue across a range of different scenarios.

Methods Directed acyclic graphs (DAGs) were used to depict a range of potential causal relationships between an exposure measured longitudinally (BMI: body mass index; e.g. at ages 2, 15 and 64 years; BMI₂, BMI₁₅ and BMI₆₄), and a later-life outcome (CRP: C-reactive protein, a marker for cardiovascular disease) measured once around the time of the last exposure measurement (CRP_{> 64}). The covariance structure and mean values of each variable were obtained from the literature to inform simulations for three scenarios where: (i) CRP_{> 64} is not caused by BMI₂, BMI₁₅ or BMI₆₄; (ii) CRP_{> 64} is caused only by BMI₆₄; and (iii) CRP_{> 64} is caused by more than one of BMI₂, BMI₁₅ and/or BMI₆₄.

Results In scenario (i), BMI₂, BMI₁₅ and BMI₆₄ are not causally related to CRP_{> 64}, yet ‘trajectories’ will be seen when plotting BMI conditioned on the outcome, generating spurious causal interpretation. If CRP_{> 64} is caused by BMI₆₄ only, as in scenario (ii), the estimated direct impact of BMI₂ and/or BMI₁₅ on CRP_{> 64} is spurious. If CRP_{> 64} is directly caused by more than one BMI exposure, as in scenario (iii), the estimated causal impact of BMI at any given time point on CRP_{> 64} can be biassed, the extent of which depends on the specific correlation structure amongst all measures of BMI and CRP_{> 64}. These results arise due to RTM caused by conditioning BMI ‘trajectories’ on CRP_{> 64} and indicate considerable scope for misinterpretation of the direct causal impact of one or other measure of BMI on CRP_{> 64}.

Conclusion With no causal relationship between changes in an exposure measured longitudinally, and a later-life outcome, spurious relationships will emerge if any of the individual measures of exposure are correlated with the outcome. With a genuine causal relationship between any measure of a longitudinal exposure and later-life outcome, the estimated relationship of the exposure at any one time point may be biassed, leading to erroneous causal interpretations.