Estimating the Causal Effect of an Exposure on Change from Baseline Using Directed Acyclic Graphs and Path Analysis

Abstract

When estimating the causal effect of an exposure of interest on change in an outcome from baseline, the choice between a linear regression of change adjusted or unadjusted for the baseline outcome level is regularly debated. This choice mainly depends on the design of the study and the regression-to-the-mean phenomena. Moreover, it might be necessary to consider additional variables in the models (such as factors influencing both the baseline value of the outcome and change from baseline). The possible combinations of these elements make the choice of an appropriate statistical analysis difficult. We used directed acyclic graphs (DAGs) to represent these elements and to guide the choice of an appropriate linear model for the analysis of change. Combined with DAGs, we applied path analysis principles to show that, under some functional assumptions, estimations from the appropriate model could be unbiased. In the situation of randomized studies, DAG interpretation and path analysis indicate that unbiased results could be expected with both models. In the case of confounding, additional (and sometimes untestable) assumptions, such as the presence of unmeasured confounders, or effect modification over time should be considered. When the observed baseline value influences the exposure ("cutoff designs"), linear regressions adjusted for baseline level should be preferred to unadjusted linear regression analyses. If the exposure starts before the beginning of the study, linear regression unadjusted for baseline level may be more appropriate than adjusted analyses.

Assessment of channeling bias among initiators of glucose-lowering drugs: A UK cohort study

“…Crude analyses were performed based on all identified initiators, in addition to analyses of the propensity score-matched initiators, to take confounding into account. In the analyses of change in HbA 1c and body weight, baseline levels of HbA 1c and body weight, respectively, were included as covariates in the outcome model 20. The relative effect estimate and their 95% confidence intervals (CIs) were plotted for graphical inspections of changes in effect estimates over time.…”

Long-Term Effectiveness of a Lifestyle Intervention for the Primary Prevention of Type 2 Diabetes in a Low Socio-Economic Community – An Intervention Follow-Up Study on Reunion Island

“…However, the post-intervention length was three times shorter than ours (24–28 months versus 84 months, respectively). Moreover, the two populations had a higher baseline risk level (mean-BMI = 32.6–35.2 kg/m² versus median BMI = 29.3 kg/m², respectively) which increases probability of regression to the mean without a control group to deal with [ 38 ].…”

“…This cohort study is a quasi-experiment with multiple groups , ‘a mixed design combining elements of both internal and external comparisons, which enhances the potential for making a causal inference’ [ 16 ]. In particular, the control group allows adjustment for regression to the mean [ 38 ], an expected statistical phenomenon in populations with a high level of baseline outcome value.…”

Protocol for a randomised trial testing a community fibrosis assessment service for patients with suspected non-alcoholic fatty liver disease: LOCal assessment and triage evaluation of non-alcoholic fatty liver disease (LOCATE-NAFLD)

“…For health-related quality of life we will model the participants' 12-month result whilst adjusting for their baseline, which is the equivalent to examining the withinparticipant change in quality of life and helps adjust for regression to the mean [19,20]. We will use a linear regression model and check the model residuals.…”