Skip to main content


Identifying underlying mechanisms between intrinsic and variable prognosticators and clinical outcomes: A structural modelling approach

Date and Location




Sunday 22 September 2013 - 10:30 - 12:00


Presenting author and contact person

Presenting author

Brent Opmeer

Contact person

Brent Opmeer
Abstract text
Background: It is often implicitly assumed in clinical practice as well as in research that prognosis of outcomes and effect of medical interventions to improve outcomes are homogeneous across fixed biological factors, such as gender, ethnicity and age. Thereby the process and underlying mechanisms, through which fixed and variable risk factors and subsequent intermediate outcomes result in the final outcome, are unclear. For instance, risk of preterm birth is different for boys and for girls, and women of different ethnic origin have different normalcy curves for pregnancy. Consequently, to appropriately evaluate interventions aiming at reducing preterm delivery, the definition of preterm – as abnormal in comparison with at term – should take these different intrinsic prognoses into consideration. In addition, it is unclear whether and how intrinsic prognosis and variable risk factors interact. Knowledge and understanding of such mechanisms is pivotal to appropriately design clinical trials and interpret their results. Objectives: To present the rationale and analytical methods and illustrate this approach with an example from the PRO-TWIN and AMPHIA studies. Methods: In the PRO-TWIN and AMPHIA studies, women with a twin-pregnancy were randomized between pessary versus no-pessary and between progesterone versus placebo, to evaluate whether they reduce the risk of preterm delivery and poor neonatal outcome. In doing so, we model the causal chain between ethnicity, fetal gender, acquired risk factors, cervical length and signs and symptoms for preterm birth. Relevant variables will be selected to specify directed acyclic graphs (DAGs) representing causal relations between prognostic variables and outcome(s) (see example shown in figure 1). Structural equations modeling will be used to statistically test whether the model adequately fits the data, also allowing specification of latent (unobserved) variables that represent a combination of several manifest (observed) variables.