Primary Submission Category: Propensity Scores
Using Fractional polynomials in the Marginal Structural Model: An Application to Thoracic Aortic Endovascular Repair
Authors: Hang Nguyen, Daniel Heitjan, Haekyung Jeon-Slaughter,
Presenting Author: Hang Nguyen*
Thoracic aortic aneurysm is a potentially suddenly lethal condition that may manifest no symptoms and is often detected only during medical visits for other reasons. Once identified, the surgeon can treat the aneurysm with thoracic endovascular aortic repair (TEVAR), a procedure that can be done electively. Aneurysm size is known to have a significant impact on the risk of rupture; thus, size at repair is expected to be associated with survival. We have used the marginal structural model (MSM), estimated with inverse propensity score weighting, to identify the effect of aneurysm size on post-surgery survival. Particularly, we first estimate stabilized weights as a function of aneurysm size. We then fit an MSM that is parameterized as a fractional polynomial function. This method provides a flexible parameterization for a continuous exposure that avoids some of the issues associated with spline analysis — in particular, the selection of the number and locations of knots. The use of fractional polynomials is evidently novel in applied causal inference. We illustrate our approach by applying our method to data from the Vascular Quality Initiative TEVAR registry. Our results show that the lowest hazard occurs with operations at an aneurysm size of 60mm.