Causal inference from longitudinal data in vascular epidemiology Seminar

Although mortality rates of abdominal aortic aneurysm (AAA) operations have decreased in the last decade, the long term survival of these patients has not improved. In general people with AAA present a higher risk of developing cardiovascular problems. There has been an increasing interest in the use of statins to reduce the cardiovascular risk among the elderly. We aim to evaluate whether prescribing statins and other cardiovascular risk modifiers improves the survival of AAA patients, but determining effects on survival is challenging due to complex time-dependent confounding between treatments and survival that is present in available longitudinal data. We used the Health Improvement Network database (THIN) to compare the effect of statins, anti-hypertensives, and anti-platelets on survival of patients with a diagnosed AAA. We applied the g-computation algorithm to correctly adjust for time-dependent confounding between patient comorbidities, treatments and the outcome.

Results showed all these cardiovascular drugs have a positive effect in AAA patients. The prescription levels of anti-hypertensives are currently optimal, however there is room for improvement in the cases of anti-platelets and statins. An intervention where all patients receive statins would decrease the hazard ratio 21% and 8.6% if antiplatelets were prescribed to all AAA patients.

We also briefly illustrate how mediation analysis can be used to decompose an overall treatment effect into direct and indirect components with a case study of 90 day mortality of patients with AAA in England and Sweden. Specifically we show how an effect of hospital type on mortality may be mediated by a type of surgery performed. If time permits, we discuss the difficulties of applying mediation analysis to longitudinal settings.