OHDSI Home | Forums | Wiki | Github

Study on treatment strategies for atrial fibrillation: OHDSI prospective/retrospective cohort study to accompany emulated clinical trials

Hi fellow OHDSI partners-on-journey,

My colleague Saman Nazarian (Director, Arrhythmia Imaging Research Laboratory, UPenn) and Eliseo Guallar (Epidemiology, Johns Hopkins) are planning an emulated clinical trial on atrial fibrillation treatments, which will prospectively follow treatment-naive patients and collect heart and brain image (cardiac and brain MRI) and rhythm data (implantable rhythm monitors). The idea is that, while treatments won’t be randomized, the study will collect the additional data that are not captured by ordinary EHR. (We might call this an “EHR+” prospective observational study.)

While getting a pre-approval on their NIH proposal, the reviewers suggested that they run a parallel, more standard observational study using only the routinely collected data; this way, we can see if the additional data actually yield more clinical insight. In concrete terms, this means that, among all the treatment naive AF patients who show up at Johns Hopkins and UPenn during the study periods, 1,000 of them will be assigned to the EHR+ study that will collect the additional data while the rest will be followed up only via the routine EHR data. Both Hopkins and Penn now have EHR data in OMOP, so we will be using OHDSI tools to run these EHR(+) prospective studies.

We think it would be great to further expand the scope of this study and run a corresponding LEGEND-like retrospective cohort study on AF patients. Besides its potential clinical impacts, these parallel studies will give us opportunities to find out the advantages and disadvantages of prospective(+) and retrospective observational studies.

Do you think it is possible to identify a corresponding cohort/outcome in retrospective data to run an accompanying retrospective study ? We are particularly interested in comparing rate/rhythm control (alternatives: medical rate control, rate control with pacing and atrioventricular node ablation, rhythm control with antiarrhythmic drugs (AAD), or rhythm control with ablation) and stroke prophylaxis (alternatives: no anticoagulation, warfarin, direct acting oral anticoagulants [DOAC)] or left atrial (LA) appendage occlusion [LAAO]) strategies in enhancing cardiac remodeling(EHR+)/CHF hospitalizations and white matter hyperintensities (EHR+)/TIA/stroke prevention. Maybe, @Patrick_Ryan will have some insight? Also, would there be enough interest and bandwidth in the community to coordinate this study?


Generally, this sounds like a very good idea.

That’s what observational studies have to do all the time: Create comparable cohorts. Often, you don’t have the clinical detail do a one-to-one copy, but have to substitute criteria.

Yes, low level rhythm information is in the EKG system, but not always copied to the EHR system, which is the basis for our analytical data. Same is true with imaging results.

You might use diagnostic codes that are provided. For example, ICD10 has 30 arrhythmia concepts in the categories atrioventricular and left bundle-branch block, paroxysmal tachycardia, atrial fibrillation and flutter and other cardiac arrhythmias. So, the comparison can be done through the rates of these diagnoses.

Folks will probably look for you to take on the coordinate the study. No?