Authors :
Presenting Author: Samhitha Rai, BS – Case Western
Jingdi Shen, MS – Cleveland Clinic
Richard Rammo, MD – Cleveland Clinic
Demitre Serletis, MD, PhD – Cleveland Clinic Epilepsy Center, USA
Dileep Nair, MD – Cleveland Clinic
Imad Najm, MD – Cleveland Clinic
William Bingaman, MD – Cleveland Clinic
Daniel Rotroff, PhD – Cleveland Clinic
Lara Jehi, MD – Cleveland Clinic
Rationale:
Neuromodulation options carry variable invasiveness and management complexity. No observational or experimental head-to-head comparisons of effectiveness or risk exist to guide decision-making. We leverage global real-world datasets and advanced computational approaches to compare post-surgical outcomes and complications of VNS versus RNS/DBS.
Methods:
We used two large research datasets that collect longitudinal de-identified structured elements from the electronic health record (EHR): TriNetX (250 million patients from >120 healthcare systems in 19 countries) and Epic Cosmos (300 million patients from 1,744 hospitals and 40.7k clinics in the United States). Neither accesses text in clinic notes. Applying validated combinations of EHR-based diagnostic and procedure codes, we identified patients who had neuromodulation for epilepsy. We then applied propensity-matching to create study cohorts balanced for potential confounders (sex, age, ethnicity, race, co-morbidities, smoking, body-mass-index, seizure medications, MRI abnormality, and epilepsy etiology) and emulate a randomized clinical trial design. As code-based distinction between RNS and DBS is unreliable, and both procedures are distinct in complexity from VNS, we generated two comparison cohorts: VNS vs RNS/DBS. Study endpoints included effectiveness measures (mortality and emergency room visits to approximate seizure control), and safety measures (post-operative hemorrhagic and ischemic stroke). We used Cox-proportional hazard modeling to analyze survival, and linear statistics otherwise. All p-values were corrected for multiple testing.
Results:
Of 4,187, 677 patients with epilepsy in Cosmos and 1,888,362 in TriNetX, we identified 11,236 propensity matched patients in Cosmos and 5,834 in TriNetX, equally split between the two study arms (Figure 1). Balance of the study cohorts was excellent (absolute standardized mean difference < 0.1). Analyzing the Cosmos dataset, short-term post-surgical mortality favored VNS (6-month all-cause mortality risk 0.534% risk with VNS vs 1.104% risk RNS/DBS; Odds ratio=2.079; adjusted p-value =0.0042); however, long-term survival favored RNS/DBS with 10-year mortality of 11% with RNS/DBS vs 14% with VNS (adjusted p-value=0.034).