Comparison of utilization and cost of healthcare services and pharmacotherapy following implantation of vagus nerve stimulation vs. responsive neurostimulation or deep brain stimulation for the treatment of drug-resistant epilepsy: analyses of a large United States healthcare claims database

Use of Healthcare Claims Data to Generate Real-World Evidence on Patients With Drug-Resistant Epilepsy: Practical Considerations for Research

“…Approximately one-third of patients with DRE have been reported to have comorbid depression; excluding these patients would therefore risk reducing study generalizability. 24 In studies of NM for DRE, we recommend including patients with comorbid depression and subsequently conducting stratified analyses to better understand how NM impacts those with and without this comorbidity.…”

“…In our study, we used propensity score matching with a “greedy” nearest neighbor approach 54 to match (1:1) a maximal number of VNS patients to their RNS/DBS counterparts; a specific radius (0.2 times the SD of the log-transformed propensity score distribution) was used for matching. 24 Standardized differences were used to assess the balance of variables across the 2 cohorts following propensity score matching. One example of an important confounding variable in our study was age.…”

“…10,[17][18][19][20][21] We identify and describe some challenges associated with RWE generation in epilepsy based on healthcare claims and offer potential solutions with illustrative examples based on our recent experience focused on patients who received NM for DRE. [22][23][24] Our goal was to provide researchers with a holistic view of designing an RWE generation study (eg, allocation of exposure[s] and assessment of outcomes), specific to patients who underwent NM for DRE, using healthcare claims data.…”

“…Because our goal was to focus on a DRE cohort undergoing initial NM device implantation, we dropped all patients with evidence of (1) NM at any time during the baseline period (eg, procedure codes for neurostimulator removal, replacement, repair, analysis, programming, or complication); (2) multiple devices (ie, VNS, RNS, DBS) implanted on the index date; or (3) evidence of cranial epilepsy surgery (an alternative surgical intervention for DRE) at any time during baseline. We reviewed relevant product manuals [36][37][38] and published studies [22][23][24] to define NM-related codes (implantation, revision, repair, device [re]programming, battery repair) (Table 1). As shown in Table 1, some billing codes were specific to individual devices; some were insufficiently specific to differentiate between NM products; some covered implantation of any NM device; others could differentiate between implantation sites (eg, chest, cranium/skull); and others were insufficiently specific to identify insertion site or NM type.…”

“…We ultimately opted to ensure a diagnosis of epilepsy was present on the index date, and to exclude all of these patients with "competing" explanations for NM to avoid risk of misclassification ( one-third of patients with DRE have been reported to have comorbid depression; excluding these patients would therefore risk reducing study generalizability. 24 In studies of NM for DRE, we recommend including patients with comorbid depression and subsequently conducting stratified analyses to better understand how NM impacts those with and without this comorbidity.…”