Our study also has additional limitations. First, our analytical datasets are derived from individuals with commercial and/or Medicare supplemental health insurance plans. Our findings cannot be generalized to individuals without the corresponding health insurance plans. Second, we used specific algorithms to define phenotypes including neuroinflammation-related conditions (AUD, epilepsy, HS, and TBI), AD, and comorbidities. Despite the algorithms are derived by subject matter experts, the phenotypes could be misclassified. Additionally, patients with the aforementioned neuroinflammation-related conditions may have heterogeneous severity of neuroinflammation both within and between the neuroinflammation-related conditions. For instance, patients with TBI may range from mild to severe TBI, and patients with AUD may range from recovery (i.e., in remission) to alcoholism, both of which may left heterogeneous severity of neuroinflammation. We assume patients with neuroinflammation-related conditions have a higher severity of neuroinflammation than the matched individuals in general. However, the presence of neuroinflammation could be misclassified, and the severity of neuroinflammation is not adjusted. Third, we use pharmacy claims data to define drug exposure, while we cannot confirm drug adherence and/or cumulative exposure. We largely assume pharmacy claims
