Medical systems generate massive amounts of Electronic Health Record (EHR) data and researchers have analyzed these data to derive new insights and improve healthcare1,2. Stanford University has established a novel and secure data platform: Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM). Alzheimer’s disease (AD) is well-suited for analysis with OMOP given its multifaceted complexity, prevalence, and the multitude of small size studies that claim benefit for certain interventions.
AD is a neurodegenerative disorder of uncertain cause and pathogenesis. In the United States, as many as 1 in 9 people (10.7%) older than 65 have AD3. Recently, repurposing bumetanide as an AD medication was proposed based on data that showed bumetanide “reversed” APOE genotype-dependent transcriptomic signatures in mouse and cell culture models.4 This finding was investigated in two EHR-based cohorts demonstrating that in individuals over 65 years of age, bumetanide exposure was associated with lower AD prevalence4. This finding warrants further validation as bumetanide is more expensive than the commonly prescribed loop diuretic (furosemide) and thus potential socioeconomic status (SES) confounding such as insurance coverage needs to be investigated. Both furosemide and bumetanide are indicated, and often interchangeably used, for patients with hypertension, congestive heart failure (CHF), and kidney disease. In this study, using Stanford’s EHR data, we sought to replicate the bumetanide findings in an independent dataset accounting for SES, hypertension, and CHF.
A total of 1,759 patients (54 cases, 1,705 controls) were exposed to bumetanide during any of their visits (Table 1). For the full matched cohort, the bumetanide prevalence among AD cases was 0.92% compared to 1.47% in the controls. The unadjusted odds ratio (OR) for AD diagnosis among bumetanide exposed was 0.63 (95% CI, 0.48-0.82; p = 7.6×10-4, Table 2).
We had some missingness in our data for the SES variables – insurance information was available for 94.0% of cases and 84.5% of controls. In addition, for the zip code data informing the median income component, we focused on patients from California, and due to deidentification reasons the data were only available for 76.8% of cases and 74.1% of controls. Since SES estimates were not easily imputable from our data, a sensitivity analysis was performed as a complete case analysis. In this sensitivity analysis we restricted the cohort to patients with complete insurance and median income data (N=77,688) and repeated the unadjusted analysis prior to fitting a multivariable model adjusted for CHF, insurance, and median income. In the complete case restricted cohort, 45/4238 (1.06%) AD cases and 1,321/7,3450 (1.8%) matched controls were exposed to bumetanide. The unadjusted OR remained similar to the one in our primary analysis (OR=0.59; 95% CI, 0.43-0.79; p = 4.5×10-4, Table 2). After adjusting for CHF, insurance, and median income the estimated OR for AD diagnosis among bumetanide exposed was 0.50 (95% CI, 0.37-0.68; p = 9.9×10-6, Table 2).
For the full matched cohort, the furosemide prevalence among AD cases was 18.8% compared to 17.2% in the controls. The unadjusted OR for AD diagnosis among furosemide exposed was 1.11 (95% CI, 1.03-1.18; p = .003, Table 2); however, this effect was not replicated when adjusting for CHF.
Most clinical trials in AD suffer from an inherent shortfall regarding primary prevention as they do not give insights on whether a compound reduces the incidence of AD since medications are tested after disease onset. Studying EHR using OMOP allows us to derive insight on possible primary prevention of AD5.
In an independent dataset, our results replicate the original study that found a protective effect of Bumetanide exposure on AD risk4. We further investigated whether this effect is generalizable to the more commonly used and less expensive medication in the same class and adjusted for potential confounding variables such as SES and CHF. In our study, despite their similar indications and mechanism of action, only bumetanide exposure associated with reduced future AD diagnosis. Notably, both bumetanide and furosemide seem to penetrate the blood brain barrier6–9. Potential explanations for these results include unique effects of bumetanide on the APOE genotype-dependent transcriptomic signature,4 bumetanide being a more potent medication than furosemide with a dose ratio of 1:40 mg, or potential protective molecular modulation of neuronal transmembrane chloride gradients by blocking NKCC1 in the central nervous system.10 A mechanism that led to its proposed investigations to treat autism,11 schizophrenia,12 and epilepsy12,13.
Our OMOP EHR dataset analysis demonstrated potential association between past bumetanide exposure and AD onset. This effect remained significant even after correcting for SES and CHF indicating that the results are not driven by differences in SES or severity of cardiac disease.
These results should be treated cautiously since they are based on retrospective data. Bumetanide is a potent loop diuretic which, if given in excessive amounts, can lead to a profound diuresis with water and electrolyte depletion which is particularly problematic in the elderly population. Additionally, insurance and income were modeled through proxies available in OMOP and may not fully account for difference in SES. Last, additional functional studies are warranted to investigate the biological mechanism through which bumetanide exposure is associated with reduced AD risk. The current findings do not support the use of bumetanide for the prevention or treatment of AD. There is a need for prospective, randomized, double-blinded, placebo-controlled clinical trials to confirm the findings in patients without comorbidities and determine the lowest effective dose that may reduce the risk of AD without causing intolerable side effects. Given the lack of effect of furosemide and its similar side effect profile, one could also consider using it as an active placebo to increase the efficacy of double blinding.