Abstract
Background:
Previous studies suggested that glucagon-like peptide-1 receptor agonists (GLP-1RAs) may have a disease-modifying effect in the development of Parkinson’s disease (PD), but population studies yielded inconsistent results.
Objective:
To compare the risk of PD associated with GLP-1RAs compared to dipeptidyl peptidase 4 inhibitors (DPP4i) among older adults with type 2 diabetes (T2D).
Methods:
Using U.S. Medicare administrative data from 2016 to 2020, we conducted a population-based cohort study comparing new use of GLP1-RA with new use of DPP4i among adults aged ≥ 66 years with T2D. The primary endpoint was a new diagnosis of PD. A stabilized inverse probability of treatment weighting (sIPTW) – adjusted Cox proportional hazards regression model was employed to estimate the hazard ratio (HR) and 95% confidence intervals (CI) for PD between GLP-1RA and DPP4i users.
Results:
This study included a total of 89,074 Medicare beneficiaries who initiated either GLP-1RA (n=30,091) or DPP4i (n=58,983). The crude incidence rate of PD was lower among GLP-1RA users than DPP4i users (2.85 vs. 3.92 cases per 1000 person-years). An sIPTW-adjusted Cox model showed that GLP-1RA users were associated with a 23% lower risk of PD than DPP4i users (HR, 0.77; 95%CI, 0.63–0.95). Our findings were largely consistent across different subgroup analyses such as sex, race, and molecular structure of GLP-1RA.
Conclusion:
Among Medicare beneficiaries with T2D, new use of GLP-1RAs was significantly associated with a decreased risk of PD compared to new use of DPP4i.
Keywords: GLP-1RA, DPP4i, Parkinson’s disease, type 2 diabetes
Introduction
Parkinson’s disease (PD) is the second-most common neurodegenerative disease, characterized by a wide range of debilitating motor and non-motor symptoms1. PD represents a major public health challenge, affecting nearly one million people in the United States (U.S.) alone2, with this number projected to double by 20403. Moreover, PD imposes a substantial economic burden on U.S. society, with costs estimated at $51.9 billion in 20172. Nonetheless, there is still no available pharmacologic therapy to cure or slow the progression of PD. While the exact etiology underlying the development of PD remains unknown, accumulative evidence has suggested linking type 2 diabetes (T2D) to PD4. Both conditions share common pathogenic mechanisms, such as insulin dysregulation, mitochondrial dysfunction, and neuroinflammation5–7. This raises the intriguing possibility that certain glucose-lowering drugs (GLDs) used to treat T2D, may also hold the potential for preventing or treating PD.
Glucagon-like peptide-1 receptor agonists (GLP-1RAs) are a newer class of GLDs that have gained popularity due to their benefits beyond glycemic control, including cardiorenal and weight loss benefits8. Importantly, preclinical studies have shown neuroprotective effects of GLP-1RAs, including improvements in motor function and cognition, mediated through their ability to ameliorate insulin resistance and inflammation9. However, population studies examining the association between GLP-1RAs and risk of PD yielded conflicting results10–12, which may be attributable to the selection of comparator. Dipeptidyl peptidase 4 inhibitors (DPP4i) shared similar mechanisms of action with GLP-1RAs, lowering glucose levels, and both drug classes are recommended as second-line treatments for T2D13, making DPP4i an ideal comparator for minimizing confounding by indication. Furthermore, DPP4i are not associated with an increased risk of PD14. Given the additional benefits of GLP-1RAs such as weight loss and cardiovascular and renal benefits8, it remains unclear whether GLP-1RAs could confer additional neuroprotective benefits and subsequently reduce the risk of PD to a greater extent compared to DPP4i. Therefore, we conducted a population-based cohort study to assess the risk of PD associated with GLP-1RAs among older individuals with T2D compared to DPP4i.
Materials and Methods
Study Design and Data Source
This study was a retrospective population-based cohort study using an active-comparator, new user study design to evaluate the risk of PD associated with GLP-1RAs as compared to DPP4i in Medicare administrative data (Medicare) (Supplement Figure 1).
Medicare is a federal health insurance program that primarily provides medical coverage for the U.S. population aged ≥ 65 years, including Parts A (inpatient), B (outpatient physician services), and D (dispensed prescription drugs) coverage. The Medicare database included longitudinal, individual-level data such as demographics, inpatient and outpatient diagnoses and procedures, as well as pharmacy claims. For this study, we accessed data from a 15% random sample of all Medicare beneficiaries with fee-for-service coverage of Medicare Parts A, B, and D between January 2016 and December 2020. This study was approved by the University of Florida Institutional Review Board.
Study population
To ensure comprehensive medical records for patients, this study included patients aged 66 years or older who were continuously enrolled in Medicare Parts A, B, and D for at least one year before the cohort entry date (index date). The study population consisted of patients diagnosed with T2D who initiated treatment with a GLP-1RA or a DPP4i between 1 January 2017 and 31 December 2020. The drug lists included in GLP-1RA and DPP4i are summarized in Supplement Table 1. The index date was the day of the first prescription for a GLP-1RA or DPP4i defined as without a previous prescription for either drug within the preceding year. To identify individuals with T2D, we identified those with a diagnosis of diabetes using the Chronic Conditions Warehouse (CCW) data and excluded those with an International Classification of Diseases (ICD) code of type 1 diabetes15. DPP4i were selected as the active comparator due to their similarity in clinical indications and mechanisms of action compared to GLP-1RAs13, aiming to minimize potential confounding by indication. Also, previous evidence suggests that DPP4i are not associated with an increased risk of PD14, making them a suitable comparator for evaluating the potential neuroprotective effects of GLP-1RAs.
Individuals were excluded if they had the following diagnoses and treatments during the baseline period: any form of parkinsonism, Lewy body dementia, end-stage renal disease, and prior exposure to anti-PD medications (e.g., levodopa, dopamine agonist, monoamine oxidase inhibitors, and entacapone). The definitions of the aforementioned conditions are summarized in Supplement Table 2. The individuals who started treatment with both GLP-1RA and DPP4i on the index date were also excluded.
Study outcome and Follow-up
The outcome of interest in this study was a new diagnosis of PD, determined by having at least two diagnosis codes for PD per individual. Using this approach to identify the PD cases yielded a sensitivity of 89.6% and a positive predictive value (PPV) of 79.4%16. The first diagnosis date during the follow-up defined the outcome date. The ICD diagnosis codes used for identifying PD are available in Supplement Table 2.
Since PD is an irreversible and chronic disease, we followed the “intention-to-treat” (ITT) principle of randomized controlled trial analysis, which did not censor data on the discontinuation of the index drug (switching to or addition of the comparator). The individuals were followed up from the day after cohort entry until the first occurrence of the following events: a study outcome, death, disenrollment from Parts A, B, or D, and the end of the study period (31 December 2020).
Statistical analysis
We calculated the incidence rate of PD for GLP-1RAs and DPP4i groups and employed a Cox proportional hazards regression model to estimate the hazard ratio (HR) and 95% confidence interval (CI) of PD between the two groups17. We included a broad set of baseline covariates as potential confounders, including demographic characteristics, comorbidities, and co-medications. These covariates were selected based on clinical experience and literature18, and were obtained 1 year before or on the index date (Table 1). To account for the nonrandom allocation of individuals receiving the treatment, a stabilized inverse probability of treatment weight (sIPTW) was applied to reduce the effects of confounding. The sIPTW created a pseudo population in that the distribution of measured baseline covariates was independent of treatment selection17. sIPTW was derived from propensity-score (PS) which was calculated using a multivariable logistic regression model that modeled the probability of each patient initiating a GLP-1RA including baseline covariates as outlisted in Table 1. We assessed the balance of baseline covariate before and after weighted cohorts using standardized mean differences (SMD), with a value less than 0.1 suggesting a negligible imbalance between the two groups19. We also plotted the cumulative incidence of PD using an sIPTW-adjusted Kaplan–Meier plot.
Table 1.
Baseline characteristics of patients included in the study
| Characteristic | Original cohort | SMD* | |||
|---|---|---|---|---|---|
|
| |||||
| All (n=89,074) | GLP-1RAs (n=30,091) | DPP4i (n=58,983) | Before sIPTW | After sIPTW | |
| Age, yrs, mean(sd) | 75.1 (6.8) | 72.7 (5.3) | 76.3 (7.2) | −0.57 | 0.004 |
| Female | 48740 (54.7%) | 15868 (52.7%) | 32872 (55.7%) | −0.06 | −0.005 |
| Race/ethnicity | |||||
| Non-Hispanic | 63984 (71.8%) | 23423 (77.8%) | 40561 (68.8%) | 0.21 | 0.000 |
| Whites | |||||
| Non-Hispanic Black | 8755 (9.8%) | 2543 (8.5%) | 6212 (10.5%) | ||
| Hispanic | 9767 (11.0%) | 2406 (8.0%) | 7361 (12.5%) | ||
| Others | 6568 (7.4%) | 1719 (5.7%) | 4849 (8.2%) | ||
| Medicare and Medicaid dual eligibility | 24826 (27.9%) | 6192 (20.6%) | 18634 (31.6%) | −0.253 | 0.011 |
| Low-income Subsidy | 27845 (31.3%) | 7167 (23.8%) | 20678 (35.1%) | −0.249 | 0.012 |
| Diabetes-related conditions | |||||
| Diabetes retinopathy | 9740 (10.9%) | 3905 (13.0%) | 5835 (9.9%) | 0.097 | 0.004 |
| Diabetic neuropathy | 25787 (29.0%) | 9675 (32.2%) | 16112 (27.3%) | 0.106 | 0.003 |
| Peripheral vascular disease | 18958 (21.3%) | 5645 (18.8%) | 13313 (22.6%) | −0.094 | 0.001 |
| Hypoglycemia | 2413 (2.7%) | 664 (2.2%) | 1749 (3.0%) | −0.048 | −0.014 |
| Hyperglycemic emergency |
343 (0.4%) | 120 (0.4%) | 223 (0.4%) | 0.003 | 0.004 |
| Comorbid conditions | |||||
| Acute myocardial infarction | 6065 (6.8%) | 1731 (5.8%) | 4334 (7.3%) | −0.065 | 0.008 |
| Alzheimer’s disease and related dementias | 14009 (15.7%) | 2878 (9.6%) | 11131 (18.9%) | −0.269 | 0.013 |
| Atrial fibrillation | 15734 (17.7%) | 4375 (14.5%) | 11359 (19.3%) | −0.126 | −0.001 |
| Cataract | 57031 (64.0%) | 17445 (58.0%) | 39586 (67.1%) | −0.190 | 0.008 |
| Chronic kidney disease | 63930 (71.8%) | 22050 (73.3%) | 41880 (71.0%) | 0.051 | 0.007 |
| Chronic obstructive pulmonary disease | 26044 (29.2%) | 7707 (25.6%) | 18337 (31.1%) | −0.122 | 0.012 |
| Chronic heart failure | 30180 (33.9%) | 8889 (29.5%) | 21291 (36.1%) | −0.140 | 0.010 |
| Glaucoma | 23363 (26.2%) | 6793 (22.6%) | 16570 (28.1%) | −0.127 | −0.010 |
| Hip or Pelvic Fracture | 2284 (2.6%) | 417 (1.4%) | 1867 (3.2%) | −0.120 | 0.015 |
| Ischemic heart disease | 53146 (59.7%) | 16802 (55.8%) | 36344 (61.6%) | −0.118 | 0.000 |
| Depression | 34801 (39.1%) | 11731 (39.0%) | 23070 (39.1%) | −0.003 | 0.005 |
| Osteoporosis | 15540 (17.4%) | 3859 (12.8%) | 11681 (19.8%) | −0.190 | −0.001 |
| Rheumatoid arthritis/osteoarthritis | 56390 (63.3%) | 18440 (61.3%) | 37950 (64.3%) | −0.063 | 0.010 |
| Stroke/TIA | 15310 (17.2%) | 3883 (12.9%) | 11427 (19.4%) | −0.177 | 0.011 |
| Breast cancer | 5115 (5.7%) | 1528 (5.1%) | 3587 (6.1%) | −0.044 | 0.002 |
| Colorectal cancer | 2573 (2.9%) | 659 (2.2%) | 1914 (3.2%) | −0.065 | 0.005 |
| Prostate cancer | 5250 (5.9%) | 1599 (5.3%) | 3651 (6.2%) | −0.038 | 0.004 |
| Lung cancer | 1292 (1.5%) | 300 (1.0%) | 992 (1.7%) | −0.060 | −0.006 |
| Endometrial cancer | 1110 (1.2%) | 352 (1.2%) | 758 (1.3%) | −0.011 | 0.004 |
| Anemia | 54572 (61.3%) | 16357 (54.4%) | 38215 (64.8%) | −0.214 | 0.007 |
| Asthma | 15379 (17.3%) | 5049 (16.8%) | 10330 (17.5%) | −0.020 | 0.002 |
| Hyperlipidemia | 84295 (94.6%) | 28423 (94.5%) | 55872 (94.7%) | −0.012 | 0.002 |
| Benign prostate hyperplasia | 20363 (22.9%) | 6479 (21.5%) | 13884 (23.5%) | −0.048 | 0.002 |
| Hypertension | 85481 (96.0%) | 28754 (95.6%) | 56727 (96.2%) | −0.031 | 0.003 |
| Acquired hypothyroidism | 29357 (33.0%) | 9615 (32.0%) | 19742 (33.5%) | −0.032 | −0.005 |
| Inflammatory bowel disease | 1027 (1.2%) | 327 (1.1%) | 700 (1.2%) | −0.009 | −0.001 |
| Obesity | 26827 (30.1%) | 12277 (40.8%) | 14550 (24.7%) | 0.349 | −0.003 |
| Medications | |||||
| Antidepressants | 29845 (33.5%) | 10846 (36.0%) | 18999 (32.2%) | 0.349 | −0.001 |
| Angiotensin-converting enzyme inhibitors | 33372 (37.5%) | 11239 (37.4%) | 22133 (37.5%) | 0.081 | 0.003 |
| Angiotensin receptor blockers | 32609 (36.6%) | 11520 (38.3%) | 21089 (35.8%) | −0.004 | −0.015 |
| Beta Blockers | 47230 (53.0%) | 15496 (51.5%) | 31734 (53.8%) | 0.052 | 0.001 |
| Calcium channel blockers | 33747 (37.9%) | 10498 (34.9%) | 23249 (39.4%) | −0.046 | −0.003 |
| Diuretics | 35671 (40.0%) | 12133 (40.3%) | 23538 (39.9%) | −0.094 | −0.002 |
| Opioids | 23882 (26.8%) | 8757 (29.1%) | 15125 (25.6%) | 0.009 | 0.007 |
| Antibiotics | 14447 (16.2%) | 4863 (16.2%) | 9584 (16.2%) | 0.078 | −0.006 |
| Statins | 70072 (78.7%) | 24283 (80.7%) | 45789 (77.6%) | −0.002 | −0.003 |
| Antipsychotics | 1714 (1.9%) | 501 (1.7%) | 1213 (2.1%) | 0.076 | 0.009 |
| NSAIDS | 19599 (22.0%) | 6869 (22.8%) | 12730 (21.6%) | −0.029 | −0.008 |
| Oral Steroids | 35139 (39.4%) | 11841 (39.4%) | 23298 (39.5%) | 0.030 | −0.004 |
| Antiplatelets | 2132 (2.4%) | 723 (2.4%) | 1409 (2.4%) | −0.003 | 0.006 |
| Aldosterone receptor antagonists | 5595 (6.3%) | 1984 (6.6%) | 3611 (6.1%) | 0.001 | −0.003 |
| Anticoagulants | 13492 (15.1%) | 3985 (13.2%) | 9507 (16.1%) | 0.019 | 0.002 |
| Immunosuppressants | 387 (0.4%) | 120 (0.4%) | 267 (0.5%) | −0.081 | 0.002 |
| Tumor necrosis factor inhibitors | 218 (0.2%) | 71 (0.2%) | 147 (0.2%) | −0.008 | 0.000 |
| Other GLDs | |||||
| GLD use at baseline | |||||
| Insulin | 23465 (26.3%) | 12709 (42.2%) | 10756 (18.2%) | 0.565 | 0.052 |
| No GLD | 8942 (10.0%) | 2100 (7.0%) | 6842 (11.6%) | ||
| 1 GLD (excluding insulin) | 30301 (34.0%) | 7173 (23.8%) | 23128 (39.2%) | ||
| ≥ 2 GLDs (excluding insulin) |
26366 (29.6%) | 8109 (26.9%) | 18257 (31.0%) | ||
| Metformin | 59119 (66.4%) | 20099 (66.8%) | 39020 (66.2%) | 0.014 | 0.003 |
| Sulfonylureas | 37346 (41.9%) | 11703 (38.9%) | 25643 (43.5%) | −0.093 | 0.022 |
| SGLT2 inhibitors | 8890(10.0%) | 4099(6.9%) | 4791(15.9%) | 0.285 | 0.004 |
| Thiazolidinediones | 6746 (7.6%) | 2611 (8.7%) | 4135 (7.0%) | 0.062 | 0.006 |
| Meglitinides | 1695 (1.9%) | 525 (1.7%) | 1170 (2.0%) | −0.018 | 0.008 |
| Alpha-glucosidase inhibitors |
460 (0.5%) | 139 (0.5%) | 321 (0.5%) | −0.012 | 0.004 |
After sIPTW with an SMD ≤ 0.1 indicating a balance between the 2 groups.
GLP-1RAs, glucagon-like peptide-1 receptor agonists; DPP4i, dipeptidyl peptidase 4 inhibitors; SGLT2 Inhibitors, sodium-glucose cotransporter 2 inhibitors; GLDs, glucose-lowering drugs; NSAIDS, non-steroidal anti-inflammatory drugs; TIA, transient ischemic attack; SMD, standardized mean difference; sIPTW, stabilized inverse probability of treatment weight.
We conducted several sensitivity analyses to assess the robustness of our findings. First, we created a 1:1 PS-matched cohort using a nearest-neighbor matching without replacement approach within a maximum caliper width of 0.0520. Second, we applied an “as-treated” analysis that accounted for treatment discontinuation of index drug (defined as 60 days elapsed after the expiration date of the last prescription’s supply without the prescription being refilled) or switching to or addition of the comparator. An additional follow-up up to 1 year after censoring was applied. Third, to address the competing risk of all-cause mortality, we employed a Cox proportional hazards model with the Fine and Gray method to estimate the adjusted sub-distribution HR for PD21. Fourth, to minimize potential reverse causality bias, where underlying prodromal PD may have influenced treatment selection, we excluded patients who had a PD diagnosis within the first 6 months after the index date. Moreover, we also quantified the association between GLP-1Ras and PD and tested the potential interaction in the following subgroups:1) age ( ≥ 75 years vs. < 75 years); 2) sex (female vs. male); 3) race/ethnicity (non-Hispanic White population vs. non-Hispanic Black population vs. Hispanic population vs. others); 4) GLD use at baseline (insulin vs. no GLD vs 1 GLD (excluding insulin) vs ≥ 2 GLDs (excluding insulin)); 5) obesity at baseline (yes vs. no); 6) chronic kidney disease at baseline (CKD) (yes vs. no); 7) atherosclerotic cardiovascular disease (ASCVD) at baseline (yes vs. no); 8) molecular structure of GLP-1RA (exenatide vs. dulaglutide vs. liraglutide vs. semaglutide).
In addition, to assess the potential unmeasured confounder between GLP-1RAs and risk of PD, we calculated the E-value 22. The E-value provides an assumption-free estimate of an unmeasured confounder that would be necessary to negate the observed results22. A P-value <0.05 was considered statistically significant. All statistical analyses were performed using SAS version 9.4 (SAS Institute Inc, NC).
Results
Study population
This study included a total of 89,074 individuals who initiated either GLP-1RA (n=30,091) or DPP4i (n=58,983) (Figure 1). The baseline characteristics of study population are presented in Table 1. GLP-1RA initiators were younger than DPP4i initiators, with mean ages of 72.7 and 76.3 years, respectively. Additionally, new users of GLP-1RA had a higher proportion of non-Hispanic Whites (77.8% vs. 68.8%), obesity (40.8% vs 24.7%), insulin use (42.2% vs 18.2%), diabetic retinopathy (13.0% vs. 9.9%), and diabetic neuropathy (32.2% vs. 27.3%), while a lower proportion of Alzheimer’s Disease and Related Dementias (ADRD) (9.6% vs. 18.9%) and stroke/ transient ischemic attack (TIA) (12.9% vs 19.4%) than new users of DPP4i. The median duration of follow-up for PD was 1.54 years (interquartile range,0.75 to 2.53) in the GLP-1RA group and 1.75 (interquartile range,0.83 to 2.77) in the DPP4i group. After sIPTW (Table 1), all baseline covariates were well-balanced with SMDs < 0.123.
FIGURE 1.
Flowchart of patient selection. GLP-1RAs, glucagon-like peptide-1 receptor agonists; DPP4i, dipeptidyl peptidase 4 inhibitors; T1D, type 1 diabetes; ESRD, end-stage renal disease.
Risk of PD
Of the study cohort, 143 individuals among 30,091 GLP-1RA users developed PD (IR, 2.85 cases per 1000 person-years) whereas 424 occurred among 58,983 DPP4i users (IR, 3.92 cases per 1000 person-years) (Table 2). This led to an unadjusted HR for incident PD associated with GLP-1RA compared to DPP4i was 0.73 (95% CI, 0.60 to 0.88). The sIPTW-adjusted Kaplan-Meier plot depicted the cumulative incidence of PD over time (Figure 2), showing a significantly lower risk of PD in the GLP-1RA group compared to the DPP4 inhibitor group (log-rank test, p-value = 0.02). Within the sIPTW-adjusted Cox proportional hazards model, GLP-1RAs were significantly associated with a lower risk of PD than DPP4i (HR, 0.77; 95%CI, 0.63–0.95).
Table 2.
Association between GLP-1RAs and risk of Parkinson’s disease.
| Analysis | Parkinson’s disease |
|---|---|
| No. of cases/no.of patients at risk(%) | |
| GLP-1RAs | 143/30,091(0.48) |
| DPP4i | 424/58,983 (0.72) |
| Incidence rate (No. of cases/1,000 person-years) | |
| GLP-1RAs | 2.85 |
| DPP4i | 3.92 |
| Crude HR (95%CI) | 0.73 (0.60–0.88) |
| sIPTW adjusted HR(95%CI) | 0.77 (0.62–0.95) |
GLP-1RAs, glucagon-like peptide-1 receptor agonists; DPP4i, dipeptidyl peptidase 4 inhibitors; HR, hazard ratio; CI, confidence interval; sIPTW, stabilized inverse probability of treatment weight; PS, propensity score.
FIGURE 2.
Cumulative incidence of Parkinson’s disease in sIPTW GLP-1RA and DPP4i cohort. sIPTW, stabilized inverse probability of treatment weight.
Sensitivity and Subgroup analyses
In the sensitivity analyses, the were consistent when using a 1:1 PS matching Cox model (HR, 0.73; 95%CI, 0.58 – 0.92), an “as-treated” approach (HR, 0.65; 95%CI, 0.52 – 0.82), and using the Fine and Gray method (HR, 0.80; 95%CI, 0.64–0.98). However, the association between GLP-1RA and decreased risk of PD was attenuated when excluding those with a diagnosis of PD within the first 6 months after the index date (HR, 0.80; 95%CI, 0.57–1.14).
The results of subgroup analyses are presented in Figure 3. The treatment effects were consistent across the subgroups by factors such as age, sex, race/ethnicity, having a diagnosis of obesity at baseline, having a diagnosis of CKD at baseline, having a diagnosis of ASCVD at baseline, and molecular structure of GLP-1RA. However, GLD use at baseline appeared to be a potential modifier of the association between GLP-1RA and risk of PD (p=0.01 for interaction). Patients using insulin at baseline seemed to derive greater benefits from GLP-1RAs (HR, 0.51; 95%CI, 0.36–0.72) compared to those using 1 GLD at baseline (excluding insulin) (HR, 1.12; 95%CI, 0.75 – 1.68).
FIGURE 3.
Subgroup analyses of the association between GLP-1RA and risk of Parkinson’s disease in sIPTW GLP-1RA and DPP4i cohort. NHW, non-Hispanic White; NHB, non-Hispanic Black; GLD, glucose-lowering drug; CKD, chronic kidney disease; ASCVD, atherosclerotic cardiovascular disease.
E-value
The E-value for the risk of PD between GLP-1RA and DPP4i was 1.92. This suggests that the observed association could be explained away by an unmeasured confounder that was associated with both GLP-1RA use and PD, with at least risk ratio of 1.92-fold each.
Discussion
In this population-based cohort study of U.S. older adults with T2D, we found that new users of GLP-1RAs had a significantly decreased risk of new-onset PD compared to new users of DPP4i. However, the association between GLP-1RA and decreased risk of PD was attenuated in a sensitivity analysis that excluded patients with a diagnosis of PD within the first 6 months after the index date. Our findings were generally consistent across various subgroups stratified by factors such as sex, race, and molecular structure of GLP-1RA. GLD use at baseline appeared to potentially modify the association between GLP-1RA and risk of PD.
Our finding of a lower risk of PD associated with GLP-1RAs is supported by emerging evidence from multiple mechanistic studies24–27. In preclinical models of PD, GLP-1RAs manifested neuroprotective effects by improving motor function, rescuing dopaminergic neuronal loss and motor impairment, restoring dopamine synthesis, and increasing cortical activity and energy utilization in the brain24–27. Importantly, GLP-1RAs may attenuate dyskinesia, a complication of chronic levodopa replacement therapy28. Early clinical trials showed promising signals of potential disease modification with exenatide (a GLP-1RA) among individuals with PD29–32. A trial involving 62 individuals with moderate PD reported positive and sustained improvements in motor function over 12 weeks following administration of exenatide29. Then, a post hoc analysis indicated the potential benefits of exenatide for non-motor symptoms like mood and emotional well-being, although these effects were transient30. However, another trial found no benefits in motor or non-motor symptoms in individuals with early untreated PD receiving NLY01 (a brain-penetrant, pegylated, long-lasting version of exenatide), though a possible motor benefit was observed in younger individuals (age < 60 years)33. In a phase 2 trial including participants with early PD, lixisenatide therapy resulted in less progression of motor disability at 12 months34.
Our observation of a significantly lower risk of PD in GLP-1RA users than in DPP4i users, aligns with the enhanced ability of GLP-1RAs to activate GLP-1 receptors, surpassing the effects achieved through increasing endogenous GLP-1 levels with DPP4i35. Previous research has indicated that GLP-1RAs can cross the blood-brain barrier and exert their neuroprotective properties36. Thus, it is suggested that GLP-1RAs with greater brain penetrance, such as exenatide and lixisenatide, may be more likely to modify the clinical course of PD. However, our subgroup analysis did not detect a statistically significant difference in risk of PD across different molecular structures of GLP-1RA (p=0.38). These findings add to the growing body of evidence supporting the potential neuroprotective benefits of GLP-1RAs in mitigating PD development and symptom progression. However, our findings must be contextualized in light of the recent negative clinical trial results for NLY0133, and further research is warranted to elucidate the mechanisms and clinical implications of our observations.
We observed an interaction effect between GLP-1RA use and other GLD use on the risk of PD. GLD use at baseline may modify the association between GLP-1RA and risk of PD. Individuals with insulin use at baseline may have greater benefits from GLP-1RAs on reducing risk of PD than those using 1 GLD use at baseline (excluding insulin). Insulin use itself was associated with an increased risk of PD when compared to those not using insulin37. This suggests that GLP-1RAs may mitigate any adverse effects associated with insulin use, consequently reducing the risk of PD. Moreover, insulin use and/or number of GLDs/ have been considered to be proxies for the severity of diabetes37,38. Insulin treatment is typically prescribed for individuals with T2D who are insulin-deficient and/or have failed other GLDs, and is therefore linked to severe diabetes38. Notably, diabetes severity has been identified as a crucial factor that significantly increases the risk of developing PD37. Our findings suggest that GLP-1RAs possibly attenuate the risk for PD through improved glycemic control and management of complications of T2D, as well as related deleterious neuroinflammatory effects, though this remains speculative. It is intriguing to find a decreased risk of PD associated with GLP-1RAs among individuals with no GLD use at baseline, indicating that the first-line use of GLP-1RA among individuals with T2D may have beneficial effects on risk of developing PD, though this required further investigation.
Existing observational studies exploring the association between GLP-1RAs and risk of PD have yielded mixed results10–12. Two case-control studies found a non-significant difference between GLP-1RAs and risk of PD10,12, while one population-based cohort study using primary care data from the Health Improvement Network showed an inverse association between GLP-1RAs and onset of PD when compared with other oral GLDs in individuals with diabetes (adjusted incidence rate ratio, 0.38; 95%CI, 0.17–0.60)11. It should be noted that these studies had several inherent limitations, such as time-related bias10–12, residual confounding10,12, and potential exposure misclassification10,12. Additionally, the non-significant findings in the case-control study may be explained by the limited sample size of GLP-1RA users10,12.
Our study addressed these limitations with several strategies. First, we applied an active comparator study design using GLP-1RAs versus DPP4i, which mitigates the susceptibility to confounding bias (e.g., confounding by indication)39. The choice of DPP4i as the active comparator was appropriate given their similar mechanisms of action to GLP-1RAs and the clinical practice of using these drug classes at similar stages of diabetes13. Also, Previous studies have not associated DPP4i use with an increased risk of PD 40,41. Our finding of a significantly decreased risk of PD among GLP-1RA users compared to DPP4i users, indicates a potential protective role of GLP-1RAs against the development of PD. Second, we defined cohort entry as the first prescription of a GLP-1RA or a DPP4i and identified newer users based on a 1-year washout period, not only reducing the immortal time bias but also minimizing the influence of pre-exposure on study outcomes. These improvements in study design strengthen our findings and contribute to a more nuanced understanding of the potential relationship between GLP-1RAs and PD risk.
However, our results should be interpreted with caution in light of several important limitations. First, while our analyses adjusted for a comprehensive set of potential confounders, we recognize the inherent limitations of overadjustment in observational studies. Correcting for an extensive number of factors may introduce biases and lead to overcorrection. To provide transparency, we included the crude (unadjusted) HR in addition to the IPTW adjusted estimates, allowing for the evaluation of the potential impact of the adjustments on the effect estimates. Nevertheless, residual confounding due to unmeasured covariates cannot be entirely ruled out. For instance, certain important confounders, such as the severity of diabetes, HbA1c, and body mass index (BMI) were unavailable in the claims data. To address this challenge, we adjusted for GLD use at baseline (e.g., insulin use at baseline), a proxy for the severity of the diabetes. A previous study indicated that employing an active comparator and new user design with PS matching to proxies of diabetes severity using claims-based data yielded an enhanced balance in unmeasured baseline covariates42. Despite our efforts to balance baseline characteristics through IPTW, there remained a higher proportion of insulin users in the GLP-1RA (SMD = 0.052). This small imbalance warrants cautious interpretation, as it may indicate poorer baseline glycemic control and cardiovascular outcomes, as well as more severe diabetes in the GLP-1RA group, potentially leading to an underestimation of the association between GLP-1RA and risk of PD. In this study, we also used the E-value to assess the potential effect of unmeasured confounding with a value of 1.92, suggesting that a moderately strong unmeasured confounder associated with both treatment and PD could potentially nullify the observed association22. Second, there was a potential for misclassification of PD diagnosis in this study. Incident PD was defined as having at least two medical claims with a PD diagnosis code. While this approach seemed to be a reasonable algorithm with a sensitivity of 89.6% and a PPV of 79.4%43, some degree of misclassification is likely. Certain cases may have been missed or falsely classified as non-cases. This misclassification could have biased our effect estimates toward or away from the null, potentially underestimating underestimating or overestimating the true effect. Third, our study has a relatively short follow-up with a median of 1.54 years for GLP-1RA group and 1.75 years for DPP4i group. This limited follow-up period would have impacted our ability to fully capture the long-term effects of GLP-1RAs and DPP4i on risk of PD. Also, a shorter observation window increases the potential for protopathic bias, where prodromal disease symptoms or characteristics could have influenced the initial treatment selection. Individuals with early, undiagnosed PD have a higher likelihood of receiving DPP4i rather than receiving GLP-1RA, as evidenced by the higher prevalence of ADRD and stroke/TIA among the DPP4i group than GLP-1RA group in the original cohort. This channeling of patients with prodromal disease into the comparator group could systematically bias the results, potentially overestimating the protective association observed with GLP-1RA. Fourth, not all available GLP-1RAs were available in Medicare claim data. Lixisenatide, a GLP-1RA with greater brain penetrance, would be more likely to confer neuroprotective benefits. Lixisenatide was one of the exposures of interest, but it was infrequently prescribed in Medicare Part D44 and no patients using lixisenatide were included in this study. Although no significant difference across the molecular structures of GLP-1RA was observed in this study, future research is warranted to clarify whether the GLP-1RAs with higher brain penetrance could have more substantial benefits on reducing the development of PD. Fifth, this study employed a sIPTW as the primary analysis method to mitigate confounding bias while preserving the sample size. However, this approach has inherent limitations, including the potential for extreme weights and bias amplification. To corroborate our findings, we also performed 1:1 PS matching which yielded similar results. Finally, this study included older individuals with T2D, thus, the generalizability of our findings to younger individuals or those without T2D remains uncertain.
Conclusions
In summary, this population-based study found that older individuals with T2D who initiated GLP-1RA therapy had a reduced risk of PD compared to those who initiated a DPP4i drug. However, these findings should be interpreted with caution due to several limitations, such as short follow-up duration, potential unmeasured confounders, and possible misclassification of outcome. Future research with longer follow-up and more diverse real-world populations could further clarify this association.
Supplementary Material
Acknowledgments:
We thank Yujia Li help extract the original cohort.
Financial Disclosures:
The work is supported by NIDDK (R01DK133465), American Foundation for Pharmaceutical Education (AFPE), and The Pharmaceutical Research and Manufacturers of America (PhRMA) Foundation.
Funding Sources for study:
This study was supported by the American Foundation for Pharmaceutical Education (AFPE) Predoctoral Fellowship, National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) (R01DK133465), and The Pharmaceutical Research and Manufacturers of America (PhRMA) Foundation. The funding organizations had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Footnotes
Relevant conflicts of interest/financial disclosure: The authors declare no conflicts of interest relevant to this work.
Data Sharing Statement :
The Medicare Administrative data could be obtained through ResDAC (email, resdac@umn.edu)
References
- 1.Jankovic J. Parkinson’s disease: clinical features and diagnosis. J Neurol Neurosurg Psychiatry. 2008;79(4):368–376. doi: 10.1136/jnnp.2007.131045 [DOI] [PubMed] [Google Scholar]
- 2.Yang W, Hamilton JL, Kopil C, et al. Current and projected future economic burden of Parkinson’s disease in the U.S. NPJ Park Dis. 2020;6:15. doi: 10.1038/s41531-020-0117-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Parkinson’s Disease: Challenges, Progress, and Promise | National Institute of Neurological Disorders and Stroke. Accessed November 27, 2023. https://www.ninds.nih.gov/current-research/focus-disorders/parkinsons-disease-research/parkinsons-disease-challenges-progress-and-promise
- 4.Sabari SS, Balasubramani K, Iyer M, et al. Type 2 Diabetes (T2DM) and Parkinson’s Disease (PD): a Mechanistic Approach. Mol Neurobiol. 2023;60(8):4547–4573. doi: 10.1007/s12035-023-03359-y [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Kleinridders A, Cai W, Cappellucci L, et al. Insulin resistance in brain alters dopamine turnover and causes behavioral disorders. Proc Natl Acad Sci U S A. 2015;112(11):3463–3468. doi: 10.1073/pnas.1500877112 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Joers V, Tansey MG, Mulas G, Carta AR. Microglial phenotypes in Parkinson’s disease and animal models of the disease. Prog Neurobiol. 2017;155:57–75. doi: 10.1016/j.pneurobio.2016.04.006 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Hong CT, Chen KY, Wang W, et al. Insulin Resistance Promotes Parkinson’s Disease through Aberrant Expression of α-Synuclein, Mitochondrial Dysfunction, and Deregulation of the Polo-Like Kinase 2 Signaling. Cells. 2020;9(3):740. doi: 10.3390/cells9030740 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Marx N, Federici M, Schütt K, et al. 2023 ESC Guidelines for the management of cardiovascular disease in patients with diabetes. Eur Heart J. 2023;44(39):4043–4140. doi: 10.1093/eurheartj/ehad192 [DOI] [PubMed] [Google Scholar]
- 9.Nowell J, Blunt E, Gupta D, Edison P. Antidiabetic agents as a novel treatment for Alzheimer’s and Parkinson’s disease. Ageing Res Rev. 2023;89:101979. doi: 10.1016/j.arr.2023.101979 [DOI] [PubMed] [Google Scholar]
- 10.Svenningsson P, Wirdefeldt K, Yin L, et al. Reduced incidence of Parkinson’s disease after dipeptidyl peptidase-4 inhibitors-A nationwide case-control study. Mov Disord Off J Mov Disord Soc. 2016;31(9):1422–1423. doi: 10.1002/mds.26734 [DOI] [PubMed] [Google Scholar]
- 11.Brauer R, Wei L, Ma T, et al. Diabetes medications and risk of Parkinson’s disease: a cohort study of patients with diabetes. Brain J Neurol. 2020;143(10):3067–3076. doi: 10.1093/brain/awaa262 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Sunnarborg K, Tiihonen M, Huovinen M, Koponen M, Hartikainen S, Tolppanen A. Association between different diabetes medication classes and risk of Parkinson’s disease in people with diabetes. Pharmacoepidemiol Drug Saf. 2022;31(8):875–882. doi: 10.1002/pds.5448 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.American Diabetes Association Professional Practice Committee. 9. Pharmacologic Approaches to Glycemic Treatment: Standards of Care in Diabetes—2024. Diabetes Care. 2023;47(Supplement_1):S158–S178. doi: 10.2337/dc24-S009 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Xie Y, Wang J, Jiang J, Liu F, Zhang Y. Do oral antidiabetic medications alter the risk of Parkinson’s disease? An updated systematic review and meta-analysis. Neurol Sci Off J Ital Neurol Soc Ital Soc Clin Neurophysiol. 2023;44(12):4193–4203. doi: 10.1007/s10072-023-06965-9 [DOI] [PubMed] [Google Scholar]
- 15.Clements JM, West BT, Harissa B, Hayden N, Khan MM, Palepu R. Race Disparities in the Use of Prevention, Screening, and Monitoring Services in Michigan Medicare Beneficiaries With Type 2 Diabetes and Combinations of Multiple Chronic Conditions. Clin Diabetes. 2020;38(4):363–370. doi: 10.2337/cd19-0088 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Szumski NR, Cheng EM. Optimizing algorithms to identify Parkinson’s disease cases within an administrative database. Mov Disord Off J Mov Disord Soc. 2009;24(1):51–56. doi: 10.1002/mds.22283 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Xu S, Ross C, Raebel MA, Shetterly S, Blanchette C, Smith D. Use of stabilized inverse propensity scores as weights to directly estimate relative risk and its confidence intervals. Value Health J Int Soc Pharmacoeconomics Outcomes Res. 2010;13(2):273–277. doi: 10.1111/j.1524-4733.2009.00671.x [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Belvisi D, Pellicciari R, Fabbrini A, et al. Risk factors of Parkinson disease. Neurology. 2020;95(18):e2500–e2508. doi: 10.1212/WNL.0000000000010813 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Franklin JM, Rassen JA, Ackermann D, Bartels DB, Schneeweiss S. Metrics for covariate balance in cohort studies of causal effects. Stat Med. 2014;33(10):1685–1699. doi: 10.1002/sim.6058 [DOI] [PubMed] [Google Scholar]
- 20.Ripollone JE, Huybrechts KF, Rothman KJ, Ferguson RE, Franklin JM. Implications of the Propensity Score Matching Paradox in Pharmacoepidemiology. Am J Epidemiol. 2018;187(9):1951–1961. doi: 10.1093/aje/kwy078 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Austin PC, Fine JP. Practical recommendations for reporting Fine-Gray model analyses for competing risk data. Stat Med. 2017;36(27):4391–4400. doi: 10.1002/sim.7501 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.VanderWeele TJ, Ding P. Sensitivity Analysis in Observational Research: Introducing the E-Value. Ann Intern Med. 2017;167(4):268–274. doi: 10.7326/M16-2607 [DOI] [PubMed] [Google Scholar]
- 23.Austin PC. Balance diagnostics for comparing the distribution of baseline covariates between treatment groups in propensity-score matched samples. Stat Med. 2009;28(25):3083–3107. doi: 10.1002/sim.3697 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Elbassuoni EA, Ahmed RF. Mechanism of the neuroprotective effect of GLP-1 in a rat model of Parkinson’s with pre-existing diabetes. Neurochem Int. 2019;131:104583. doi: 10.1016/j.neuint.2019.104583 [DOI] [PubMed] [Google Scholar]
- 25.Badawi GA, Abd El Fattah MA, Zaki HF, El Sayed MI. Sitagliptin and liraglutide reversed nigrostriatal degeneration of rodent brain in rotenone-induced Parkinson’s disease. Inflammopharmacology. 2017;25(3):369–382. doi: 10.1007/s10787-017-0331-6 [DOI] [PubMed] [Google Scholar]
- 26.Ma D, Liu X, Liu J, et al. Long-term liraglutide ameliorates nigrostriatal impairment via regulating AMPK/PGC-1a signaling in diabetic mice. Brain Res. 2019;1714:126–132. doi: 10.1016/j.brainres.2019.02.030 [DOI] [PubMed] [Google Scholar]
- 27.Zhang L, Zhang L, Li L, Hölscher C. Neuroprotective effects of the novel GLP-1 long acting analogue semaglutide in the MPTP Parkinson’s disease mouse model. Neuropeptides. 2018;71:70–80. doi: 10.1016/j.npep.2018.07.003 [DOI] [PubMed] [Google Scholar]
- 28.Badawi GA, Abd El Fattah MA, Zaki HF, El Sayed MI. Sitagliptin and Liraglutide Modulate L-dopa Effect and Attenuate Dyskinetic Movements in Rotenone-Lesioned Rats. Neurotox Res. 2019;35(3):635–653. doi: 10.1007/s12640-019-9998-3 [DOI] [PubMed] [Google Scholar]
- 29.Athauda D, Maclagan K, Skene SS, et al. Exenatide once weekly versus placebo in Parkinson’s disease: a randomised, double-blind, placebo-controlled trial. The Lancet. 2017;390(10103):1664–1675. doi: 10.1016/S0140-6736(17)31585-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Athauda D, Maclagan K, Budnik N, et al. What Effects Might Exenatide have on Non-Motor Symptoms in Parkinson’s Disease: A Post Hoc Analysis. J Park Dis. 2018;8(2):247–258. doi: 10.3233/JPD-181329 [DOI] [PubMed] [Google Scholar]
- 31.Aviles-Olmos I, Dickson J, Kefalopoulou Z, et al. Exenatide and the treatment of patients with Parkinson’s disease. J Clin Invest. 2013;123(6):2730–2736. doi: 10.1172/JCI68295 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Aviles-Olmos I, Dickson J, Kefalopoulou Z, et al. Motor and cognitive advantages persist 12 months after exenatide exposure in Parkinson’s disease. J Park Dis. 2014;4(3):337–344. doi: 10.3233/JPD-140364 [DOI] [PubMed] [Google Scholar]
- 33.McGarry A, Rosanbalm S, Leinonen M, et al. Safety, tolerability, and efficacy of NLY01 in early untreated Parkinson’s disease: a randomised, double-blind, placebo-controlled trial. Lancet Neurol. 2024;23(1):37–45. doi: 10.1016/S1474-4422(23)00378-2 [DOI] [PubMed] [Google Scholar]
- 34.Meissner WG, Remy P, Giordana C, et al. Trial of Lixisenatide in Early Parkinson’s Disease. N Engl J Med. 2024;390(13):1176–1185. doi: 10.1056/NEJMoa2312323 [DOI] [PubMed] [Google Scholar]
- 35.He D, Aleksic S. Is it time to repurpose geroprotective diabetes medications for prevention of dementia? J Am Geriatr Soc. 2023;71(7):2041–2045. doi: 10.1111/jgs.18405 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Hunter K, Hölscher C. Drugs developed to treat diabetes, liraglutide and lixisenatide, cross the blood brain barrier and enhance neurogenesis. BMC Neurosci. 2012;13:33. doi: 10.1186/1471-2202-13-33 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Han K, Kim B, Lee SH, Kim MK. A nationwide cohort study on diabetes severity and risk of Parkinson disease. Npj Park Dis. 2023;9(1):1–8. doi: 10.1038/s41531-023-00462-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Zghebi SS, Panagioti M, Rutter MK, et al. Assessing the severity of Type 2 diabetes using clinical data-based measures: a systematic review. Diabet Med. 2019;36(6):688–701. doi: 10.1111/dme.13905 [DOI] [PubMed] [Google Scholar]
- 39.Lund JL, Richardson DB, Stürmer T. The active comparator, new user study design in pharmacoepidemiology: historical foundations and contemporary application. Curr Epidemiol Rep. 2015;2(4):221–228. doi: 10.1007/s40471-015-0053-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40.Qin X, Zhang X, Li P, et al. Association Between Diabetes Medications and the Risk of Parkinson’s Disease: A Systematic Review and Meta-Analysis. Front Neurol. 2021;12. doi: 10.3389/fneur.2021.678649 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41.Lin YH, Hsu CC, Liu JS, Chang KC, Huang JA. Use of dipeptidyl peptidase-4 inhibitors was associated with a lower risk of Parkinson’s disease in diabetic patients. Sci Rep. 2023;13(1):22489. doi: 10.1038/s41598-023-49870-z [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42.Patorno E, Gopalakrishnan C, Franklin JM, et al. Claims-based studies of oral glucose-lowering medications can achieve balance in critical clinical variables only observed in electronic health records. Diabetes Obes Metab. 2018;20(4):974–984. doi: 10.1111/dom.13184 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43.Bujang MA, Adnan TH. Requirements for Minimum Sample Size for Sensitivity and Specificity Analysis. J Clin Diagn Res JCDR. 2016;10(10):YE01-YE06. doi: 10.7860/JCDR/2016/18129.8744 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 44.Luo J, Feldman R, Rothenberger SD, Hernandez I, Gellad WF. Coverage, Formulary Restrictions, and Out-of-Pocket Costs for Sodium-Glucose Cotransporter 2 Inhibitors and Glucagon-Like Peptide 1 Receptor Agonists in the Medicare Part D Program. JAMA Netw Open. 2020;3(10):e2020969. doi: 10.1001/jamanetworkopen.2020.20969 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The Medicare Administrative data could be obtained through ResDAC (email, resdac@umn.edu)