Introduction
Parkinson's disease is a progressive neurodegenerative disorder. Type 2 diabetes has been associated with an increased risk of Parkinson's disease as well as faster disease progression [1, 2, 3]. There are no curative or disease‐modifying therapies available for Parkinson's disease for clinical use. Identifying potential interventions that may prevent the disease is therefore of high priority [4]. Shared pathophysiological features exist between type 2 diabetes and Parkinson's disease [3, 5].
Glucagon‐like peptide‐1 (GLP‐1) receptor agonists are glucose‐lowering drugs commonly prescribed for type 2 diabetes. Growing evidence suggests that GLP‐1 receptor agonists modulate pathophysiological mechanisms in the brain of relevance for the development of Parkinson's disease, including neuroinflammation, mitochondrial dysfunction, amyloid aggregation and insulin signalling [6, 7, 8, 9].
In support of the hypothesis that GLP‐1 receptor agonists may provide benefits to patients with Parkinson's disease, a randomized controlled trial of 62 patients with Parkinson's disease showed that treatment with the GLP‐1 receptor agonist exenatide had a positive effect on motor function in comparison to placebo [10]. Moreover, a randomized controlled trial including 156 patients with early Parkinson's disease showed slower progression of motor symptoms in patients treated with the GLP‐1 receptor agonist lixisenatide compared to placebo [11].
Whether GLP‐1 receptor agonists reduce the risk of incident Parkinson's disease remains uncertain. No clinical trials have addressed this question. Observational studies report a lower risk among GLP‐1RA users, but limited outcome events and methodological limitations preclude firm conclusions [12, 13, 14].
We aimed to examine the association between use of GLP‐1 receptor agonists and the risk of incident Parkinson's disease.
Methods
Data Sources
We used nationwide health and administrative registers in Denmark, Norway and Sweden including population registers (vital status, demographics), patient registers (comorbidities, outcomes), prescribed drug registers (study drugs, co‐medications), Statistics Denmark and Statistics Sweden (socioeconomic variables) and the Swedish National Diabetes Register (glycated haemoglobin, blood pressure, albuminuria, estimated glomerular filtration rate [eGFR], body mass index and smoking). Data sources are described in detail in the. Supporting Information
Study Design and Study Population
We conducted an active‐comparator new‐user cohort study [15]. We included patients, aged 45 years or older, who initiated a GLP‐1 receptor agonist or the comparator, sulfonylureas, between January 1st, 2007, and December 31st, 2021, in Denmark and Sweden; and between January 1st, 2010, and December 31st, 2018, in Norway. New use was defined as no use of a GLP‐1 receptor agonist at any time before cohort entry or a sulfonylurea 1 year before cohort entry. This allowed patients to contribute separate treatment episodes, meaning they could enter the cohort first upon initiating a sulfonylurea and later upon initiating a GLP‐1 receptor agonist. The anatomic therapeutic chemical codes for the study drugs are provided in Table S1. The date of filling the first prescription constituted cohort entry.
We excluded patients who had a history of Parkinson's disease, secondary parkinsonism and Lewy body dementia at any time before cohort entry and patients who used any anti‐Parkinson drug within the last year before cohort entry. Further exclusion criteria were history of dialysis or renal transplantation, end stage illness including dementia, drug misuse, severe pancreatic disorders and use of liraglutide with obesity indication at any time before cohort entry, neither use of any prescription drug nor any specialist care contact in the previous year and hospital admission for any reason within 30 days before cohort entry (Table). S2
Outcome
The outcome was a first diagnosis of Parkinson's disease (ICD‐10 code: G20) or dementia in Parkinson's disease (ICD‐10 code: F02.3), registered as diagnoses in the patient registers during any type of hospital contact (hospitalization or outpatient visit; primary or secondary diagnosis).
Follow Up
Patients were followed from treatment initiation until the study outcome or until they were censored due to emigration, death, end of the study period, or a switch in therapy (initiation of a sulfonylurea among patients who entered the study on a GLP‐1 receptor agonist and vice versa).
Statistical Analyses
We used standardized mortality ratio (SMR) weighting based on propensity score to adjust for confounding, estimating the average treatment effect among the treated [16]. The probability of starting a GLP‐1 receptor agonist versus a sulfonylurea was estimated using a logistic regression model containing 54 variables, as registered at cohort entry. The variables comprised sociodemographic characteristics, diabetes complications, co‐morbidities, antidiabetic medications, non‐diabetes medications and measures of co‐morbidity burden, frailty and healthcare utilization (Table S3). The propensity score was estimated in each country separately and patients with propensity scores outside of the common range of the propensity score distribution were excluded. Analyses were performed in a pooled dataset from the three countries. To limit the influence of extreme propensity score weights, we applied weight truncation at the 1st and 99th percentile of the weight distribution [17]. The covariate balance after weighting was assessed with standardized difference; differences below 10% were considered as good balance.
A Cox proportional hazards regression model with time since start of treatment as the time scale was used to estimate hazard ratios (HRs) [16]. A 95% confidence interval (CIs) that did not overlap 1 was considered as a statistically significant difference. We described the cumulative incidence using Kaplan–Meier curves. The reported incidence rates are unadjusted. Absolute incidence rate differences were calculated using Poisson regression with an identity link function. Additional analyses were conducted to assess incidence rate differences at prespecified time intervals (3, 5 and 7 years after treatment initiation).
We conducted prespecified subgroup analyses by age group (45–69 and ≥ 70 years) and sex. A separate propensity score was estimated within each subgroup. Effect modification by subgroup status was examined by including an interaction term between treatment status and subgroup in the Cox model; in these analyses, p‐values of < 0.05 were considered statistically significant. We also conducted analyses by country.
We conducted several prespecified sensitivity analyses. First, as Parkinson's disease may develop slowly before becoming manifest and diagnosed [18, 19], we performed an analysis using a lag‐time period of 2 years between treatment initiation and the start of follow‐up. Second, we restricted the outcome definition to primary diagnoses in the patient registers. Third, to address the temporal treatment trends of increased GLP‐1 agonist use and decreased sulfonylurea use during the study period we conducted two complementary analyses: (1) a propensity score model incorporating calendar time as a covariate, and (2) a calendar time‐specific propensity score analysis in which we stratified the study period into biennial calendar intervals and estimated separate propensity scores within each time stratum. Fourth, since both GLP‐1 receptor agonists and dipeptidyl peptidase‐4 (DPP‐4) inhibitors target the incretin system and it has been suggested that DPP‐4 inhibitors may have a beneficial effect in Parkinson's disease [20], we performed an analysis in which patients with use of DPP‐4 inhibitors at any time before cohort entry were excluded and patients who initiated DPP‐4 inhibitors during follow‐up were censored. Fifth, to address potential residual confounding by unmeasured frailty, or other age‐related factors that may not be fully captured by the covariates included in the propensity score model, we performed a sensitivity analysis excluding participants aged > 80 years. Sixth, to evaluate the robustness of our primary asymmetric lookback approach, we conducted a sensitivity analysis applying a strict new‐user definition that required no prior exposure to either study drug class at any time before cohort entry. Within this cohort, we also performed an analysis employing an intention‐to‐treat exposure definition, where patients were not censored at treatment switch. Seventh, we performed an inverse probability of censoring weighting analysis to account for potential informative censoring. Eighth, to assess outcome validity, we conducted a sensitivity analysis in which the outcome was defined as a Parkinson's disease diagnosis recorded at two separate visits. Ninth, to account for the competing risk of death, a sensitivity analysis was performed using a Fine and Gray proportional subdistribution hazards model with death from any cause treated as a competing event. Finally, in the Swedish part of the cohort we expanded the propensity score to include additional variables including glycated haemoglobin, blood pressure, albuminuria, eGFR, body mass index and smoking. Given the proportion of missing values for the additional variables (Table S4), multiple imputation (fully conditional specification imputation) with 10 imputed datasets was used [21]. Imputation was based on all variables included in the propensity score, the additional variables and the outcome variable.
The study was approved by the Regional Ethics Committee in Stockholm, Sweden, and the Regional Committee for Medical and Health Research Ethics, Norway. In Denmark, approval by an ethics committee is not required for register‐based research.
Results
Study Population
Our analysis included 158 961 eligible patients initiating GLP‐1 receptor agonists and 188 065 initiating sulfonylureas, providing 559 979 and 1 156 732 patient‐years of follow‐up, respectively (Figure 1). Population characteristics before and after propensity score weighting are shown in Table 1; covariates in the two groups were well‐balanced after weighting. Mean age of the study population was 65 years and 43% were female. Median follow‐up time was 2.5 years (interquartile range 1.0–5.2) for GLP‐1 receptor agonist initiators and 5.8 years (2.9–9.0) for sulfonylurea initiators. The proportion of total follow‐up time by type of GLP‐1 receptor agonist was 72.9% for liraglutide, 13.4% for semaglutide, 7.3% for exenatide, 5.1% for dulaglutide, 1.3% for lixisenatide.
Flow chart of patient inclusion in the study cohort, Sweden, Denmark and Norway.
| Unweighted(%)n | Propensity score weighted % | |||||
|---|---|---|---|---|---|---|
| GLP‐1 receptor agonists (= 158 961)N | Sulfonylureas (= 188 065)N | Standardized difference (%) | GLP‐1 receptor agonists | Sulfonylureas | Standardized difference (%) | |
| Male | 90 992 (57) | 108 090 (57) | 0.5 | 57 | 57 | 1 |
| Age, mean (SD) in years | 63 (10) | 67 (11) | — | 63 (10) | 62 (9) | |
| Age group in years | ||||||
| 45–49 | 16 314 (10) | 14 114 (8) | 9.7 | 10 | 11 | 3.7 |
| 50–54 | 22 533 (14) | 18 802 (10) | 12.8 | 14 | 15 | 2 |
| 55–59 | 26 199 (16) | 23 481 (12) | 11.4 | 16 | 17 | 1.7 |
| 60–64 | 27 368 (17) | 28 901 (15) | 5 | 17 | 17 | 0.5 |
| 65–69 | 25 441 (16) | 30 088 (16) | 0 | 16 | 16 | 0.8 |
| 70–74 | 21 143 (13) | 26 422 (14) | 2.2 | 13 | 12 | 2.8 |
| 75–79 | 12 923 (8) | 20 555 (11) | 9.5 | 8 | 7 | 3 |
| 80–84 | 5103 (3) | 14 688 (8) | 20.3 | 3 | 3 | 1 |
| ≥ 85 | 1937 (1) | 11 014 (6) | 25.3 | 1 | 1 | 1 |
| Place of birth | ||||||
| Scandinavia | 136 653 (86) | 156 009 (83) | 8.3 | 86 | 85 | 3.4 |
| Rest of Europe | 9982 (6) | 13 266 (7) | 3.1 | 6 | 7 | 1.7 |
| Outside Europe | 12 326 (8) | 18 790 (10) | 7.9 | 8 | 9 | 2.9 |
| Civil status | ||||||
| Married/living with partner | 88 851 (56) | 102 683 (55) | 2.6 | 56 | 56 | 0.4 |
| Single | 69 856 (44) | 84 142 (45) | 1.6 | 44 | 44 | 0.3 |
| Education 70760 | ||||||
| Primary‐/secondary school· vocational training | 112 607 (71) | 126 598 (67) | 7.6 | 71 | 70 | 1.6 |
| Short tertiary education | 11 439 (7) | 9679 (5) | 8.5 | 7 | 7 | 0.4 |
| Medium or long tertiary education | 19 757 (12) | 15 116 (8) | 14.5 | 12 | 13 | 1.5 |
| Missing | 2726 (2) | 6625 (4) | 28.6 | 10 | 10 | 0.5 |
| Year of cohort entry 70760 | ||||||
| 2007–09 | 2411 (2) | 50 344 (27) | — | 2 | 22 | — |
| 2010–11 | 12 449 (8) | 40 765 (22) | — | 8 | 19 | — |
| 2012–13 | 12 272 (8) | 31 990 (17) | — | 8 | 15 | — |
| 2014–15 | 13 728 (9) | 26 867 (14) | — | 9 | 14 | — |
| 2016–17 | 21 217 (13) | 21 250 (11) | — | 13 | 13 | — |
| 2018–19 | 35 784 (23) | 10 470 (6) | — | 23 | 8 | — |
| 2020–21 | 61 100 (38) | 6379 (3) | — | 38 | 7 | — |
| Comorbidities | ||||||
| Ischemic heart disease | 24 967 (16) | 28 625 (15) | 1.3 | 16 | 15 | 3.1 |
| Heart failure/cardiomyopathy | 10 768 (7) | 12 757 (7) | 0 | 7 | 6 | 3.1 |
| Stroke/cerebrovascular disease | 7243 (5) | 10 861 (6) | 5.5 | 5 | 4 | 1.1 |
| Head trauma | 1755 (1) | 2422 (1) | 1.7 | 1 | 1 | 0 |
| Other neurologic disease | 26 707 (17) | 19 543 (10) | 18.8 | 17 | 16 | 1.8 |
| Arrythmia | 16 458 (10) | 20 222 (11) | 1.3 | 10 | 10 | 2.7 |
| Peripheral arterial disease (incl. amputation) | 8369 (5) | 8303 (4) | 4 | 5 | 5 | 3 |
| Kidney disease | 16 521 (10) | 12 245 (7) | 14 | 10 | 9 | 3.4 |
| Diabetes complications | 38 890 (24) | 29 244 (16) | 22.4 | 24 | 22 | 5.9 |
| COPD | 6414 (4) | 7911 (4) | 0.9 | 4 | 4 | 1 |
| Other lung disease | 10 618 (7) | 9500 (5) | 6.9 | 7 | 6 | 1.3 |
| Venous thromboembolism | 3593 (2) | 3576 (2) | 2.5 | 2 | 2 | 0.2 |
| Cancer (excl. non‐melanoma skin cancer) | 9959 (6) | 14 099 (7) | 4.9 | 6 | 6 | 1 |
| Melanoma | 667 (0) | 672 (0) | 1 | 0 | 0 | 0.1 |
| Liver disease | 3174 (2) | 2748 (1) | 4.1 | 2 | 2 | 0.5 |
| Thyroid disease | 2167 (1) | 1715 (1) | 4.3 | 1 | 1 | 0.5 |
| Osteoporosis | 3650 (2) | 5606 (3) | 4.3 | 2 | 2 | 0.7 |
| Fracture in the previous year | 3060 (2) | 4332 (2) | 2.6 | 2 | 2 | 0.2 |
| Alcohol‐related disorders | 2577 (2) | 3040 (2) | 0 | 2 | 2 | 0.6 |
| Health care utilization in previous year | ||||||
| Hospitalization due to neurological causes | 1422 (1) | 1619 (1) | 0.4 | 1 | 1 | 0.6 |
| Hospitalization due to other causes | 36 181 (23) | 43 832 (23) | 1.3 | 23 | 22 | 1.9 |
| Outpatient hospital contact due to neurological causes | 7143 (4) | 5240 (3) | 9.1 | 4 | 4 | 0.6 |
| Outpatient hospital contact due to other causes | 106 582 (67) | 107 202 (57) | 20.8 | 67 | 66 | 1.9 |
| Diabetes drugs in previous 6 months | ||||||
| Metformin | 120 226 (76) | 138 504 (74) | 4.6 | 76 | 78 | 5.2 |
| Any 2nd line anti‐diabetic | 106 776 (67) | 40 595 (22) | 103.3 | 67 | 68 | 1.1 |
| Prescription drug use in previous year | ||||||
| ACE‐inhibitors or ARB | 111 618 (70) | 111 557 (59) | 23 | 70 | 69 | 2.8 |
| Calcium‐channel blocker | 55 459 (35) | 54 042 (29) | 13.2 | 35 | 34 | 2.9 |
| Spironolactone | 10 264 (6) | 9594 (5) | 5.8 | 6 | 6 | 2.4 |
| Loop diuretic | 29 009 (18) | 32 910 (17) | 2 | 18 | 17 | 4 |
| Other diuretic | 30 979 (19) | 36 108 (19) | 0.7 | 19 | 18 | 2.9 |
| Beta‐blocker | 60 286 (38) | 70 660 (38) | 0.7 | 38 | 36 | 3.5 |
| Other cardiovascular drugs | 15 215 (10) | 23 663 (13) | 9.6 | 10 | 9 | 1.8 |
| Antiarrhythmic drugs | 1087 (1) | 1146 (1) | 0.9 | 1 | 1 | 0.5 |
| Platelet inhibitors | 54 265 (34) | 71 996 (38) | 8.6 | 34 | 32 | 3.6 |
| Anticoagulants | 16 681 (10) | 16 575 (9) | 5.7 | 10 | 10 | 2.5 |
| Lipid lowering drug | 115 663 (73) | 112 831 (60) | 27.3 | 73 | 71 | 3.4 |
| Lithium | 630 (0) | 706 (0) | 0.3 | 0 | 0 | 0 |
| Antidepressants | 30 102 (19) | 28 920 (15) | 9.5 | 19 | 19 | 0.4 |
| Typical antipsychotics | 2022 (1) | 3639 (2) | 5.3 | 1 | 1 | 0.1 |
| Atypical antipsychotics | 3962 (2) | 3906 (2) | 2.8 | 2 | 3 | 0.4 |
| Anxiolytic, hypnotic or sedative | 27 754 (17) | 40 346 (21) | 10.1 | 17 | 18 | 0.3 |
| Beta‐2 agonist inhalant | 17 948 (11) | 15 924 (8) | 9.5 | 11 | 11 | 0.4 |
| Anticholinergic inhalant | 5311 (3) | 6809 (4) | 1.5 | 3 | 3 | 0 |
| Glucocorticoid inhalant | 17 668 (11) | 17 681 (9) | 5.6 | 11 | 11 | 0.6 |
| Oral glucocorticoid | 13 613 (9) | 18 159 (10) | 3.8 | 9 | 9 | 0.1 |
| Opioid | 30 708 (19) | 34 589 (18) | 2.4 | 19 | 19 | 0.2 |
| Antiepileptic | 8356 (5) | 6232 (3) | 9.6 | 5 | 5 | 0.8 |
| Urate‐lowering drug | 9920 (6) | 10 117 (5) | 3.7 | 6 | 6 | 1.4 |
| No. of prescription drugs in last year | ||||||
| 0–5 | 17 792 (11) | 39 769 (21) | 27.3 | 11 | 12 | 2.4 |
| 6–10 | 61 643 (39) | 79 697 (42) | 7.3 | 39 | 40 | 2.4 |
| 11–15 | 47 799 (30) | 44 135 (23) | 15 | 30 | 29 | 1.6 |
| > 15 | 31 727 (20) | 24 464 (13) | 18.8 | 20 | 19 | 3.1 |
Primary Analysis
Figure 2 shows the adjusted cumulative incidence of Parkinson's disease. During follow‐up, 290 users of GLP‐1 receptor agonists and 927 users of sulfonylureas received a first diagnosis of Parkinson's disease. The incidence rate for incident Parkinson's disease was 5.2 per 10 000 person‐years with use of GLP‐1 receptor agonists and 8.0 per 10 000 person‐years with use of sulfonylureas. The unadjusted hazard ratio (HR) was 0.73 (95% CI 0.64–0.83) and the adjusted HR was 0.81 (95% CI 0.68–0.96). The absolute rate difference was −2.1 events per 10 000 person‐years (95% CI −3.2 to −1.0). Time‐specific incidence rate differences are presented in Table S5.
Adjusted cumulative incidence of incident Parkinson's disease among users of GLP‐1 receptor agonists compared with users of sulfonylureas.
Subgroups and Additional Analyses
Figureshows subgroup analyses. Hazard ratios in analyses by sex (women: adjusted HR 0.86 [95% CI 0.62–1.18] and men: adjusted HR 0.77 [95% CI 0.62–0.95]) and age group (45–69 years: adjusted HR 0.84 [95% CI 0.67–1.05] and ≥ 70 years: adjusted HR 0.89 [95% CI 0.67–1.16]) were similar to those of the primary analyses. In the additional analysis stratified by time since treatment initiation the adjusted HR was 0.70 for < 2 years, 0.84 for 2 to < 5 years and 0.85 for > 5 years (Table). Results by country are shown in Table. S1 S6 S7
Sensitivity Analyses
Table 2 shows sensitivity analyses. Findings remained consistent when using a lag‐time period of 2 years between treatment initiation and the start of follow‐up (adjusted HR 0.84 [95% CI 0.70–1.02]) and when restricting the outcome definition to primary diagnoses in the patient registers (adjusted HR 0.84 [95% CI 0.69–1.02]). Incorporation of calendar time into the propensity score yielded an adjusted HR of 0.82 (95% CI 0.68–0.98), while use of a calendar time‐specific propensity score model resulted in an adjusted HR of 0.84 (95% CI 0.66–1.06). In the analysis excluding and censoring patients at DPP‐4 inhibitor use, the adjusted HR was 0.74 (95% CI 0.60–0.93). Consistency persisted when we excluded patients older than 80 years (adjusted HR, 0.79 [95% CI 0.66–0.94]). When applying a strict new‐user definition, the adjusted HR was 0.85 (95% CI 0.70–1.04); within this cohort, an analysis employing an intention‐to‐treat exposure definition where patients were not censored at treatment switch resulted in an adjusted HR of 0.90 (95% CI 0.74–1.09). The analysis with inverse probability of censoring weighting produced an identical hazard ratio as the primary analysis, adjusted HR 0.81 (0.68–0.96). In the analysis requiring a Parkinson's disease diagnosis at two separate encounters, the adjusted HR was 0.76 (95% CI, 0.63–0.93). The competing risk analysis accounting for death produced a consistent adjusted subdistribution HR of 0.80 (95% CI 0.67–0.95). The total numbers of deaths during follow‐up for each treatment group are provided in Table S8. In the analysis with additional adjustment for glycated haemoglobin, blood pressure, albuminuria, eGFR, body mass index and smoking in the Swedish part of the cohort (patient characteristics are shown in Table S9), the point estimate for the HR was similar to the country‐specific analysis without such adjustment (Table S10).
| Analysis | GLP‐1 receptor agonists | Sulfonylureas | Adjusted hazard ratio (95% CI) 70760 | ||||
|---|---|---|---|---|---|---|---|
| No. of patients | Events | Incidence rate (events per 10 000 person‐years) | No. of patients | Events | Incidence rate (events per 10 000 person‐years) | ||
| Primary analysis | 158 961 | 290 | 5.2 | 188 065 | 927 | 8 | 0.81 (0.68–0.96) |
| Sensitivity analysis | |||||||
| 24 months lag‐time analysis | 89 673 | 215 | 6.7 | 154 878 | 746 | 9.2 | 0.84 (0.70–1.02) |
| Outcome definition restricted to primary diagnoses | 158 961 | 233 | 4.2 | 188 065 | 690 | 6 | 0.84 (0.69–1.02) |
| Additional adjustment for calendar time | 158 961 | 290 | 5.2 | 188 065 | 927 | 8 | 0.82 (0.68–0.98) |
| Calendar‐time specific propensity score | 158 524 | 290 | 5.2 | 181 030 | 890 | 7.9 | 0.84 (0.66–1.06) |
| Exclusion of patients with previous use of DPP‐4 inhibitors and censoring of patients initiating DPP‐4 inhibitors during follow‐up 70760 | 99 581 | 163 | 4.9 | 162 072 | 679 | 8.2 | 0.74 (0.60–0.93) |
| Exclusion of patients > 80 years | 151 925 | 277 | 5.1 | 162 285 | 841 | 8.1 | 0.79 (0.66–0.94) |
| Strict new‐user definition with unrestricted lookback period | 125 568 | 205 | 4.8 | 162 534 | 767 | 7.6 | 0.85 (0.70–1.04) |
| Strict new‐user definition with unrestricted lookback period and ITT exposure criteria | 125 568 | 205 | 4.8 | 162 534 | 812 | 7.6 | 0.90 (0.74–1.09) |
| Inverse probability of censoring weighting analysis | 158 961 | 290 | 5.2 | 188 065 | 927 | 8 | 0.81 (0.68–0.96) |
| Outcome definition limited to diagnoses recorded at two separate visits | 158 961 | 222 | 4 | 188 065 | 725 | 6.3 | 0.76 (0.63–0.93) |
Discussion
In this large cohort study using nationwide registers from three countries to include patients seen in routine clinical practice, use of GLP‐1 receptor agonists compared with use of sulfonylureas was associated with a 19% relative risk reduction of incident Parkinson's disease. Study findings were consistent across a range of predefined sensitivity and subgroup analyses.
Clinical trials have shown that GLP‐1 receptor agonists may be beneficial in patients with established Parkinson's disease. A randomized controlled trial evaluating exenatide versus placebo in patients with moderately severe Parkinson's disease demonstrated significantly decreased motor symptoms [10]. Moreover, in a randomized controlled trial including patients with early Parkinson's disease, treatment with lixisenatide resulted in slower progression of motor disability in comparison to placebo [11]. In contrast, in a randomized controlled trial including patients with early, untreated Parkinson's disease, treatment with a pegylated, brain‐penetrant version of exenatide (NLY01) did not affect motor symptom progression [22].
No clinical trials have investigated the effect of GLP‐1 receptor agonists on incident Parkinson's disease, and observational evidence remains limited. In a cohort study based on primary care data in the UK, use of GLP‐1 receptor agonists was associated with a reduced risk of incident Parkinson's disease (incidence rate ratio 0.38, 95% CI 0.17–0.60) in comparison to other non‐insulin glucose‐lowering drugs. However, the study was limited by the lack of a new‐user design, a heterogeneous comparator group and only 17 outcome events [12]. Two subsequent active‐comparator new‐user studies using dipeptidyl peptidase‐4 (DPP‐4) inhibitors as the reference reported directionally consistent findings: a US Medicare study observed a hazard ratio of 0.77 (95% CI 0.63–0.95) based on 143 events, and a Danish cohort study reported an HR of 0.57 (95% CI 0.37–0.85) in its primary on‐treatment analysis [13, 14]. However, the early separation of Kaplan–Meier curves in these studies raises questions regarding a neuroprotective mechanism and could potentially reflect protopathic bias or residual confounding. Observational studies indicate that DPP‐4 inhibitors may themselves be associated with a reduced risk of Parkinson's disease. Consequently, their use as an active comparator may have attenuated the estimated treatment effects [12, 23].
A strength of our study is the use of a robust study design restricting the cohort to new users and the use of a suitable active comparator. Sulfonylureas were selected as the active comparator because both drug classes were recommended as second‐ or third‐line glucose‐lowering agents during the study period, ensuring comparable disease stage and clinical indication at cohort entry. Sulfonylureas act through a mechanism confined to peripheral pancreatic insulin secretion and are regarded as pharmacologically neutral with respect to Parkinson's disease risk [24, 25]. DPP‐4 inhibitors and SGLT‐2 inhibitors were considered less appropriate comparators, as observational data have suggested protective associations with Parkinson's disease risk for both drug classes, which would attenuate any estimated effect of GLP‐1 receptor agonists toward the null [12, 20, 26]. Differential temporal trends in the use of the two drug classes during the study period were addressed through calendar‐time adjustment in sensitivity analyses, with no material change in the estimates (Table 2).
Although additional research is needed to determine whether the observed association reflects a true neuroprotective effect of GLP‐1 receptor agonists, our findings are consistent with the hypothesis that this drug class may reduce the risk of Parkinson's disease and support the rationale for further investigation. In perspective, while widespread use of GLP‐1 receptor agonists for the prevention of Parkinson's disease in the general population is unrealistic because of the relatively low background risk of the disease, future research might target the role of GLP‐1 receptor agonists in specific populations at particularly high risk of Parkinson's disease.
This study also had limitations. First, residual confounding from unmeasured variables remains a potential limitation despite adjustment for a broad range of baseline covariates. Data on HbA1c, BMI and other clinical parameters were unavailable for the full cohort; however, estimates remained consistent in the Swedish subcohort after additional adjustment for these variables, alongside blood pressure, albuminuria, eGFR and smoking status (Table S10). Second, Parkinson's disease has an insidious onset, and the shorter median follow‐up among GLP‐1 receptor agonist users (2.5 years) compared with sulfonylurea users (5.8 years) reflects the increasing uptake of GLP‐1 receptor agonists during the latter part of the study period. Both treatment groups contributed person‐time throughout the entire study period, and the Cox proportional hazards model accounted for time since treatment initiation. Moreover, one quarter of GLP‐1 receptor agonist users were followed for at least 5.2 years, providing sufficient follow‐up time for Parkinson's disease to emerge in a substantial proportion of the exposed cohort. Randomized trials have observed effects on motor symptoms within 12 months of treatment initiation, suggesting that a biologically meaningful signal would be detectable within the available follow‐up window [10, 11]. Third, although high sensitivity and positive predictive values have been observed for diagnoses recorded in Scandinavian health registers, there is a risk of outcome misclassification [27, 28]. A Swedish validation study of the ICD‐10 codes used for Parkinson's disease diagnosis in this study demonstrated a positive predictive value of 71% [29], which increased to 83% when restricting the outcome to primary diagnoses. Fourth, the exposure definition was based on filled prescriptions. Low adherence may have biased the findings toward the null. Fifth, the study was performed in Scandinavia and its generalizability to other populations and health care systems is unknown. Sixth, Protopathic bias may be a concern if prodromal symptoms of Parkinson's disease influenced treatment decisions, leading to differential avoidance of injectable GLP‐1 receptor agonists prior to diagnosis. However, randomized trials have indicated effects on motor symptoms within 12 months of treatment initiation, supporting the decision not to impose a lag‐time in the primary analysis. The sensitivity analysis imposing a 24‐month lag yielded a consistent point estimate (adjusted hazard ratio, 0.84; 95% CI, 0.70–1.02) and the cumulative incidence curves diverged gradually over time rather than separating early after treatment initiation, as would be expected under protopathic bias. Time‐stratified analyses showed some attenuation with longer follow‐up (less than 2 years: HR 0.70; 2 to less than 5 years: HR 0.84; 5 or more years: HR 0.85), though a directionally consistent association persisted across all strata (Table S6). The modest attenuation beyond 5 years may reflect the intention‐to‐treat nature of the analysis, as accumulating treatment discontinuation over time would be expected to attenuate estimates toward the null independent of any true neuroprotective effect. Seventh, patients who initiated liraglutide contributed with most of the follow‐up time for GLP‐1 receptor agonists. Therefore, the results of this study mainly apply to this specific drug.
Conclusion
In this large cohort study, use of GLP‐1 receptor agonists compared with use of sulfonylureas was associated with a lower risk of incident Parkinson's disease. These results support a potential neuroprotective role of GLP‐1 receptor agonists and warrant further investigation.
Author Contributions
Concept and design: Arvid Engström, Henrik Svanström, Peter Ueda and Björn Pasternak. Acquisition, analysis or interpretation of data: all authors. Drafting of the manuscript: Arvid Engström, Peter Ueda and Björn Pasternak. Critical revision of the manuscript for important intellectual content: all authors. Statistical analysis: Henrik Svanström. Obtained funding: Peter Ueda and Björn Pasternak. Study supervision: Peter Ueda and Björn Pasternak. Drs. Henrik Svanström and Björn Pasternak are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Funding
The study was supported by grants from the Swedish Research Council, Region Stockholm (ALF) and Dr. Margaretha Nilsson's Foundation for Medical Research. Dr. Björn Pasternak was supported by a consolidator investigator grant from Karolinska Institutet. Dr. Peter Ueda was supported by a grant from the Strategic Research Area Epidemiology programme and a Faculty Funded Career Position at Karolinska Institutet. Prof. Anders Hviid was supported by an investigator grant and holds a faculty recruitment grant from the Novo Nordisk Foundation; by an investigator grant from the Lundbeck Foundation; and by a research grant from the Independent Research Foundation Denmark. Dr. Björn Eliasson was supported by Konung Gustaf V:s och Drottning Victorias Frimurarestiftelse. Prof Mads Melbye was supported by a research grant from the Independent Research Foundation Denmark and a grant from the Danish Cancer Society.
Disclosure
The funding sources 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.
Conflicts of Interest
All authors have completed the ICMJE uniform disclosure form at www.icmje.org/coi_disclosure.pdf and have the following declarations. Dr. Christian Jonasson is an employee of NordicRWE. Dr. Björn Eliasson reports personal fees from Amgen, AstraZeneca, Boehringer Ingelheim, Eli Lilly, Merck Sharp & Dohme, Mundipharma, Navamedic, Novo Nordisk and Sanofi outside the submitted work. Dr. Henrik Svanström is a former employee of IQVIA. Prof. Anders Hviid is a scientific advisory board member of VAC4EU. The other authors declare no conflicts of interest.