Metformin versus DPP-4 inhibitors and risk of parkinsonism in type 2 diabetes: an active-comparator cohort study with a landmark design

BACKGROUND: Type 2 diabetes mellitus (T2DM) is a significant risk factor for Parkinson's disease (PD). While metformin shows neuroprotective potential in preclinical models, clinical evidence remains conflicting, likely due to limited follow-up, comparator heterogeneity, and insufficient consideration of differential survival. METHODS: We conducted a retrospective cohort study (observing data from 2005 to 2025) using the TriNetX global network. Using an incident user design and a 1-year landmark period to ensure sustained therapy and minimize immortal time bias, we matched 75,535 metformin initiators with 75,535 dipeptidyl peptidase-4 inhibitor (DPP-4i) users via propensity scores. The primary outcome was a composite of new-onset idiopathic PD and secondary parkinsonism. Exploratory time-stratified landmark analyses assessed whether the association differed among patients alive and outcome-free at 5 and 10 years. RESULTS: Following the landmark period, over a median follow-up of 5.2 years (IQR, 2.8-8.4; total 785,564 person-years), the overall risk of the composite neurologic outcome was similar between groups (aHR, 0.97; 95% CI, 0.87-1.09). Metformin users had lower observed all-cause mortality than DPP-4 inhibitor users (aHR, 0.75; 95% CI, 0.73-0.77). In a 5-year landmark sensitivity analysis, no association was observed (aHR, 0.98; 95% CI, 0.88-1.09). In exploratory analyses, a lower observed risk was seen among individuals surviving beyond 10 years (aHR, 0.85; 95% CI, 0.75-0.97; p = 0.0152), but this analysis was conditioned on long-term survival and should be interpreted as exploratory and hypothesis-generating. CONCLUSIONS: The primary analysis showed no overall difference in parkinsonism risk. Metformin users had lower observed all-cause mortality, which may reflect residual confounding or treatment-selection mechanisms. The lower observed parkinsonism risk among patients surviving beyond 10 years was exploratory and hypothesis-generating, and may be affected by survivor or collider bias.