In a recent cohort study published in JAMA Network Open, researchers evaluated whether metformin treatment is associated with a reduced risk of all-cause dementia in people with type 2 diabetes (T2D).
The researchers examined whether mechanisms other than improved glucose control, such as hemoglobin A1c (HbA1c) levels, mediate this association.
Metformin (dimethylbiguanide) has been the preferred first-line treatment for T2D since 2006. Studies have suggested that initiating metformin treatment may reduce dementia risk in those with type 2 diabetes.
However, patients frequently terminate metformin treatment for reasons such as gastrointestinal adverse effects and kidney dysfunction. In fact, one-fifth of early users use alternative antidiabetic drugs instead of metformin.
According to current recommendations, people should carefully weigh the benefits versus risks of metformin therapy when the estimated glomerular filtration rate (eGFR) plummets below a certain threshold.
About the study
In the present study, researchers investigated whether cessation of metformin treatment for reasons unrelated to kidney dysfunction (eGFR rates falling below the safe threshold) was associated with increased dementia incidence.
They examined electronic health records (EHRs) of a cohort of metformin users enrolled in Kaiser Permanente Northern California (KPNC), one of the largest integrated healthcare delivery systems in the United States of America (USA), with over 4.6 million members.
These individuals were born before January 1, 1955 (older adults), used metformin, were not suffering from dementia, and completed one of the two health surveys before enrolment into this study.
The follow-up for dementia incidence began with the implementation of their EHRs in 1996 and continued till 2020. They censored participants at age 90, death, dementia diagnosis, or at the start of a 90-day membership gap.
They used the KPNC mortality database to obtain death dates, while data on race and ethnicity were self-reported. Participants self-endorsed as Asians, Blacks, Hispanic/Latino, White, and other or uncertain. KPNC health plan databases confirmed the race and ethnicity of those individuals who did not endorse a category.
This cohort study used an emulated trial design, which replicated critical features of an RCT. Accordingly, the team estimated associations analogous to intent-to-treat (ITT) estimates while comparing early metformin terminators with the routine metformin users’ group.
The team matched early terminators and routine metformin users in a 1:4 ratio. They had the same age, gender, and diabetes for the same duration.
All calculated P values were two-sided, and the statistical threshold of significance for proportional hazard assumptions was set at α = .05. The data was analyzed between November 2021 and September 2023.
Causal mediation analysis was performed to determine whether changes in HbA1c levels or insulin prescription status mediated the relationships between early metformin termination and all-cause dementia.
Changes in mean HbA1c levels were measured during the 8-16 months post metformin termination and at the same age among matched routine users. For additional analyses, they used changes in insulin prescription status and HbA1c levels measured five years after early metformin termination as mediators.
The researchers conducted mediation analyses using accelerated failure time (AFT) models, which provided a more straightforward interpretation of the observed associations.
Further, they performed Cox proportional regression modeling to determine hazard ratios (HRs) for dementia diagnosis using time since metformin termination as the time scale, stratified by age at metformin treatment initiation and gender.
The team also conducted sensitivity analyses using creatinine instead of eGFR, with gender-specific cutoffs. They also carried out data analysis restricted to participants with medication adherence of over 80%.
In further sensitivity analyses, the researchers restricted differences between early and routine users, wherein they discarded differences in HbA1c level of more than 0.1% and diabetes duration of over one year.
They also analyzed data from those early terminators who had initiated metformin within the past two years; thus, they showed no diabetes progression.
Results and conclusions
In total, 12,220 early terminators (46.2% women) and 29,126 (46.6% women) routine users, with a mean age of 59.4 and 61.1 years at the start of the first metformin prescription, respectively, were analyzed.
Early terminators were at 1.21 times higher risk of dementia diagnosis than matched routine users.
Mediation analysis revealed that changes in HbA1c level or insulin use contributed to the association between early metformin termination and all-cause dementia, with no contribution for insulin use five years after metformin termination to 0.07 years for HbA1c level at one year after termination, suggesting they minimally mediated the association. The results of sensitivity analyses were similar to that of the primary analysis.
Overall, this cohort study of metformin in older adults largely corroborated prior observational research that initiating metformin was associated with a reduced risk of dementia.
Its large sample size and long follow-up period helped the researchers make precise estimates of the association between early metformin termination and all-cause dementia. Moreover, the study design mitigated the potential for cohort effects and confounding by indication.
This study has important implications for the clinical management of diabetes, especially in older adults.
For individuals with diabetes at a higher risk of dementia, such as carriers of apolipoprotein E (APOE) ε4 allele, mitigating adverse effects rather than terminating metformin use or using alternative drugs might be more effective.
These patients might consider switching to other metformin formulations (e.g., slower-release metformin) to overcome gastrointestinal adverse effects.
Future studies could evaluate the heterogeneity of the observed association of metformin across known risk factors for dementia. Studies could also extrapolate the study findings to prediabetic groups.