Comparative Effectiveness of Metformin Versus Sulfonylureas on Exceptional Longevity in Women With Type 2 Diabetes: Target Trial Emulation

Abstract

Background

The association of metformin with mortality has been mixed, and no prior study has determined whether metformin initiation is associated with exceptional longevity, defined as survival to ages 90 and older.

Methods

We performed a new-user, active comparator cohort study using the target trial emulation framework among the Women’s Health Initiative cohort to determine whether metformin versus sulfonylurea initiation was associated with exceptional longevity (survival to age 90). We identified participants ≥60 years with incident type 2 diabetes and no history of hypoglycemic agents or insulin prior to treatment initiation to perform intention-to-treat analyses. We used 1:1 propensity score matching on demographic characteristics, lifestyle behaviors, diabetes duration, comorbidities (hypertension, cardiovascular disease, chronic obstructive pulmonary disease, and cancer), body mass index, and concomitant medications to balance treatment groups on key confounders.

Results

Among 438 propensity score-matched women with type 2 diabetes, the incidence rate of death before age 90 per 100 person-years in women initiating metformin monotherapy was 3.7 (95% CI: 3.1–4.4) compared with 5.0 (95% CI: 4.2–5.8) for sulfonylurea monotherapy. The adjusted risk of death before age 90 was 30% lower for initiation of metformin monotherapy versus sulfonylurea monotherapy (hazard ratio, 0.70; 95% CI: 0.56–0.88).

Conclusions

In this first target trial emulation of metformin and exceptional longevity, we found that metformin initiation increased exceptional longevity compared with sulfonylurea initiation among women with type 2 diabetes. Because this comparison was not made to placebo in a randomized controlled trial and given the observational design with potential for residual confounding, causality cannot be inferred.

There is increasing interest in identifying gerotherapeutics, medications that target pathways of aging to extend human longevity (1,2). Metformin, a first-line diabetes medication, has been proposed as a potential gerotherapeutic, as it targets several mechanisms of aging (1). Metformin decreases insulin levels and insulin-like growth factor 1 signaling, inhibits mammalian target of rapamycin, reduces generation of reactive oxygen species, activates AMP-activated kinase, and reduces DNA damage (1). Metformin also activates FOXO3, a key gene involved in the insulin/insulin-like growth factor-1 receptor pathway that is implicated in longevity (3), and has been shown to positively affect other mechanisms that contribute to age-related diseases, including inflammation, autophagy, and cellular senescence (1). The association between metformin and lifespan in mice has been mixed. Some studies found that metformin increased mean lifespan in mice by 14% when initiated early in life, delayed onset of carcinoma and extended lifespan by 8% in a breast cancer model, and improved cognitive functioning and other healthspan indices (1). However, The National Institute on Aging Interventions Testing Program found that metformin alone did not significantly extend lifespan in mice but when combined with rapamycin, metformin was associated with a significant increase in lifespan (4).

Evidence on the associations of metformin with age-related mortality and morbidity has been mixed. In the UKPDS trial of overweight adults with newly diagnosed diabetes, metformin lowered the risk of all-cause mortality relative to conventional control (diet alone) or intensive control with chlorpropamide, glibenclamide, or insulin; there was a nonsignificant 5% reduction in diabetes-related death for the combination of metformin and sulfonylurea relative to all other treatments (5). A systematic review and meta-analysis observed that people with diabetes taking metformin had lower all-cause mortality compared to those without diabetes or those with diabetes taking insulin, other diabetes medications, or sulfonylurea (6). On the other hand, the Diabetes Prevention Program and Diabetes Prevention Program Outcomes Study observed no difference in all-cause or cause-specific mortality in metformin versus placebo during a median follow-up of 21 years (7). A few studies found that metformin users had lower risk of age-related diseases, including cancer, cardiovascular disease, and dementia, compared to those without diabetes or those with diabetes receiving nonmetformin therapies (6,8–11). Yet, in a Danish study among case–control pairs matched on birth year and sex or familial factors, there was an increase in mortality for initiation of metformin monotherapy among individuals with type 2 diabetes relative to those without diabetes diagnosis (12). While some observational studies observed reduced incidence of cancer with metformin use, randomized trials of metformin for the treatment of type 2 diabetes and as adjuvant therapy for the treatment of different cancers did not find any benefit in cancer incidence or outcomes, suggesting that findings from observational studies may be explained by time-related biases (eg, immortal time bias) (13).

As summarized above, prior studies were largely focused on examining the association of metformin with mortality or disease, and no prior observational study or randomized controlled trial (RCT) has determined whether metformin increases exceptional longevity (ie, survival to advanced ages such as 90 and older). Few observational cohorts have followed individuals from midlife into ages 90 and older to be able to examine this relationship. Moreover, performing an RCT of exceptional longevity is not feasible, given the decades of follow-up that would be required. To address this gap in knowledge, we performed a target trial emulation study to determine the association of metformin with exceptional longevity. Target trial emulation seeks to quantify the effect of a treatment on disease risk by emulating a hypothetical RCT using observational data and causal inference (14,15). Target trial emulation helps avoid several biases in pharmacoepidemiologic studies, including selection bias and immortal time bias, and provides a good approximation of results observed in an actual RCT (14–18).

We leveraged a large, well-characterized, national cohort study with over 30 years of follow-up, the Women’s Health Initiative (WHI), to perform the first target trial emulation study of metformin and exceptional longevity. The extensive follow-up of women from midlife into ages 90 and older in the WHI provided the unique opportunity to mimic an RCT of exceptional longevity to test the potential of metformin as a longevity-enhancing drug.

Method

Target Trial Emulation

We emulated a hypothetical target trial using a new-user, active comparator cohort design to compare women with type 2 diabetes initiating metformin monotherapy versus sulfonylurea monotherapy. Metformin is currently the main first-line treatment for type 2 diabetes, with sulfonylureas a popular second choice (19); thus, sulfonylureas served as a useful comparator group to minimize confounding by indication, similar to recent comparative effectiveness studies (20–22).

We performed target trial emulation in 2 steps. In the first step, we specified the study protocol of a hypothetical pragmatic unblinded target trial that aimed to determine the effect of initiation of metformin monotherapy versus sulfonylurea monotherapy on exceptional longevity among individuals with type 2 diabetes. Table 1 shows the key design characteristics of this trial. The population of the target trial included adults >60 years with type 2 diabetes who were randomized to receive metformin monotherapy (treatment arm) or sulfonylurea monotherapy (control arm). The main outcome was death before age 90. In the second step, we used data from the WHI to emulate the target trial. The characteristics of the emulated trial are summarized in Table 1. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines.

Table 1.

Specification and Emulation of a Target Trial Evaluating Initiation of Metformin Monotherapy Versus Sulfonylurea Monotherapy on Exceptional Longevity

Target Trial Specification	Emulation Using WHI Data
Eligibility criteria
>60 years old	Same
No hypoglycemics or insulin	No recorded prior exposure to hypoglycemic agents or insulin
No chronic kidney disease (contraindication for metformin)	No self-reported kidney dialysis or a kidney machine for kidney or renal failure
Treatment strategies
Treatment arm: metformin monotherapy Control arm: sulfonylurea monotherapy	Initiation of metformin monotherapy or sulfonylurea monotherapy between baseline and year 3 visit
Treatment assignment
Double-blind, randomized treatment assignment	Emulate randomization by balancing baseline confounders with 1:1 propensity score matching
Outcome
Death prior to age 90 (primary outcome)	Same
Follow-up
From baseline to death prior to age 90	Same
Causal contrast
Intention-to-treat effect	Observational analog of intention-to-treat effect
Statistical analysis
Intention-to-treat analysis of primary outcome using Cox proportional hazards regression	Same

Note: WHI = Women’s Health Initiative.

Data Source

The WHI is a 31-year prospective study investigating major determinants of chronic diseases among postmenopausal women (23). The WHI recruited 161 808 postmenopausal women ages 50–79 years from 1993 to 1998 across 40 U.S. clinical centers into an observational study (OS; N = 93 676) or 1 or more of 3 clinical trials (CT; N = 68 132). Women were followed for mortality and other health outcomes through February 17, 2024. The present study was approved by the Institutional Review Board of the Fred Hutchinson Cancer Research Center. Each participant provided written informed consent.

Eligibility Criteria and Treatment Strategies

At the baseline and year 3 clinic visits, women were asked about current prescription and over-the-counter medication use in the past 2 weeks. Information on medication use was confirmed from medication containers that women brought to the clinic visit and entered into the WHI database. The brand or generic name on the medication label was matched to the corresponding item in the Master Drug Database (Medi-Span, Indianapolis, IN). Participants also provided the duration (in days) of their medication use. We used this information to determine the date of medication initiation, which was designated as the index date for analysis. Information on medication dose was not collected.

Treatment initiation was defined as new use of metformin or sulfonylurea (glipizide, glyburide, gliclazide, or glimepiride) among women newly diagnosed with type 2 diabetes between baseline and the year 3 visit. Incident diabetes was defined as a new self-report of physician diagnosis of diabetes treated with oral hypoglycemic agents or insulin, which was previously validated for its reliability and accuracy in WHI (24). Women with a history of type 2 diabetes at baseline and those who reported prior exposure to insulin or glucose-lowering medications (metformin, sulfonylureas, meglitinides, thiazolidinediones, alpha-glucosidase inhibitors, DPP-4 inhibitors, SGLT2 inhibitors, cycloset) prior to the year 3 visit were excluded. Our analysis focused on women initiating metformin monotherapy or sulfonylurea monotherapy; thus, women reporting concomitant use of insulin or other glucose-lowering medications at the year 3 visit were excluded. Additional eligibility criteria included having the opportunity, due to birth year, to survive to ages 90 and older during follow-up ending February 17, 2024 (ie, > 60 years old at medication initiation), and no history of chronic kidney disease (CKD), which is a contraindication for metformin (25); no participant reported history of CKD.

Treatment Assignment

To emulate randomization in the target trial, we used propensity score (PS) matching methods to adjust for any imbalance in participant characteristics at the index date. To achieve this, we used logistic regression to model the probability of initiating metformin according to the following characteristics collected prior to initiation of treatment: observational study or clinical trial participation, age at metformin or sulfonylurea initiation, race, ethnicity, education, dietary quality, smoking status, comorbidities (hypertension, coronary heart disease, stroke, congestive heart failure, chronic obstructive pulmonary disease, and cancer), body mass index (BMI), total physical activity (summarized into metabolic-equivalent hours per week), and concomitant medications for other indications, including angiotensin-converting enzyme inhibitors, angiotensin II receptor blockers, beta-blockers, calcium channel blockers, histamine type 2 receptor agonists, nonsteroidal anti-inflammatory drugs, statins, proton pump inhibitors, and diuretics (loop diuretics, thiazide diuretics, and potassium-sparing diuretics). Metformin was launched in the United States in 1995, or a minimum of 1 year prior to the year 3 visit during 1996–2001; its use in the United States doubled during this period (26). Accordingly, to control for the possibility of confounding by trends in diabetes prescription drug use, we additionally matched participants on diabetes duration, defined as the time from type 2 diabetes diagnosis (between WHI baseline and the year 3 visit) to treatment initiation.

Weight and height were measured at the clinic using standardized protocols, with BMI calculated as weight in kilograms divided by height in meters squared. Dietary quality was assessed using a validated, semiquantitative food frequency questionnaire and summarized using the Healthy Eating Index 2015 (HEI-2015), which measures conformance to recommendations from the 2015–2020 Dietary Guidelines for Americans (27). Comorbidities were assessed with annual questionnaires. Incident coronary heart disease, stroke, congestive heart failure, and cancer, which are main outcomes of interest in the WHI trials, were adjudicated through physician review of medical records (28). Other baseline characteristics were collected through questionnaires administered during the baseline period prior to the index date. Information on concomitant medications was collected as described above.

Outcomes

Follow-up time was calculated from the date of treatment initiation to occurrence of death before age 90, date of turning age 90 for those who survived to at least this age, last contact, or February 17, 2024, whichever came first. Deaths were verified by trained physician adjudicators using hospital records, autopsy or coroner’s reports, or death certificates. Periodic linkage to the National Death Index was performed for all participants, including those lost to follow-up, for verification if death certificates or medical records were not available.

Statistical Analysis

Participants were matched 1:1 using greedy nearest neighbor matching with a caliper of 0.25 times the standard deviation (SD) (29). A caliper of 0.25 was selected as previously recommended to achieve covariate balance across treatment groups while allowing us to identify adequate numbers of matched pairs for analysis (30). We examined baseline characteristics prior to and after matching. Covariate imbalance was assessed using the absolute standardized mean difference (SMD), with an absolute SMD > 0.1 indicating cohort imbalance (31,32). Our analysis focused on participants without missing data on the exposures, covariates, and study outcomes.

The intention-to-treat (ITT) effect of initiation of metformin monotherapy versus sulfonylurea monotherapy on death prior to age 90 was evaluated using Cox proportional hazards regression models with a robust variance estimator to account for clustering within matched pairs. Hazard ratios (HRs) and their corresponding 95% confidence intervals (CIs) were estimated. The ITT analysis compared the risk of the outcome for metformin versus sulfonylurea initiators according to treatment assignment at baseline, irrespective of medication adherence, medication switching, or adding on medications. The proportional hazards assumption was assessed using Kolmogorov-type supremum tests using cumulative sums of martingale residuals; no violations of this assumption were noted. Cumulative incidence curves were calculated using Kaplan–Meier survival analysis. We used Poisson regression to determine incidence rates per 100 person-years (with 95% CIs) of death before age 90 by treatment group.

We performed secondary analyses to assess the robustness of our findings. First, we excluded women who died within the first two years of follow-up to determine whether a lag period after treatment initiation affected the findings. To explore within class differences among sulfonylurea users, metformin use was compared to glyburide use and glipizide use in separate analyses with new PS-matched samples. We included income and region of the country in which the participant resided as additional confounders for PS matching, given that these factors may be associated with longevity; a new PS-matched sample was created after inclusion of these factors. Finally, we determined E-values for HRs. E-values represent the minimum strength of association that unmeasured confounders would need to have with both the exposure and the outcome to explain away an observed association (33,34). A large E-value suggests that there would need to be a strong association between unmeasured confounding and exposure and outcome to fully explain the association, whereas a small E-value suggests that weak unmeasured confounding would be enough to explain the observed association (34). In other words, a small E-value suggests that the findings could be more likely explained by residual confounding than causality relative to a higher E-value.

Analyses were performed using SAS OnDemand for Academics (SAS Institute Inc., Cary, NC) and R. Statistical significance was defined as a 95% CI excluding 1 for the HRs.

Results

We identified 726 metformin monotherapy initiators and 567 sulfonylurea monotherapy initiators who met inclusion and exclusion criteria (Supplementary Figure 1). After 1:1 PS matching, 219 metformin initiators were matched to 219 sulfonylurea initiators. Supplementary Table 1 and Table 2 show the distribution of baseline characteristics prior to and after matching. Prior to matching, metformin and sulfonylurea initiators differed in a few characteristics as indicated by absolute SMD values >0.1 (Supplementary Table 1). For example, the metformin group was younger at treatment initiation, more likely to be college graduates, more likely to use statins, and less likely to use diuretics. After 1:1 PS matching, baseline characteristics were well balanced across metformin and sulfonylurea groups, with all absolute SMD <0.1 (Supplementary Table 1, Table 2). Participants in the final sample had a mean age at treatment initiation of 70.0 (SD 4.3) years. Overall, 2.3% were Asian, 11.9% Black, 0.5% more than one race, 84.9% White, 0.5% unknown/not reported race, and 1.4% Hispanic/Latino.

Table 2.

Baseline Characteristics of Metformin Monotherapy Initiators Versus Sulfonylurea Monotherapy Initiators After Propensity-score Matching Among Women With Type 2 Diabetes

Characteristic	Metformin (n = 219) N (%)	Sulfonylurea (n = 219) N (%)	SMD
Age at medication initiation, years, mean (SD)	69.8 (4.1)	70.1 (4.5)	−0.07
Diabetes duration, years, mean (SD)	0.98 (0.84)	1.0 (0.8)	−0.05
WHI component Clinical trial Observational study	102 (46.6) 117 (53.4)	107 (48.9) 112 (51.1)	0.05 −0.05
Race Asian Black More than one race White Unknown/not reported	5 (2.3) 26 (11.9) 1 (0.5) 186 (84.9) 1 (0.5)	5 (2.3) 26 (11.9) 1 (0.5) 186 (84.9) 1 (0.5)	0 0 0 0 0
Ethnicity Hispanic/Latino Not Hispanic/Latino	3 (1.4) 216 (98.6)	3 (1.4) 216 (98.6)	0 0
Education Less than high school High school Some college College graduate	14 (6.4) 52 (23.7) 82 (37.4) 71 (32.4)	16 (7.3) 57 (26.0) 77 (35.2) 69 (31.5)	0.04 0.05 −0.05 −0.02
Body mass index, kg/m2, mean (SD)	31.9 (5.6)	31.9 (5.7)	−0.003
Smoking behavior Never smoked Past smoker Current smoker	117 (53.4) 91 (41.6) 11 (5.0)	115 (52.5) 93 (42.5) 11 (5.0)	−0.02 0.02 0
Total physical activity, MET-hours/week, mean (SD)	9.5 (12.9)	8.8 (10.7)	0.06
Healthy Eating Index-2015 score, mean (SD)	64.7 (10.3)	64.5 (9.4)	0.02
Coronary heart disease Yes No	12 (5.5) 207 (94.5)	12 (5.5) 207 (94.5)	0 0
Stroke Yes No	4 (1.8) 215 (98.2)	5 (2.3) 214 (97.7)	0.03 −0.03
Heart failure Yes No	9 (4.1) 210 (95.9)	7 (3.2) 212 (96.8)	−0.05 0.05
Cancer Yes No	22 (10.1) 197 (90.0)	23 (10.5) 196 (89.5)	0.02 −0.02
Chronic obstructive pulmonary disease Yes No	9 (4.1) 210 (95.9)	12 (5.5) 207 (94.5)	0.06 −0.06
Hypertension Yes No	166 (75.8) 53 (24.2)	166 (75.8) 53 (24.2)	0 0
ACE inhibitor use Yes No	63 (28.8) 156 (71.2)	59 (26.9) 160 (73.1)	−0.04 0.04
Angiotensin II receptor blocker use Yes No	24 (11.0) 195 (89.0)	26 (11.9) 193 (88.1)	0.03 −0.03
Beta-blocker use Yes No	50 (22.8) 169 (77.2)	51 (23.3) 168 (76.7)	0.01 −0.01
Calcium channel blocker use Yes No	51 (23.3) 168 (76.7)	51 (23.3) 168 (76.7)	0 0
Histamine type 2 receptor agonist use Yes No	12 (5.5) 207 (94.5)	12 (5.5) 207 (94.5)	0 0
Nonsteroidal anti-inflammatory drug use Yes No	45 (20.6) 174 (79.5)	53 (24.2) 166 (75.8)	0.09 −0.09
Statin use Yes No	78 (35.6) 141 (64.4)	79 (36.1) 140 (63.9)	0.01 −0.01
Proton pump inhibitor use Yes No	22 (10.1) 197 (90.0)	24 (11.0) 195 (89.0)	0.03 −0.03
Any diuretic use Yes No	44 (20.1) 175 (79.9)	46 (21.0) 173 (79.0)	0.02 −0.02

Notes: ACE = angiotensin-converting enzyme; SD = Standard deviation; SMD = standardized mean difference.

Primary Analyses

The cumulative incidence curve comparing metformin with sulfonylurea is shown in Figure 1. In the 1:1 PS-matched cohort, the incidence rate of death before age 90 (per 100 person-years) in the metformin group was 3.7 (95% CI: 3.1–4.4) compared with 5.0 (95% CI: 4.2–5.8) in the sulfonylurea group, during a mean (SD) follow-up of 15.5 (5.8) and 14.3 (6.0) years, respectively. This finding corresponded to an adjusted HR of 0.70 (95% CI: 0.56–0.88; Figure 2). The E-value for this association was moderately strong (E-value: 1.88). The primary causes of death in the metformin group (n = 125 deaths) were CVD (40.0%) and non-CVD (60.0% including 16.8% cancer and 11.2% dementia). The primary causes of death in the sulfonylurea group (n = 156 deaths) were CVD (39.1%) and non-CVD (60.9% including 14.1% cancer and 14.1% dementia).

Figure 1.

Cumulative incidence of death before age 90 between propensityscore-matched participants with type 2 diabetes initiating metformin monotherapy or sulfonylurea monotherapy. Participants were matched on sociodemographic characteristics, medications, lifestyle behaviors, and comorbidities. Outcomes were followed from treatment initiation to the occurrence of death prior to age 90, date of turning age 90, last contact, or February 17, 2024, whichever came first.

Figure 2.

Comparison of death before age 90 between propensityscore-matched metformin monotherapy initiators versus sulfonylurea monotherapy initiators among postmenopausal women with type 2 diabetes. Participants were matched on sociodemographic characteristics, medications, lifestyle behaviors, and comorbidities. Outcomes were followed from treatment initiation to the occurrence of death prior to age 90, date of turning age 90, last contact, or February 17, 2024, whichever came first. CI = confidence interval; HR = hazard ratio; IR = incidence rate; PY = person-years.

Secondary Analyses

After removing those who died within the first 2 years of follow-up, metformin remained associated with lower risk of death before age 90 relative to sulfonylurea (HR, 0.72; 95% CI: 0.57–0.91; E-value: 1.82). When examining individual sulfonylureas, metformin was associated with lower risk of death before age 90 relative to glipizide (HR, 0.63; 95% CI: 0.46–0.87), whereas metformin was not significantly associated with risk of death before age 90 when compared with glyburide (HR, 0.78; 95% CI: 0.58–1.06), with moderately strong E-values of 2.10 and 1.66, respectively (Figure 2). When including income and region of residence as additional confounders in a new PS-matched sample (N = 408; Supplementary Table 2), the HR for metformin versus sulfonylurea was 0.81 (95% CI: 0.64–1.03; Supplementary Figure 2). Further, the HR for metformin versus glipizide was 0.57 (95% CI: 0.41–0.79), and the HR for metformin versus glyburide was 0.79 (95% CI: 0.59–1.05).

Discussion

Metformin has been shown to target multiple pathways of aging and therefore has been postulated as a drug that may extend human longevity (1). However, no prior observational study or RCT has examined the link between metformin and exceptional longevity. Using data from the WHI, this target trial emulation study found that initiation of metformin monotherapy versus sulfonylurea monotherapy was associated with increased exceptional longevity among women 60 years and older with type 2 diabetes.

A key advantage of our analysis was the long follow-up period after treatment initiation enabled by examination of a cohort with extensive follow-up from midlife to ages 90 and older, which is not feasible in typical RCTs. Our use of the target trial framework allowed us to mimic an RCT that would be performed in the real-world setting to determine the effect of metformin on exceptional longevity. We mimicked randomization through PS matching to achieve balance on key characteristics and performed ITT analyses. Importantly, our active comparator, new user cohort design is stronger than designs that compare treatment initiators to noninitiators, as it minimized confounding by indication (35).

A systematic review and meta-analysis of observational studies found that individuals with diabetes taking metformin had lower risk of all-cause mortality relative to those without diabetes (HR, 0.93; 95% CI: 0.88–0.99) or taking nonmetformin therapies (HR, 0.72; 95% CI: 0.65–0.80), insulin (HR, 0.68; 95% CI: 0.63–0.75), or sulfonylurea (HR, 0.80; 95% CI: 0.66–0.97) (6). In the only other target trial emulation study of metformin and lifespan, initiation of metformin monotherapy relative to sulfonylurea monotherapy was associated with reduced risk of all-cause mortality, with an HR of 0.57 (95% CI: 0.48–0.67) in ITT analyses among men and women (22). However, some randomized trials have shown that metformin does not reduce risk of mortality or age-related diseases such as cancer (7,13). For example, in the Diabetes Prevention Program, there was no mortality difference in metformin versus placebo during a median 21-year follow-up (7). These prior studies, however, did not examine exceptional longevity. Meta-analyses have yielded mixed findings on the association of sulfonylureas with all-cause and cardiovascular mortality, with a meta-analysis of RCTs showing that sulfonylureas do not increase mortality risk in individuals with type 2 diabetes (36–38). Because we were not able to compare metformin to placebo, our findings do not suggest that metformin promotes longevity but rather that it may be better for long survival relative to sulfonylureas. Future RCTs examining metformin versus placebo in relation to survival to ages 90 and older are needed.

Biological aging is characterized by a progressive loss in physiological functions and tissues, thereby increasing the risk of chronic diseases and geriatric syndromes (39). The geroscience hypothesis posits that biological aging is malleable and that slowing biological aging may delay or prevent the onset of multiple age-related diseases and disability (40). A key goal of geroscience is to identify novel therapeutic and preventive interventions that slow biological aging (41). With the aging of the population, identifying existing medications that can be repurposed to target biological aging has profound implications for public health. Our study contributes to the discussion on the development of future RCTs to test the impact of metformin on healthspan indices in populations without diabetes. The Targeting Aging with MEtformin (TAME) trial aims to determine whether metformin delays mortality as well as the onset of age-related chronic diseases and conditions including CVD, cancer, dementia, mobility decline, and cognitive decline (41). The TAME trial aims to recruit individuals 65–80 years and follow them for six years; thus, it will not be able to determine whether metformin increases exceptional longevity, underscoring the importance of future target trial emulation studies in large databases to replicate our findings. The TAME trial has not been launched to date due to need for funding.

Strengths of this study include the target trial emulation design and causal inference methodology applied to a well-characterized, national cohort study with decades of follow-up and comprehensive data collection. Unlike prior studies focused on all-cause mortality, our specific focus on death prior to age 90 as the primary outcome allowed us to determine the association between metformin and exceptional longevity. The rich WHI dataset enabled us to emulate randomization using PS methods by taking into consideration multiple potential confounders including demographic characteristics, medical history, medication use, and lifestyle behaviors, which are not available in administrative databases that are often used to conduct target trial emulation studies. Our sample included 14.7% who were racial minorities, including 11.9% Black women, which is over twofold higher than the minority enrollment rate typically achieved in RCTs (42). WHI medication use was confirmed with pill bottles that women brought to the clinic, thereby minimizing exposure misclassification. The use of a new-user design minimized prevalent user bias (43).

This study also has several limitations. There are sex differences in type 2 diabetes, with women having a higher cardiometabolic risk factor burden at the time of diagnosis than men (44). Further, women with diabetes have higher risk of mortality than men (45). Younger versus older age at diabetes diagnosis is associated with higher risk of mortality and vascular diseases (46). Accordingly, because our study was limited to postmenopausal women 60 years and older with type 2 diabetes, its generalizability to men and younger populations is unknown. In the WHI, occurrence of type 2 diabetes was based on self-report and in-person medication inventories; thus, misclassification of diabetes is possible. However, self-reported diabetes has been previously found to be reliable and accurate in WHI when compared against medication inventories and fasting glucose levels (24). Residual confounding is a possibility given the observational nature of our study. For example, it is possible that findings were confounded by indication (eg, due to diabetes severity), as we did not have information on the reasons for prescribing metformin or sulfonylurea as first-line treatment. However, as noted above, we adjusted our analyses for a comprehensive set of confounders using PS matching and focused only on new users of study medications. Furthermore, the E-values for the analyses were moderately strong, suggesting that an unmeasured confounder would need to be associated with both the treatment and outcome at a minimum risk ratio of 1.66–2.10 to fully explain away the associations between metformin and death before age 90 (33). The year 3 visit occurred during 1996–2001, and during that period, metformin prescription trends in the United States increased, whereas sulfonylurea prescription trends decreased (26). To minimize confounding by prescription trends, participants were matched on diabetes duration (ie, time from diabetes diagnosis to medication initiation). We were unable to examine a comparison group without diabetes, given that the medications we examined are primarily used to treat diabetes. We also lacked a comparison group taking placebo typically given in the setting of an RCT. Although prescription medications were inventoried in person, the timing of medication initiation relied upon participant recall at the year 3 visit. Additionally, metformin users may have stopped using this medication or switched to other medications during follow-up. However, due to lack of regularly collected longitudinal medication data, we were unable to perform per-protocol (ie, as-treated) analyses to account for medication adherence or adding on additional hypoglycemic agents during follow-up. Finally, we lacked information on medication dosage, measures of disease severity (eg, hemoglobin A1c), and contraindications to metformin use (eg, serum creatinine to determine renal function), which may have confounded the relationship between metformin and longevity. We were unable to perform stratified analysis by demographic factors, such as age or race/ethnicity, because we were unable to generate new PS-matched samples for each subgroup with adequate sample size for analysis.

The findings from this target trial emulation study among women with type 2 diabetes showed that initiation of metformin monotherapy compared with sulfonylurea monotherapy reduced the risk of death before age 90. However, because this comparison was not made to placebo in an RCT and given the observational design with potential for residual confounding, causality cannot be inferred, and findings should be interpreted with caution. These findings support future target trial emulation studies examining intention-to-treat and per-protocol analyses as well as randomized trials to determine the effect of metformin on human longevity.

Funding

This work was supported by the National Heart, Lung, and Blood Institute, National Institutes of Health, U.S. Department of Health and Human Services (grant numbers 75N92021D00001, 75N92021D00002, 75N92021D00003, 75N92021D00004, and 75N92021D00005).

Conflict of Interest

A.H.S., M.A.E., A.O.O., J.E.M., K.C.J., Z.C., B.Z., and A.Z.L. declare no conflicts of interest. B.H. has received fees from Bristol Myers Squibb, Boehringer Ingelheim, Inari, and Pfizer for lectures on topics unrelated to the content of this article.

Author Contributions

Concept and design: Shadyab, LaCroix.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Shadyab.

Critical review of the manuscript for important intellectual content: All authors.

Statistical analysis: Shadyab.

Obtained funding: Manson, LaCroix.

Administrative, technical, or material support: Espeland, Odegaard, Zhang.

Supervision: Shadyab.