Metformin‐based oral antihyperglycemic combination therapy and risk of dementia in patients with newly diagnosed type 2 diabetes: A population‐based study

Abstract

INTRODUCTION

This study examined the association between metformin‐based oral antihyperglycemic combination therapy and the risk of developing dementia in patients with newly diagnosed type 2 diabetes mellitus (T2DM).

METHODS

We conducted a retrospective cohort study using data from a random sample of 2,000,000 individuals in Taiwan's National Health Insurance Research Database from 2000 to 2018. A total of 44,073 patients with newly diagnosed T2DM were identified and categorized into four groups based on initial treatment: metformin (MET)‐sulfonylureas (SU), MET–dipeptidyl peptidase‐4 inhibitor (DPP4i), MET–thiazolidinedione (TZD), and MET–glinides. The participants were followed until December 2018.

RESULTS

The hazard ratio of MET–TZD was 0.73 (95% confidence interval [CI]: 0.57–0.94), of MET–DPP4i was 0.64 (95% CI: 0.52–0.79), and of MET–glinides was 1.14 (95% CI: 0.94–1.39).

DISCUSSION

The MET–DPP4i group had the lowest risk of dementia, followed by the MET–TZD group.

Highlights

• Newly diagnosed type 2 diabetes mellitus (T2DM) patients using metformin (MET)–glinides and MET–sulfonylureas are at an increased risk of dementia.

• The MET–glinides group had the highest risk of dementia.

• The MET–dipeptidyl peptidase‐4 inhibitor group had the lowest risk of dementia, followed by the MET–thiazolidinedione group.

1. BACKGROUND

The prevalence of dementia increases significantly with age. ¹ According to the World Alzheimer Report 2023, > 55 million people globally were living with dementia in 2019, and this number is projected to rise to 139 million by 2050.¹ In Taiwan, by 2018, > 14% of the population was aged > 65 years, with ≈ 7.86% of this group having dementia.² The number of individuals with dementia in Taiwan is expected to exceed 800,000 by 2051. ²

Dementia encompasses a range of neurodegenerative and non‐neurodegenerative disorders. ³ , ⁴ , ⁵ , ⁶ Alzheimer's disease (AD) is the most common form, mainly characterized by progressive neuronal degeneration. ⁴ Other causes of dementia include cerebrovascular disease, autoimmune diseases, drug toxicity, liver and kidney dysfunction, diabetes mellitus (DM), and hypothyroidism. ³ , ⁴ , ⁵ , ⁶ , ⁷ , ⁸ , ⁹ Notably, compared to the general population, patients with DM have a 1.5 to 2.5 times higher risk of developing dementia. ⁷ , ⁸ , ⁹

Dementia is closely linked to the regulation of blood sugar levels. ⁷ , ⁸ , ⁹ Patients with DM often experience difficulties in controlling blood sugar, resulting in episodes of hypoglycemia, hyperglycemia, or fluctuating blood sugar levels. ¹⁰ , ¹¹ , ¹² , ¹³ These conditions can potentially damage brain nerve cells, contributing to the development of dementia. ⁸ , ⁹ , ¹⁰ , ¹¹ , ¹²

The 2022 Type 2 Diabetes Mellitus (T2DM) Treatment Guidelines issued by the Diabetes Association of the Republic of China stratifies patients based on baseline glycated hemoglobin (HbA1c) levels, with a 7.5% treatment threshold. ¹³ For newly diagnosed patients with T2DM and HbA1c levels < 7.5%, metformin (MET) monotherapy is recommended as the first‐line treatment. ¹³ If adequate glycemic control is achieved, MET should be continued. Conversely, if glycemic control remains suboptimal, the addition of other oral antihyperglycemic agents should be considered. ¹³ For patients with newly diagnosed T2DM and baseline HbA1c levels ≥ 7.5%, MET‐based oral antihyperglycemic combination therapy is recommended ¹³ , ¹⁴ , ¹⁵

Compared to individuals with HbA1c < 7.5%, those with HbA1c ≥ 7.5% are at a higher risk of developing dementia later in life. ¹¹ , ¹² , ¹⁶ Hsu et al. reported that the combination oral hypoglycemic therapy may reduce the risk of dementia. ¹⁷ However, the selection of appropriate antidiabetic agents is critical, as inappropriate regimens may lead to hypoglycemia, hyperglycemia, or glycemic variability—factors that have been associated with an increased risk of dementia. ¹¹ , ¹² , ¹⁶

Notably, certain medications, such as sulfonylureas (SUs), are associated with an increased risk of hypoglycemia. ¹³ , ¹⁴ , ¹⁵ , ¹⁷ , ¹⁸ , ¹⁹ Although their combination with MET is effective in lowering blood glucose levels, concerns persist regarding their potential long‐term effects on cognitive decline and dementia risk. ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ , ²¹ , ²² , ²³ , ²⁴ , ²⁵ To date, previous studies have not conclusively demonstrated that specific combinations of oral antihyperglycemic agents can reduce the incidence of dementia. ²³ , ²⁴ , ²⁵ Therefore, this study aimed to examine whether different MET‐based oral antihyperglycemic combination therapies are associated with a reduced risk of dementia in patients with newly diagnosed T2DM.

RESEARCH IN CONTEXT

1. Systematic review: According to previous literature, patients with diabetes mellitus (DM) have a 1.5 to 2.5 times higher risk of developing dementia compared to the general population. Dementia is closely linked to the regulation of blood sugar levels. Individuals with severe initial glycated hemoglobin (HbA1c) levels have a higher risk of developing dementia later in life compared to those with normal levels. However, prescribing combination medications can potentially reduce the risk of dementia. The correlation between combination therapies and dementia risk in newly diagnosed DM patients with HbA1c ≥ 7.5% levels is lacking.

2. Interpretation: Our study found that the metformin (MET)‐dipeptidyl peptidase‐4 inhibitor combination was associated with the lowest risk of dementia, followed by the MET–thiazolidinedione group. In contrast, the MET–glinides group had a higher risk of dementia among DM patients.

3. Future directions: These findings provide valuable insights for clinicians when prescribing DM medications, potentially reducing the risk of dementia associated with DM treatments.

2. METHODS

2.1. Study population

This retrospective cohort study used 2,000,000 random datasets from the National Health Insurance Research Database (NHIRD) from 2000 to 2018, provided by the Health and Welfare Statistics Application Center, Ministry of Health and Welfare, Taiwan. The NHIRD encompasses claims data for ambulatory care, inpatient services, contracted pharmacy records, and demographic information of enrolled beneficiaries. DM was diagnosed using the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD‐9‐CM) codes starting with 250. ²⁶ A total of 345,059 T2DM patients were selected (Figure S1 in supporting information). This study used the NHIRD to identify cases of T2DM diagnosed for the first time since January 2000, requiring at least three outpatient or emergency records within 1 year of the initial diagnosis. ²⁷ , ²⁸ , ²⁹ The index date was defined as the date of the first DM diagnosis. Due to the time required for dementia to develop, only cases diagnosed with T2DM by the end of 2014 were included as new cases. In this study, the primary analysis cohort comprised 236,153 patients with newly diagnosed T2DM (Figure S1).

In addition, eligible patients were required to have been prescribed two oral antihyperglycemic agents at the time of diagnosis. Specifically, patients who received MET in combination with another oral antihyperglycemic agent were categorized as MET–SU, MET–thiazolidinedione (TZD), MET–dipeptidyl peptidase‐4 inhibitor (DPP4i), and MET–glinide (Figure 1) groups. ¹³ , ¹⁴ , ¹⁵ , ¹⁷ , ¹⁸ , ²⁰ , ²¹ , ²⁵ Patients were excluded if they were < 18 years of age, had type 1 DM, had used insulin, were prescribed only a single oral antihyperglycemic agent at the time of diagnosis, had a diagnosis of dementia before or within 1 year after the diabetes diagnosis, or changed their oral antihyperglycemic drug combinations during the observation period. After applying these criteria, a total of 44,073 patients were included in the final analysis (Figure S1).

FIGURE 1.

Classification of four metformin‐based oral antihyperglycemic combination therapies. DM, diabetes mellitus; DPP4i, dipeptidyl peptidase‐4 inhibitor; TZD, thiazolidinedione.

2.2. Data collection

The patients were categorized into four groups according to the drug use guidelines as follows: MET–SU, MET–DPP4i, MET–TZD, and MET–glinides. ¹³ , ¹⁴ , ¹⁵ , ¹⁷ , ¹⁸ , ²⁰ , ²¹ , ²⁵ Each study group consisted of patients prescribed both MET (ATC code A10BA02) ³⁰ and another medication at the time of initial DM diagnosis, with treatment records confirming prescriptions of these medications at least three times within a 90 day period. Specifically, the groups included MET and one of the other anti‐diabetics such as SU (ATC code A10BB), DPP4 inhibitors (ATC code A10BH), TZD (ATC code A10BG), and glinides (ATC code A10BX02). ³⁰

2.3. Dependent variables

2.3.1. Dementia incidence

Dementia incidence in this study is defined using specific diagnostic codes: for ICD‐9‐CM, dementia is identified by codes 290, 294, and 331. ²⁶ For ICD‐10‐CM, dementia is identified by codes F03.90, F05, F01.50, and F01.51. ²⁶ Additionally, patients must have had more than three outpatient visits or more than one hospitalization record related to dementia within 1 year to meet the criteria. ²⁷ , ²⁸ , ²⁹ Given the potential influence of DM medication use, a long‐term observation period is necessary to establish causation. Therefore, patients who developed dementia within 4 years after their DM diagnosis were excluded from the study.

2.3.2. Risk factors

This study considers several factors such as age; sex; hypertension;, hyperlipidemia; stroke; ischemic heart disease, including myocardial infarction, angina pectoris, and heart failure; arrhythmia, including atrial fibrillation; cerebrovascular disease, including transient ischemic attack, ischemic stroke, and intracranial hemorrhage; kidney disease; retinal detachment; Charlson Comorbidity Index (CCI); and the use of calcium channel blockers (CCBs), direct thrombin inhibitors, warfarin, and β‐hydroxy β‐methylglutaryl‐CoA (HMG‐CoA) reductase inhibitors. ³¹ , ³² , ³³ , ³⁴ , ³⁵ , ³⁶ , ³⁷ , ³⁸ , ³⁹ , ⁴⁰ , ⁴¹

Comorbidity is defined as having at least one inpatient or two outpatient diagnoses at least 4 years before the initial dementia diagnosis. The CCI assigns weighted scores to different chronic diseases based on severity (1, 2, 3, or 6 points). ⁴² The total score predicts the 10 year mortality risk and reflects the burden of multiple comorbidities, which may increase dementia risk. ²⁷ , ²⁸ , ²⁹

2.4. Statistical analysis

Descriptive statistics present the age, sex distribution, and CCI scores across four MET‐based oral antihyperglycemic combination groups. Continuous variables are shown as mean (standard deviation), whereas categorical variables are presented as frequencies (%). Subsequently, the study compares the incidence rates (IRs) per 10,000 person‐years among the groups and calculates the relative risk (RR) of dementia for MET–DPP4i, MET–TZD, and MET–glinides compared to MET–SU. Cox regression, competing risk regression, and cumulative incidence curves are used to analyze dementia risk, adjusting for other factors. Results are reported as hazard ratios (HRs) with 95% confidence intervals (CIs). Sensitivity analysis examines the RR between medication initiation and dementia onset over 1 to 4 years to observe differences in IR and RR. Data processing and analyses were conducted using SAS version 9.4 (SAS Institute Inc.). A two‐tailed P value of < 0.05 was considered statistically significant.

3. RESULTS

This study included 44,073 patients with DM who met the inclusion criteria (Figure S1). These patients were categorized into four groups based on the received MET‐based oral antihyperglycemic combination therapy: MET–SU (n = 35,443), MET–TZD (n = 2046), MET–DPP4i (n = 4519), and MET–glinides (n = 2065; Table 1).

TABLE 1.

Baseline characteristics of newly diagnosed T2DM patients by four metformin‐based oral antihyperglycemic combinations.

Characteristics	MET–SU (n = 35,443)	MET–TZD (n = 2046)	MET–DPP4i (n = 4519)	MET–glinides (n = 2065)
Sex
Male	18,926 (53.4)	1023 (50)	2351 (52.02)	1083 (52.45)
Female	16,517 (46.6)	1023 (50)	2168 (47.98)	982 (47.55)
Age	60.23 (12.6)	59.41 (12.58)	60.62 (12.9)	62.32 (13.69)
≥ 65 years	13,073 (36.88)	694 (33.92)	1696 (37.53)	913 (44.21)
< 65 years	22,370 (63.12)	1352 (66.08)	2823 (62.47)	1152 (55.79)
CCI	2.5 (1.18)	2.43 (1.17)	2.53 (1.2)	2.67 (1.24)
≥ 3	8534 (24.08)	438 (21.41)	1159 (25.65)	676 (32.74)
< 3	26,909 (75.92)	1608 (78.59)	3360 (74.35)	1389 (67.26)
Hypertension	31,607 (89.18)	1852 (90.52)	3998 (88.47)	1856 (89.88)
Hyperlipidemia	29,637 (83.62)	1771 (86.56)	3819 (84.51)	1656 (80.19)
Hypertensive heart disease with or without renal disease	22,546 (63.61)	1357 (66.32)	2757 (61.01)	1355 (65.62)
Ischemic heart disease	18,774 (52.97)	1126 (55.03)	2345 (51.89)	1115 (54)
Myocardial infarction	9348 (26.37)	568 (27.76)	1285 (28.44)	530 (25.67)
Angina	11,608 (32.75)	745 (36.41)	1596 (35.32)	673 (32.59)
Heart failure	10,419 (29.4)	557 (27.22)	1105 (24.45)	671 (32.49)
Cardiac dysrhythmia	10,647 (30.04)	609 (29.77)	1340 (29.65)	660 (31.96)
Atrial fibrillation and flutter	3413 (9.63)	180 (8.8)	423 (9.36)	236 (11.43)
Cerebrovascular disease	15,380 (43.39)	825 (40.32)	1749 (38.7)	929 (44.99)
Transient ischemic attack	7326 (20.67)	448 (21.9)	989 (21.89)	424 (20.53)
Ischemic stroke	12,878 (36.33)	685 (33.48)	1428 (31.6)	768 (37.19)
Intracranial hemorrhage	3255 (9.18)	179 (8.75)	329 (7.94)	228 (11.04)
Chronic kidney disease	25,063 (70.71)	1529 (74.73)	3032 (67.09)	1478 (71.57)
Retinopathy	18,320 (51.69)	1111 (54.3)	1974 (43.68)	989 (47.89)
CCBs	30,322 (85.55)	1749 (85.48)	3773 (83.49)	1814 (87.85)
Antiplatelet drugs	1043 (2.94)	62 (3.03)	214 (4.74)	58 (2.81)
Direct thrombin inhibitors	390 (1.1)	25 (1.22)	76 (1.68)	32 (1.55)
Warfarin	2981 (8.41)	159 (7.77)	347 (7.68)	190 (9.2)
HMG‐CoAs	27,516 (77.63)	1682 (82.21)	3659 (80.97)	1505 (72.88)

Note: Categorical variables were represented using frequencies and percentages. Continuous variables used means and standard deviations (SDs).

Abbreviations: CCB, calcium channel blocker; CCI, Charlson Comorbidity Index; DPP4i, dipeptidyl peptidase‐4 inhibitor; HMG‐CoA, β‐hydroxy β‐methylglutaryl‐CoA inhibitor; MET, metformin; SU, sulfonylurea; T2DM, type 2 diabetes mellitus; TZD, thiazolidinedione.

Demographic variables were calculated according to the four MET‐based oral antihyperglycemic combination groups, including the sex ratio, average age, and average CCI. Regarding sex ratios, males were predominantly higher than females in all four groups, and the proportions between each group are similar. The average age of the four groups is ≈ 60 years. Notably, the mean CCI score ranged between 2 and 3 across all groups. The proportions of other related risk factors in each group are similar (Table 1). Subsequently, the cumulative incidence of dementia per 10,000 person‐years was analyzed using the MET–SU group as the reference. The MET–glinide group exhibited the highest RR at 1.32 (95% CI: 1.08–1.56), whereas the MET–DPP4i group showed the lowest (RR = 0.67; 95% CI: 0.56–0.78; Figure 2).

FIGURE 2.

Cumulative incidence of dementia per 10,000 person‐years by four metformin‐based oral antihyperglycemic combination therapies. CI, confidence interval; DPP4i, dipeptidyl peptidase‐4 inhibitor; IR, incidence rate; MET, metformin; PY, 10,000 person‐years; RR, relative risk; SU, sulfonylurea; TZD, thiazolidinedione.

Moreover, this study performed collinearity testing among the four oral antihyperglycemic combination groups, age, sex, CCI, risk factors, and non‐diabetic medications. Variables with a variance inflation factor of > 3, including ischemic heart disease, myocardial infarction, angina pectoris, atrial fibrillation, cerebral ischemia, and retinopathy, were excluded prior to subsequent analyses.

The Cox proportional hazards model and the competing risk model demonstrated that the MET–TZD and MET–DPP4i groups had a significantly lower risk of developing dementia than the MET–SU group. Although the MET–glinide group showed a higher risk of developing dementia than the MET–SU group, the difference was not significant (Figure 3). Several additional risk factors for dementia were identified. Male patients had a lower risk compared to female patients (Table S1). Moreover, patients aged > 65 years had a > 2‐fold increased risk of developing the disease compared to those aged < 65 years. The CCI used a score of 3 as the cut‐off point and found that the risk of developing dementia in the group with a score > 1‐fold increased risk was higher than the control group. Heart failure, arrhythmia, cerebrovascular disease, ischemic stroke, intracranial hemorrhage, and chronic kidney disease all showed higher risk ratios compared to the control group. Regarding the use of other types of drugs, patients who used CCB drugs had a higher risk of developing dementia than the control group. Conversely, those who used antiplatelet drugs, direct thrombin inhibitors, and HMG‐CoA inhibitors had a lower risk of developing dementia than the control group.

FIGURE 3.

Hazard ratios for dementia associated with four metformin‐based oral antihyperglycemic combination therapies. CI, confidence interval; Cox, Cox regression model; CRR, competing risk regression; DPP4i, dipeptidyl peptidase‐4 inhibitor; HR, hazard ratio; IR, incidence rate; MET, metformin; SU, sulfonylurea; TZD, thiazolidinedione.

The Kaplan–Meier curves of cumulative dementia incidence illustrate the risk of dementia associated with the four MET‐based combination therapies over a 15 year follow‐up period. At 15 years, the MET–glinides group showed the highest risk of dementia, followed by the MET–SU and MET–TZD groups. Conversely, the MET–DPP4i group exhibited the lowest risk (Figure 4). Moreover, the Kaplan–Meier curves from the sensitivity analyses, conducted with different follow‐up exclusion periods, demonstrated no significant differences in cumulative dementia incidence among the four treatment groups when the interval between drug exposure and dementia onset ranged from 1 to 4 years (Figure 5).

FIGURE 4.

Kaplan–Meier curves of cumulative dementia incidence in patients treated with four metformin‐based oral antihyperglycemic combination therapies. DPP4i, dipeptidyl peptidase‐4 inhibitor; MET, metformin; SU, sulfonylurea; TZD, thiazolidinedione.

FIGURE 5.

Kaplan–Meier curves of cumulative dementia incidence by four metformin‐based oral antihyperglycemic combination therapies in sensitivity analyses with different follow‐up exclusion periods. DPP4i, dipeptidyl peptidase‐4 inhibitor; MET, metformin; SU, sulfonylurea; TZD, thiazolidinedione.

In the sensitivity analyses with different follow‐up exclusion periods, the cumulative IRs per 10,000 person‐years were evaluated across the four MET‐based combination groups. A shorter exclusion interval was associated with higher IR. However, the RRs of the three groups compared to the control group (MET–SU) remained relatively stable across exclusion periods. All three groups showed significant associations: MET–TZD (RR: 0.8), MET–DPP4i (RR: 0.6–0.7), and MET–glinides (RR: 1.2–1.3; Table 2).

TABLE 2.

Cumulative incidence of dementia per 10,000 person‐years in sensitivity analysis of four metformin‐based combination therapies.

Groups	No. of events	PY	IR (95% CI)	RR (95% CI)
Exclude events within 2 years
MET–TZD	149	18,765.7	79.35 (62.60–96.09)	0.79 (0.59–0.99)
MET–DPP4i	225	33,682.6	66.81 (55.32–78.33)	0.66 (0.56–0.76)
MET–glinides	214	16,563.5	129.24 (106.38–152.04)	1.28 (1.12–1.45)
MET–SU (reference)	3110	30,7920.8	101.00 (96.0–105.3)	–
Exclude events within 3 years
MET–TZD	131	18,767.9	69.84 (54.10–85.58)	0.77 (0.61–0.93)
MET–DPP4i	196	33,677.0	58.24 (47.51–68.96)	0.64 (0.50–0.77)
MET–glinides	187	16,548.7	113.01 (91.65–134.37)	1.25 (1.01–1.48)
MET–SU (reference)	2792	308,849.6	90.40 (86.02–94.86)	–
Exclude events within 4 years
MET–TZD	117	18,749.2	62.40 (47.54–77.25)	0.78 (0.61–0.94)
MET–DPP4i	167	33,635.2	49.65 (39.7–59.63)	0.62 (0.50–0.78)
MET–glinides	167	16,500.3	101.21 (81.03–121.39)	1.26 (1.02–1.51)
MET–SU (reference)	2477	30,8084.6	80.40 (76.24–84.49)	–

Abbreviations: CI, 95% confidence interval; DPP4i, dipeptidyl peptidase‐4 inhibitor; IR, incidence rate; MET, metformin; PY, 10,000 person‐years; RR, relative risk; SU, sulfonylurea; TZD, thiazolidinedione.

In the sensitivity analyses using different follow‐up exclusion periods, the risk of dementia associated with various MET‐based oral antihyperglycemic combinations was compared to that of the MET–SU group. The MET–TZD group showed no significant difference in HRs compared to the control. Conversely, the MET–DPP4i group consistently demonstrated a significantly lower risk of dementia across all exclusion periods, with HRs indicating a protective effect relative to the control group. The MET–glinides group exhibited HRs slightly > 1.0, but none of the results reached statistical significance during any exclusion period (Table 3).

TABLE 3.

Sensitivity analysis for dementia hazard ratio by four metformin‐based oral antihyperglycemic combinations.

Groups	HR of Cox	P	HR of CRR	P
Exclude events within 2 years
MET–TZD	0.75 (0.58–0.97)	0.026	0.78 (0.60‐1.00)	0.051
MET–DPP4i	0.60 (0.48–0.75)	<0.001	0.58 (0.46–0.73)	<0.001
MET‐glinides	1.12 (0.91–1.37)	0.275	1.02 (0.83–1.26)	0.839
MET–SU (reference)	–	–	–	–
Exclude events within 3 years
MET–TZD	0.74 (0.56–0.97)	0.029	0.77 (0.58–1.00)	0.054
MET–DPP4i	0.60 (0.47–0.77)	<0.001	0.59 (0.46–0.75)	<0.001
MET–glinides	1.17 (0.95–1.45)	0.147	1.06 (0.95–1.32)	0.592
MET–SU (reference)	–	–	–	–
Exclude events within 4 years
MET–TZD	0.75 (0.56–0.99)	0.046	0.78 (0.58–1.03)	0.083
MET–DPP4i	0.63 (0.49–0.81)	<0.001	0.61 (0.47–0.78)	<0.001
MET–glinides	1.16 (0.92–1.45)	0.21	1.04 (0.83–1.31)	0.713
MET–SU (reference)	–	–	–	–

Abbreviations: CI, 95% confidence interval; Cox, Cox regression model; CRR, competing risk regression; DPP4i, dipeptidyl peptidase‐4 inhibitor; HR, hazard ratio; MET, metformin; SU, sulfonylurea; TZD, thiazolidinedione.

4. DISCUSSION

The management of antihyperglycemic agents generally adheres to established diabetes treatment guidelines, which stratify patients based on HbA1c levels into mild (HbA1c < 7.5%) and severe (HbA1c > 7.5%) categories. ¹³ , ¹⁴ , ¹⁵ For patients with mild hyperglycemia, MET is typically initiated as the first‐line therapy. ¹³ , ¹⁴ , ¹⁵ If adequate glycemic control is not achieved within 3 to 6 months of MET monotherapy, a second‐line oral antihyperglycemic agent is usually added. Conversely, patients presenting with more severe hyperglycemia are often prescribed MET combined with another oral antihyperglycemic agent at the time of diagnosis to achieve more effective glycemic control. ¹³ , ¹⁴ , ¹⁵

For adverse drug effects, SUs and meglitinides are associated with a higher risk of hypoglycemia than newer agents such as TZDs and DPP4i. ¹³ , ¹⁴ , ¹⁵ , ¹⁷ As a result, these antihyperglycemic agents may be less suitable for individuals at a high risk of hypoglycemia. Furthermore, their use may be linked to a modestly increased risk of developing dementia. ¹⁶

This study observed that compared to the combination of MET–SU, both MET–TZD and MET–DPP4i were associated with a lower risk of dementia, while MET–glinides had a slightly higher risk. This may be due to the side effects of SU and glinides, which can lead to hypoglycemia. ¹⁷ , ²³ , ⁴³ , ⁴⁴ Glucose is a crucial energy source for normal brain function, and hypoglycemia can impair the brain's energy supply, potentially leading to a decline in cognitive function. ⁷ , ⁸ , ¹¹ Long‐term use of these medications may, therefore, increase the risk of dementia. ¹⁷ , ²³ , ⁴³ , ⁴⁴

This result is consistent with the findings of Cho and Cho, which showed that MET–DPP4i has a lower incidence of hypoglycemia and cardiovascular disease compared to MET–SU. ²³ The adjusted HRs were 0.39 and 0.72, respectively. Both hypoglycemia and cardiovascular disease are significant risk factors for dementia, suggesting that the lower incidence of these conditions with MET–DPP4i could contribute to a reduced risk of dementia.

The study by Mamza et al. highlighted that the risk of uncontrolled blood sugar was 15%, 23%, and 8% in the MET–SU, MET–TZD, and MET–DPP4i groups, respectively. ⁴³ Blood sugar control was poorest in the MET–SU group, followed by the MET–TZD group, whereas MET–DPP4i demonstrated the best control. Maintaining blood sugar levels is crucial, as hyperglycemia can cause vascular damage and inflammatory reactions, adversely affecting brain structure and function, thus increasing the risk of dementia. ⁴³

However, the study by Cheng et al. concluded that SU had a lower dementia risk compared to TZD, which contrasts with our findings. ⁸ This discrepancy may arise from differences in study design; Cheng et al. included single oral antihyperglycemic agents in their analysis, whereas our study focused on MET‐based combination therapies used from the initial diagnosis. ⁸ Additionally, variations in sample size might contribute to these differences. Smaller sample sizes in the Cheng et al. study could have led to an overestimation of results, whereas our strict screening criteria for combined antihyperglycemic agents might have influenced our findings. ⁸ Additionally, the observed discrepancy may be explained by variations in antihyperglycemic treatment strategies Lu et al. (2018) reported that such strategies encompass initiating therapy with a single oral agent followed by the sequential addition of a second drug, as well as commencing combination therapy at the time of diagnosis. ²⁵ These differences in study design and screening criteria may account for variations in subsequent analysis of results.

This study adopted more stringent inclusion criteria, requiring patients to initiate MET‐based combination therapy at the time of diabetes diagnosis and maintain the same regimen throughout the observation period. These criteria likely contributed to a smaller sample size and may partially explain discrepancies in findings compared to previous studies. Additionally, the definition of dementia included diagnoses made within 1 year before or after the DM diagnosis, allowing for a more comprehensive identification of dementia. Furthermore, sensitivity analyses using various follow‐up exclusion periods consistently yielded non‐significant results, suggesting the robustness of the findings and further refining the analytical cohort.

Our collinearity test revealed high collinearity among six diseases, such as ischemic heart disease, myocardial infarction, angina pectoris, atrial fibrillation, cerebral ischemia, and retinopathy. This may result from treatment for these conditions, such as platelet aggregation inhibitors for myocardial infarction and anticoagulants for atrial fibrillation, which can reduce dementia risk by preventing blood clots. ⁴⁴ Conversely, glucocorticoids used for retinopathy might affect brain tissue and neurons, increasing dementia risk. ⁴⁵ To avoid skewed results, these diseases were excluded from subsequent analyses.

Additionally, heart failure, arrhythmia, cerebrovascular disease, ischemic stroke, and intracranial hemorrhage all have a higher risk of dementia compared to healthy individuals. ³¹ , ³⁵ , ³⁶ , ³⁷ , ⁴⁶ , ⁴⁷ , ⁴⁸ , ⁴⁹ Cerebrovascular disease, ischemic stroke, and intracranial hemorrhage can impair brain blood vessels, leading to cognitive decline and dementia. ³¹ , ³⁵ , ³⁶ , ³⁷ , ⁴⁹ Cheng et al. observed that both cerebrovascular disease and hypertension increase dementia risk, whereas hyperlipidemia reduces it, which is slightly different from our study. ⁸ Huang et al. reported higher dementia risk in patients with hypertension (HR 1.30), hyperlipemia (HR 1.06), and ischemic stroke (HR 1.79), aligning with our findings, except for heart failure (HR 0.74). ¹⁰ This may be because our study included more severe diabetic patients, whose underlying conditions differ. Heart failure reduces effective blood pumping, leading to poor circulation and potential brain cell damage, increasing dementia risk. ¹⁰ Similarly, arrhythmia can disrupt blood and oxygen flow to the brain, affecting brain cell health and function, though further research is needed to confirm this association.

Among the other medications, CCBs have been associated with a higher risk of dementia compared to patients not using these drugs, consistent with the Wu et al. ⁵⁰ However, the relationship between CCBs and dementia remains unclear and requires further clinical study. Conversely, antiplatelet drugs and HMG‐CoA inhibitors, which primarily treat vascular diseases, are associated with a lower risk of dementia. ⁴⁹ , ⁵¹ Because dementia often arises from brain vascular diseases or blockages, maintaining good vascular health with these medications may reduce the risk of dementia. ⁴⁹ , ⁵⁰ , ⁵¹

In our sensitivity analyses comprising the different follow‐up exclusion periods, the MET–TZD group showed no significant difference in HRs compared to the control group. Conversely, the MET–DPP4i group consistently exhibited a significantly lower risk of dementia across all exclusion periods, with HRs suggesting a protective effect. The MET–glinides group demonstrated HRs slightly > 1.0, although none reached significance under any exclusion condition. However, the stringent inclusion criteria used in this study may have led to the exclusion of a substantial number of patients during data processing, potentially resulting in a slight overestimation of effect sizes and introducing selection bias.

This study has several inherent limitations that should be considered when interpreting the findings. First, the use of administrative claims data from the NHIRD precludes access to important non‐clinical variables, including medication adherence, glycemic control indicators (e.g., HbA1c), lifestyle behaviors, and environmental exposures. The absence of these covariates may have led to residual confounding and potentially attenuated the observed associations between antihyperglycemic regimens and dementia risk. Second, due to the retrospective cohort design, the study can only establish associations, not causality. Third, T2DM was identified solely based on ICD‐9‐CM diagnostic codes. Although prior validation studies using the NHIRD have demonstrated good diagnostic accuracy, with a sensitivity of 90.9% and a positive predictive value of 92%, ⁵² the absence of biochemical confirmation such as fasting plasma glucose or HbA1c raises the possibility of misclassification bias. Furthermore, dementia subtypes could not be distinguished in this claims‐based dataset, limiting our ability to assess subtype‐specific effects.

Despite these limitations, this study provides clinically relevant insights. Among patients with newly diagnosed T2DM, the MET–DPP4i combination was associated with the lowest risk of developing dementia, followed by the MET–TZD combination. Conversely, the MET–glinides group exhibited a comparatively higher risk. These findings may inform treatment decisions when considering long‐term cognitive outcomes in patients with T2DM. Although the current analysis emphasized dementia of vascular origin owing to its established link with diabetes, we acknowledge that AD is the most common subtype. Future prospective studies incorporating clinical assessments or biomarker data are warranted to further elucidate the effects of specific antihyperglycemic therapies on distinct dementia subtypes, particularly AD.

AUTHOR CONTRIBUTIONS

Tsung‐Cheng Hsieh, Hsiang‐Hao Chen, and Chen‐Pei Ho contributed to the development of the theoretical framework. Tsung‐Cheng Hsieh and Hsiang‐Hao Chen carried out the analytical calculations and numerical simulations. All three authors—Tsung‐Cheng Hsieh, Hsiang‐Hao Chen, and Chen‐Pei Ho—were involved in manuscript writing and approved the final version. Tsung‐Cheng Hsieh and Chen‐Pei Ho supervised the overall project. Wei‐Chuan Chang contributed to data collation, software implementation, figure and table revisions, forest plot generation, and manuscript review and editing in response to comments from the reviewers.

CONFLICT OF INTEREST STATEMENT

All authors declare no conflict of interest with respect to the research, authorship, and/or publication of this article. Author disclosures are available in the supporting information.

During the preparation of this work, the author(s) used ChatGPT to correct the grammar of the article. After using this tool/service, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the content of the publication.

Supporting information

Supporting Information

Supporting Information

Hsieh T‐C, Chen H‐H, Chang W‐C, Ho C‐P. Metformin‐based oral antihyperglycemic combination therapy and risk of dementia in patients with newly diagnosed type 2 diabetes: A population‐based study. Alzheimer's Dement. 2025;21:e70590. 10.1002/alz.70590