The effects of improved access to diabetes medications on risk of Alzheimer's disease and related dementias

Abstract

INTRODUCTION

Diabetes is associated with elevated risk of Alzheimer's disease and related dementias (ADRD). Disease management with pharmaceutical therapies may mitigate this increased risk, but access and use of diabetes therapies are limited, especially for persons at highest risk of ADRD.

METHODS

Using 10 years of data on Americans aged 51 and older with diabetes, we employed dynamic microsimulation to analyze the impact of increased utilization of glucagon‐like peptide 1 receptor agonists (GLP1s) and sodium‐glucose cotransporter‐2 inhibitors (SGLT2s) on ADRD incidence.

RESULTS

Over 10 years, utilization of GLP1s and SGLT2s by people with diabetes and HbA1c > 7 could avert 76,000–201,000 ADRD cases. The risk reduction is largest for Black and Hispanic individuals and individuals with HbA1c > 8.

DISCUSSION

This study shows that improving diabetes management can lead to large reductions in ADRD risk. Improved adherence and access for those with higher HbA1c and higher risk of ADRD will have the largest benefit.

Highlights

• Use of glucagon‐like peptide 1 receptor agonists (GLP1s) and sodium‐glucose cotransporter‐2 inhibitors (SGLT2s) by people with diabetes can reduce 10‐year Alzheimer's disease and related dementias (ADRD) risk by 7.1%.

• Over 40 years, ADRD risk would be reduced by 9.2%.

• The benefit is larger for people at higher risk of ADRD, like people with HbA1c > 8.

• Good adherence is necessary to receive the full risk reduction.

1. BACKGROUND

Approximately 45% of Alzheimer's disease and related dementia (ADRD) cases are attributable to modifiable risk factors, including diabetes, ¹ and among incident ADRD cases in the United States, 39% have diabetes. ² Diabetes is thus an especially important ADRD risk factor, affecting 38.1 million American adults (14.7%), and 16.5 million Americans aged 65 and older (29.2%). ³ Importantly, better diabetes management is associated with reduced ADRD risk, ⁴ but limited access to prescription drugs inhibits effective management for disadvantaged individuals with diabetes. ⁵ , ⁶ , ⁷ This is especially relevant for Black and Hispanic adults, who have higher prevalence of diabetes than White adults (Black 17.4%, Hispanic 15.5%, White 13.6%), ³ and higher prevalence of Alzheimer's dementia among older adults (Black 19%, Hispanic 14%, White 10%). ⁸ , ⁹ , ¹⁰ Diabetes and lack of access to diabetes management therapies in both mid‐life and late‐life, therefore, are contributors to the large and disparate burden of ADRD on patients, caregivers, and the health system. ⁸ , ⁹ , ¹¹ , ¹²

For people with diabetes, higher blood glucose (HbA1c of 8.25) is associated with 51% higher risk of ADRD, as compared to HbA1c of 7.2. ⁴ Prescription drug therapy plays an important role in glycemic control, ¹³ emphasizing the importance of prescription drug access for ADRD risk. This is especially true for glucagon‐like peptide 1 receptor agonists (GLP1s) and sodium‐glucose cotransporter‐2 inhibitors (SGLT2s), which are notable for their high cost, ¹⁴ effective glycemic control (and consequent benefits for ADRD risk), ⁴ and their beneficial effects on cardiovascular and renal outcomes. ¹⁵ , ¹⁶ Additionally, GLP1s promote protective effects on vascular, microglial, and neuronal functions, and thus may be effective against ADRD, independent of their role in diabetes management. ¹⁷ , ¹⁸

Despite the potential for reduced ADRD risk, utilization of GLP1s and SGLT2s remains relatively low: among Americans with diabetes in 2024, 12.1% used SGLT2s and 18.1% used GLP1s. ¹⁹ Black and Hispanic individuals are even less likely to initiate these classes. ¹⁵ , ²⁰ , ²¹ , ²² Compared to White individuals with diabetes, and after adjusting for personal and clinical characteristics, Hispanic individuals with diabetes have 31% lower odds of GLP1 use and 13% lower odds of SGLT2 use; Black individuals with diabetes have 14% lower odds of GLP1 use, and 11% lower odds of SGLT2 use. ²² Medicaid eligibility, lower income, and rural location were also associated with less likelihood of initiating a GLP1 or SGLT2. ¹⁵ , ²¹ These access differences have many well‐documented causes rooted in socioeconomic determinants of health. ²³ Many mechanisms that affect access can be modified with policies, such as reductions in out‐of‐pocket costs for prescription drugs, but the long‐term effects of improved access to diabetes management therapies on ADRD risk have not been studied.

In this study, we applied rigorous estimates of the relationships between diabetes therapies, HbA1c, and ADRD, to examine the long‐term effects of access to GLP1s and SGLT2s. Using dynamic microsimulation over 40 years, we compared real‐world ADRD projections with what could be achieved under scenarios with improved access to GLP1s and SGLT2s. We examined two scenarios that increased utilization of these classes for all people with diabetes and HbA1c above a certain threshold. We show that the targeting of new access, dependent on HbA1c levels, has important implications for ADRD risk. We also examined the role of medication adherence in reducing ADRD risk. This work can inform policy efforts and care decisions that seek to prevent or delay ADRD by improving the management of its risk factors, and also illuminate the benefits of improving access to treatments for chronic conditions.

RESEARCH IN CONTEXT

1. Systematic review: The authors examined prior literature using traditional (e.g., PubMed, Google Scholar) sources. There is established evidence showing a link between diabetes medications and HbA1c, and between HbA1c and risk of Alzheimer's disease and related dementias (ADRD). The relevant citations are appropriately cited. There are no projections of the expected effects of better diabetes management and long‐term ADRD incidence.

2. Interpretation: Quantifying the ADRD cases that can be averted from better management of diabetes for specific people can inform care decisions and policy strategies aimed at preventing ADRD via management of its risk factors.

3. Future directions: This work connects diabetes pharmacotherapy to long‐term ADRD incidence. A critical next step is to incorporate a broader set of ADRD risk factors and their treatments, and to connect them with both costs and outcomes of care.

2. METHODS

2.1. Summary

The Future Elderly Model (FEM) is a dynamic microsimulation of health and economic outcomes for the US population ages 51 years and older. ²⁴ , ²⁵ The model simulates individuals’ life trajectories under various intervention scenarios, enabling individual‐level comparisons (e.g., the same person with and without utilization of GLP1s and SGLT2s). The outcomes are then aggregated and scaled to be representative of the entire US population. In this study, we examined a sample of individuals with diabetes, ages 51 years and older, drawn from a cohort of Health and Retirement Study (HRS) respondents. We simulated their future outcomes (including ADRD incidence) in 2‐year time steps for 40 years, under scenarios featuring both status quo utilization of diabetes medications, and increased access to GLP1s and SGLT2s. Figure 1 depicts a framework of how use of these drugs could affect ADRD incidence. The simulations rely on rigorously measured relationships between diabetes medication utilization and HbA1c, ²⁶ , ²⁷ , ²⁸ and between HbA1c and ADRD risk for people with diabetes. ⁴ We considered 100% adherence among new GLP1 and SGLT2 users to be the upper bound of the benefit, and 40% adherence to be the lower bound. ²⁹ In a sensitivity analysis, we simulated results for 80% adherence. The outcomes of interest were ADRD cases and years in the hypothetical scenarios, as compared to the status quo scenario. Full methodological details are available in the Technical Appendix.

FIGURE 1.

Scenario frameworks. ADRD, Alzheimer's disease and related dementias; GLP1s, glucagon‐like peptide 1 receptor agonists; HbA1c, glycated hemoglobin; SGLT2s, sodium‐glucose cotransporter‐2 inhibitors.

2.2. Data and setting

The simulation sample is drawn from a cohort of HRS respondents, who participated in the biomarker portion of the core data collection from 2006 to 2016. The HRS is sponsored by the National Institute on Aging (grant numbers NIA U01AG009740 and NIA R01AG073289) and is conducted by the University of Michigan. ³⁰ , ³¹ We included respondents with a self‐reported diagnosis of diabetes, no ADRD (as measured by the Telephone Interview of Cognitive Status [TICS]), non‐missing HbA1c, and no death in the wave of observation. ADRD status was verified in the subsequent survey wave, and did not distinguish between subtypes (see Section 3 of the Technical Appendix). Data collection occurred from 2006 to 2016; for individuals observed in multiple waves, we used their most recent wave available. The final sample was 4,731 individuals (1,051 Black, 788 Hispanic, 2,700 White, 192 Other race indivuduals), who represent approximately 21 million Americans. Measurement of HbA1c, race, ethnicity, and ADRD are described in Section 3 of the Technical Appendix. Additional data came from the Adult Changes in Thought (ACT) Study, ⁴ which was used to characterize the relationship between HbA1c and ADRD risk (details in Section 8 of the Technical Appendix). Finally, we used the Komodo Healthcare Map (KHM) research network data, ¹⁹ to measure current (2024) utilization of GLP1s and SGLT2s, among people with diabetes (details in Section 5 of the Technical Appendix).

2.3. Scenarios

In addition to simulating ADRD outcomes under status quo utilization of diabetes medications, we tested two scenarios with increased utilization of GLP1s and SGLT2s. The two scenarios assume initiation of these drugs for individuals with HbA1c above a certain threshold. Although we did not condition eligibility on race and ethnicity, these scenarios disproportionately provide access to Black and Hispanic people with diabetes, because they have higher HbA1c on average. ³²

• Scenario A: All people with HbA1c ≥ 7 receive drugs. All respondents with HbA1c > 7 were assumed to initiate GLP1s or SGLT2s. In order to account for the fact that some individuals were already using GLP1s or SGLT2s, we deflated the probability of initiation by the probability of current use (24.9%, as measured in 2024, and described in Section 5 of the Technical Appendix ¹⁹ ). HbA1c of 7.0 is widely considered to be the target goal of diabetes medication therapy, ¹⁶ and this scenario is designed to depict a “best‐case scenario,” defined as the maximum of what could be achieved with widespread use of GLP1s and SGLT2s.

• Scenario B: All people with HbA1c ≥ 8 receive drugs. This is the same as scenario A, except the eligibility threshold was HbA1c > 8. This scenario is designed to depict the effects of a smaller intervention, with fewer people initiating GLP1s and SGLT2s, and restricting access to those with worse glycemic control.

2.4. Intervention: Effects of GLP1s and SGLT2s on HbA1c and ADRD risk

In scenarios A and B, individuals deemed eligible for the intervention in each scenario initiated GLP1s or SGLT2s. These individuals then experienced reductions in their HbA1c, as measured in recent meta‐analyses. ²⁶ , ²⁷ , ²⁸ These findings were converted to 2‐year effects on HbA1c, and are reported in Table 1A. The expected 2‐year reduction in HbA1c from using either GLP1s or SGLT2s, conditional on the probability of also using metformin, is 1.6473. The simulations did not allow HbA1c to drop below 7, as this is widely considered to be the target goal of diabetes medication therapy. ¹⁶ Details of the conversion calculation are available in Section 6 of the Technical Appendix, and reported in our earlier work. ³³

TABLE 1.

Parameters of interest.

	GLP1 (60% of users)		SGLT2 (40% of users)
	Monotherapy (20% of GLP1 users)	With Metformin (80% of GLP1 users)	Monotherapy (18% of SGLT2 users)	With Metformin (82% of SGLT2 users)
1A. Effects of GLP1s and SGLT2s on HbA1c over 2 years
Expected HbA1c reduction	−0.904	−1.96	−1.56	−1.624
	−1.67066		−1.61239
	−1.6473*
1B. Relationship between glucose and ADRD risk, among individuals with diabetes
Daily average glucose	HbA1c	HR for ADRD
140	6.51	1.04
145	6.68	1.02
150	6.85	1.00
155	7.03	0.99
160	7.20	1.00
165	7.38	1.02
170	7.55	1.07
175	7.72	1.14
180	7.90	1.24
185	8.07	1.36
190	8.25	1.51
195	8.42	1.68
200	8.60	1.88

Note: ^*−1.6473 is the expected 2‐year effect on HbA1c, conditional on the real‐world utilization rates of GLP1s and SGLT2, and how often they are used in combination with metformin (details of these calculations are in Section 6 of the Technical Appendix). We did not allow individuals’ HbA1c to drop below 7.0, as this is widely considered to be the target goal of diabetes medication therapy ¹⁶ . Effects of GLP1s are as measured in Alexander et al (2022a). Effects of SGLT2 monotherapy is as measured in Musso et al (2012). Effects of SGLT2 combined with metformin is as measured in Alexander et al (2022b). Percent of GLP1 and SGLT2 users with combination metformin utilization is as measured in Rodriguez et al (2024). Baseline utilization of GLP1s and SGLT2s is 2024 utilization in Komodo Healthcare Map data. Relationship between HbA1c and dementia is as measured in Figure S5 of Crane et al (2013).

Abbreviations: ADRD, Alzheimer's disease and related dementias; GLP1, glucagon‐like peptide 1 receptor agonists; HbA1c, glycated hemoglobin; HR, hazard ratio; SGLT2, sodium‐glucose cotransporter‐2 inhibitors.

The effect of HbA1c on ADRD risk among people with diabetes was as measured in the ACT Study, a decades‐long cohort study of aging and cognition, and reported in Figure S5 of Crane et al (2013). ⁴ Our simulations use the point estimates from Crane et al. For example, an individual with diabetes and HbA1c equal to 8.25 has a 51% higher risk of ADRD than someone with diabetes and HbA1c equal to 7.20 (Table 1B). These results are supported by other studies, which also show that higher HbA1c is associated with elevated ADRD risk among people with diabetes. ³⁴ In scenarios A and B, the eligible individuals were thus given a reduction in their ADRD probability that was commensurate with their reduction in HbA1c (full details in Sections 8, 9, and 10 of the Technical Appendix).

For ineligible individuals, who could not receive the intervention, we made no changes to HbA1c; thus, there were no changes in their probability of ADRD incidence. This includes all the individuals in the status quo scenario.

2.5. Simulations

For each scenario, we simulated 40 years of health outcomes for people with diabetes, both with (eligibility scenarios A and B) and without (status quo) the intervention. Each simulation was repeated in 500 Monte Carlo replications. ADRD, and all other outcomes in the FEM, were simulated in a first‐order Markov chain with 2‐year time steps, with transition probabilities estimated from the HRS public use dataset. ³⁰ , ³¹ ADRD was an absorbing state, with its probability in each time period predicted as a function of health and demographic characteristics (Section 9 of the Technical Appendix); this probability was reduced for eligible individuals in scenarios A and B (Section 10 of the Technical Appendix). The initial health states and personal characteristics in the FEM come from HRS data, and each individual was weighted according to demographic characteristics, so that the simulation results represent the entire US population with diabetes aged 51 and older. In our simulation, we predict outcomes for a single cohort of individuals up until their death, and report on cumulative ADRD outcomes in 10‐year intervals. The technical details of the FEM are described in Goldman et al (2018), ²⁵ and the details of our simulation are in the Section 11 of the Technical Appendix.

3. RESULTS

Our simulation predicted outcomes for a cohort of 4,731 individuals with diabetes, whose initial characteristics were measured in 2006–2016 (Table 2). Scaled to the US population, this sample represents 21 million people. The average age was 66.4 years, and 1,051 (22%) were non‐Hispanic Black, 788 (17%) were Hispanic (any race), and 2,700 (57%) were non‐Hispanic White adults. Compared to White individuals, Black and Hispanic individuals were younger on average, but exhibited several characteristics consistent with a higher risk of ADRD: lower educational attainment, lower TICS, and higher body mass index (BMI). ¹ Black and Hispanic individuals also had higher baseline HbA1c and higher rates of insulin use than White individuals, suggesting more advanced diabetes. ¹⁶

TABLE 2.

Sample characteristics of HRS respondents with diabetes, 2006–2016.

N (%)	All	Black	Hispanic	White
N	4,731	1,051	788	2,700
Female	2,549 (53.9%)	633 (60.2%)	449 (57.0%)	1,371 (50.8%)
Memory‐related diseases	109 (2.3%)	17 (1.6%)	18 (2.3%)	67 (2.5%)
Light physical activity	4,372 (92.4%)	971 (92.4%)	728 (92.4%)	2,496 (92.4%)
Moderate physical activity	3,575 (75.6%)	803 (76.4%)	603 (76.5%)	2,011 (74.5%)
Vigorous physical activity	1,619 (34.2%)	355 (33.8%)	294 (37.3%)	889 (32.9%)
Hypertension	3,673 (77.6%)	917 (87.3%)	582 (73.9%)	2,032 (75.3%)
Smoked (ever)	2,759 (58.3%)	629 (59.8%)	431 (54.7%)	1,595 (59.1%)
Stroke (ever)	472 (10.0%)	115 (10.9%)	56 (7.1%)	288 (10.7%)
Cholesterol medication use	3,108 (66.0%)	674 (64.4%)	494 (62.7%)	1,814 (67.6%)
Insulin use	904 (20.6%)	271 (27.2%)	142 (19.9%)	454 (18.1%)
Mean (SD)
Age (years)	66.4 (9.66)	63.5 (8.19)	63.1 (8.58)	68.9 (9.85)
Education (years)	12.4 (3.15)	12.6 (2.55)	9.6 (4.30)	13.0 (2.51)
HbA1c	6.9 (1.50)	7.2 (1.66)	7.3 (1.76)	6.7 (1.27)
TICS score	14.8 (3.76)	13.6 (3.62)	13.8 (3.45)	15.6 (3.71)
BMI	31.6 (6.78)	32.4 (7.26)	31.6 (6.30)	31.3 (6.67)
ADL—no limitations	0.8 (0.43)	0.7 (0.44)	0.7 (0.45)	0.8 (0.42)
ADL ‐ 1 limitation	0.1 (0.32)	0.1 (0.31)	0.1 (0.32)	0.1 (0.32)
ADL ‐ 2 limitation	0.1 (0.22)	0.1 (0.24)	0.1 (0.25)	0.0 (0.21)
ADL ‐ 3+ limitations	0.1 (0.26)	0.1 (0.29)	0.1 (0.31)	0.1 (0.23)
IADL—No limitations	0.8 (0.38)	0.8 (0.39)	0.8 (0.41)	0.8 (0.37)
IADL ‐ 1 limitation	0.1 (0.30)	0.1 (0.30)	0.1 (0.31)	0.1 (0.29)
IADL ‐ 2+ limitations	0.1 (0.27)	0.1 (0.28)	0.1 (0.31)	0.1 (0.24)

Note: Sample is HRS respondents with diabetes, without dementia, who participated in the biomarker portion of the core data collection from 2006–2016.

Abbreviations: ADL, activities of daily living; HbA1c, glycated hemoglobin; HRS, Health and Retirement Study; IADL, instrumental activities of daily living; TICS, telephone interview for cognitive status.

The number of new users of GLP1s or SGLT2s in each of the scenarios is shown in Table 3. Scenario A has 5.317 million new users (1.403 million Black, 1.130 million Hispanic, and 2.507 million White adults). Scenario B has 2.577 million new users (783,000 Black, 640,000 Hispanic, and 1.003 million White adults).

TABLE 3.

ADRD cases and years for people with diabetes in the United States, simulated for 40 years in scenarios with increased utilization of GLP1s and SLGT2s.

All races and ethnicities			Access improvement eligibility
	Year	Status quo	A. HbA1c > 7	B. HbA1c > 8
New users (thousands, %)		0	5317 (25.3%)	2577 (12.3%)
ADRD cases (thousands, % of original sample)	+10	2818 (13.4%)	2617–2742 (12.5%–13.1%)	2644–2752 (12.6%–13.1%)
	+20	5738 (27.3%)	5278–5572 (25.1%–26.5%)	5335–5594 (25.4%–26.6%)
	+30	8285 (39.4%)	7552–8031 (36.0%–38.2%)	7638–8064 (36.4%–38.4%)
	+40	9717 (46.3%)	8819–9417 (42.0%–44.8%)	8925–9458 (42.5%–45.0%)
ADRD person years (thousands)	+10	4448	4137–4330	4179–4346
	+20	10345	9537–10050	9637–10089
	+30	15689	14321–15212	14483–15274
	+40	19086	17331–18496	17539–18576

Black			Access improvement eligibility
	Year	Status quo	A. HbA1c > 7	B. HbA1c > 8
New users (thousands, %)		0	1403 (30.1%)	783 (16.8%)
ADRD cases (thousands, % of original sample)	+10	565 (20.2%)	511–545 (18.3%–19.5%)	517–547 (18.5%–19.6%)
	+20	1176 (42.1%)	1045–1128 (37.4%–40.4%)	1056–1133 (37.8%–40.5%)
	+30	1684 (60.2%)	1479–1614 (52.9%–57.7%)	1496–1620 (53.5%–58.0%)
	+40	1925 (68.9%)	1678–1842 (60.0%–65.9%)	1698–1850 (60.7%–66.2%)
ADRD person‐years (thousands)	+10	889	806–857	814–860
	+20	2112	1883–2029	1904–2036
	+30	3201	2818–3069	2850–3081
	+40	3795	3312–3633	3351–3648

Hispanic			Access improvement eligibility
	Year	Status quo	A. HbA1c > 7	B. HbA1c > 8
New users (thousands, %)		0	1130 (32.3%)	640 (18.3%)
ADRD cases (thousands, % of original sample)	+10	557 (20.1%)	492–533 (17.8%–19.2%)	498–535 (18.0%–19.3%)
	+20	1216 (43.9%)	1053–1158 (38.0%–41.8%)	1066–1163 (38.5%–42.0%)
	+30	1825 (65.9%)	1555–1735 (56.1%–62.6%)	1575–1743 (56.9%–62.9%)
	+40	2161 (78.0%)	1827–2056 (65.9%–74.2%)	1849–2065 (66.7%–74.5%)
ADRD person‐years (thousands)	+10	866	767–829	775–833
	+20	2163	1883–2063	1905–2072
	+30	3435	2934–3266	2972–3281
	+40	4243	3590–4035	3634–4054

White			Access improvement eligibility
	Year	Status quo	A. HbA1c > 7	B. HbA1c > 8
New users (thousands, %)		0	2507 (20.9%)	1003 (8.4%)
ADRD cases (thousands, % of original sample)	+10	1708 (11.1%)	1622–1675 (10.5%–10.9%)	1638–1682 (10.6%–10.9%)
	+20	3418 (22.1%)	3226–3347 (20.9%–21.7%)	3259–3360 (21.1%–21.8%)
	+30	4939 (32.0%)	4615–4822 (29.9%–31.2%)	4667–4842 (30.2%–31.4%)
	+40	5857 (37.9%)	5440–5711 (35.2%–37.0%)	5508–5737 (35.7%–37.2%)
ADRD person‐years (thousands)	+10	2708	2574–2656	2600–2666
	+20	6180	5843–6054	5902–6077
	+30	9341	8743–9124	8840–9162
	+40	11483	10675–11198	10807–11247

Note: Results of microsimulations of the US population with diabetes, projected for 40 years after hypothetical intervention. Status quo projects ADRD outcomes under 2024 utilization of diabetes medications. Columns A and B show cases if all people with HbA1c above the noted threshold use GLP1s and SGLT2s. The ranges in columns A and B reflect adherence varying from 100% to 40% (percent of new users maintaining PDC ≥ 80% after 18 months of use).

Abbreviations: ADRD, Alzheimer's disease and related dementias; GLP1, glucagon‐like peptide 1 receptor agonists; HbA1c, glycated hemoglobin; PDC, proportion of days covered; SGLT2, sodium‐glucose cotransporter‐2 inhibitors.

The results of our microsimulation, scaled to represent the entire US population of people with diabetes, are shown in Figure 2 and Table 3. Figure 2 depicts cumulative ADRD cases averted from increasing utilization of GLP1s and SGLT2s, in scenarios A and B, as compared to the status quo. Among all races and ethnicities, eligibility for individuals with HbA1c > 7 (scenario A), led to 76,000–201,000 cases averted in 10 years, compared the status quo (2.818 million cases). With eligibility for individuals with HbA1c > 8 (scenario B), there were 66,000–174,000 cases averted in 10 years. After 40 years, the cases averted grew to 300,000–898,000 in scenario A, and 259,000–792,000 in scenario B, as compared to the status quo (9.717 million cases). Scenarios A and B also led to large reductions in the total number of person‐years lived with ADRD: after 40 years, the status quo projection predicts 19.086 million ADRD person‐years, compared to 17.331–18.496 million in scenario A, and 17.539–18.576 million in scenario B (Table 3). Results did not meaningfully change when increasing  from 250 to 500 Monte Carlo replications. The 95% confidence intervals of these estimates are reported in Table S1. Additionally, simulation results assuming 80% adherence are shown in Table 2.

FIGURE 2.

Cumulative ADRD cases averted from increasing utilization of GLP1s and SGLT2s. ADRD cases averted after hypothetical increased utilization of GLP1s and SGLT2s, as compared to ADRD projections under status quo utilization of GLP1s and SGLT2s. Results are from microsimulations of the US population with diabetes, projected for 40 years after hypothetical intervention. Scenarios A and B show cases averted if all people with HbA1c above the noted threshold use GLP1s and SGLT2s. 95% confidence intervals calculated from 500 Monte Carlo replications. Adherence is the percent of new users with PDC ≥ 80% after 18 months. ADRD, Alzheimer's disease and related dementias; GLP1s, glucagon‐like peptide 1 receptor agonists; HbA1c, glycated hemoglobin; PDC, proportion of days covered; SGLT2s, sodium‐glucose cotransporter‐2 inhibitors.

Scenarios A and B gave increased utilization for all races and ethnicities, but the benefits regarding ADRD risk were the largest for Black and Hispanic individuals. In the status quo simulation, the percent of people with diabetes having ADRD incidence in 40 years was 69% for Black, 78% for Hispanic, and 38% for White (Table 3) individuals. But in scenario A, with increased use of GLP1s and SGLT2s, the percent with ADRD incidence after 40 years was 60%–66% for Black, 66%–74% for Hispanic, and 35%–37% for White individuals.

Additional results of the microsimulation are shown in Table 4, which depicts the average treatment effect on the treated in scenario B, in terms of the ADRD incidence over 10 and 40 years. Among individuals who received the intervention in scenario B, 10‐year ADRD incidence dropped from 15.8% to 7.5% among Black individuals (a 53% decrease), from 15.9% to 9.0% among Hispanic individuals (a 43% decrease), and from 9.3% to 5.8% among White individuals (a 37% decrease).

TABLE 4.

ADRD incidence among individuals with HbA1c > 8 who ever received intervention (Scenario B).

Parameter	All (n = 753)	Black (n = 228)	Hispanic (n = 189)	White (n = 291)
10‐year ADRD incidence
Status quo	12.6%	15.8%	15.9%	9.3%
Increased utilization	7.0%	7.5%	9.0%	5.8%
40‐year ADRD incidence
Status quo	45.5%	56.1%	58.7%	31.8%
Increased utilization	28.6%	36.8%	36.0%	19.9%

Note: Results of microsimulations of the US population with diabetes. Status quo projects ADRD under 2024 utilization of diabetes medications. Increased utilization depicts ADRD if all individuals with HbA1c > 8 used GLP1s or SGLT2s.

Abbreviations: ADRD, Alzheimer's disease and related dementias; GLP1, glucagon‐like peptide 1 receptor agonists; HbA1c, glycated hemoglobin; SGLT2, sodium‐glucose cotransporter‐2 inhibitors.

4. DISCUSSION

This study used microsimulation to examine scenarios with increased access to GLP1s and SGLT2s for people with diabetes, and measured resulting reductions in ADRD incidence that occur via lower HbA1c. We show that increased utilization of these drugs can lead to as high as a 7.1% reduction in 10‐year ADRD incidence, which amounts to 201,000 cases averted. The benefit was largest for those at higher risk of ADRD, including Black and Hispanic individuals and people with HbA1c > 8. Also, good adherence was critical to achieving the maximum benefit. These results emphasize the importance of ADRD risk factor management as a mechanism for preventing ADRD.

For individuals who initiated GLP1s or SGLT2s, the benefit was large: 10‐year ADRD incidence dropped from 12.6% to 7.0%, and 40‐year incidence dropped from 45.5% to 28.6% (Table 4). However, our results show that medication adherence is important to achieving the benefits of diabetes pharmacotherapy. In our simulation, 100% adherence for new users led to much larger reductions in ADRD risk than 40% adherence (Figure 1). A total of 40% reflects real‐world estimates of the percent of GLP1 users who maintain a proportion of days covered (PDC) > 80% for at least 18 months. ²⁹ Given the large difference in benefit between these two adherence levels, it is imperative that diabetes management efforts acknowledge and adjust for factors associated with adherence to diabetes medications, including adverse events, baseline HbA1c, patient costs, insurance status, health literacy, and comorbidities. ³⁵ , ³⁶ The use of newer non‐injectable or long‐acting modalities for GLP1s and SGLT2s are also likely to improve adherence, ³⁷ , ³⁸ and thus lead to larger reductions in ADRD risk.

In both simulated scenarios, the benefits of the intervention were greatest for those at the greatest risk of ADRD. Black and Hispanic individuals with diabetes have a 20% risk of ADRD in 10 years, compared to 11% for White people with diabetes (Table 3). This is partially because they have higher HbA1c, on average, than White individuals (Table 2), and higher HbA1c is associated with ADRD. ⁴ The reduction in ADRD risk from increased access to GLP1s and SGLT2s is thus much larger for Black and Hispanic individuals, especially if the access is targeted at individuals with higher HbA1c. Both scenarios also showed a slightly larger benefit from improved access for Hispanic than for Black individuals. This is primarily because among American ages 51 and higher with diabetes, Hispanic individuals have higher HbA1c on average (Table 2), meaning that they were more likely to receive the intervention in both scenarios. Additionally, since the effect of reduced HbA1c on ADRD risk is larger at higher levels of HbA1c (Table 1B), ⁴ Hispanic individuals enjoyed a larger benefit of increased utilization.

The increased benefit for people with higher HbA1c is reflected in the results of scenarios A and B. Scenario A, which initiated use for individuals with HbA1c > 7 (5.317 million people), led to 76,000–201,000 cases averted in 10 years. Scenario B had less than half as many new users (2.577 million), but still averted 66,000–174,000 cases in 10 years. In other words, despite scenario B having 52% fewer new users, the ADRD risk reduction benefit decreased by only 13%. This reflects the higher ADRD risk of people with higher HbA1c, and emphasizes the importance of targeting interventions toward individuals who will benefit the most. In both scenarios, the number of new users is large, and we acknowledge that achieving these changes would require interventions that alleviate multiple access barriers (e.g., financial, geographic, socio‐cultural, health system) and alter clinical decision‐making.

In this study, the effects of improved GLP1 access on ADRD risk only operated through an intermediate effect on HbA1c. This may be an underestimate of the benefits of these drugs, as GLP1s promote protective effects on vascular, microglial, and neuronal functions, and thus may be effective against ADRD, independent of their role in diabetes management. ¹⁷ Indeed, there is new empirical evidence that found an association between GLP1s and reduced ADRD risk, independent of their effects on HbA1c and body mass index. ¹⁸ If the additional neuroprotection suggested in these studies were incorporated into our simulation, the number of ADRD cases averted in scenarios A and B would likely be higher.

This paper contributes to the literature on the benefits of preventing or delaying ADRD. Such efforts can lead to large economic effects; ³⁹ for example, by 2050, a 1‐year delay in AD onset will result in savings of $160 billion in formal healthcare costs, and $59 billion in informal caregiving costs. ⁴⁰ There are also significant benefits to partners and spouses of people with ADRD, who experience higher healthcare expenditures 5 years post‐onset, ⁴¹ and face a large caregiving burden. ⁴² Despite a recent decline in ADRD incidence, the gains have been unevenly distributed. ⁴³ Our results suggest that prevention efforts will have the largest benefit if they target disadvantaged communities, who tend to have worse management of ADRD risk factors like diabetes, and the highest risk of ADRD.

Strengths of this study include the application of rigorous estimates of real‐world relationships between diabetes therapy, HbA1c, and ADRD, and the use of a well‐validated microsimulation model. ⁴ , ²⁶ , ²⁷ , ²⁸ , ³⁴ Limitations include that our simulations do not incorporate all the benefits of GLP1s and SGLT2s. This is because we only allow utilization of these drugs to affect ADRD via HbA1c. We do not incorporate the effects of the drugs on other outcomes that may affect ADRD risk (e.g., BMI, blood pressure), and we do not incorporate the effects of HbA1c control on diabetes complications and other outcomes that may affect ADRD risk (e.g., stroke). This implies that our predicted benefits with regard to ADRD risk are lower bounds. Relatedly, this study did not examine healthcare costs; future studies should do so, and account for expected reductions in the costs of GLP1s and SGLT2s as they become generically available. Another limitation is that we cannot observe if individuals in our sample had previously used GLP1s or SGLT2s; we therefore discounted the population‐level probability of initiating by 0.751, which reflects the percentage of people with diabetes not using either drug (see Section 4 of the Technical Appendix). Also, we did not allow initiation of GLP1s or SGLT2s to affect an individual's probability of death, except through ADRD. This ignores potential other benefits of the drugs that might affect mortality (e.g., via blood pressure control). The relationship between HbA1c and ADRD in this study, as measured in Crane et al (2013), did not distinguish between subtypes of dementia. ⁴ While AD and probable AD made up 77% of the ADRD cases in Crane et al, it is possible that the association would be different for vascular dementia or other subtypes. Other recent evidence has indeed suggested that glucose control is even more protective against vascular dementia than AD dementia. ⁴⁴ , ⁴⁵ , ⁴⁶ Another limitation of the HbA1c‐ADRD relationship measurement in Crane et al (2013) is the racial and ethnic homogeneity of that study sample (82% of participants with diabetes were White individuals); it is possible that this relationship varies across race and ethnicity.

5. CONCLUSIONS

Widespread use of GLP1s and SGLT2s by people with diabetes can lead to large reductions in long‐term risk of ADRD. The relationships between these drugs and HbA1c, and between HbA1c and ADRD, are well established; simulation modeling allows us to quantify the benefits from scenarios that increase utilization of these drugs for specific subsets of adults with diabetes. Importantly, these benefits are especially large for Black and Hispanic individuals and people with HbA1c > 8. This work can guide policy initiatives and care strategies aimed at preventing or delaying ADRD by enhancing the management of its risk factors, while also highlighting the advantages of expanding access to treatments for chronic conditions more generally. Subsequent research should more completely model the complex relationships between a larger set of ADRD risk factors, their treatments, and long‐term costs and outcomes.

CONFLICT OF INTEREST STATEMENT

The authors have no conflicts of interest to declare. Author disclosures are available in the Supporting Information.

CONSENT STATEMENT

This study did not involve human subjects (no interaction or intervention with subjects, and all data was de‐identified); this study therefore did not require Institutional Review Board approval.

Supporting information

Supporting Information

Supporting Information