Cost-effectiveness of the National Diabetes Prevention Program: A Real-world, 2-Year Prospective Study Free

In 2021, diabetes affected 14.7% of the U.S. adult population (1). The cost of diabetes has increased over time in the U.S., and in 2022, the total direct medical cost of diagnosed diabetes was $307 billion (2). After adjustment for age and sex, people with diabetes have medical expenditures 2.6 times higher than those without diabetes (2). Prevention of diabetes and its complications in people with prediabetes can decrease morbidity and mortality and reduce medical expenditures associated with diabetes. In 2002, the Diabetes Prevention Program (DPP) clinical trial demonstrated that over 3 years, an intensive lifestyle intervention compared with placebo led to a 58% reduction in diabetes incidence in people with prediabetes (3). A within-trial cost-effectiveness analysis found that over 3 years, the lifestyle intervention was cost-effective with an incremental cost-effectiveness ratio (ICER) of $32,029 per quality-adjusted life-year (QALY) gained from a health system perspective (4). Subsequently, two within-trial cost-effectiveness analyses reported that over 10 years, the lifestyle intervention remained cost-effective with an ICER of $12,878 to $19,988 per QALY gained (5,6).

In 2021, prediabetes affected 38% of the U.S. adult population (1), providing an opportunity to intervene for diabetes prevention. Given the importance of diabetes prevention and the apparent cost-effectiveness of the DPP, the Centers for Disease Control and Prevention (CDC) launched the National DPP (NDPP) in 2010 to create a national lifestyle change program for diabetes prevention (7). Since April 2018, the Centers for Medicare & Medicaid Services have paid for CDC-recognized DPPs for eligible Medicare beneficiaries in both clinical and community settings (8).

Several cost-effectiveness evaluations of the DPP and translational diabetes prevention interventions have been reported. In a systematic review of 16 studies by Li et al. in 2015 (9), they reported a median ICER of $13,761 per QALY gained from a health system perspective. Three studies that evaluated translational implementation of the DPP reported a median ICER of $5,494 per QALY gained. These estimates are considered cost-effective by the standard threshold of $100,000 per QALY gained. Zhou et al. (10), in a meta-analysis, reported that from a health system perspective, lifestyle interventions following the DPP curriculum had a median ICER of $6,212 per QALY gained, whereas those not following the DPP curriculum had a median ICER of $13,228 per QALY gained. Lifestyle interventions provided in group settings or by a combination of health professionals and lay health workers had lower ICERs compared with those delivered one-on-one or by health professionals alone (10). Another cost evaluation of a translational DPP provided by lay health educators in rural senior centers found it to be effective at producing weight loss at a low implementation cost of $165 per participant (11). More recently, the NDPP has been translated into digital platforms. A simulation study comparing a digital DPP with a small group educational intervention found the digital DPP to be cost-saving for patients with prediabetes from a health system perspective over a 10-year time horizon (12).

Previous cost-effectiveness evaluations often were conducted as a part of clinical trials or using simulation modeling of clinical trial results (9,10,13). There is limited information on the cost-effectiveness of the NDPP in real-world settings. In this study, we evaluated the 2-year cost-effectiveness of the NDPP using real-world empirical data. The results are important in clarifying the NDPP’s real-world economic impact and informing decisions about its implementation, scalability, and sustainability.

The University of Michigan (U-M) is a large public research university in Ann Arbor, MI, with satellite campuses in Flint and Dearborn. Beginning in 2015, the U-M offered the NDPP at no out-of-pocket cost to university employees, dependents, and retirees ≥18 years of age with prediabetes and overweight or obesity who belonged to the university’s self-funded health insurance program. Prediabetes was defined by either a health plan claims diagnosis of prediabetes (Supplementary Table 1) or a HbA_1c level of 5.7%–6.4% in individuals without a history of diabetes defined as at least one inpatient encounter or two outpatient encounters with a diagnosis of diabetes or one dispensing of a glucose-lowering medication other than metformin on an ambulatory basis.

Beginning in 2015 and approximately every 6 months thereafter, the health plan sent letters to individuals newly identified with prediabetes encouraging them to enroll in the NDPP. The dates when enrollees attended their first NDPP session were used as their enrollment dates. The dates that nonenrollees were sent letters inviting them to enroll in the NDPP plus the median number of days to enrollment after the date of the invitation letter for individuals from the same cohort who enrolled in the NDPP were used as the index dates for nonenrollees. We used the enrollment dates for enrollees and the index dates for nonenrollees to establish baseline, 1-year, and 2-year follow-up time windows.

A total of 7,846 individuals with prediabetes were identified and 753 enrolled in the NDPP (Fig. 1). For enrollees and nonenrollees, we used data from the health plan and surveys to characterize the individuals. Baseline comorbidities were defined as having a claims diagnosis for the conditions in the 3 years before the date of the initial contact letter. The codes used to identify comorbidities are shown in Supplementary Table 1. Baseline values of BMI, blood pressure, lipids, and HbA_1c were the last measured values in the year before the enrollment/index date. One-year and 2-year follow-up values were the last measured values in the interval between the enrollment/index date plus 365 days and between 365 and 730 days, respectively.

We surveyed all enrollees and a random sample of nonenrollees at baseline, 1, and 2 years to assess their EuroQol 5-dimension 5-level questionnaire (EQ-5D-5L) health utility scores and EQ-5D-5L visual analog scale scores (14,15) (Fig. 1). This economic analysis focused on the 575 enrollees and 5,373 nonenrollees identified between 1 August 2015 and 28 February 2018 (Fig. 1). We excluded individuals whose 2-year follow-up overlapped with the COVID-19 pandemic, because health care utilization and costs changed dramatically during that time.

This study was reviewed and approved by the U-M Institutional Review Board for Human Research (HUM no. 00108065) and was granted a waiver of documented informed consent.

The primary outcomes were the change in total annual direct medical costs in the 2 years after the enrollment/index date (baseline) compared with the year before baseline, and the change in QALYs calculated using EQ-5D-5L utility scores at baseline, 1, and 2 years of follow-up between NDPP enrollees and nonenrollees. Direct medical costs were provided by the health plan using claims data and allowable amounts paid by the health plan and by patients out-of-pocket. Costs were categorized into eight all-inclusive and mutually exclusive categories: inpatient, outpatient, emergency room, laboratory, diagnostic testing, durable medical equipment, prescription medications, and other (e.g., dental/vision, hearing evaluation/aids, and home health care). The cumulative incidence of diabetes was defined on the basis of an HbA_1c level ≥6.5% or a self-reported diagnosis of diabetes and was assessed over 2 years of follow-up among enrollees and nonenrollees.

The health plan contracted with four organizations that offered CDC-recognized NDPPs including 1) an in-person classroom-based program led by certified diabetes educators in an endocrinology outpatient clinic, 2) an in-person classroom-based program led by trained peer instructors in community settings, 3) an in-person program in a YMCA led by trained lifestyle coaches, and 4) an online digital program with virtual group meetings led by personal health coaches. Organizations received payments from the health plan based on contracted rates and the milestones that each enrollee met. The minimum payments ranged from $50 to $250 per enrollee and the maximum payments ranged from $500 to $960 per enrollee. The weighted cost per enrollee, considering both payment rates and milestones achieved, ranged from $417 to $668.

Enrollees and nonenrollees differed in several characteristics (Supplementary Table 2). We used propensity score weighting to adjust for possible bias arising from self-selection for NDPP enrollment, multiple imputation to handle missing data, and the bootstrap method to produce CIs (16). All analyses were based on the mean of 1,000 point estimates, and the SEs and 95% CIs were produced using 1,000 bootstrapped results.

First, we selected baseline variables associated with enrollment (17) and included them in the propensity score model. The variables included age, sex, race, history of smoking, atrial fibrillation, angina pectoris, BMI, high-density lipoprotein cholesterol, total annual direct medical costs, and the number of childcare services and the percentage of families receiving Supplemental Nutrition Assistance Program benefits in the neighborhood based on 5-digit residential zip codes. We sampled 1,000 bootstrapped data sets with missing data and, based on each of the bootstrapped data sets, we imputed five data sets. For each of the five imputed data sets within each bootstrapped data set, we fit the propensity score model and computed the propensity scores using the step-forward selection procedure based on combined inference at each step with Rubin’s rule (18). We then converted the propensity scores into weights and produced the weighted point estimates for subsequent analyses. We calculated the weighted summary statistics and standardized differences to examine the balance between the two groups after weighting and compared them with the original data. A standardized difference of <0.1 indicates an insignificant between-group difference.

If an individual was not enrolled in the health plan during any days of the month, the monthly cost was set to missing. To impute missing cost data and health utility data, we used multivariate imputation by chained equations and the predictive mean matching method for multiple imputation. All baseline information (Supplementary Table 2) and changes in clinical outcomes were included in the imputation as auxiliary variables. By assuming missing at random, we were able to adjust for potential biases due to differences between survey respondents and nonrespondents. If an individual was enrolled in the health plan at least 1 day during a month but had no cost data, the monthly cost was set to zero. If nonmissing cost data existed for some months during a year, we used the average monthly cost multiplied by 12 months to estimate the annual cost. These annual cost data were considered true observed data.

To compare the enrollees and nonenrollees, the inverse propensity weighted difference-in-differences analysis was performed on patient-year-level data to estimate the effects of the NDPP on costs and health utility scores over 2 years of follow-up. The inverse propensity weighted between-group differences in diabetes incidence and QALYs over 2 years of follow-up were calculated to estimate the NDPP’s effects on diabetes prevention and quality-adjusted life expectancy. Because the baseline inverse propensity weighted health utility scores differed between enrollees and nonenrollees, we used a regression model to adjust for baseline health utility scores to assess the difference in EQ-5D-5L utility scores at 2 years and QALYs over 2 years between the two groups. The ICER as cost per QALY gained was estimated by the difference in change of average costs between the two groups divided by the difference in average QALYs between the two groups over 2 years. The cost per case of diabetes prevented was estimated by applying the incremental costs between the two groups to the number needed to treat for preventing one case of diabetes over 2 years. The net monetary benefit (NMB) was calculated by multiplying the incremental QALYs between the two groups by the willingness-to-pay threshold ($100,000 per QALY gained) and then subtracting the incremental costs between the two groups.

We adopted a health care sector perspective and discounted costs and QALYs at 3% annually. All costs were expressed in 2020 U.S. dollars. In uncertainty analyses, we plotted bootstrapped incremental cost-QALY pairs on the cost-effectiveness plane to graphically illustrate uncertainty around ICERs (19). We further constructed a cost-effectiveness acceptability curve to illustrate the probability that the NDPP was cost-effective at a range of cost values or thresholds that decision makers are willing to pay per QALY gained (20,21). All analyses were performed using R, version 4.2.3.

An impact inventory (22) for the components considered in the economic analyses is provided in Supplementary Table 3. The economic analyses were reported in compliance with the Consolidated Health Economic Evaluation Reporting Standards (23), which are listed in Supplementary Table 4.

Of the 7,846 adults with prediabetes, 753 (9.6%) enrolled and 7,093 did not enroll in a NDPP between 2015 and 2019. After excluding three individuals who did not have any cost data available and 1,895 individuals whose enrollment/index dates were after 1 March 2018 (to exclude the COVID-19 period), 575 enrollees and 5,373 nonenrollees remained and were included in this cost-effectiveness analysis (Fig. 1). Supplementary Table 5 compares the characteristics of individuals whose enrollment/index dates were before with those whose dates were after 1 March 2018. Those whose enrollment/index dates were before 1 March 2018 were slightly older, more likely to be women, less likely to be White, had higher BMIs, and were less likely to have atrial fibrillation and to smoke.

Compared with nonenrollees, NDPP enrollees were slightly older, more likely to be women, and less likely to be White (Supplementary Table 2). Enrollees had higher BMIs, but blood pressure, lipids, and HbA_1c levels did not differ between the groups. Enrollees were less likely to have atrial fibrillation and to smoke. Total direct medical costs and health utility scores at baseline did not differ between the groups. After propensity score weighting, the baseline characteristics of enrollees and nonenrollees were similar, except that enrollees had higher health utility scores than nonenrollees (Table 1).

Table 2 shows the discounted costs, QALYs, and diabetes incidence for enrollees and nonenrollees over 2 years. For each NDPP enrollee, the 2-year direct medical costs were $5,549 (95% CI −14,447, 640) lower compared with the costs in the year before enrollment. Cost savings ranged from $3,393 (95% CI −10,993, 966) for hospitalization to $10 (95% CI −45, 23) for laboratory testing. Each nonenrollee also spent $479 (95% CI −2,573, 1,526) less in the 2-year direct medical costs, ranging from $382 less (95% CI −1,564, 806) for hospitalization to $184 more (95% CI −41, 491) for other costs. Compared with nonenrollees, each NDPP enrollee saved an average of $5,070 (95% CI −13,755, 1,504) in 2-year direct medical costs. Even after accounting for the cost of the NDPP ($518 per enrollee), each enrollee saved an average of $4,552 (95% CI −13,231, 2,014) in 2-year total discounted direct medical costs compared with each nonenrollee. The lower costs were mainly attributable to lower costs of hospitalizations, outpatient visits, and emergency room visits. Compared with nonenrollees, each enrollee, on average, had a small increase in EQ-5D-5L utility scores of 0.005 (95% CI −0.027, 0.038) at 2 years. Based on changes in EQ-5D-5L utility scores, enrollees and nonenrollees accrued 1.703 and 1.681 QALYs over 2 years, respectively, with a trivial net change of −0.001 (95% CI −0.022, 0.019) discounted QALY per enrollee over 2 years. The probability of developing diabetes was significantly lower in enrollees (5.09%) compared with nonenrollees (7.85%) over 2 years. The absolute risk reduction was 2.77 percentage points and the number needed to treat with the NDPP to prevent one case of diabetes over 2 years was approximately 36. Compared with nonenrollees, the ICER of NDPP enrollees was approximately $4.6 million saved per QALY lost and $164,000 saved per case of diabetes prevented. Given a willingness-to-pay threshold of $100,000 per QALY gained, the NMB of enrollment in the NDPP was $4,452. Supplementary Table 6 provides estimates of the disaggregated, annual direct medical costs in the year before the enrollment/index date, and in the first and second years after the enrollment/index date. It also shows health utility scores at baseline, 1, and 2 years after the enrollment/index date.

The undiscounted cost-effectiveness analysis results in Supplementary Table 7 are similar to the discounted results reported in Table 2. Compared with nonenrollment, the ICER of enrollment in the NDPP was approximately $4.6 million saved per QALY lost and $160,000 saved per case of diabetes prevented. Given a willingness-to-pay threshold of $100,000 per QALY-gained, the NMB of enrollment in the NDPP was $4,470.

The cost-effectiveness plane based on the 2-year discounted cost-effectiveness analyses (Fig. 2A) shows that 87.6% of incremental cost-QALY points fell in the southeast and southwest quadrants, indicating that enrollment in the NDPP is less costly (cost-saving) than nonenrollment. The cost-effectiveness acceptability curve (Fig. 2B) shows that at a willingness-to-pay threshold of $100,000 per QALY gained, enrollment in the NDPP has an 84% probability of being cost-effective compared with nonenrollment. The results are similar for the 2-year undiscounted cost-effectiveness analyses (Supplementary Fig. 1).

Compared with nonenrollment, enrollment in the NDPP was associated with a significant reduction in the absolute risk of developing diabetes by 2.8 percentage points over 2 years of follow-up. These results are consistent with our previous analyses (24). In the present study, we evaluated the cost-effectiveness of the NDPP among adult employees, dependents, and retirees with prediabetes and employer-sponsored health insurance from a U.S. health care sector perspective. The average reduction in total direct medical costs was approximately $4,600 over 2 years for each enrollee versus each nonenrollee. Cost savings were primarily related to reduced costs for hospitalizations ($3,000 less), outpatient visits ($1,200 less), and emergency room visits ($300 less). The average improvement in EQ-5D-5L utility scores was 0.005 at 2 years for each enrollee compared with each nonenrollee. There was no difference in QALYs over 2 years between enrollees and nonenrollees. Compared with nonenrollment, the ICER of NDPP enrollment was about $4.6 million saved per QALY lost and $160,000 saved per case of diabetes prevented. The NMB of NDPP enrollment was approximately $4,500. The uncertainty analyses indicated that over 2 years, there was an 88% probability that enrollment in the NDPP was cost-saving and an 84% probability that the cost per QALY gained was <$100,000. A willingness-to-pay threshold of <$100,000 per QALY gained (incrementally cost-effective) or a saving of >$100,000 per QALY lost (decrementally cost-effective) is considered to represent good value for money in the U.S. (25–27). In addition, a positive NMB value implies that the additional value of an intervention outweighs the extra cost and, therefore, the NDPP is considered cost-effective.

As shown in the cost-effectiveness plane (Fig. 2A), 87.6% of incremental cost-QALY points fell in the southwest and southeast quadrants, indicating that enrollment in the NDPP was less costly (cost-saving) compared with nonenrollment. There was, however, a larger proportion of incremental cost-QALY points in the southwest quadrant (46.9%; less effective and less costly) than in the southeast quadrant (40.7%; more effective and less costly) or the northeast quadrant (6.3%; more effective and more costly). For this reason, when we increase the willingness-to-pay threshold from $0 (for cost saving) to $50,000 per QALY gained, $100,000 per QALY gained, or $150,000 per QALY gained, we lose more incremental cost-QALY points in the southwest quadrant and gain fewer incremental cost-QALY points in the northeast quadrant. As a result, the slope in the cost-effectiveness acceptability curve (Fig. 2B) is downward.

Although the differences in direct medical costs and in QALYs between enrollees and nonenrollees did not reach statistical significance, it should be remembered that the estimates of incremental costs and benefits are surrounded by substantial uncertainty. Faced with this uncertainty, decision makers will often deliberate on the probability that a new treatment is cost-effective compared with the alternative (20,28). Cost-effectiveness planes and cost-effectiveness acceptability curves provide guidance because they explicitly incorporate the joint uncertainty around costs and benefits to inform decisions (28–30). The smaller the cloud of points in the cost-effectiveness plane, the more certain we are about the estimate of the ICER. As shown in Fig. 2 and Supplementary Fig. 1, the majority of incremental cost-QALY pairs fell in the southwest and southeast quadrants of the cost-effectiveness plane, indicating that enrollment in the NDPP was likely to be less costly (cost saving) than nonenrollment. Moreover, the cost-effectiveness acceptability curves showed that the decision uncertainty surrounding the adoption of the NDPP remained small (<20%) as the willingness-to-pay threshold increased, indicating that there is a high degree of certainty that the NDPP is a cost-effective treatment option for people with prediabetes and overweight or obesity.

Several factors may have contributed to our failure to find statistically significant differences in costs and QALYs. These include small sample size, short duration of follow-up, healthy worker effect, and heterogeneous treatment effects by program type. Generally, medical costs are highly skewed, and large sample sizes are required to detect statistically significant differences even when clinical benefits are apparent (31,32). In our study, we based the sample size calculation on changes in BMI and HbA_1c (24), which may have resulted in being underpowered to detect significant differences in costs. The statistical significance of an inferential test can be scientifically important but not necessarily relevant to pragmatic decisions (33). We demonstrated a compellingly high probability of cost savings or cost-effectiveness over 2 years. Overall net benefits should be given priority when evaluating the cost-effectiveness of a new treatment (33).

We followed participants for only 2 years, which included the 1-year intervention and 1-year observational follow-up. This is a short time frame in which to observe significant changes in costs and health-related quality of life. DPP-like interventions produce improvements in health-related quality of life by delaying or preventing diabetes and diabetes-related complications, which reduce downstream health care costs (9,10). Cost-effectiveness outcomes are sensitive to the analytical time horizon. Studies of effective interventions that adopt a longer time horizon often demonstrate a lower ICER (9,10,34,35). Most previous studies that evaluated the cost-effectiveness of the DPP adopted 3-year to lifetime time horizons (9,10), allowing more time for beneficial effects to become apparent. Our short time horizon may have contributed to our inability to observe statistically significant changes in costs and QALYs.

Although both NDPP enrollees and nonenrollees were at risk for developing diabetes, they were in relatively good health as judged by their high baseline health utility scores (Supplementary Table 6). Our failure to observe a significant impact of NDPP enrollment on health utility scores and QALYs may be due, in part, to the fact that participants’ baseline health utility scores were good, leaving little room for improvement. This phenomenon is termed the healthy worker effect. A systematic review concluded that the EQ-5D is less able to detect health improvements in people with better baseline health (36).

Finally, four NDPPs were offered to our population. There were differences in the focus, location, flexibility, and support provided by these programs (37). Our inability to observe significant differences in costs and QALYs may be explained, in part, by heterogeneous treatment effects due to the differences in programs used by different participants.

We could not identify other real-world economic evaluations of the NDPP performed in a workforce population with prediabetes using individual-level, empirical health insurance claims, and survey data (9,10,13). There are, however, a few studies that examined the cost impact of the NDPP for individuals with prediabetes in real-world settings. Alva et al. (38) used claims data to compute total direct medical costs for fee-for-service Medicare participants in the YMCA-DPP versus a matched comparison group of nonparticipants. They reported an average quarterly savings of $278 per person during the first 3 years after the intervention. Khan et al. (39) studied a cohort of commercially insured adults with prediabetes and compared medical expenditures for those who were newly diagnosed with diabetes and those who were not. They estimated that participation in the NDPP would yield an average annual saving of $2,671 for 3 years. Barthold et al (40). used claims data to examine the effects of a digital DPP on health care costs among Medicare Advantage participants and reported that there were no significant differences in medical or prescription drug costs between enrollees and nonenrollees at 2 years. Sweet et al. (41) studied adults with private insurance and found that each enrollee in a digital DPP had a reduction in health care expenditures of $1,169 in the first year and $630 in the second year after enrollment compared with a matched group of nonenrollees. None of these studies assessed the impact of the NDPP on health utility scores or QALYs. Compared with these studies, the average reduction in total direct medical costs of $4,600 observed over 2 years in our study was at the upper end of the range of reported savings. This may be because both enrollees and nonenrollees in our study had laboratory- or claims diagnosis–confirmed prediabetes and, thus, were at higher risk of progressing to type 2 diabetes than those identified based on the CDC–American Diabetes Association risk screener alone.

Our study had several strengths. First, demographic, diagnosis, laboratory, medical utilization, and cost data were available from the health plan, and health utility scores and diagnoses were available from surveys (42). Second, unlike model-based economic analyses, our study prospectively assessed disease progression, costs, and health utility scores. Our study had several limitations. First, uptake of the NDPP was low (9.6%) but was nearly four times higher than has been observed among U.S. adults with prediabetes who reported being told by health professionals that they had prediabetes (2.5%) (43–45). Second, the study was nonrandomized, and NDPP enrollees were likely to be more highly motivated than nonenrollees. To address this limitation, we applied propensity score weighting to minimize baseline differences between enrollees and nonenrollees. Third, due to the unavailability of data on time lost from work and usual activities, we only examined the cost-effectiveness of the NDPP from a health care sector perspective and not from a societal perspective. The health care sector perspective does, however, have policy implications because most administrators make decisions regarding the allocation of resources based on an intervention’s impact on their enrollees’ health outcomes and downstream medical costs. Fourth, the unavailability of health care utilization data limited our ability to explore the mechanism of cost savings associated with the NDPP. Cost savings were primarily related to reduced costs for hospitalizations, but we were unable to determine whether this was due to lower rates of hospitalization, shorter hospital stays, or differences in hospital treatments. Fifth, we only assessed EQ-5D-5L and visual analog scale scores in a random sample of nonenrollees, and not all surveyed enrollees and nonenrollees responded to the survey. To address these limitations, we applied multiple imputation to handle missing data and the bootstrap method to estimate CIs. Finally, our study was performed in a single employer group with prediabetes and, thus, its generalizability to other populations is unknown.

In summary, the DPP and NDPP have been proven to delay or prevent the development of diabetes among people with prediabetes in clinical trial and real-world study settings (24,37,46–49). We have extended these findings using individual-level, empirical data to assess the cost-effectiveness of the NDPP in a self-insured population with prediabetes. Our results suggest that the NDPP enrollees, compared with nonenrollees, are likely to save direct medical costs over 2 years. If delivered to a larger prediabetes population with longer-term follow-up, the NDPP could potentially reduce health care costs and improve QALYs for millions of American adults with prediabetes. Further research is warranted to confirm our findings and their generalizability to other populations and settings.

Acknowledgments. The authors thank Marsha Manning, Medical Benefits and Strategy, University of Michigan; Ashley Weigl, MHealthy; and Marc D. Keshishian and Dawn Beaird, Blue Cross Blue Shield of Michigan, for their contributions to this project.

Funding. This study was supported by grants R01DK109995 and P30DK092926 (to the Michigan Center for Diabetes Translational Research) from the National Institute of Diabetes and Digestive and Kidney Diseases and by grant U18DP006712 from the Centers for Disease Control and Prevention.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. S.K. and W.Y. researched the data, contributed to writing the manuscript, and reviewed and edited the manuscript. D.W. researched the data, performed statistical analyses, assisted in writing the manuscript, and reviewed the manuscript. L.N.M. researched the data, contributed to writing the manuscript, and reviewed and edited the manuscript. C.V.S. contributed to writing the manuscript, contributed to the discussion, and reviewed the manuscript. W.H.H. designed the study, researched the data, contributed to the discussion, and reviewed and edited the manuscript. All authors approved the final version of the manuscript. S.K. and W.H.H. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analyses.

Prior Presentation. This work was presented at the 84th Scientific Sessions of the American Diabetes Association, Orlando, FL, 21–24 June 2024.

Handling Editors. The journal editors responsible for overseeing the review of the manuscript were Elizabeth Selvin and Neda Laiteerapong.