Equity-Focused Evaluation of a Medicaid-Funded Statewide Diabetes Quality Improvement Project Collaborative

The prevalence of diabetes in the U.S. is 11.6%, with 29.7 and 8.7 million with diagnosed and undiagnosed diabetes, respectively (1). Over the last decade, the incidence has been increasing and prevalence of diabetes has been higher in historically marginalized racial and ethnic minority groups including non-Hispanic Black (NHB) and Hispanic American communities throughout the U.S. (2,3). Unfortunately, only approximately 20% of individuals meet recommended diabetes targets for glycemia (hemoglobin A_1c [A1C] <7%, non-HDL cholesterol <130 mg/dL, and blood pressure <140/90 mmHg), and the percentage is lower with use of the current blood pressure target (<130/80 mmHg) (4). Long-standing disparities in attaining glycemic goals exist in the U.S., with worse A1C control in racial and ethnic minority populations (5,6). Hypothesized root causes of diabetes disparities include differences in socioeconomic, clinical, health care, and self-management factors, along with higher levels of diabetes distress impacting racial and ethnic minority groups, but confirmatory evidence remains limited and mixed (3,5,7–9).

The prevalence of adults diagnosed with diabetes in Ohio was 13.1% in 2022 (10). From 2016 to 2020, diabetes was the 7th leading cause of death in Ohio, and Ohio’s diabetes age-adjusted death rate ranked 9th of 50 states with an age-adjusted diabetes mortality of 25.8 per 100,000 (11). Consistent with U.S. population data significant racial disparities exist, with the highest prevalence of diabetes among NHB adults at 17.5%, compared with 12.8% among non-Hispanic White (NHW) adults, 11.7% Hispanic or Latino adults, and 10.7% multiracial non-Hispanic adults in 2020 (10). Mortality rate is also 1.8-fold higher for NHB Ohioans than for NHW adults (11).

Given the significant impacts of diabetes on health and health equity among Ohioans, the Ohio Department of Medicaid (ODM) partnered with the Ohio Colleges of Medicine Government Resource Center (GRC), Ohio Medicaid managed care plans (MCPs), and seven medical schools in Ohio to develop a statewide diabetes collaborative focused on improving diabetes outcomes and advancing diabetes equity. The Diabetes Quality Improvement Project (QIP) overall goal was to improve A1C levels and reduce the rate of complications and costs for Medicaid enrollees with diabetes, due to 44.5% of individuals having A1C >9% among Ohio Medicaid enrollees in 2022 (12). The SMART (specific, measurable, achievable, relevant, and time-bound) aims of the project were for participating practices to 1) reduce the percentage of patients with diabetes who have A1C >9% by 15% from 25% to 21% (13) and 2) reduce the percentage of NHB and Hispanic patients with A1C >9% by 20% from 23.0% to 18.4% and 28.6% to 22.9%, respectively, by 30 June 2022. Practice engagement reflected an emphasis on serving high volumes of patients enrolled in or eligible for Medicaid. The overall results have previously been reported and showed that the percentage of patients with A1C >9% improved from 25% to 20% (13). In the current report we focus on the disparities SMART aim, discussing the strategies to improve control in racial and ethnic minority populations, along with the resulting process and clinical outcomes across racial and ethnic populations.

The statewide diabetes collaborative (ODM, GRC, MCPs, and the seven medical schools in Ohio) participated in a planning year using the collective impact model as a mechanism for shared power to address complex challenges (13–15), which includes five key elements: 1) a common agenda (i.e., shared vision, mission, and project aims), 2) shared measurement (electronic health record [EHR] data queries), 3) mutually reinforcing activities (intervention and implementation strategies described further below), 4) continuous communication (routine Steering Committee meetings every 2 weeks), and 5) backbone support (organizational structure and roles for all partners) (15). The process led to the development of a key driver diagram and toolkit (Supplementary Material) to reflect the theory for improvement and guide the quality improvement (QI) activities (13,15). An overview and timeline of the project are shown in Fig. 1, and details of the aims, key drivers, and interventions can be found in Supplementary Fig. 1.

Collaborative partners recruited 20 primary care clinics across 11 academic and nonacademic health systems to participate in the Diabetes QIP from July 2019 through May 2020. We used purposeful sampling of clinics with ≥30% Medicaid-insured population or ≥50% Medicaid- or Medicare-insured or uninsured populations across the state of Ohio. The recruitment period included engagement at the site level, discussing project activities, and execution of needed data use agreements. All participating clinics served a high volume of Medicaid patients, and health systems were given a range of stipends to submit EHR data (15). The stipends were calculated based on the no. of practices participating within a health system, and invoices were paid in 6-month increments based on timely data submission and completion of deliverables for the project. MCPs and primary care practices worked together to align their respective QI projects to catalyze improvements in outcomes. For example, in response to barriers to diabetes care voiced by providers and patients, Medicaid payers added coverage for diabetes self-management education and support, piloted home A1C testing, removed prior authorization requirements for continuous glucose monitors within select populations, aligned quantity limits on diabetes supplies, and aligned formularies with no step therapy or prior authorization required for at least one drug in each class.

Implementation strategies included audit and feedback, peer-to-peer learning, QI coaching/practice facilitation, and subject matter expert consultation (15). After the planning year, a half-day virtual kickoff was held with the participating primary care practices, Medicaid MCPs, and other partners. This was followed by QI coaching calls with each of the practices and monthly 1-h webinars or “action period calls” with the practices and MCPs to share aggregate practice-level data, discuss evidence-based best practices, and promote peer-to-peer learning (15). Additionally, there were two virtual learning sessions held to increase peer-to-peer learning. Lastly, we held monthly 1-h collaborative calls with the MCPs and a subset of clinical practices and partners to enhance payer and clinic collaborative planning.

For addressing race- and ethnicity-based disparities in diabetes follow-up visits, A1C testing, and achievement of A1C goals, in the Diabetes QIP emphasis was placed on 1) standardized processes, 2) provider-patient engagement, 3) addressing nonmedical health-related social needs, and 4) addressing the impact of racism on care. First, standardized processes (16) were aimed to reduce implicit biases (17,18), with use of both the Five Whys to identify root causes (19) and positive deviance approaches (20). The Five Whys is an interrogative problem-solving technique used to explore cause and effect relationships and determine the root cause by repeating the question “Why?” five times (19). With use of the positive deviance approach we capitalized on beneficial practices in similar settings and tested them in the Diabetes QIP. For standardized processes the focus was on having a process for scheduling a follow-up visit within 1 month in team-based care using a standardized visit template, outreach to engage patients with A1C >9%, and testing A1C for anyone missing this value in the last 3–6 months. Additionally, practices were encouraged to establish standardized processes for all Plan-Do-Study-Act cycles to enhance sustainability including elements of the toolkit (Supplementary Material), visit templates, A1C individualized targets, etc. Second, provider-patient engagement included approaches such as common ground, self-disclosure, shared decision-making aimed at increasing culturally competent and humble health promotion through the toolkit, action period calls, and QI coaching. Culturally and linguistically appropriate services (CLAS) guidelines were reviewed to promote culturally and linguistically appropriate care (21). Third, methods were included to address nonmedical health-related social needs through health care community linkages with the following: 1) Pathways Community HUB models (22–24); 2) MCP approaches, involving community health workers, patient navigators, and case managers; 3) health system and practice-level social workers (25); and 4) tasks that providers were already performing including completing forms for utilities (25). The toolkit also addressed healthy environments of care (community health worker, nonmedical health-related social needs) and effective supportive relationships (peer/social support, standardized team care processes). Fourth, the impact of racism on care was reviewed and tangible strategies to address biases, discrimination, and racism were reviewed during action period calls and summarized in a best practice report, “Addressing Racism to Achieve Equity in Health Care” (3,9,26,27).

Practices submitted EHR data every 2 weeks to an honest broker (the GRC), including demographic and clinical information along with other data elements. The data were aggregated, cleaned, and made available for analysis by the GRC. Data specifications and the EHR query and submission schedule can be found in Supplementary Tables 1 and 2, respectively. Where appropriate, patients were linked to the ODM registry based on Medicaid identifier as a means of identifying patient insurance classification. For accurate identification of project patients who were Medicaid enrollees and for obtaining additional information about diagnoses and health services from Medicaid claims, the EHR data were linked with Medicaid enrollment records with use of a probabilistic linkage process. Match scores were generated with a set of identifiers including the patient’s first and last names, soundex codes capturing the phonetic pronunciations of these names, date of birth, and residential zip code. Through the linkage process the record pairs with the highest match scores were then identified, with each record in the EHR data set tentatively linked with the highest-scoring candidate match in the Medicaid enrollment file. Record pairs with perfect match scores (meaning that there are no differences across any of the identifiers in both files) were considered linked, while those with a score above a minimum threshold but falling short of a perfect match were manually reviewed to confirm accuracy.

Practice data were aggregated at the clinic level and shared back via a curated online dashboard for practices to monitor progress in conducting continuous QI. Key interventions tested by the practices included 1) A1C testing for those with no test within the last 3–6 months, 2) timely follow-up in team-based care—defined as follow-up scheduled virtually or in person with a primary care provider or team member (e.g., a clinical pharmacist, nurse, dietitian, or diabetes educator) at least every 30 days until the glucose level was at goal, 3) outreach to reengage patients with A1C >9% and no upcoming appointment, and 4) social drivers of health interventions (e.g., community health worker engagement, referrals to resources for healthy food, mobile vans, and/or virtual care).

The Diabetes QIP started in June 2020, with the prior 12 months considered the preintervention period (1 June 2019–31 May 2020). Year 1 of the QIP extended from 1 June 2020 through April 2021, and year 2 extended from 1 May 2021 to 30 June 2022. We evaluated the Diabetes QIP impact using the following parameters: A1C testing in the last 12 months, follow-up visit scheduled virtually or in person in the case of A1C >9%, and most recent A1C >9% (primary outcome). We measured outcomes longitudinally using repeated cross sections of EHR data and presented results as statistical process control charts with upper and lower confidence limits during the Diabetes QIP (28).

The analytic cohort includes all patients with at least one visit and a biologically valid A1C value (A1C between 4% and 20%) in the baseline period and at least one visit (at the same practice as at baseline) and a biologically valid A1C value in the study period (years 1 or 2). The patient exclusion cascade is shown in Supplementary Fig. 2. The analytic cohort consisted of 7,689 patients from 19 practices that submitted EHR data for the duration of the study period (Supplementary Table 3). Descriptive statistics were used to report baseline characteristics of all patients (nonstratified) and by race and ethnicity. The included and excluded patients were similar across most sociodemographic and biometric variables. The exception is that the excluded patients were 3 years younger (Supplementary Table 4). The prevalence of process measures was analyzed overall and by race and ethnicity and insurance status. For the primary outcome (reducing the proportion of patients with A1C >9%), the last A1C value within the baseline window and study periods (years 1 and 2) was used to assess the change in the proportion of participants with A1C >9%, overall and by race and ethnicity.

Annualized A1C change was examined as a secondary outcome. Annualized A1C change refers to the per-year difference between the most recent A1C within the baseline period and the most recently observed A1C within the study period (year 1 or 2 of diabetes QIP), so that a negative change indicates an improvement in A1C control. Self-identified race and ethnicity were classified into categories of NHW, NHB, non-Hispanic Asian (NHA), Hispanic, and other (e.g., refused to report/unreported, American Indian or Alaska Native, multiple races, etc.), and primary insurance was divided into the categories private/commercial, Medicare, Medicaid, and uninsured.

Multivariable linear regression models were used to examine the association of demographic factors with annualized A1C change. The last available A1C value within each study period (baseline, year 1, and/or year 2) was used in assessing improvements over time. The primary model controlled for age (10-year increments), biological sex (male or female), self-identified race and ethnicity, insurance class, and baseline A1C value. Statistical significance was defined as a two-sided α < 0.05. Analyses were completed with R, version 4.4.0 (29).

Overall, the subject group of 7,689 patients was sociodemographically and racially and ethnically diverse, as shown in Table 1. Among the population of patients, 54% were female and 46% male. Patients self-identified as NHW (42%), NHB (43%), Hispanic (8%), NHA (4%), or other (3%). Most patients were Medicaid insured (39%), followed by Medicare (36%) and commercial/private (24%), and <2% did not have insurance. Average no. of A1C tests over the course of the Diabetes QIP was 3.4 and average no. of visits 6.7. Evaluation of sociodemographic variables by race and ethnicity status in Table 1 revealed that NHW patients were older, a greater proportion having Medicare insurance, and had the least number of A1C tests and visits during the Diabetes QIP (all P < 0.0001). NHB patients were younger with higher levels of Medicaid insurance in comparison with the NHW patients in the Diabetes QIP.

In continuous A1C analyses (Table 2), overall mean (SD) A1C at baseline was 7.9% (2.1%). Mean A1C was 7.8% and 7.5% at years 1 and 2, respectively, with annualized A1C decreasing by 0.15 points (95% CI −0.18, −0.12). NHA patients had the lowest A1C at baseline (7.5%), followed by other (7.8%), NHW (7.8%), NHB (8.0%), and Hispanic (8.3%) (P < 0.0001). The nonadjusted annualized reduction in A1C decreased significantly (P < 0.0001) in all racial and ethnic groups except the NHA group and other (Hispanic −0.19, NHW −0.17, NHB −0.15, other −0.11, NHA 0.02).

In categorical A1C analyses (Table 2), overall 21% of patients had A1C >9%, with the proportion varying from 13% for NHA patients to 29% for Hispanic patients. In year 2 compared with baseline, there were decreases in the proportion of patients with A1C >9% among NHW, NHB, and Hispanic patients (NHW from 19% to 12% [37% reduction], NHB 23% to 18% [22% reduction], and Hispanic 29% to 23% [20% reduction]). There was no decrease among NHA patients: 13% to 13%.

Table 3 explores the change in process measures over the Diabetes QIP including A1C testing within 12 months, scheduled follow-up within 12 months, and scheduled follow-up within 30 days for patients with A1C >9%. There was an increase in all process measures overall from baseline to year 2, with increases in A1C check within 12 months (from 87% to 93%), scheduled follow-up within 12 months (49% to 68%), and scheduled follow-up within 30 days (28% to 30%). Rates for A1C check within 12 months increased in all racial and ethnic groups at baseline, year 1, and year 2; follow-up scheduled within 12 months increased in all racial and ethnic groups; and scheduled follow-up within 30 days increased in the NHA, other, and NHW groups only. The overall rates were numerically higher among racial and ethnic minority groups. The percent change from preintervention to 2 years for the process measures was mixed, due to higher baseline levels for some of the process measures among racial and ethnic minority groups.

The fully adjusted annualized A1C changes are shown in Supplementary Table 5. Older age and higher baseline A1C were associated with greater reductions in A1C during the Diabetes QIP, while male sex, Medicaid insurance (vs. commercial/private insurance), and self-identification as Hispanic (vs. NHW) were associated with less A1C improvement. The model predicts that a Hispanic patient would have an annualized A1C change of 0.108% (95% CI 0.007, 0.209) points less than an NHW individual with similar age, sex, insurance type, and baseline A1C value.

In the Diabetes QIP, among 19 practice sites across Ohio with 7,689 patients, the proportion of patients with A1C >9% was relatively reduced by 22% in NHB and 20% in Hispanic patients. A1C numerically improved over 2 years in all racial and ethnic groups with absolute reductions of 0.43%, 0.38%, 0.28%, 0.08%, and 0.33% among NHW, NHB, Hispanic, Asian, and other populations, respectively. NHB and Hispanic patients had less robust A1C lowering in the Diabetes QIP in comparison with NHW patients with confounders controlled for, including age, sex, insurance type, and baseline A1C. Importantly, among racial and ethnic minority patients rates were higher for levels of A1C checks within 12 months and 12-month and 30-day scheduled follow-up.

In 2023, in a Cochrane review 553 trials with >400,000 participants were analyzed to determine the most effective combination of effective QI strategies using hierarchical multivariable meta-regression models in a Bayesian framework (30). The included trials contained one system-targeted approach (case management, team changes, electronic patient registry, facilitated relay of clinical information, continuous QI) or provider-targeted approach (audit and feedback, clinician education, clinician reminders, financial incentives), alone or in combination with a patient-targeted strategy (patient education, promotion of self-management, patient reminders). In the models, the most effective A1C-lowering strategy in groups with baseline A1C <8.3% was a combination of promotion of self-management, patient reminders, and clinician education, with A1C reduction of 0.41%. The findings were consistent with those of earlier meta-regression and meta-analysis by Shojania et al. (31) and Tricco et al. (32), respectively. The Diabetes QIP attained an A1C reduction consistent with the “optimized” QI strategy from the meta-regression model in real-world practice in NHW and NHB populations and was slightly less effective in Hispanic populations. Thus, the Diabetes QIP design with payer-, system-, provider-, and patient-targeted approaches and implementation has the potential for meaningful population-level improvements across diverse racial and ethnic populations.

Although a ≥20% reduction in the proportion of patients with A1C >9% was attained among the NHB and Hispanic populations, the Diabetes QIP did not reduce or eliminate racial and ethnic disparities gaps in A1C control. At baseline there was a 20% (19% vs. 23%) and 50% (19% vs. 29%) higher relative disparity in the proportion of patients with A1C >9% among the NHB and Hispanic populations compared with NHW, respectively, which was maintained during the Diabetes QIP, even with improvements in A1C control in all groups. However, process measures indicated that care efforts (such as scheduling additional visits and conducting repeated A1C measurements) were high at baseline and were intensified for racial and ethnic minority patients. For future diabetes QI projects, there are opportunities to center projects on baseline inequities including a target for reducing A1C disparities. A critical need remains to develop best practices for disparity reduction across racial, ethnic, and socioeconomic groups in the U.S., with fewer than one-third of QI interventions including equity considerations (3,33,34). Academic-community-government-industry partnerships represent one potential avenue to address drivers of disparities in A1C control including the social drivers of health and nonmedical health-related social needs (24,35–37).

Across many QI interventions standardized processes are used to increase quality and in projects targeting disparities to reduce potential implicit biases. Theories of health inequality suggest that standardized processes may also elicit “intervention-generated inequalities” whereby more advantaged groups experience greater benefits than less advantaged groups (38,39). Downstream (closer to the patient) standardized processes have been associated with a greater potential for intervention-generated inequalities than upstream or macro-level changes and interventions (built environment, policy, etc.) (38,39). Thus, careful consideration of the role of equity and the potential to generate inequities will be critical in the design of future QI interventions.

Strengths of the Diabetes QIP include 1) that the Diabetes QIP was codesigned and implemented by a statewide diabetes collaborative with ODM, GRC, MCPs, and seven Ohio medical schools, along with >7,000 patients and 19 primary care practices; 2) that the Diabetes QIP design and implementation included payer-, system-, provider-, and patient-targeted approaches, which is a unique multilevel, multifaceted approach; and 3) the sociodemographic diversity of the patients at the primary care clinics (42% NHW, 43% NHB, 8% Hispanic, 4% NHA, and 3% other) along with the mix of insurance types (24% commercial, 39% Medicaid, 36% Medicare, and 2% uninsured).

The Diabetes QIP must be considered in light of some potential limitations. The majority of the Diabetes QIP occurred during the coronavirus disease 2019 pandemic, which led to changes in many processes for all partners and patients involved in the Diabetes QIP. Notably, there were more frequent staff changes at primary care sites, along with shifting priorities and deployments for QI experts within health systems. Additional factors are as follows: Practices shifted between in-person and telehealth care during this time. Data were collected from practice EHRs that may have omitted data collected via a non–Diabetes QIP provider, which may have led to an underestimation of increases in A1C collection or follow-up. All included patients in analysis of A1C testing were seen in either year 1 or year 2, but not all patients were seen in both years, and thus the sample of patients was smaller at years 1 and 2 and does not account for patients who were entirely lost to follow-up, which may be a source of bias. The cultural background of the team member performing follow-up was not assessed, thus limiting the ability to determine the impact of cultural concordance. Information on diabetes medication and technology, nonmedical health-related social needs, and mental health referrals was not collected. Due to a limited number of NHA patients involved in the Diabetes QIP, we cannot make strong comparisons involving that group. Major shifts in diabetes care have occurred since the pandemic, notably including changes in the evidence for, availability of, and coverage of newer medications associated with weight loss and A1C improvements (especially glucagon-like peptide 1 agonists); thus, we cannot exclude that observed A1C improvements may be due to temporal change or unrecognized cointerventions. Additional work to promote equity will be needed amidst these changes to diabetes care.

The novel statewide Ohio Diabetes QIP reduced the proportion of patients with A1C >9% and improved process measures among NHW, NHB, and Hispanic patient populations across 19 primary care practice sites through a focus on payer-, system-, provider-, and patient-targeted approaches. Statewide approaches offer an innovative avenue allowing for cross-sector academic-community-government-industry partnerships to advance equity in diabetes treatment and outcomes. Advancements in attaining glycemic goals in racial and ethnic minority groups may be catalyzed by applying multisector collaboration–based approaches, but gaps in disparities necessitate further tailoring and attention.

Acknowledgments. The authors gratefully acknowledge the administrative support of the GRC and partnership of the ODM, MCPs, and Ohio medical schools; QI consultants; and primary care clinics throughout Ohio. The authors acknowledge the contributions of Joseph Ballas and Bhavya Appana at The Ohio State University for creating Fig. 1.

Funding. Funding for the Diabetes QIP was provided by the ODM Medicaid Technical Assistance and Policy Program and General Revenue Fund.

The authors are solely responsible for the contents, findings, and conclusions of this article, which do not represent the views of the state of Ohio, ODM, or any federal programs. Readers should not interpret any statement in this report as an official position of ODM or of Health and Human Services.

Duality of Interest. J.J.J. is a board member for Buckeye Health Plan. K.M.D. discloses research support from ViaCyte, Abbott, Insulet, and Dexcom; consulting with Eli Lilly, Dexcom, Oppenheimer, and Insulet; and honorarium from UptoDate, Elsevier, Med Learning Group, and Medscape. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. J.J.J. and A.T.P. wrote the manuscript and researched data. K.M.D., E.A.B., D.E., K.J., A.L., S.J.U., K.M.A., M.W.K., M.S.A., and S.D.B. contributed to the discussion and reviewed and edited the manuscript. J.F., D.S., and T.E.L. researched data and reviewed and edited the manuscript. J.J.J. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented at the Annual Meeting of the Society of General Internal Medicine, Orlando, FL, 6–9 April 2022, and AcademyHealth Annual Research Meeting, Seattle, WA, 24–27 June 2023. A video presentation can be found in the online version of the article at https://doi.org/10.2337/dci24-0025.

Handling Editors. The journal editor responsible for overseeing the review of the manuscript was Steven E. Kahn.