An official website of the United States government
NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.
Joan O'Connell, PhD, Jennifer Rockell, PhD, Margaret Reid, MS, Kathleen Harty, MPH, Marcelo Perraillon, PhD, and Spero Manson, PhD.
Author Information and AffiliationsAmerican Indians and Alaska Natives (AI/ANs) experience disparities in diabetes-related morbidity and mortality. The Indian Health Service (IHS) and Tribal health programs provide education, case management, and advanced practice pharmacy (ECP) services for patients with diabetes to improve health outcomes.
The purpose of this study was to evaluate patient outcomes associated with the use of ECP services by AI/AN adults with diabetes by comparing the outcomes of ECP users with those of nonusers.
We analyzed fiscal year (FY) 2011-2013 data for AI/AN adults with diabetes from the IHS Improving Health Care Delivery Data Project, which includes data on nearly 30% of the IHS service population. The diabetes and cardiovascular disease (CVD) measures were used to create 3 study cohorts: all adults with diabetes; and 2 subgroups, adults with diabetes without CVD and adults with both diabetes and CVD. The analyses were conducted for each study cohort.
Using an observational study design and propensity score models that employed inverse probability weighting, to control for the nonrandom assignment of patients to the treatment group, we evaluated FY2013 outcomes for patients who used ECP services during FY2012, controlling for baseline characteristics in FY2011. The outcomes for ECP users were compared with those for patients who obtained usual care (ie, patients who did not use ECP services), using multivariable regressions. Baseline characteristics included age, sex, health coverage, and health status measures. Other characteristics included in the propensity model were drive times to ECP services and county-level measures of educational attainment and household income from the American Community Survey. Health outcomes included high hemoglobin A1c (HbA1c), systolic blood pressure (SBP), and low-density lipoprotein cholesterol (LDL-C) levels; onset of CVD and end-stage renal disease (ESRD) among those who did not have those conditions; and use of hospital emergency and inpatient services.
The study population included 28 578 adults with diabetes. During FY2012, 41.0% of adults with diabetes had ≥1 ECP visits. ECP use, compared with no use, among adults with diabetes was associated with lower odds of high SBP (odds ratio [OR], 0.85; 95% CI, 0.79-0.93; P < .001) and high LDL-C (OR, 0.89; 95% CI, 0.84-0.98; P < .01). Among adults with diabetes without CVD, ≥3 ECP visits vs no visits was associated with lower odds of CVD onset (OR, 0.79; 95% CI, 0.63-0.99; P < .05). ECP was associated with lower odds of ESRD onset (OR, 0.60; 95% CI, 0.39-0.93; P < .05) among adults with both diabetes and CVD. ECP use, compared with no use, among adults with diabetes was also associated with lower odds of having ≥1 hospitalizations (OR, 0.80; 95% CI, 0.71-0.89; P < .001) and having ≥1 potentially preventable hospitalizations (OR, 0.79; 95% CI, 0.64-0.91; P < .05). ECP use was also associated with fewer hospital emergency visits and inpatient days.
Despite differences in morbidity between adults with diabetes by CVD status (ie, with and without CVD), some evidence of a positive association between ECP use and patient outcomes (ie, significant improvement) was found among those with and without CVD for 5 of the 7 outcomes examined in both populations. When examining the level (or dose) of ECP service use, we found that for 5 of the 9 total outcomes studied, higher levels of ECP use (ie, ≥3 visits) were associated with greater improvement in patient outcomes.
ECP use was associated with improvements in blood pressure and cholesterol control, lower onset of CVD and ESRD, and lower use of hospital emergency and inpatient services. These results may inform IHS, Tribes, and other organizations in allocating resources for ECP and other services to address the needs of AI/AN adults with diabetes.
Study limitations included the use of an observational study design and a propensity score model to control for observed differences between ECP users and nonusers. An assumption associated with the use of a propensity score model is that all potential confounders are included in the analysis, or that possible unobservable factors are correlated with the observable factors. We conducted analyses to assess the sensitivity of our findings to model specifications to address, in part, this potential limitation.
American Indian and Alaska Native (AI/AN) peoples, compared with other races, experience some of the greatest health disparities, including those associated with diabetes and related complications, such as cardiovascular disease (CVD) and end-stage renal disease (ESRD).1-5 The prevalence of diabetes among AI/ANs ≥18 years was 15.1% in 2015, more than double that of non-Hispanic White adults, and the highest among US racial/ethnic groups.1 Among AI/ANs, nearly 50% of those with CVD have diabetes.6 The AI/AN all-cause mortality rate is 46% higher than that of non-Hispanic White adults and is largely attributable to disparities in heart disease, stroke, diabetes, and kidney disease mortality.2,7-9 Diabetes and CVD contribute not only to higher mortality among AI/ANs of all ages but also to higher rates of premature mortality (ie, mortality among those aged <65 years).7-9 Forty-one percent of AI/AN diabetes-related deaths occur among adults aged <65 years, whereas 23.4% of diabetes-related deaths occur among non-Hispanic White adults aged <65 years.7 A similarly high percentage of AI/AN deaths attributable to heart disease occur in adults aged <65 years.8
The Indian Health Service (IHS) and Tribal health programs have implemented many programs and policies to improve health outcomes among AI/ANs. Here, we describe findings from a comparative effectiveness study conducted to evaluate patient outcomes associated with the use of IHS and Tribal education, case management, and advanced practice pharmacy (ECP) services among adults with diabetes. We start by describing IHS and Tribal health programs, followed by a description of ECP services.
The federal government has a federal trust responsibility to provide health care to members of federally recognized Tribes; IHS fulfills this responsibility through programs and policies outlined in numerous federal statutes.10 The IHS delivery system includes hospitals, clinics, and health programs operated by the federal government, by Tribal organizations through contracts and compacts with IHS, and by urban Indian health organizations that receive modest IHS support. Combined, they serve approximately 2.3 million AI/ANs who live throughout the United States.11 Unfortunately, this system of care has very limited funding: per-capita health spending was $3851 in 2017.11,12 Although this amount did not include all spending associated with patient care—for example, it excluded costs of some specialty outpatient and inpatient services not provided by IHS, Tribes, or urban Indian clinics (known collectively as I/T/U services)—it is substantially lower than per-capita spending for the US general population ($10 739) in 2017.13 Here, we use IHS to refer to I/T/U providers and the IHS service delivery system as a whole.
IHS resources are further compromised by provider shortages and community-level factors that affect service use and health (eg, low household income and educational attainment, rural geography).6,14-21 Consequently, AI/ANs with diabetes who use IHS services require delivery models that can effectively address their risks and complex needs.6,11,15-20
Similar to other health systems, IHS has implemented many initiatives to prevent the onset of diabetes and to reduce complications among those with diabetes.22 The Special Diabetes Program for Indians (SDPI) began in 1998 and provides grants to over 400 I/T/Us to fund an array of services that include diabetes prevention programs, exercise programs, health personnel training on clinical guidelines, ECP services, and data collection.5 SDPI funds diabetes educators, nutritionists, other health educators, case managers, and advance practice pharmacy (APP) providers. Although APP delivery models vary, they can include patient assessment, medication reconciliation, and health education conducted by certified pharmacists who may also order laboratory tests and modify prescriptions under the supervision of a physician.6,20,23-25 Some ECP services evaluated in this study, and provided by these 5 types of health professionals, were funded via SDPI.
Using evidence-based best practices, SDPI grant programs have made improvements in diabetes treatment and prevention in both clinical settings and community-based programs. Since SDPI's implementation, improvements in blood pressure and cholesterol among AI/ANs with diabetes and decreases in the incidence of ESRD have been reported.22,26,27 In addition, SDPI supports the annual collection of data through the IHS Diabetes Care and Outcomes Audit to guide this work and measure outcomes.5
ECP services provided to prevent or delay the onset of diabetes-related complications address patient lifestyles (eg, weight management, exercise, smoking cessation), blood sugar (ie, hemoglobin A1c [HbA1c]), systolic blood pressure (SBP), low-density lipoprotein-cholesterol (LDL-C), and pharmacy management.22,28-31 Among non-AI/AN populations, studies have shown that ECP services have improved outcomes (eg, in SBP, LDL-C) among individuals with diabetes.32-49 Patient outcomes vary by the types of ECP provided, patient risk factors, the intensity of services (eg, number of ECP visits), the length of the intervention time period studied (eg, 6 months, 1 year), and the health delivery system in which the services were employed.20,24,32,34-49
ECP may play an especially important role in addressing the health needs of AI/ANs with diabetes, yet there is limited information about the provision and use of ECP services, as well as related outcomes, among AI/ANs with diabetes who access IHS services. SDPI implemented a demonstration project that included intensive case management services, primarily provided by nurses, educators, and pharmacists, for approximately 3400 adults with diabetes.50 Data from an annual assessment of Healthy Heart participants indicated that CVD risk factors (eg, HbA1c, SBP, and LDL-C) were reduced after 1 year of program participation.50 Although the study documented a successful translation of an intensive case management program to many IHS and Tribal health programs, it did not include a comparison population to provide context for the results.
A second study examined data for AI/AN adults with diabetes who obtained health services through 9 commercial integrated health delivery systems in the United States that provided ECP services. The study found that AI/AN adults with diabetes had rates of annual testing for HbA1c, SBP, and LDL-C comparable with those of non-Hispanic White adults with diabetes who obtained services from the same health systems.51 However, this study did not examine the use of ECP services nor the outcomes associated with their use.
To address this gap in knowledge, we aimed to conduct a comparative effectiveness study to examine the association between ECP service use and patient outcomes among AI/ANs with diabetes. We hypothesized that there would be a positive relationship between the use of ECP services and patient outcomes and that the relationship would vary by patient outcome (eg, blood pressure control, use of hospital inpatient services), CVD status, and level of ECP use. Two goals related to this aim were to (1) describe the health status and service use of AI/AN adults with diabetes to inform the analysis; and (2) work with the study's Collaborative Network, described in the “Participation of Patients and Other Stakeholders” section, to develop strategies to facilitate patients' ability to make informed choices about using ECP services and to enhance the provision of these services to address patients' needs.
The initial study aim was to examine patient outcomes among AI/AN adults with diabetes and CVD. However, the Collaborative Network recommended that analyses also be conducted for AI/AN adults with diabetes without CVD and for all AI/AN adults with diabetes so that study findings could improve understanding of the needs, ECP use, and outcomes of each study cohort; inform efforts to prevent the onset of CVD; and interpret results for IHS and Tribal health programs that served all adults with diabetes.
Since 2010, IHS and Tribes have collaborated with the Centers for American Indian and Alaska Native Health (CAIANH) at the Colorado School of Public Health through the IHS Improving Health Care Delivery Data Project (Data Project) to create and update a data infrastructure that synthesizes existing electronic data from multiple IHS platforms. It currently includes data for 7 years (fiscal year [FY] 2007-FY2013) on the health status, service use, and treatment costs for >640 000 AI/ANs who live throughout the United States and represent nearly 30% of AI/ANs who use IHS services.6 This ECP comparative effectiveness study was conducted using data from FY2011 to FY2013 extracted from the IHS Data Project's longitudinal data infrastructure for recording and storing data on AI/AN adults with diabetes. Before describing the methods employed to conduct the study, the next section of this report provides information on patient and other stakeholder involvement in the study.
CAIANH works in a collaborative manner with IHS and the Tribal organizations that participate in the IHS Data Project. This collaboration takes place through the project's Collaborative Network, through CAIANH personnel who travel to the project sites, and through the process used to obtain approvals from the IHS national IRB, Tribal IRBs, and Tribal Councils. Collaboration with IHS and Tribal representatives and patients/caregivers is a vital component of any study to ensure that the study's aim is relevant to their needs. Network members provide advice and guidance for all studies that include analyses of data from the IHS Data Project.
Our Collaborative Network includes 3 advisory committees: Steering, Project Site, and Patient. The Steering and Project Site Committees were formed at the start of the IHS Data Project in 2010 in order to receive advice and guidance from a broad IHS and Tribal perspective as well as the perspective of the specific project sites. The Patient Committee was formed as part of this study to provide patients' and caregivers' perspectives and is now a permanent project committee. Table 1 provides information on each committee, including their membership, meeting frequency and format, and overall goals.
Descriptions of Health Care Delivery Data Project's Collaborative Network Committees and Their Goals.
The IHS Data Project protocols were reviewed by the IHS National IRB, the Colorado Multiple IRB (ie, university IRB), 1 regional IHS IRB, and 6 Tribal IRBs. Tribal Councils without IRBs have provided approval though resolutions and letters for project involvement. In addition, a data agreement was put in place with IHS and 1 Tribal organization. CAIANH met with project site Tribal Council representatives not only to discuss approvals for project participation but also to provide project updates, obtain feedback and guidance, and understand their concerns about the health of their Tribal members. Thus, these representatives are an informal part of the Collaborative Network.
In this study, the Collaborative Network contributed to the research in many ways. First, its members influenced the makeup of the study population. The Collaborative Network recommended that analyses be conducted not only for AI/AN adults with diabetes and CVD but also for AI/AN adults with diabetes without CVD and for all AI/AN adults with diabetes. Second, they influenced the development of the intervention measure, patient outcome measures, and covariate measures included in the multivariable analyses. The intervention measure included 5 types of ECP services. The provision of ECP services by site varied with respect to the supply of those services and the type of ECP services provided based on the availability of providers, patient needs, and site administrator preferences. Due to this variation, we considered the use of services provided by any of the 5 types of health professionals to represent ECP use for the purposes of this study.
Patients also influenced the types of outcomes evaluated in this study. Based on their knowledge of diabetes and available data, they recommended outcomes related to HbA1C, blood pressure, and cholesterol control. A member of the Steering Committee requested that we include 2 additional patient outcomes: the onset of CVD and ESRD. The Patient Committee recommended that outcomes related to the use of hospital emergency and inpatient services be included because the use of these services may be due to an exacerbation of diabetes or a related comorbidity.
Collaborative Network members recommended that we conduct several analyses to improve understanding of patients' health status and service use. The Project Site Committee requested specific health status analyses; for example, they requested analyses concerning the relationship between behavioral health disorders and patient outcomes. Patient Committee members stressed the importance of examining service use and outcomes by age and sex to inform efforts to improve patients' ability to manage diabetes and limit the onset of complications.
Collaborative Network members translated project results into recommendations on service and policy enhancements to address the study goal of developing strategies to facilitate patients' ability to make informed choices about using ECP services and to enhance the provision of these services to address patients' needs. These strategies were developed and refined during each committee's meetings and during 2 webinars held in May 2017, which included representation from each committee. The strategies are provided in the Results section. Committee members also provided recommendations concerning the competencies they believe are important to facilitate patients' ability to make informed choices about service use and advocate for their health needs, as well as outreach methods and materials that may be used to achieve such competencies.
Collaborative Network members are included as authors in manuscripts that are under development. Committee members, IHS IRB and Tribal IRBs, and Tribal Council members review documents that will be publicly disseminated, such as this report. Tribes and I/T/U providers work with us to disseminate project information via Tribal meetings and other methods. Patient Committee members have worked and will continue to work with us to disseminate information via community summaries targeted at specific populations.
In this section, we provide information on the comparative effectiveness study population, measures, and analyses conducted to examine the relationship between ECP use and patient outcomes among AI/AN adults with diabetes. The comparative effectiveness study was conducted using inverse probability weighted (IPW) propensity models to control for patient self-selection into the intervention group, compared with usual care, in this observational study.
This ECP comparative effectiveness study was conducted using data from FY2011 to FY2013 for AI/AN adults with diabetes that were extracted from the IHS Data Project's longitudinal data infrastructure. The data infrastructure includes information for a purposeful sample of AI/ANs who lived in 15 IHS service units, which are geographic classifications, located throughout the United States (see Figure 1). Based on geographic regions employed elsewhere, 1 service unit is located in the East, 4 in the Northern Plains, 2 in the Southern Plains, 5 in the Southwest, 2 in the Pacific Coast, and 1 in Alaska.2
Locations of the 15 IHS Service Units by Geographic Region.
The project population was identified by geographic area (ie, service units) rather than by random sampling so that we could create important community-level (eg, drive time to services), county-level (eg, household income), and site-level (eg, measure of ECP supply) measures not available elsewhere. The project population was comparable with the national IHS service population by age and sex. According to the national IHS Diabetes Audit, in 2014, the mean values for HbA1c, SBP, and LDL-C among patients with diabetes were 8.1%, 131 mm Hg, and 92 mg/dL, respectively.5 These findings mirror those for FY2013 for the 15 IHS service units (here, we refer to service units as Data Project sites).
Inclusion criteria for the study population were AI/AN adults who had diabetes in FY2011, lived in the Data Project sites, and used health services during 3 consecutive years (FY2011-FY2013, N = 36 751). Among these adults, we excluded those who lived in a Data Project site that did not provide ECP services or who lived in 2 sites with extremely limited FY2012 ECP data (eg, no data for ECP providers who provided the majority of services, n = 3898). Of these 12 Data Project sites, 4 sites served an AI/AN population that included <12 000 persons, 4 sites served populations between 12 000 and 30 000 persons, and 4 sites served populations of >30 000 persons.
Other exclusion criteria included (1) being treated for malignant cancer, ESRD, or any kind of transplant during FY2011 to FY2013 (n = 3771, except for the analysis of ESRD onset in FY2013, in which adults with ESRD onset in FY2013 were included); (2) having missing data for community- and county-level variables (n = 1142); (3) having missing FY2011 morbidity burden scores (n = 102); and (4) evidence of having died during the study period (n = 583). The final study population included 28 578 AI/AN adults with diabetes.
The study population of adults with diabetes included ECP users and nonusers who met these inclusion and exclusion criteria. See Appendix A for additional information on the study population. The diabetes and CVD measures, defined below, were used to create 3 study cohorts: all adults with diabetes; and 2 subgroups, adults with diabetes without CVD and adults with both diabetes and CVD. Due to the number of adults with diabetes who met these criteria, the observational nature of the study, and data availability, we did not conduct power analyses.
The IHS Data Project's longitudinal data infrastructure includes 4 types of IHS electronic data: (1) National Data Warehouse (NDW), (2) Purchased/Referred Care (PRC) data, (3) Centers for Medicare & Medicaid Services Cost Report (Cost Reports) data, and (4) procurement cost data for prescribed medications. We drew upon the Chronic Care Model52,53 and the existing diabetes and ECP literature to identify patient and community, county, and facility characteristics to include in the comparative effectiveness study. Summary information on the study measures is provided below; data for these measures were extracted from the Data Project's longitudinal data infrastructure.
NDW data provided information on age, sex, and health insurance coverage in FY2011. Data on diagnosed conditions (ICD-9-CM), procedure codes, medication use, and HbA1C control, included in the NDW and PRC inpatient and outpatient service use records, were used to create health condition measures (eg, diabetes, CVD, depression) for each FY. Project algorithms, developed from national references, were used to identify adults with diabetes (type 1 and type 2), CVD, and ESRD.54-56 We used the Sightlines DxCG Risk Solutions software57 to identify adults diagnosed with other conditions. To identify adults with diabetes, we adopted a validated algorithm employed by other researchers; the algorithm included data on diagnostic codes, medication use, and HbA1c values.55,58 The CVD algorithm was based on CVD classifications employed in the IHS Diabetes Audit to identify types of CVD that may be related to diabetes.5 ESRD was identified using data on ESRD diagnoses, kidney transplants, and dialysis.59 The DxCG software classifies the ICD-9-CM diagnostic codes recorded in medical service use records into nearly 800 disease categories.
The DxCG software also provides an annual measure of morbidity burden, or risk score, for individuals based on their age, sex, and all diagnosed conditions. The morbidity burden score typically ranges from 0 to 100, and the measure was benchmarked to a US commercially insured population with a mean morbidity burden score of 1.57 The morbidity burden mean scores for adults with diabetes, adults with diabetes without CVD, and adults with both diabetes and CVD were 5.6, 4.4, and 8.7, respectively.
For each study cohort, we categorized the morbidity burden scores into quartiles; adults with low morbidity were assigned to quartile 1 and those with the highest morbidity to quartile 4. For example, among adults with diabetes, those with a morbidity value of ≤2.0 were assigned to quartile 1, those with scores between >2.0 and ≤3.7 were assigned to quartile 2, those with scores between >3.7 and ≤6.3 were assigned to quartile 3, and those with scores >6.3 were assigned to quartile 4.
Patient outcome measures included 5 health status measures. Based on national and IHS guidelines for diabetes management and input from the Collaborative Network, we created measures that identified adults with an HbA1c ≥8% as having uncontrolled long-term blood glucose, high SBP as ≥140 mm Hg, and high LDL-C as ≥100 mg/dL or higher.28,60-62 New or recurring onset of CVD and ESRD onset in FY2013 was defined as a diagnosis during that FY with no diagnosis of the condition during the previous 2 FYs.
ECP includes visits for individual or group diabetes education, provided by nurses or health educators in diabetes clinics; nutrition education; other types of education (eg, smoking cessation); case management, provided primarily by nurses; and APP. ECP services are provided to supplement primary care services provided by physicians and midlevel providers. Providers may refer patients to ECP services. In addition, patients may learn of ECP through other means and choose to use them. We identified I/T-provided outpatient ECP visits using NDW data on provider type, clinic, and procedure codes and created 2 ECP intervention measures: (1) any ECP use, a dichotomous measure with a value of 1 indicating ≥1 ECP office visits during FY2012; and (2) the number of ECP visits during FY2012. We refer to I/T ECP services, rather than I/T/U ECP services, due to the very low provision of ECP services by urban Indian clinics located in the 12 Data Project sites.
The ECP measure was created in consultation with our Collaborative Network to account for the variation across sites in the provision of different types of services. Our intent was to develop 1 broad measure of ECP that could assess outcomes for the array of services provided. Differences across sites appeared to be associated with personnel availability (eg, in rural areas, it may be easier to hire and retain a nurse or health educator than a nutritionist), personnel experience (eg, pharmacists with APP experience), administrator and provider preferences, and precedence. ECP users included adults with diabetes with and without CVD, CVD but not diabetes, high body mass index (BMI), behavioral health disorders, and other health needs.
Patient outcomes include 3 hospital inpatient use measures and a measure of hospital emergency department (ED) use (ie, number of I/T ED visits) using NDW data. The 3 hospital inpatient use measures were (1) one or more hospitalizations (ie, admissions) recorded in the NDW or PRC data; (2) the total number of hospital inpatient days recorded in the NDW and PRC data; and (3) one or more I/T hospitalizations for potentially preventable (PP) conditions, defined using a nationally recognized algorithm, recorded in the NDW data.63 Urban Indian clinics do not provide hospital services and are not referenced here.
To better understand the factors that may influence health status and service use, we created measures of ECP access (ie, the availability of ECP services and patient drive times to obtain ECP services) and county-level measures of AI/AN educational attainment and household income using data from IHS and other sources.
The ECP availability measure was a facility-level measure of the supply of ECP services at an I/T facility for its service population. For each facility, the FY2012 facility ECP supply rate was calculated by dividing the number of ECP visits the facility provided during FY2012 by the number of adults in the facility's service population for that year. The service population was defined by geographic regions, referred to as a community of residence, that were smaller than a county and identified where adults lived in the project sites. Most sites had ≥10 communities of residence. Each community was assigned to a facility in the site based on patient use patterns and drive times to obtain services. Patient drive times from a central location in their community to an I/T facility that provided ECP were estimated using geocodes (latitude and longitude) and OpenStreetMap.64
County-level measures of AI/AN educational attainment and household income were derived from 2010-2014 American Community Survey (ACS) county-level data from the US Census Bureau for AI/AN respondents who reported access to IHS services. For each county, we defined the percentage of households with low incomes as the percentage of households with incomes <139% of the federal poverty level (FPL), a threshold that is currently used in many states to determine Medicaid eligibility. We defined the county-level educational attainment measure as the percentage of adults who did not complete high school.
For each multivariable analysis, the number of adults included in the analysis was based on data availability of the outcome measure (see Appendix A). Due to the manner in which the health condition measures were created and the inclusion/exclusion criteria for the study population (ie, used health services during 3 consecutive years), there were no missing data for the health condition measures, including CVD and ESRD onset. Adults who lived in 1 of the 2 Data Project sites where IHS did not provide hospital emergency or inpatient services were excluded from analyses of I/T ED visits and hospitalizations. Adults who lived in project sites with a low percentage of data for an outcome (ie, based on comparisons with historical data and data from other sites) were excluded from analyses of that outcome. Two sites had very limited clinical outcomes data (ie, HbA1c, SBP, and LDL-C data) due to electronic health record (EHR) system conversions. An additional site had a large percentage of missing SBP data. One site had very limited PRC hospital data due to EHR system conversion, and 2 sites had limited PRC hospital data for other reasons. These 3 sites were excluded from the analysis of hospitalizations.
Finally, adults who lived in Data Project sites with data for the 3 clinical outcomes (ie, HbA1c, SBP, and LDL-C) may have had missing data for ≥1 of those outcomes. For individuals who lived in sites with HbA1c, SBP, and LDL-C data yet did not have an FY2011 value, we created measures that indicated that the value was missing and included their data in the multivariable analyses. However, adults at these sites with a missing clinical measure for FY2013 were excluded from multivariable analyses of that measure.
Due to the number of AI/AN adults with diabetes in the data, we do not believe that these exclusions compromised our ability to conduct the comparative effectiveness study. We believe that this approach to addressing missing data provides a framework to ensure that analyses are conducted in a manner that maximizes use of the IHS longitudinal data for a large number of AI/AN adults with diabetes who use IHS services. More information on the IHS Data Project is reported elsewhere.6,65,66
We used SAS (SAS Institute) and Stata (StataCorp) statistical software programs to conduct descriptive and multivariable analyses. This comparative effectiveness study employed an observational study design, using propensity score analyses and IPW, to compare outcomes associated with ECP use, compared with usual care, among AI/ANs with diabetes.
The inclusion criteria for the study population required patients to use health services during 3 consecutive years. Multivariable models were estimated using data for 3 discrete time periods (FY2011, FY2012, and FY2013) to control for baseline differences in health status and other patient characteristics in FY2011 that may have influenced ECP use in FY2012, the second year. We evaluated patient outcomes during the third year (FY2013), the year following FY2012 ECP use.
Selection bias is often present in observational studies, as patients are not randomly assigned to a treatment group. In this study, patients self-selected into obtaining ECP services (use or nonuse of ECP); some used ECP services based on recommendations from other providers. Personal characteristics associated with ECP use may include observable characteristics (eg, age, sex, comorbidities) and unobservable characteristics, such as a tendency for providers to strongly recommend that certain persons use ECP (eg, adults with both diabetes and CVD), or a tendency for patients to use services due to personal motivation or family encouragement.67-70 Because we did not have data measures for these tendencies, they are considered unobservable. Because both observable and unobservable characteristics may influence health outcomes, regression findings may underestimate or overestimate the relationship between ECP and patient outcomes if selection bias is not taken into account.
We employed an IPW propensity score model to adjust for nonrandom assignment of patients to the intervention group (ie, ECP users).71-81 The model includes 1 equation to model self-selection into the intervention group, which may be defined as any ECP use or level of ECP use (ie, 0 visits, 1-2 visits, or ≥3 visits). A second equation examines patient outcomes associated with ECP use. An assumption associated with the use of propensity score analysis is that all potential confounders are included in the analysis, or that possible unobservable factors are correlated with the observable factors. We considered the use of an instrumental variable (IV) model for this study, because unmeasured confounding is not a limitation of this model if a valid instrument is identified.67-70 After conducting preliminary analyses and considering the advantages and limitations of each model, we decided that the propensity score model was the most appropriate method, largely due to our inability to identify a valid instrument for the IV model.
We used IPW propensity models, assessed the balance of the covariate distributions between the intervention and control groups, and examined the overlap of propensity distributions between the intervention and control groups (see Appendix B). Based on the results, we adjusted the model to obtain a sufficient balance and overlap on key baseline characteristics. Additionally, we assessed whether trimming observations with extremely low or high propensity scores (1st and 99th percentiles, respectively) would improve the balance. Because trimming appeared to worsen the covariate balance following IPW, we did not do so.
There are multiple approaches for using propensity scores to adjust for nonrandom assignment of patients to the intervention group. We considered 2 other approaches: (1) matching ECP users with controls, and (2) subclassification or stratification of the analysis sample based on propensity scores.71-81 The findings from the matching approach were largely similar to those of IPW. Based on preliminary analyses and an assessment of the advantages and disadvantages of each approach, we concluded that IPW was the most appropriate approach for this study.72 An important advantage of IPW is that we could include data for all adults who met the inclusion/exclusion study criteria in the model, and the findings might therefore be generalizable to that population.
The first set of propensity-weighted models were estimated to examine the association between any ECP use (ie, ≥1 ECP visits vs no ECP visits) and patient outcomes. Equation 1 of these models was estimated using logistic regression. The second set of propensity-weighted models were estimated to examine the relationship between the level of ECP use and patient outcomes. For this set of models, an ordered logistic model was estimated for equation 1. Using this approach, we assessed patient outcomes associated with 1 to 2 ECP visits vs no visits, ≥3 visits vs no visits, and ≥3 visits vs 1 to 2 visits. These 3 levels of ECP use were chosen based on the empirical distribution of ECP visits and findings from other studies. The specification of equation 2 of the propensity models varied by patient outcome. Most outcomes were binary, and equation 2 was estimated using logistic regression. Outcomes associated with the number of ED visits and hospital inpatient days were modeled using negative binominal distributions. The IPW propensity models included fixed effects in both equations to control for unobserved variation across the project sites. We provide 95% CIs for the resulting estimates from both types of regressions.
Collaborative Network members recommended that we conduct several analyses not anticipated in the original study protocol to improve understanding of patients' health status and service use. In addition, they requested that we conduct analyses for 3 study cohorts, not just the one included in the original study protocol (ie, adults with diabetes and CVD). Due to project resources allocated to these additional efforts, this report does not include information on the type of CVD or treatment costs, 2 measures that were included in the original study protocol.
Tables 2 through 8 show the results from descriptive analyses conducted to inform the specification of the IPW propensity model used to conduct the comparative effectiveness study. Due to the quantity of the results, we describe selected findings from the descriptive analyses that are of particular relevance to the comparative effectiveness study. The results of the comparative effectiveness study are provided in Tables 9 to 11; we describe these results in greater detail.
The inclusion criteria for the study population required adults to have used IHS or Tribal health services during 3 consecutive FYs (FY2011-FY2013). To examine the influence of this requirement, we compared measures of health status and service use between AI/AN adults with diabetes who used services in FY2011 with those who used services during the 3 consecutive FYs for each study cohort (ie, all adults with diabetes, adults with diabetes without CVD, and adults with both diabetes and CVD).
Of all adults with diabetes who used services in FY2011, 87.7% used services during each of the 3 FYs, FY2011 to FY2013 (Table 2). The percentage among adults with diabetes without CVD was 89.0%; among adults with diabetes and CVD, the percentage was 84.4%. We found minor differences in patient characteristics when comparing all adults with diabetes who used services in FY2011 with those who continued to use services during all 3 FYs. For example, 24.6% of adults who used services in FY2011 had high SBP, compared with 23.2% among those who used services during all 3 FYs.
Baseline (FY2011) Characteristics of Adults With Diabetes by CVD Status and Data Availability.
Information on patient age, sex, health coverage, health status, and access to ECP services is provided in Table 3 for each study cohort by ECP user status (ie, adults who had no visits and adults with ≥1 visits). Adults with both diabetes and CVD were older than those without CVD, had more comorbidities, were more likely to be male, were more likely to have high SBP, and were less likely to have high LDL-C.
Baseline (FY2011) Characteristics of Adults With Diabetes by CVD Status and Use of ECP Services During FY2012.
ECP users and nonusers differed in several ways. Among adults with diabetes, ECP users were older and had lower Medicaid and higher private insurance coverage; more comorbidities, including CVD; higher DxCG risk scores; and higher hospitalization rates. A higher percentage of ECP users had SBP, HbA1c, and LDL-C test results. For example, a higher percentage of ECP users than nonusers had an HbA1c test result (91.7% vs 82.5%, respectively; P < .001; data not shown in Table 3), and among those with a test result, a higher percentage had high HbA1c (45.1% vs 42.8%, respectively; P < .01; data not shown in Table 3).
For this study, we created 2 measures of access to ECP services, the FY2012 facility-specific ECP supply rate and estimated patient drive times from their communities to facilities providing ECP services in FY2012. Information on ECP access by study cohort and ECP user status is provided in Table 3. The data in Tables 4 and 5 provide additional information on these 2 measures. According to the Table 3 results for adults with diabetes, ECP users compared with nonusers lived in communities where the average facility ECP supply rate was higher (0.35 vs 0.30, P < .001). ECP users also had shorter estimated drive times to ECP facilities (15.2 minutes vs 22.5 minutes, P < .001).
ECP Service Provision by Site and Facility for FY2012.
Selected Characteristics Among Adults With Diabetes by Use of ECP Services in FY2012.
Project sites varied in the number of facilities (ie, hospitals or clinics) that provided ECP and the supply of ECP services at those facilities (see Table 4). During FY2012, 3 project sites had 1 facility providing ECP; in 5 sites, ≥3 facilities did. Across all Data Project sites, the average ECP supply rate was 0.32 visits per adult in the facilities' service areas. The ECP supply rate varied from a low of 0.08 at 1 facility in a Data Project site to a high of 0.89 at another facility in the same site.
Table 4 also includes information on ECP use across facilities within a Data Project site by CVD status to discern how use patterns may vary by health status. Although some CVD status differences were observed (ie, use at facilities 2 and 3 in site 9 differed by CVD status), we did not identify any trends that influenced our approach to specifying the model used to conduct the comparative effectiveness study.
According to the results presented in Table 3 for each study cohort, ECP users lived in communities with statistically shorter drive times. Among adults with diabetes, the mean drive time for ECP users was 15.2 minutes. Drive times among nonusers in this study cohort averaged 22.5 minutes (P < .001). Table 5 provides the distribution of patient drive times among adults with diabetes by ECP user status. Among adults with diabetes, the distribution among ECP users differed statistically from that of nonusers (P < .001). Among ECP users, nearly 58% lived in communities with estimated drive times <10 minutes; among ECP nonusers, approximately 46% did.
According to the county-level 2010-2014 ACS data and the FY2011 study population's county weights, adults with diabetes lived in counties where 31.2% of AI/AN households had incomes <100% of the FPL, 12.2% had incomes between 100% and 138% of the FPL, 43.8% had incomes between 139% and 399% of the FPL, and 12.7% had incomes ≥400% of the FPL (Table 5). ECP users were statistically more likely than were nonusers to live in counties with higher household incomes. As summarized in Table 3, ECP users lived in counties where the mean percentage of households with low incomes was 41.1%, which is statistically lower than that for ECP nonusers (45.1%, P < .001). With respect to educational attainment, adults with diabetes lived in counties where 46.5% of AI/AN adults reported <12 years of education, 23.0% had a high school degree or equivalent, 23.6% reported some postsecondary education, and 7.0% reported completing ≥4 years of college. There were no significant differences in educational attainment between ECP users and nonusers.
During FY2012, adults with diabetes averaged 1.1 ECP visits (Table 6). Approximately 40% had at least 1 ECP visit. Among ECP users, the average number of ECP visits was 2.8; 45.4% had 1 ECP visit, 23.2% had 2 visits, 11.1% had 3 visits, and 20.3% had ≥4 visits. Average ECP use in FY2012 was higher among adults with both diabetes and CVD than among those with diabetes without CVD (1.7 visits vs 0.9 visits, respectively). The larger average among adults with both conditions stemmed from a larger percentage of these adults using ECP services (47.2% vs 38.7%) and a higher average number of visits among ECP users (3.7 visits vs 2.3 visits).
Use of ECP Services Among Adults With Diabetes for FY2012.
ECP services include diabetes education, nutrition education, APP, case management, and other health education services. The use of these 5 types of ECP services varied by study cohort (Table 7). Among adults with diabetes without CVD, 41.2% of ECP visits were nutrition visits, and 19.1% were APP visits. In contrast, 26.5% of ECP visits among adults with both diabetes and CVD were nutrition visits, and 47.6% were APP visits.
Use of ECP Services by Type of Visit Among Adults With Diabetes Who Had at Least 1 ECP Visit, FY2012.
Descriptive information on the bivariate associations between ECP use and patient FY2011 and FY2013 outcome measures is provided in Table 8 by study cohort. Across the 3 study cohorts, there were no baseline (FY2011) differences in the percentages of adults with high SBP by ECP status. Between FY2011 and FY2013, the percentage of adults with diabetes with high SBP increased among ECP users and nonusers. However, in FY2013, a statistically lower percentage of ECP users than nonusers had high SBP among all adults with diabetes and in the subgroup without CVD.
Health Status and Service Use in FY2011 and FY2013 by Use of ECP Services in FY2012, Bivariate Associations.
Among all adults with diabetes, and in the subgroup without CVD, a statistically higher percentage of ECP users had high HbA1c at baseline (FY2011). Similar to SBP, the percentage of adults with high HbA1c increased between FY2011 and FY2013 across the 3 study cohorts. Although there were no statistically significant differences between ECP users and nonusers in the percentage with high HbA1c in FY2013 across the 3 study cohorts, increases in the percentage with high HbA1c between FY2011 and FY2013 were statistically smaller in ECP users among all adults with diabetes and in the subgroup without CVD.
In contrast to SBP and HbA1c, the percentage of adults with high LDL-C decreased between FY2011 and FY2013 across the 3 study cohorts. A statistically lower percentage of ECP users, compared with nonusers, had high LDL-C in both FY2011 and FY2013. However, there were no differences between ECP users and nonusers in the change between FY2011 and FY2013 in the percentage with high LDL-C.
The last 2 health status outcome measures were onset of CVD and ESRD in FY2013. Among adults with diabetes without CVD, the onset of CVD was statistically higher among ECP users than among nonusers (9.1% vs 7.3%, respectively; P < .001). There were no observed differences between ECP users and nonusers in the onset of ESRD in any of the 3 study cohorts. Because the percentages of adults with ESRD onset among all adults with diabetes and among adults with diabetes without CVD were less than <1%, we conducted multivariable analyses for the onset of ESRD only among adults with diabetes and CVD, for whom the onset was >1%.
Differences in hospital service use by ECP use varied by hospital use measure and study cohort (Table 8). Statistically significant differences in the average numbers of ED visits during FY2011 and FY2013 were identified among all adults with diabetes and in the subgroup without CVD. Among adults with diabetes, there was a significant reduction in ED use between FY2011 and FY2013 among ECP users compared with nonusers. Among adults with diabetes without CVD, ED use increased over time but to a significantly lesser extent among ECP users than among nonusers.
The findings concerning hospitalizations and PP hospitalizations were similar. Significant differences in these 2 measures were identified only among all adults with diabetes. A higher percentage of ECP users than nonusers had ≥1 hospitalizations in FY2011 (10.3% vs 8.8%, respectively; P < .001) and ≥1 PP hospitalizations in FY2011 (2.6% vs 2.1%, respectively; P < .01). ECP users, compared with nonusers, had significant decreases in hospital use between FY2011 and FY2013 for both measures. In FY2011, the average number of hospital inpatient days among adults with diabetes was higher among ECP users than among nonusers (0.61 days vs 0.49 days, respectively; P < .01). No changes in hospital days by ECP use were observed.
In this section, we describe the results from the comparative effectiveness analyses conducted using IPW propensity models to evaluate patient outcomes in FY2013 associated with ECP use during FY2012. First, we describe detailed results from an IPW propensity model estimated to examine the relationship between any ECP use (ie, no visits compared with ≥1), using logistic regression, and high SBP among adults with diabetes (Table 9). Using the same approach, we next summarize results concerning the association between any ECP use with health status and hospital use outcomes, for each study cohort, in Table 10. As described previously, the level of ECP use during FY2012 ranged from 1 visit to ≥3 visits. To understand the association between the level of ECP use and patient outcomes, we estimated IPW propensity models using an ordered logistic regression to first predict a patient's propensity to have no ECP visits, 1 to 2 ECP visits, and ≥3 ECP visits in FY2012. We summarize our findings from the models estimated to examine the association between the level of ECP use and each health status and service use outcome in Table 11.
Association Between Any Use of ECP Services During FY2012 and High Blood Pressure (SBP ≥140 mm Hg) During FY2013 Among Adults With Diabetes, Propensity Score Model With IPW.
Patient Outcomes During FY2013 Associated With Use of ECP Services During FY2012, Propensity Score Model, Based on a Logistic Regression, Using IPW.
Patient Outcomes During FY2013 Associated With Level of ECP Service Use During FY2012, Propensity Score Model, Based on an Ordered Logistic Regression, Using IPW.
Table 9 includes findings from 1 model estimated to evaluate the association between any ECP use by A/AN adults with diabetes and the odds of high SBP. The results include findings from 2 equations. The first equation was estimated to predict the propensity to use ECP services during FY2012 by an adult with diabetes; the results presented in Table 9 describe patient and provider characteristics that influenced ECP use. Table 9 also includes results from the model's second equation estimated to evaluate the association between any ECP use and the odds of high SBP in FY2013.
According to equation 1, the treatment equation, females were found to have higher odds than males of using ECP services (odds ratio [OR], 1.12, P < .001). The influence of patient health status during FY2011 on ECP use during FY2012 was assessed through the use of 3 different types of measures. Patient morbidity burden was positively associated with greater odds of ECP use. Adults in the third and fourth highest health risk quartiles, compared with adults in the lowest health risk quartile, had 1.71 (P < .001) and 2.13 (P < .001) odds of using ECP, respectively. Patient health status, as measured by clinical measures of blood pressure, HbA1C, and cholesterol control, significantly influenced ECP use. Adults with high SBP and high LDL-C were less likely to use ECP services, while those with high HbA1c were more likely to use them, with ORs of 0.87 (P < .001), 0.86 (P < .001), and 1.33 (P < .001), respectively. Finally, adults diagnosed with a drug or alcohol use disorder, compared with adults who were not, had lower odds of using ECP services (OR, 0.84; P < .05). Other specific health conditions were not significantly associated with a patient's odds of using ECP services.
Patient community, county, and facility characteristics were associated with any use of ECP services. Each 10 minutes of additional drive time was associated with increasingly lower odds of using ECP services (OR, 0.96 per 10 minutes of drive time; P < .001). Each unit increase in the percentage of households with low incomes in a county was associated with lower odds of using ECP services (OR, 0.98 per percentage point increase in households with low income; P < .001). In contrast, each percentage point increase in the percentage of adults with less than a high school degree in a county was associated with higher odds (OR, 1.03; P < .001) of using ECP services.
The facility ECP supply rate (ie, the average number of ECP visits provided per adult) was positively associated with ECP use when equation 1 of the propensity model was estimated without fixed effects to control for site-level variation. The results presented in Table 9 were estimated using a fixed-effects model, and the relationship between the supply of ECP services and ECP use in this model was not significant.
The results from equation 2, the outcome equation, indicate that any use of ECP in FY2012, compared with no use, was associated with a 0.85 (P < .001) lower odds of having high SBP in FY2013. Other patient characteristics associated with high SBP in FY2013 included older age; a diagnosis of hypertension, chronic kidney disease, a drug or alcohol use disorder, or CVD; and high SBP, high HbA1c, or high LDL-C at baseline (FY2011).
The results of the IPW propensity model for the association between any use of ECP and patient outcomes in FY2013 are summarized in Table 10 for each study cohort. Any ECP use was associated with lower odds of having high SBP during FY2013 among all adults with diabetes (OR, 0.85; 95% CI, 0.79-0.93; P < .001). This association was observed in the subgroup without CVD (OR, 0.83; 95% CI, 0.75-0.91; P < .001) but not in the subgroup with CVD. Any use of ECP was also associated with lower odds of having high LDL-C in FY2013 among all adults with diabetes (OR, 0.89; 95% CI, 0.84-0.98; P < .01). This finding was observed in the subgroup with CVD (OR, 0.82; 95% CI, 0.71-0.96; P < .05) but not in the subgroup without CVD. There was no statistical association between any ECP use and high HbA1c.
The relationship between any ECP use and new or recurring CVD onset in FY2013 was evaluated among adults with diabetes without CVD in FY2011 and FY2012. The results indicated no statistical association between any use of ECP and CVD onset. Due to the very low rate of ESRD onset in FY2013 among all adults with diabetes (0.5%) and among adults with diabetes without CVD (0.1%), we evaluated ESRD onset only among adults with both diabetes and CVD, who had a rate of onset of 1.3%. The analysis population for this patient outcome included adults with diabetes and with ESRD in FY2013, unlike propensity-weighted analyses for other patient outcomes that excluded adults with ESRD during any FY. Any use of ECP during FY2012 was associated with lower odds (OR, 0.60; 95% CI, 0.39-0.93; P < .05) of ESRD onset among adults with both diabetes and CVD.
ECP use was significantly associated with reduced use of hospital ED and inpatient services during FY2013 among all study cohorts. However, this relationship varied by hospital use measure and CVD status. ECP use, compared with no use, was associated with 0.8 fewer ED visits during FY2013 (95% CI, −0.12 to −0.05 visits; P < .001) among all adults with diabetes. This association was similar among those without CVD (−0.09 visits; 95% CI, −0.13 to −0.05 visits; P < .001) and with CVD (−0.10 visits; 95% CI, −0.19 to −0.02 visits; P < .05).
The odds of having ≥1 hospitalizations during FY2013 were lower among all adults with diabetes who were ECP users (OR, 0.80; 95% CI, 0.71-0.89; P < .001) than among nonusers. This association was similar among those without CVD (OR, 0.77; 95% CI, 0.66-0.89; P < .001) and with CVD (OR, 0.74; 95% CI, 0.61-0.91; P < .01). Any ECP use was also associated with lower odds of having ≥1 PP hospitalizations among all adults with diabetes (OR, 0.79; 95% CI, 0.64-0.91; P < .05). This finding was significant among those with CVD (OR, 0.71; 95% CI, 0.55-0.91; P < .01) but not among those without CVD. Among all adults with diabetes, ECP use was significantly associated with 0.13 fewer hospital inpatient days (95% CI, −0.21 to −0.04 days; P < .01). This association was also significant in both subgroups, those without CVD (−0.13 days; 95% CI, −0.20 to −0.06 days; P < .001) and those with CVD (−0.32 days; 95% CI, −0.57 to −0.07 days; P < .05).
Table 11 summarizes IPW propensity model results for health status and hospital use outcomes in FY2013 associated with a patient's level of ECP use during FY2012. Using ordered logistic regressions to create the propensity model, we estimated patients' propensity to have had 1 to 2 ECP visits compared with no ECP visits, ≥3 ECP visits compared with no ECP visits, and ≥3 visits compared with 1 to 2 visits, and examined patient outcomes in FY2013 associated with each level of ECP use. In this section, we describe the relationships between the level of ECP use with the same 5 health status outcomes included in Table 10.
Among all adults with diabetes, there was a graded association between the level of ECP use and high SBP: 1 to 2 ECP visits and ≥3 ECP visits, vs no visits, were associated with lower odds of high SBP (OR, 0.89 [95% CI, 0.82-0.97; P < .05] and OR, 0.74 [95% CI, 0.63-0.86; P < .001], respectively), and ≥3 ECP visits, compared with 1 to 2 visits, was associated with lower odds of high SBP (OR, 0.82; 95% CI, 0.70-0.97; P < .05). This pattern was generally similar across subgroups with and without CVD, though in the subgroup with both diabetes and CVD, there were no statistically significant associations between the level of ECP use and high SBP (Table 11).
There was also a graded association between the level of ECP use and high LDL-C among all adults with diabetes, where adults with ≥3 visits vs no visits, and adults with ≥3 visits vs 1 to 2 visits, had lower odds of high LDL-C (OR, 0.77 [95% CI, 0.67-0.89; P < .001] and OR, 0.83 [95% CI, 0.71-0.97; P < .05], respectively). The results were generally similar in the subgroup with diabetes without CVD, but in the subgroup of adults with both diabetes and CVD, ECP use was associated with the odds of high LDL-C only among adults with 1 to 2 visits vs no visits (OR, 0.82; 95% CI, 0.68-0.97; P < .05).
There was no clear pattern indicating an association between level of ECP use and high HbA1c. We found that ECP use was significantly associated with higher odds of high HbA1c among all adults with diabetes who had 1 to 2 ECP visits vs no ECP visits (OR, 1.092; 95% CI, 1.004-1.186; P < .05) and in the subgroup with both diabetes and CVD who had ≥3 visits vs no visits (OR, 1.28; 95% CI, 1.02-1.61; P < .05).
Among adults with diabetes without CVD, adults with ≥3 ECP visits during FY2012, vs no visits, had lower odds of CVD onset (OR, 0.79; 95% CI, 0.63-0.99; P < .05) in FY2013. The relationship between ECP use and CVD onset was not significant among adults with 1 to 2 visits vs no visits.
Among adults with diabetes and CVD, adults with 1 to 2 visits vs no visits had statistically significant lower odds of ESRD onset (OR, 0.55; 95% CI, 0.33-0.92; P < .05). The OR for ESRD onset among adults with ≥3 visits vs no visits was 0.64, but it was not statistically significant. This may be due in part to the low number of adults with ≥3 ECP visits who had ESRD onset in FY2013.
The association between the level of ECP use and hospital service use during FY2013 varied by measure and study cohort. Among all adults with diabetes, those with 1 to 2 ECP visits vs no visits, and those with ≥3 visits vs no visits, had 0.08 (95% CI, −0.13 to −0.04; P < .001) and 0.11 (95% CI, −0.17 to −0.04; P < .01) fewer ED visits, respectively. The results for the subgroup without CVD were generally similar. In the subgroup with CVD, adults with 1 to 2 visits vs no visits had 0.14 (95% CI, −0.24 to −0.04; P < .01) fewer ED visits, but having ≥3 ECP visits was not associated with fewer ED visits.
Among all adults with diabetes, the odds of ≥1 hospitalizations during FY2013 were significantly lower among those with 1 to 2 ECP visits than among those with no visits and among those with ≥3 ECP visits than among those with no visits (OR, 0.82 [95% CI, 0.72-0.93; P < .01] and OR, 0.75 [95% CI, 0.61-0.91; P < .01], respectively). The OR for adults with ≥3 visits compared with 1 to 2 visits was not significant. The findings concerning the level of ECP use and hospitalization were generally similar in the subgroups with and without CVD.
The level of ECP use had a statistically significant relationship with having ≥1 PP hospitalizations among all adults with diabetes and in the subgroups with and without CVD. Among all adults with diabetes, adults with ≥3 visits vs no visits, and adults with ≥3 visits vs 1 to 2 visits, had lower odds of PP hospitalizations (OR, 0.62 [95% CI, 0.46-0.82; P < .01] and OR, 0.72 [95% CI, 0.53-0.98; P < .05], respectively). This graded association was seen mainly in the subgroup with CVD. In the subgroup without CVD, the odds of PP hospitalizations were only lower in a comparison of ≥3 visits with no visits (OR, 0.647; 95% CI, 0.420-0.997; P < .05).
Among all adults with diabetes, more ECP visits were associated with fewer hospital inpatient days. Those with ≥3 visits vs no visits, and those with ≥3 visits vs 1 to 2 visits, had 0.23 (95% CI, −0.34 to −0.12; P < .001) and 0.14 (95% CI, −0.26 to −0.01; P < .05) fewer hospital days, respectively. The relationship between hospital days and having 1 to 2 visits vs no visits was not significant. The results for the subgroup analyses differed somewhat. Among adults with diabetes without CVD and among adults with both diabetes and CVD, having 1 to 2 visits vs no visits, and having ≥3 visits vs no visits, were associated with fewer hospital days (Table 11).
IHS and Tribes provide an array of services for A/AN adults with diabetes to prevent complications and improve health outcomes. An important component is the provision of ECP services, in addition to primary and specialty care services. In this comparative effectiveness study, we examined the relationship between ECP use, compared with usual care, and patient outcomes using an observational study design that included the analysis of data for nearly 28 000 AI/AN adults with diabetes from 12 Data Project sites located throughout the country. Using data from FY2011 to FY2013, propensity score models were estimated to control for the nonrandom assignment of patients into the treatment group (ie, patients who used ECP services during FY2012).
The comparative effectiveness study results suggest that the use of ECP services by adults with diabetes was associated with short-term and long-term health status outcomes (ie, lower odds of high SBP, high LDL-C, and CVD and ESRD onset). The study results also indicate that ECP use was associated with lower use of hospital emergency and inpatient services. The hospital use findings may have important implications not only for patients' health but also for resource use by IHS and Tribal health programs due to their financial constraints.11,12 In this section, we summarize and interpret key study findings, provide the recommendations of our Collaborative Network, highlight the implications of these results for important subpopulations of adults with diabetes, describe study limitations, and identify topics for future research.
During FY2012, 41.0% of AI/AN adults with diabetes used ECP services, averaging 2.8 visits during the year. Among adults with diabetes, any ECP use during FY2012, compared with no use, was associated with lower odds of high SBP and high LDL-C in FY2013. Because high SBP and high LDL-C influence the risk for diabetes-related complications,28 the impact of these findings may be substantial, particularly in light of the high percentages of ECP users who had high SBP and high LDL-C at baseline (approximately 23% and 42%, respectively).
Among adults with diabetes without CVD, ≥3 ECP visits vs no visits was associated with lower odds of CVD onset during FY2013. Among adults with diabetes and CVD, any use of ECP services vs no use was associated with lower odds of ESRD onset during FY2013. These findings should be interpreted with caution. CVD and ESRD are conditions that evolve over years, and the onset of both conditions was assessed in our study during a 12-month period (FY2013) following ECP use (during FY2012). Moreover, the statistical significance of the findings was borderline. However, given our findings and the high morbidity and mortality associated with CVD and ESRD, future studies that include longer follow-up periods are needed to better understand the potential impact of ECP use on these outcomes. In addition, we examined the onset only of ESRD, the final stage of chronic kidney disease. Future analyses could examine the onset of and progression through earlier stages of chronic kidney disease.
We did not identify an association between ECP use and high HbA1c. Other studies have found varying relationships between ECP services and clinical outcomes such as SBP and HbA1c.43,50 There are many factors that could explain this finding. Among adults with diabetes, a higher percentage of ECP users than nonusers had an HbA1c test and a high HbA1c value at baseline. The difference between ECP users and nonusers in the percentage with a high value at baseline was not observed for SBP and LDL-C. Despite the baseline difference in high HbA1c, no difference in high HbA1c was found between ECP users and nonusers at follow-up. Due to the importance of controlling HbA1c levels for adults with diabetes, future research should address patient characteristics (eg, age, sex, comorbidity status) associated with having an HbA1c test, ECP referrals for patients with high HbA1c levels, and how HbA1c control is addressed during ECP and other outpatient visits.
We compared our health outcome results with those of AI/AN adults with diabetes who participated in the SDPI Healthy Heart demonstration project that provided case management services.50 The program participants with outcomes data after 1 year of enrollment (n = 2259) in that study had, on average, 7 case management visits during that 12-month period. Although our comparative effectiveness study addressed the limitations of the Healthy Heart study by including a large number of AI/ANs with diabetes and a comparison population, the average number of ECP visits among ECP users in our study was 2.8. Approximately a third of ECP users had ≥3 visits. Despite the lower level of service use in our study, we found ECP to be associated with improvements in SBP and LDL-C levels, similar to the findings from the Healthy Heart case management study. Although we found no association between ECP use and high HbA1c levels, Healthy Heart participants had a small improvement in HbA1c levels. The difference in HbA1c results between the 2 studies may be due, in part, to differences in the outcome measure (ie, HbA1c level vs high HbA1c) and, in our study, baseline differences in high HbA1c between ECP users and nonusers. Healthy Heart participants were found to have lower CVD risk at follow-up; the study, however, did not report on CVD or ESRD onset. To our knowledge, our study is the first to report a relationship between the use of ECP services and the onset of CVD and ESRD among AI/ANs with diabetes. The ESRD results from this study align with findings by Bullock and colleagues that documented reductions in ESRD onset following the implementation of SDPI; the study by Bullock and colleagues did not include information on ECP services.27
As described previously, ECP use during FY2012 was associated with reductions in hospital service use in FY2013. Adults with diabetes averaged 0.79 ED visits in FY2011. Any use of ECP services, compared with no use, was associated with nearly a 10% reduction in ED use in FY2013. In FY2011, the percentage of adults with diabetes with ≥1 hospitalizations was 9.4%, and the percentage with ≥1 PP hospitalizations was 2.3%; they averaged 0.53 days in the hospital. Any use of ECP was also associated with lower odds of hospitalization and a reduction in hospital inpatient days of approximately 25%. Future research is needed to assess treatment costs associated with the provision of ECP and potential cost savings from reduced use of hospital and other services.
The propensity model results indicated that for 5 (high SBP, high LDL-C, onset of CVD, ≥1 PP hospitalizations, hospital inpatient days) of the 9 outcomes studied, higher levels of ECP use (ie, ≥3 visits) were associated with greater improvement in patient outcomes. Among all adults with diabetes, ≥3 visits vs no visits and ≥3 visits vs 1 to 2 visits were associated with lower odds of high SBP, high LDL-C, ≥1 PP hospitalizations, and fewer hospital days, suggesting a dose-response association between ECP use and these outcomes.
We conducted a series of analyses to understand the use of ECP among adults with diabetes with and without CVD, the 2 subgroups of interest. Adults with both diabetes and CVD, vs those with diabetes without CVD, were, on average, older and had a higher morbidity burden. A higher percentage had high SBP, and a lower percentage had high LDL-C. They had higher use of ECP and hospital services. For example, the percentage of adults with both diabetes and CVD with any hospitalizations in FY2011 was 20.5%, compared with 5.5% among adults with diabetes without CVD. Despite these differences, some evidence of a positive association between ECP use and patient outcomes (ie, significant improvement) was found among those with and without CVD for 5 (high LDL-C, ED visits, ≥1 hospitalizations, ≥1 PP hospitalizations, hospital inpatient days) of the 7 outcomes examined in both populations. Below, the results for each subgroup are described in greater detail; we did not test for treatment response heterogeneity in the analyses.
Any use of ECP services, compared with no use, was associated with lower odds of high SBP among adults with diabetes without CVD, but not among those with both diabetes and CVD. This may be due in part to the higher percentage of adults with both diabetes and CVD, compared with those without CVD, who had high SBP at baseline, were of older age, and had higher morbidity burden.
LDL-C results by CVD status differed from those of SBP. Any use of ECP was associated with lower odds of high LDL-C among adults with both diabetes and CVD. Although the relationship between any ECP use and high LDL-C was not significant among adults with diabetes without CVD, it was significant in that subgroup for those with ≥3 ECP visits vs no visits, suggesting that the impact of ECP on this outcome may depend on consistent ECP use.
The results concerning the level of ECP use and high HbA1c among adults with and without CVD differed from the data for high SBP and high LDL-C, as they did when examining any ECP use and high HbA1c. There was no clear pattern indicating an association between the level of ECP use and high HbA1c. The propensity model results indicated that adults with diabetes who were experiencing poor glycemic control were more likely to use ECP services, possibly explaining the observed association between more ECP use and higher HbA1c. These results further support the need for future research on the relationship between ECP use and HbA1c.
ECP use was associated with fewer ED visits among adults with diabetes with and without CVD. Among adults with diabetes without CVD, ED visits were lower among those with 1 to 2 ECP visits and ≥3 visits than among those with no visits. However, among adults with both diabetes and CVD, statistically fewer ED visits were observed only for those with 1 to 2 visits. It may be that reasons for ED use among adults with diabetes and CVD differed from those for adults with diabetes without CVD (eg, higher medical complexity), leading to differences in the relationship between ECP use and ED visits.
ECP use was associated with lower odds of having ≥1 hospitalizations and fewer hospital days among adults with diabetes with and without CVD. However, any ECP use was statistically associated with lower odds of ≥1 PP hospitalizations among adults with both diabetes and CVD but not among adults with diabetes without CVD. However, in the latter group, those with ≥3 ECP visits had significantly fewer PP hospitalizations than those with no visits. Because the percentage of adults with any PP hospitalizations was lower among those without CVD, it was not surprising that the relationship between ECP use and this outcome also differed.
The IPW propensity models provided important information about other patient characteristics associated with ECP use; the results identified 3 subpopulations of adults with diabetes who merit specific attention by efforts to increase ECP use. Men, compared with women, and adults with an alcohol or drug use disorder compared with those without, were less likely to use ECP services. Higher morbidity burden, compared with a lower burden, and having CVD were associated with a greater likelihood of ECP use. Younger and healthier adults with diabetes had lower ECP use.
Each subpopulation has unique characteristics to consider. Men with diabetes, compared with women with diabetes, had a higher prevalence of CVD, kidney disease, and substance use disorders. Future research will address differences not only in health status but also in health service use patterns more broadly. Adults with diabetes with an alcohol or drug use disorder may benefit not only from ECP use but also from coordination with behavioral health services.
The prevalence of CVD, morbidity burden, and ECP use increased with age among adults with diabetes. Although adults aged 18 to 44 years averaged <1 ECP visit in FY2012, adults aged ≥65 years averaged 1.5 visits. In addition, a larger percentage of adults aged 18 to 44 years, compared with adults aged ≥65 years, did not have a test result for HbA1c during FY2011 (nearly 20% vs approximately 10%, respectively), and a higher percentage of the younger adults had high HbA1c (ie, 30% vs 18%) during that year. The relationship between age and ECP use merits consideration, particularly given that a goal of providing ECP services is to prevent the onset of diabetes-related complications.
The project's Collaborative Network reviewed study findings concerning patterns of ECP use and patient outcomes associated with its use and assisted with the interpretation of study results. Based on the premise that ECP use improves patient outcomes by improving patients' knowledge to assist in their management of diabetes and navigation of the health system to obtain recommended services (eg, medical testing, dental care, specialty services, and smoking cessation support) at health facilities and through community-based organizations, the Collaborative Network worked with CAIANH personnel to develop strategies to facilitate patients' ability to make informed choices about using ECP services and to enhance the provision of ECP services to address patients' needs. The Collaborative Network suggested that providers consider the help-seeking behaviors and lifestyles of younger adults, men, and individuals with alcohol or drug use disorders when implementing strategies to help adults with diabetes understand the benefits of using ECP services and enhancing ECP access to promote increased ECP use. Although we could not study outcomes among adults who were newly diagnosed with diabetes due to a lack of data on this topic, the Collaborative Network identified this population as another subpopulation meriting specific attention.
The Collaborative Network discussed the lifestyles of younger adults and the emotions associated with a diabetes diagnosis; they highlighted the importance of providing ECP in a supportive and accessible manner to improve patients' ability to cope. Existing examples of such support include conducting a prediabetes group that meets during evenings and provides incentives for participation, having an “open door” policy for patients newly diagnosed with diabetes, and co-locating ECP services within primary care clinics to improve access and care coordination. Collaborative Network members commented on the different types of health personnel who provided ECP services to accommodate Data Project site characteristics. Regardless of the type of personnel, they deemed phone and other types of patient outreach between visits important. The Collaborative Network suggested that ECP outreach and service delivery accommodate differences according to gender and health status. Although gender-specific or health-status-specific ECP outreach, such as educational materials and community-based strategies, may require additional provider resources, the relationship between those resources and ECP use should be studied to understand their effectiveness.
In this study, we found that adults with diabetes and an alcohol or drug use disorder were less likely to use ECP services. Other studies have found a relationship between depression and the ability of patients with diabetes to manage their condition. Collaborative Network members recommended the evaluation and enhancement of services provided to improve the diagnosis and treatment of substance use and mental health disorders, such as screening for these disorders in primary care clinics and other settings. To improve the understanding of behavioral health services, including support groups, by patients with these disorders, they recommended that providers consider co-locating behavioral health services within primary care clinics, or nearby, to improve access and coordination between providers. In several Data Project sites, the behavioral health services are located in buildings separate from those housing primary care services, or the behavioral health services are located in the same building but in a location that is unfamiliar to many patients. Finally, Collaborative Network members noted the importance of providing information to patients with diabetes on the high prevalence of behavioral health disorders among adults with diabetes as a means to reduce any stigma associated with such diagnoses and with seeking related treatment.
The Collaborative Network recommended that health care providers consider options to address identified barriers to ECP use. According to study results, patients with longer drive times between their communities and facilities that provided ECP services and patients living in counties with high rates of poverty were less likely to use ECP services. Collaborative Network members suggested that providers and other community organizations increase patient transportation services for patients who live in such areas and consider options for meeting such needs through the provision of home visits and mobile medical units.
Although not evaluated in our comparative effectiveness study, Collaborative Network members noted the importance of addressing variations in learning styles among patients. Members highlighted the benefits of providing culturally relevant services, including the use of Native languages for some patients. They also noted that some adults may prefer individual vs group ECP services and that efforts should be made to offer both types of ECP services.
Many study limitations need to be considered when interpreting the results of this comparative effectiveness study. In many studies of patient outcomes associated with the use of education, case management, and/or APP services for adults with diabetes, the interventions were designed to include >3 visits during the intervention time period (eg, a 12-month period), as described previously for the Healthy Heart demonstration project on case management services.50 Thus, the level of ECP use by some adults in our study may not have been sufficient to understand ECP's potential influence on patient outcomes. At the same time, these data reflect actual ECP use patterns in I/T clinics, in which some patients only used ECP services 1 to 2 times during a 12-month period.
Our use of 3 discrete 12-month time periods and the measurement of ECP use during 1 of these periods did not allow for intervention periods specific to patients' ECP use. Because patients may or may not have used ECP services during previous years, this aspect of our study design may have biased the relationships between ECP use and patient outcomes toward the null. For example, a patient may have initiated ECP use during late FY2011 and had the last ECP visit during mid-FY2012. The FY2011 visits would not have been included in our measure of ECP use. We may consider allowing individually defined intervention periods in future studies.
Other limitations included our inability to control for all differences between users and nonusers in the analysis and Data Project site variation in the provision of ECP and other services. Although the IPW propensity model controlled for observed differences between ECP users and nonusers, there may have been important unobserved provider and patient characteristics that were associated with ECP use and patient outcomes, for which we did not and could not control, and that could have biased our results. For example, project sites varied in resources allocated to the provision of ECP, ECP data quality and methods, and coordination between ECP and primary care providers. We did not have measures of these characteristics, and these differences may have influenced ECP use as well as outcomes associated with its use. With respect to patient characteristics, motivation to maintain or improve one's health status may influence not only ECP use but also patient outcomes. Thus, lacking a measure of motivation, we may have overestimated the influence of ECP use on patient outcomes. For this reason, we estimated the IPW propensity models with and without fixed effects in both propensity model equations to control for project site variation. The fixed effects also controlled for patient differences across project sites. For many patient outcome measures, we found that the results from analyses with and without fixed effects were largely consistent. However, both the propensity score analysis and the outcomes analysis, which used the propensity score, are subject to bias associated with unmeasured confounders that in many cases are unmeasurable.
Another consideration related to project site variation is that we attempted to examine the comparative effectiveness of ECP compared with usual care without ECP use. It might be difficult to statistically disentangle the benefits of ECP from those of usual care at some sites, as many services may have interacted (eg, coordination among primary care providers, laboratory services, diabetes education) and provided benefits to the patients. We plan to explore the influence of these and other factors, such as ECP use over a longer time period (eg, 24 months compared with 12 months), on outcomes in future analyses.
We conducted another set of analyses to examine the sensitivity of our results to assumptions associated with the use of an IPW propensity model in this comparative effectiveness study. We estimated patient outcomes using a matched propensity model (creating an ECP population similar to a usual care population by matching ECP patients to usual care patients with the same propensity score) and found that the results from that analysis were largely consistent with the results from the IPW propensity model. It is important to recognize that we were not able to control for unobserved patient variation in motivation to maintain or improve their health status that may not only have influenced their use of ECP but also influenced other behaviors (eg, diet, exercise, medication adherence), regardless of ECP use. Because of this limitation, we may have overestimated the influence of ECP use on patient outcomes, but the influence of this and other unmeasured confounders on outcomes is unknown.
Our analyses required that an individual lived in the Data Project sites and used IHS or Tribal health services during 3 consecutive FYs (FY2011-FY2013) and did not have a condition that excluded them from the study population (eg, malignant cancer, ESRD in most analyses). Within each study cohort, analyses indicated that the study population may have been, on average, healthier than those who did not meet inclusion criteria. In analyses of SBP, HbA1c, and LDL-C outcomes, an FY2013 value was required for inclusion in the analyses; the percentages of all adults with diabetes with FY2013 missing data for SBP, HbA1c, and LDL-C were 6%, 15%, and 25%, respectively. Some patients who did not meet study inclusion criteria may have moved to another geographic area, due to poor health or employment, or had health coverage that provided reimbursement for non-IHS services. The percentage of adults who met these criteria (ie, the number who met these criteria divided by those who lived in one of the project sites in FY2011) varied by study cohort; it was lowest among adults with both diabetes and CVD. Patient mortality during FY2012 or FY2013 most likely contributed to this finding. Thus, the inclusion criteria may have biased study findings by including healthier adults with diabetes, compared with the actual patient population of adults with diabetes.
Data from a limited number of sites were excluded from some patient outcome analyses due to missing data, largely associated with EHR system conversions. It was not possible to assess any potential bias that may have been associated with these exclusions.
Other study limitations may have influenced project findings more broadly. First, the study included data for IHS active users who lived in 12 Data Project sites. They represent a large proportion of individuals eligible for I/T health services. Nevertheless, the findings may not reflect the health status of AI/AN peoples who live elsewhere or who do not obtain health services from I/T providers. There are more than 550 federally recognized Tribes throughout the United States, with considerable variations in culture, traditions, and history.22,82
Second, similar to other projects that involve the use of administrative health data, we reported the prevalence of conditions based on diagnoses included in medical service use records. Although using such data allowed us to include a large number of AI/ANs in the analysis, we did not have the level of detail and accuracy that would have been available from reviews of medical records or other types of health assessments. Third, these data include information for services provided by I/T programs or paid for by those programs through PRC. We did not have data on other services used by the study population, and access to other services varied across project sites. This limitation may have biased the prevalence rates for chronic conditions downward, and this bias likely varied across the sites. Fourth, sites also varied by the types of services provided (eg, specialty and inpatient services), PRC service use, and completeness of data.
Finally, the IHS data do not include measures for a wide array of personal and community characteristics that may be important for understanding health status and service use. We attempted to address this limitation by including in our analyses county-level measures of AI/AN educational attainment and household income and community-level measures of drive times to services.
Although some priorities for future research were described previously, here, we provide additional research intentions. First, patient outcomes included high SBP, high HbA1c, and high LDL-C. IHS guidelines do not include a recommendation that LDL-C tests be conducted annually. Some adults may not have had a cholesterol test due to the required fasting before blood is drawn for the test. Future analyses will assess patient and provider characteristics associated with HbA1c and LDL-C testing and test results. Second, we conducted a comparative effectiveness study of ECP services using a combined measure of different types of ECP services. In future studies, we plan to specifically examine the influence of specific types of services (ie, education, case management, and APP). Third, we plan to examine treatment costs and potential savings associated with ECP services. Finally, we plan to update the data infrastructure used in this project with information for more recent FYs, improving our ability to conduct longitudinal analyses.
Studies have documented exceedingly high morbidity and mortality associated with diabetes among AI/ANs who use IHS and Tribal health services. ECP services are provided for patients with chronic conditions to supplement those provided in primary and specialty care clinics. Although studies indicate that ECP use is associated with improved outcomes in other populations,32-49 there are limited data on the outcomes associated with ECP use among AI/ANs with diabetes. Using observational data for a large number of AI/ANs who lived throughout the country, we conducted a comparative effectiveness study to evaluate clinical and health care use outcomes associated with ECP using data from a 3-year time period. We supplemented IHS and Tribal health data with data from other sources to improve our ability to control for factors such as geographic access, county-level education, and household income in the study.
The comparative effectiveness study results suggested that ECP use was associated with improvements in health and reduced hospital service use and that higher levels of ECP use were associated with improved outcomes for several outcome measures relative to lower levels of use. The study provided important information on patient characteristics associated with ECP use in general and actual levels of ECP use, which may inform IHS, Tribes, and other organizations in allocating resources for ECP and other services to address the needs of AI/AN adults with diabetes.
This work could not have been conducted without the guidance and advice of colleagues at the IHS, including the Division of Planning, Evaluation, and Research; the Office of Public Health Support; the Office of Information Technology; IHS subcontractor Sue Ehrhart; and members of the project's Steering, Project Site, and Patient Committees who met regularly to provide consultation on the project. Last, the project includes data for many AI/AN communities. It would not have been possible to conduct this project without the support and approval of Tribal IRBs, Tribal Councils, and Tribal authorities who educated us about the health concerns they have for their members and how they hope this project will inform their work.
Research reported in this report was funded through a Patient-Centered Outcomes Research Institute® (PCORI®) Award (#AD-1304-6451). Further information available at: https://www.pcori.org/research-results/2013/improving-health-outcomes-among-native-americans-diabetes-and-cardiovascular
Influence of Study Inclusion and Exclusion Criteria on the Sample Size of the Study Population (PDF, 183K)
Evaluation of Propensity Score Model Specification (PDF, 217K)
O'Connell J, Rockell J, Reid M, Harty K, Perraillon M, Manson S. (2020). Improving Health Outcomes among Native Americans with Diabetes and Cardiovascular Disease. Patient-Centered Outcomes Research Institute (PCORI). https://doi.org/10.25302/11.2020.AD.13046451
The [views, statements, opinions] presented in this report are solely the responsibility of the author(s) and do not necessarily represent the views of the Patient-Centered Outcomes Research Institute® (PCORI®), its Board of Governors or Methodology Committee.
This book is distributed under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License which permits noncommercial use and distribution provided the original author(s) and source are credited. (See https://creativecommons.org/licenses/by-nc-nd/4.0/
Your browsing activity is empty.
Activity recording is turned off.
See more...
