Study setting and population
This observational cohort study took place in Navajo Nation from 2010 through 2014. Navajo Nation has 332,129 members and covers 27,000 square miles in parts of Arizona, Utah, and New Mexico[i]. The Navajo Area Indian Health Service (NAIHS) is one of 12 regional administrative units of the national IHS system. The NAIHS is divided into eight sub-regional “Service Units”, comprised of a total of 6 hospitals, 7 health centers, and 15 health stations. Separate from Indian Health Services, the tribe oversees its own health programs under the auspices of the Navajo Nation Department of Health, including the Navajo CHR Program. Formally established in 1968, the Navajo CHR Program includes nearly 100 CHRs who offer crucial services for patients and families including health education and in-home health assessments. Each Service Unit has between six and eighteen CHRs, each of whom is assigned to specific communities, called Chapters.
Program intervention
COPE was started in 2010 as collaboration between the Navajo CHR Program, Brigham and Women’s Hospital, and the NAIHS. This program has been implemented in all eight Service Units of the Navajo Area, and more than 650 patients have been enrolled to date. The “COPE intervention” is comprised of three inter-related strategies designed to strengthen existing community outreach and linkage to clinic-based care. The intervention focuses on three activities: 1) training CHRs to ensure proficiency in health topics and build skills in behavior change techniques such as motivational interviewing and goal-setting; 2) developing patient coaching materials that CHRs can then use in the home; and 3) facilitating greater connection and communication between the patient and clinic-based team. Centers for Disease Control and Prevention recommendations for strengthening the role of CHWs include 1) clear definitions regarding the scope of activities, 2) rigorous training standards, 3) increased clinician awareness about the role of CHWs, and 4) enhanced professional support networks for CHWs[ii]. COPE addresses all of these recommendations through standardized CHR training, close supervision, quality control of CHR activities, and integration of CHR activities into the primary care teams in IHS facilities.
Participant Selection
Although individuals living with any chronic condition may enroll in COPE, the vast majority have Type 2 diabetes mellitus, and, as described above, diabetes is a serious chronic health condition prevalent in Navajo Nation. Therefore, our prospective cohort study of COPE enrollees included COPE enrollees who: (1) have been diagnosed with diabetes, (2) received care in one of the participating Service Units, and (3) had at least one baseline measure of HbA1c prior to enrolling in COPE. The study encompassed six of the eight Service Units in Navajo Nation that had implemented COPE; we excluded two Service Units because of low COPE enrollment at one site, and use of a different electronic health record system at the other. We abstracted data for adults with an ICD-10 diagnosis of diabetes mellitus from the IHS resource and patient management system (RPMS). RPMS is an electronic health record that is used by the majority of IHS facilities for routine clinical care; each facility maintains an individual RPMS database with administrative and clinical data[iii]. In addition to clinical diagnoses, we abstracted laboratory tests, vital signs, medications, and healthcare utilization data from RPMS.
A matching algorithm was generated to identify patients seen at multiple sites and to identify COPE patients within the database. Once patients were identified across sites and COPE cases were captured, the dataset was de-identified. Within this dataset, we matched COPE participants to non-COPE patients based on age (+/- 5 years), gender, primary health facility, hemoglobin HbA1c (+/- 1 point) and systolic blood pressure (+/- 10 mm Hg) at baseline (i.e. three months prior to the date of COPE enrollment). The matching process is presented in detail in Trevisi et al.17
Study Outcomes and Confounding Factors
Our primary study outcome was the frequency with which patients are hospitalized, visit outpatient clinics, or otherwise use healthcare resources. IHS service unit utilization is primarily organized around clinics, and each healthcare encounter in RPMS lists the clinic visited by the patient. We used the RPMS clinic variable as the basis for the frequency of utilization analysis. With approximately 120 types of “encounters” (because “clinic” may also include miscellaneous types of utilization that are not, strictly speaking, a clinic visit, such as telephone call, chart review, and outpatient use of an inpatient treatment room, for example, we chose to use the broader term “type of encounter”), we grouped utilization into seven broad categories: community encounters, counseling/behavioral, dental, emergency, inpatient, primary care, and specialty care. Of note, community encounters included other community-based services – such as public health nurse visits and school visits – but did not include CHR visits.
Each type of encounter reported in RPMS was counted as one utilization incident for the purposes of this analysis. For inpatient services, the primary data point was the presence of a DRG code indicating the patient was hospitalized. However, we also included encounter listings for labor and delivery and for observation as inpatient utilization. Table 1 presents the encounters listed and their grouping for the purposes of the utilization analysis. We excluded from the analysis utilization such as telephone calls and telemedicine because the frequency with which these were reported suggested they were inconsistently recorded. Chart review was also excluded because it did not represent a separate, distinct encounter for the patient.
A second outcome was the utilization of lab and radiology services by counting the individual number of lab or radiology tests ordered, identified using Current Procedural Terminology (CPT) codes. The use of pharmacy services was measured using the number of medications prescribed.
Our outcomes of interest were the total number healthcare utilizations by type of encounter and intervention group in each quarter. We also aggregated over encounter types to measure the frequency of total healthcare clinic utilization for each patient, but excluding lab, radiology, and pharmacy utilization.
Community Participation and Ethical Considerations:
Our study protocols were developed and interpreted collaboratively with two stakeholder groups. The COPE Advisory Group established in 2012 is comprised of local physicians, nurses, program leaders, information technology specialists, Navajo Nation Department of Health program directors, and CHR supervisors and the Community Health Advisory Panel (CHAP), established in 2013, which includes COPE participants, their relatives, and CHRs. The COPE Advisory Group and Community Health Advisory Panel meet quarterly and provided suggestions on how to group health services, ensuring that study findings could be disseminated in a comprehensible manner to patients and families, and providing interpretive feedback on results.
For this study, Community Health Advisory Panel participants requested additional data on utilization of traditional medicine services. We shared that visits to Traditional Medicine services were included as Specialty Outpatient care and accounted for 476 of 82,803 (0.57 percent) of Specialty Outpatient visits. COPE Advisory Group members suggested a sensitivity analysis excluding community encounters (e.g. joint home visits between public health nurses and CHRs), in case CHRs visits may have been included in this group.
This study was approved by Partners Healthcare Institutional Review Board (2012P001069) and the Navajo Nation Human Research Review Board (NNR-11.150T).
Statistical approach
We used a generalized linear mixed regression model for count outcomes, assuming a Poisson distribution for the outcome variable using a log-link to assess the differences in the frequency of healthcare utilization between COPE and non-COPE patients. The analysis was implemented with the SAS PROC GLIMMIX, and performed using SAS 9.3 (SAS Institute, Cary, North Carolina).
We used random effects to account for patient correlation over time, and within-site correlation at the service unit. We also adjusted for covariates with potential influences on utilization that exist at the time of the intervention: age, gender, language, primary care physician, and the following diagnoses: essential hypertension, major depression disorder, alcohol abuse, major cardiovascular disease (defined as at least one of the following diagnoses: acute myocardial infarction, coronary artery bypass surgery, coronary angioplasty, peripheral arterial disease, abdominal aortic aneurysm, carotid artery disease, cerebrovascular disease), and dyslipidemia.
In the model, we measured the outcome of interest as the number of encounters by encounter type that each participant had in each period, as well as the covariates listed above to adjust for differences between patients. The regression model measured the change (or trend) in utilization over time distinguishing the pre-intervention trend from the post-intervention trend, and utilization by COPE patients versus non-COPE patients. The model was parameterized so that the regression coefficients for utilization directly measure the difference in COPE patient utilization relative to non-COPE patients both before and after enrollment in COPE. Thus, we were most concerned with the value and statistical significance of the regression coefficient that measured the difference in utilization trend between the control group and the COPE group following the intervention. First, if it did not differ significantly from the post-intervention utilization trend of the control group, then statistically there was no difference between any post-intervention change that might have occurred to the control group and the COPE group. Even if a change occurred in the utilization trend, because the control group did not experience the COPE intervention, we could not attribute whatever change occurred in the COPE group to the COPE intervention. Second, if the COPE post-intervention coefficient was statistically significant and greater than zero, COPE patient utilization of health resources would have increased relative to control group patients; if it was statistically significant and less than zero, COPE patient utilization of health resources would have decreased relative to the control group. The model is presented in further detail in the technical appendix.