Impact of a Community Health Worker–Led Diabetes Education Program on Hospital and Emergency Department Utilization and Costs
Methods
Setting
BSWH is one of the largest nonprofit health care systems in the United States and includes 46 hospitals, > 800 patient care sites, > 6000 affiliated physicians, 35,000 employees, and an accountable care organization. This study was conducted in 5 community clinics located in the Dallas metroplex surrounding BSWH North hospitals. The community clinics serve low-income, uninsured, and chronically ill patients. The study was approved by the Baylor Research Institute institutional review board.
DEP Intervention
The DEP program consisted of 2 initial 60-minute educational sessions and quarterly clinical assessments scheduled for 30 to 60 minutes for a maximum of 6 patient-contact hours over 12 consecutive months. The DSME curriculum for DEP was adapted from CoDE, a pilot program implemented in a Dallas clinic serving a largely uninsured Mexican American population [23]. Patients who participated in CoDE for 12 months experienced a significant reduction in HbA1c [23,24]. During the 2 educational sessions, the CHWs educated DEP participants about diabetes and the importance of blood glucose control, medication adherence, diet, and exercise. In addition to the educational sessions, CHWs performed quarterly clinical assessments of HbA1c, blood pressure, weight, and foot condition (visual and monofilament assessment). They also assessed self-management behaviors and facilitated goal setting at each visit. The CHWs documented patient visits in the electronic health record and contacted the patient’s primary care provider immediately if the patient was symptomatic or had critical blood glucose or blood pressure measurements as defined by program protocol.
Participants
Study participants were recruited from the clinics or referred by Baylor care coordinators following hospital visits related to uncontrolled diabetes from September 2009 to July 2013. Participants had to be 18 years or older with a diagnosis of type 2 diabetes and be uninsured or underinsured. Although the program targeted Hispanic patients, all patients who met the inclusion criteria were eligible to participate. To control for internal threats to validity such as history and maturation biases, we created a control group consisting of clinic patients who had a diagnosis of diabetes and met the DEP inclusion criteria but who did not enroll in DEP.
Assessment
We used a retrospective pre-post study design to assess the impact of the DEP on hospital utilization trends and the program’s return on investment (ROI). The primary outcomes were number of hospital encounters, length of stay (LOS) per encounter, and direct cost per patient. The pre-enrollment period was defined as the year priod to the day of the initial DEP visit (the date of enrollment for DEP patients) or the initial clinic visit (the date of “enrollment” for control patients). All study participants were included in the analysis regardless of their number of inpatient or ED encounters. If a participant did not have a documented encounter during the analysis time period, encounters, LOS, and medical costs were set to zero. Patients with at least 2 HbA1c measurements taken during the first year post enrollment were included in the analysis.
Data Sources
Inpatient and emergency department (ED) utilization data were obtained from the Dallas Fort-Worth Hospital Council (DFWHC) database. The DFWHC captures administrative data from over 80 participating hospital systems and 9 million unique patients in North Texas. The DFWHC applied a matching algorithm using first and last name and date of birth to match study participants with encounter and length of stay detail for all hospitalizations across all DFWHC member hospitals. We obtained direct medical costs for patients treated at BSWH facilities from the BSWH Trendstar administrative database. Direct medical cost was not available for encounters at non-BSWH facilities so cost was estimated for these patients based on BSWH costs using a prediction model that accounted for LOS, primary ICD-9 diagnosis, patient age, sex, and race. The model explained approximately 66% of the variation in direct costs (R2 = 0.6581).
Statistical Analysis
All analyses were performed in SAS V9.3 using an α level of 0.05. A 2-tailed independent samples t test was used to test the mean differences in utilization outcomes one year prior to and post program enrollment. An analysis of covariance (ANCOVA) was used to assess whether the mean change in utilization outcomes was greater for DEP patients than the control group, , after controlling for age, sex, and ethnicity. The gamma distribution with a log link function was used to model direct cost and the negative binomial distribution was used to model hospital encounters and LOS.
The ROI calculation included 2 components: DEP investment cost and risk-adjusted direct medical cost savings 1 year post program enrollment. DEP investment cost included the average yearly costs per community health worker, the fixed one-time start-up costs distributed across the length of the program (4 years), and the salaries of the 5 community health workers employed for the duration of the program. These costs were divided by the number of patients who enrolled in the DEP to calculate the per patient cost (investment). Direct medical cost savings were calculated as the mean reduction in hospitalization and ED costs per patient adjusted for age, sex, and ethnicity.
