Interpersonal continuity of primary care of veterans with diabetes: a cohort study using electronic health record data

The Veteran’s Health Administration (VHA) is the largest integrated delivery system in the U.S. In 2012, the VHA provided primary care to over 6.33 million patients and had 990 outpatient clinics in 23 regionally defined integrated service networks (VISNs) [18, 19]. Patients who utilize the VHA for care tend to be sicker, older, and have lower incomes than the general population [20]. VHA’s patient-centered medical home model, Patient-Aligned Care Team (PACT), aims to provide strong ICoC within a team-based setting [18, 21]. A panel of approximately 1200 primary care patients are assigned to a PACT, which consists of one PCP (physician, nurse practitioner (NP) or physician assistant (PA)), a registered nurse care manager, a clinical associate (licensed practical nurse, medical assistant or health technologist), and a clerk.

This patient-level cohort study used centrally-available national VHA electronic health record data from fiscal years 2012 and 2013 (Fig. 1). The goal of the approach to the study cohort selection was to identify veterans with diabetes that have received a sufficient amount of primary care within the VHA system to impact diabetes outcomes. The sample included adult, pharmaceutically treated veterans with diabetes seen within VHA primary care clinics nationally. Specifically, veterans must have had a diabetes diagnosis (International Classification of Diseases 9th revision (ICD-9) codes 250.xx) associated with at least one VHA inpatient visit and/or at least two a primary care clinic visits (VHA stop codes 322,323,342, and 348) in fiscal year (FY) 2012 (N = 1,049,638) and a filled prescription for insulin and/or an oral hyperglycemic agent (VHA drug classes HS501 or HS502) the same year (N = 830,602). The combination of ICD-9 and medication criteria was chosen to maximize the likelihood that patients in the dataset have diabetes [22]. Similar algorithms in the VHA have a specificity approaching 100%. Patients were excluded if they did not also have an outpatient visit with a diabetes diagnosis in FY 2013 or were younger than 18. Each patient was assigned a “home” VHA facility as the clinic most frequently visited for primary care in FY 2012. To be retained in the cohort, patients had to have a “home” VHA facility with at least 100 diabetic patients in FY 2012 (N = 719,370). This was done to ensure that facilities have experience treating diabetes and provide an extra measure of patient confidentiality. The provider most often visited in the home VHA’s primary care clinic in FY 2012 was considered to be the veteran’s PCP. The same procedure was used to determine home clinic and PCP in FY 2013. We excluded patients whose home VHA facility was > 1000 miles from their home zip code or was not in one of the 50 U.S. states or the District of Columbia, did not have an assigned provider in FY 2012, or had missing information regarding body mass index (N = 656,368).

Community data used for explanatory variables was acquired from non-VHA sources. PA scope of practice (SOP) information by state was obtained from a tabulation of recommended key elements assembled by the American Academy of Physician Assistants [23]. NP SOP information was obtained from the 2012 Pearson Report [24]. Both data sources evaluate the extent to which SOP laws in a given state allow PAs or NPs to practice to the level of autonomy considered ideal by each profession’s standards.

The outcome of interest was ICoC with a provider between 2012 and 2013. A binary variable was constructed to indicate whether patients switched providers between 2012 and 2013. Patients assigned to the same PCP in 2012 and 2013 were considered to have maintained ICoC. Patients assigned to different PCPs in 2012 and 2013 were considered to have interpersonal discontinuity.

Explanatory variables included patient, provider, organizational and community factors previously demonstrated to be related to ICoC and determined a priori. Patient level demographic factors included age, sex, race, ethnicity, marital status, distance of home address from assigned VHA primary care clinic, and change in home ZIP code between the two years. Patient-level variables suggesting social complexity included homelessness, whether or not the patient was exempt from VHA copayments on the basis of disability or low income, and presence of mental health diagnoses (separate variables for post-traumatic stress disorder (PTSD), mood disorders, substance abuse, and other mental health conditions), and diagnosis of dementia. Patient medical complexity was measured by the Diagnostic Cost Group (DCG) comorbidity measure, originally designed to predict cost of care but validated to measure medical complexity within the VHA population [25, 26]. The algorithm uses demographic and diagnostic information to assign each patient a DCG score, normed so that the average Medicare patient (patients in the U.S. that are 65 years or older, are disabled, or have end stage renal disease) has a score equal to 1 [27]. All patient-level variables were constructed using VHA electronic health record data from FY 2012. Specific categories for each variable can be seen in Table 1.

Two provider-level variables were measured and assigned to each patient based on their assigned provider in 2012. Primary care provider type was represented by a categorical variable (staff physician, resident, NP, PA). A variable indicating provider turnover was also created. Since VHA providers can commonly take up to 3 months off work for health or professional issues, provider turn-over was considered to have occurred if a provider did not provide any visits within the clinic in a consecutive 4 month period. Primary care provider assignment to a clinic in FY 2012 was compared to visits performed at the same assigned station in FY 2013. If there was a 4 month or longer period in which a provider did not perform a single primary care visit at the station in FY 2013 then the provider was flagged as having left the clinic. If a PCP performed no visits in the final 4 months of year, they were also considered to have left the clinic (turn-over = 1).

Organizational level variables were assessed for each facility. The value for each facility variable was calculated based on the patient clinic assignment in 2012. Region of the country was categorized as West, Midwest, Northeast, South and rurality was indicated by rural urban commuting area (RUCA) of the facility location (metropolitan area core, micropolitan, other metropolitan area, small town or rural). If a facility had 500 or more clinic stops (visits) for 305 or 306 (VHA codes for an endocrinology visit) in FY 2012 then the facility was defined as having endocrinology referral capacity, designated with an indicator variable.

State level variables were identified based on the state in which the facility is located. State variables include the percent of primary care physicians who work with NPs or PAs within the state (lowest, middle and highest tertile), NP and PA SOP regulations (least restrictive, moderately restrictive, and most restrictive). The value for each state variable was based on the state of the patient’s home clinic assignment in 2012.

All analyses were conducted using SAS® 9.4 [28] and SAS Enterprise Guide 7.1 [29]. Descriptive statistics were calculated for all variables at the patient-level. The association between experiencing ICoC with a PCP and patient, provider, facility, and state contextual variables was evaluated using logistic regression fit with generalized estimating equations and an exchangeable correlation structure to account for clustering within facilities. Covariates were specified a priori and assessed for multicollinearity prior to being entered into the model. All analyses set statistical significance at p < 0.05.

This work was reviewed and approved by the Internal Review Board (IRB) of the Durham Veterans Affairs Health Care System. Reflecting that this is a secondary data study, the IRB approved a waiver of informed consent for this study.

Approximately 18% (N = 1,049,638) of VHA patients had pharmaceutically treated diabetes mellitus (Fig. 1). Among the patients with diabetes that could be assigned to a PCP (N = 656,368) in 2012, 76% were assigned to an attending physician, 16% were assigned to an NP, 6.2% were assigned to a PA, and 1.7% were assigned to a resident physician. Approximately 22.3% of VHA patients with diabetes switched providers between 2012 and 2013. Switching providers occurred in 20.9% of patients assigned to physicians, 24.9% of patients assigned to PAs, 24.3% of patients assigned to NPs, and 58.2% of patients assigned to resident physicians.

Characteristics of the study population were similar to the general VHA population (Table 1). Patients with diabetes were predominantly older (mean age = 64.9 [standard deviation (SD) = 10.1]; less than 2% were under 40 years old), and were predominantly non-Hispanic (95%), white (71%), and male (97%). The study population was medically complex with approximately 20% of patients having at least 50% higher utilization (being at least 50% more complex) than the average Medicare patient (i.e., the oldest and sickest patients in the U. S.) based on the DCG score and high rates of mental health disorders (mood disorders 24%, post-traumatic stress disorder 15%). Social complexity is also prevalent, with 55% of the sample having no copay due to disability, 27% having no copay due to income, and 2% experiencing homelessness during fiscal year 2012.