The study setting was the Veterans Health Administration (VA). Primary care serves as the “medical home” for most VA patients to help coordinate care for medical and mental health specialties, including an integrated primary care-mental health program.22 We report on one study aspect from a larger mixed methods study examining patient care and its relation to clinical processes and outcomes.
The study included survey data from PCPs matched to their patients and administrative data.
Our sample consisted of complex patients who were likely to have a high coordination burden. All patients had a diagnosis of type 2 diabetes mellitus and at least one other comorbidity, either a medical or mental health comorbidity. We chose patients with diabetes because it is a chronic condition that can involve coordination of multiple healthcare professionals within primary care and across specialists. Diabetes is common among the national population; one in ten non-Veterans has diabetes and one in four VA patients has diabetes. VA spends $1.5 billion annual for treatment.23 We selected patients based on a conceptual model that characterizes patients by disease domain (physical, mental health) and severity (low, high).24 In this framework, disease domain and severity can lead to greater specialist involvement and greater challenges in coordination.
We used a multi-step process to identify eligible patients. All patients needed at least two outpatient or one inpatient diagnosis for diabetes plus one of four conditions documented from January 1, 2014, to December 31, 2015. We categorized patients into one of four groups with all patients having diabetes plus (a) medical conditions generally managed in primary care (hypertension); (b) medical conditions generally requiring specialty care (congestive heart failure); (c) mental health conditions generally managed in primary care (anxiety or depression); or (d) mental health conditions generally requiring specialty care (schizophrenia, bipolar disorder, schizoaffective disorder, or posttraumatic stress disorder). Conditions were selected so that our sample included a range of patient complexity and associated coordination burden.
Survey Fielding Method
We first selected 29 out of 140 medical centers to reflect relatively high and low performance on the basis of multiple care coordination items using previously collected national program data from the primary care and mental health integration and patient aligned care team surveys completed by a key clinical informant. After identifying sites, we obtained administrative data by provider for patients seen during the most recent 12-month period. Using ICD-9 codes, we classified patients into one of the four categories described above. We selected providers with approximately 100 or more patients in each of the four categories. We sampled approximately three providers from each hospital-based and three providers from each community-based outpatient clinic. We first surveyed PCPs. We then sent patient surveys only to the patients of providers who responded. For each provider, we randomly selected approximately 15 patients from each of four categories per provider.
The provider survey measured three organizational coordination constructs: relational, feedback, and within-team. The survey also included demographic and practice characteristic items. The relational coordination measure included seven items that assessed PCP perceptions of shared goals; shared knowledge; mutual respect; and the frequency, timeliness, accuracy, and problem-solving aspects of communication.25 Providers rated 12 specialty services and clinics including cardiology, mental health, nutrition, and endocrinology on these seven items. Each item was scored using a 5-point Likert-type scale. Cronbach’s α values ranged between .86 and .91 among specialties. Because PCPs do not coordinate with every specialty for a particular patient, we created a weighted measure of relational coordination based on the number of patient visits to specialty clinics assessed on the PCP survey over 12 months preceding when surveys were first sent to patients. For example, if a patient had 8 visits to cardiology and 2 visits to endocrinology, then the PCP ratings for cardiology contributed 80% to the weighted relational coordination score, while ratings from endocrinology contributed 20%. We computed this measure for each patient.
We measured feedback coordination using ratings on “how helpful have consults, including (including e-consults)” been with a set of specialty clinics and services in caring for patients.10 PCPs rated each of 12 specialty services, similar to the relational coordination measure. Response options included opt-out options if the service was not available or if the provider did not use the service. Evaluative ratings were on a five-point scale ranging from “not at all helpful” to “extremely helpful.” We created a composite measure with categories for helpful (for those respondents with an average rating of “helpful or extremely helpful”), less helpful (for those respondents with a less than “helpful or extremely helpful” average), and not used.
We measured within-team coordination (k = 5, α = .89) using the average PCP response to a series of questions asking them to rate the “degree of teamwork and cooperation” demonstrated by specific team members (e.g., registered nurse, PCP, licensed practical nurse). Team members were rated using a five-point response scale from “poor” to “excellent.”
We recruited patients between April 13, 2016, and September 2, 2016, using up to four mailed invitations.26 Patients could complete the survey by mail or online. We created two versions of the patient survey, one with a small number of questions about patient experiences of cardiac care and the other with parallel questions for mental healthcare. Patients with cardiac comorbidities were sent the cardiac version and patients with mental health comorbidities were sent the mental health version. The patient survey contained items reflecting coordinated care from the Patient Perceptions of Integrated Care (PPIC) survey.27,28,29
The PPIC focuses on several dimensions of coordination, such as test result communication, transition following hospitalization, and support for self-directed care. We examined three care dimensions that focused on coordination among PCPs and specialists as dependent variables (Appendix). Dimensions were supported with analysis of the factor structure of the instrument. Items generally consisted of Likert-type response options ranging from “never” (1) to “always” (4). Knowledge integration across settings and time (k = 4, α = .81) consisted of items assessing the extent to which the PCP was well-informed of treatment the patient was receiving from other care providers and teams. Knowledge fragmentation across settings and time (k = 5, α = .69) consisted of items asking whether patients needed to repeat information to providers about their care needs; lower scores were favorable. Specialist knowledge management (k = 4, α = .73) consisted of items asking whether the specialists knew the patient’s medical history and test results from other providers. We created alternative scoring measures to reflect condition treated, including coordination for diabetes care (k = 6, α = .73); coordination for heart specialist care (k = 6, α = .69); and coordination for mental healthcare (k = 6, α = .73).
We included covariates that reflected both patient and provider characteristics. For patients, we included variables for age, sex, race, ethnicity, and marital status, based on self-reported survey question responses. Using administrative data from the VA Corporate Data Warehouse, we included a count of VA medical center outpatient visits within 12 months (e.g., primary care, mental health, surgery), a ratio of selected clinics for specialty care visits to total primary care/specialty visits, a count of comorbid conditions based on Elixhauser codes,30 and a measure to reflect diabetes control, based on whether the patient’s average hemoglobin A1C test value for a 12-month period was less than 7%. For providers, we modeled self-reported responses to VA tenure. From administrative data, we coded occupation as physician, nurse practitioner, or physician assistant. We also accounted for panel size, practice location (urban or rural), and whether providers practiced in a hospital or CBOC.
Our goal was to examine relationships between organizational coordination and patient-reported coordination. We first reviewed data for range of values, completeness, and collinearity. We created a “missing” category for demographics. We regressed our three measures of patient perceptions of coordination on the set of patient and provider characteristics. We used a random effects multi-level model with robust standard errors that accounted for patients being clustered within providers. Analysis was completed in SAS 9.2 (Cary, NC).