Study Design We conducted a retrospective cohort study comparing processes and outcomes of diabetes care before and after OA implementation. The study was approved by the Institutional Review Board of Indiana University School of Medicine.
Participating Clinics IUMG-PC is the largest primary care group in Indiana. Six clinics that implemented OA were classified as ‘OA’ clinics. Six clinics that continued to use a traditional scheduling approach were classified as ‘control clinics.’ All sites accepted patients with the Wishard Advantage health, a county tax-funded health-care program for uninsured patients in the greater Indianapolis area. Point of care testing for hemoglobin A1c (A1c) with immediate availability of results was available at all participating sites.
The patient sample consisted of all adult patients with type 2 diabetes from the OA and control clinics continuously enrolled in the Wishard Advantage Health Plan from July 1, 2004 to June 30, 2006. Patients were identified as having diabetes if they met one or more of the following three criteria: (1) diagnosis of diabetes established by International Classification of Diseases (ICD–9) codes 250.xx, 357.2, 362.0, 366.41;15,16
associated with two outpatient visits (or one inpatient visit); (2) prescription fill for one or more glucose-lowering medication without a diagnosis of polycystic ovarian disease (ICD-9 256.4x); (3) elevated glycosylated HbA1c (A1c) of 9% or higher or elevated fasting glucose level ≥200 mg/dl.16–18
We restricted our analysis to patients of the Wishard Advantage Health Plan.19
Since Wishard Health plan eligibility is based on income (<200% Federal Poverty Level), socio-economic status (SES) is less likely to differ across sites. Thus, access to care and the benefits available to these patients should be identical across clinics. This health plan also uses a primary care referral management policy that makes it less likely that patients will transition between clinical care sites with different scheduling policies. Moreover, for care to be reimbursed by the plan, it must be delivered within the IUMG-PC or Wishard Hospital system. This allows for near-complete capture of all utilization records, clinical laboratory, and prescribing data for participating patients in both the OA and control clinic sites. Similar to previous studies measuring quality of care for diabetes,20,21
we restricted our analytic sample to continuously enrolled health plan members because it provided critical “pre-intervention” data necessary for adequate case-mix adjustment.17,22
We excluded patients (2.4%) if they were missing all relevant laboratory, vital sign, or visit data over the study period.
We used two main sources of data. Patient demographics, comorbidity, laboratory and medication data, and inpatient and outpatient utilization were obtained from the Regenstrief Medical Record System (RMRS).23,24
This system interfaces with inpatient and outpatient scheduling and administrative databases for the entire Wishard Health Services system.
Continuous enrollment in the Wishard Advantage Health Plan and the organizational characteristics of the OA and control clinics, including clinic level characteristics such as the number of physician full-time equivalents (MD FTEs), the registered nurse (RN) FTEs, the number of trainee FTEs, and the annual number of clinic visits at each site, were determined from the IUMG data warehouse.25
Procedures The index date for the OA clinics was the date of OA implementation. For the control clinics the index date was June 1, 2005, the date when the last OA clinic began using OA. The ‘pre’ period was the 12 months before the index date. The ‘post’ period was the 12 months after the index date.
Outcome Measures Primary outcomes included process and intermediate outcomes for diabetes and (inpatient and outpatient) health-care utilization.
Process measures included documentation of annual A1c, low density lipoprotein cholesterol (LDL), and urine protein tests during the 12 months before and after the index date.26–29
We measured A1c, LDL, and systolic blood pressure (SBP) as intermediate outcomes of diabetes care.30–33
Each patient could have zero, one, or multiple values available for each laboratory variable. If multiple values were available on the same day, we used the mean of the measures. We used the last value of A1c, LDL and SBP prior to the ‘index’ date as the value for the ‘pre-intervention’ period and the last value between 3 months and 12 months of post ‘index’ date as the value for the ‘post- intervention’ period, under the assumption that it would take at least 3 months for system level changes to affect outcomes (especially A1c, LDL). For the measures of SBP, we used sitting BP.
Inpatient utilization (total number of hospitalizations) and outpatient utilization [ED, urgent care (UVC), and primary care visits] were determined in the 12-month period preceding and following the index date of OA implementation. No information was available for outpatient or inpatient utilization outside of the Wishard system. Such utilization, however, is rare.
Selected covariates included age, sex, and race. Since we were already including a relatively homogenous group of patients (with regards to their SES), we did not additionally control for income and education. To adjust for illness severity, we used the Charlson index of comorbidity.34,35
Each hospitalization record in the RMRS contains up to ten discharge diagnosis codes. We used the discharge diagnoses codes 2 through 10 from the hospital admission to construct the Charlson Index, according to the Deyo scheme.36
To control for DM-related severity, we assessed whether patients were on insulin or oral agents for their diabetes. Adjustment using medication intensity has become a standard in Translating Research into Action for Diabetes (TRIAD) and other studies of diabetes care quality.31,33
Clinic level variables included the number of annual visits per clinic site, ratio of support staff full time equivalent (FTEs) to MD FTE,37–39
and physician productivity, which is defined as the ratio of total Relative Value Units (RVUs) to MD FTE. Clinical systems may differ in process and outcomes of care when there are differences in the overall case mix of the clinic (i.e., provider exposure to other patients beyond the analytic sample). To adjust for these possible differences, we included percentage of managed care patients at each site. The system employs fewer than five primary care physician extenders (NP/PA) overall. Resident physicians provided care at both the open access and non-open access clinic sites. Hence, we did not control for provider type.
Baseline characteristics of the study patients were compared between the OA clinics and control clinics using a Pearson’s chi-square test for categorical data and a Student’s t-test for continuous data. Clinic characteristics were compared using either a Student’s t-test or a Wilcoxon rank-sum test. SAS version 9.1 (SAS Institute, Cary, NC) was used for all statistical analyses.
A generalized linear model (GLM) framework was used to assess the impact of OA on the processes of diabetes care, intermediate clinical outcomes, and health-care utilization.40
Within this GLM framework, generalized estimating equations (GEE)41
were used to adjust parameter estimates for within-clinic correlation (clustering) of the patients. All multivariable models were similar in the covariates that were included. For each outcome measure that was modeled, terms were included to compare the OA clinics to the control clinics and to adjust for the pre-intervention (baseline) value of each outcome. Because patients with worse outcomes attributable to disease severity would also typically have worse intermediate risk factor levels at baseline, it is likely that any residual bias is low after adjusting for differences in both medication intensity and baseline risk factor levels. Patient and clinic characteristics were also simultaneously included in the models as covariates and left in the models irrespective of statistical significance. Patient characteristics included were age, sex, race, Charlson co-morbidity index, and diabetes severity. Clinic characteristics included the percentage of managed care patients, the ratio of support staff FTEs to physician FTEs, annual number of visits at clinic site, and physician productivity (defined as the ratio of total RVU to number of MD FTE’s).
An additional term was included in the logistic models to account for the interaction of race with clinic type. This is a pre-specified hypothesis given the known disparities in both processes and outcomes of diabetes care for African Americans.42,43
Thus, race-specific adjusted odds ratios (OR) and 95% confidence intervals were estimated for the clinic types (OA or control). A multivariate logistic regression model was used to assess the effect of OA on the diabetes processes of care measures. A linear regression model was used to assess the effect of OA on the intermediate outcomes of diabetes care. The adjusted least square mean differences (95% CI) between OA and control clinics were estimated. A Poisson regression model was used to assess the effect of OA on health-care utilization (number of ED/UVC visits, number of hospitalizations, and total number of outpatient visits). Adjusted rate ratios were estimated from these models along with 95% confidence intervals (CI).