Do Diabetes Group Visits Lead to Lower Medical Care Charges? Kathryn Marley Magruder, PhD, MPH VA Medical Center Medical University of South Carolina Charleston, South Carolina Funded by: Robert Wood Johnson Foundation Agency for Healthcare and Quality
Dawn E. Clancy, MD, MSCR Clara E. Dismuke, PhD Kit N. Simpson, DrPH David Bradford, PhD
Background Efficient and effective care for patients with type 2 diabetes (DM) consistent with American Diabetes Association (ADA) Efficient and effective care for patients with type 2 diabetes (DM) consistent with American Diabetes Association (ADA) Group Visits (GVs) improve efficiency and throughput of patients by increasing access and decreasing backlogs of patients awaiting appointments. Group Visits (GVs) improve efficiency and throughput of patients by increasing access and decreasing backlogs of patients awaiting appointments. Previous managed care studies reveal GVs less costly and at least as effective as usual care for quality; review of the literature does not find they substantially reduce costs for individuals with DM. Previous managed care studies reveal GVs less costly and at least as effective as usual care for quality; review of the literature does not find they substantially reduce costs for individuals with DM.
Rationale Research on GVs remains in its infancy. Research on GVs remains in its infancy. Over a dozen articles exist describing GVs but only 6 studies reported are randomized controlled trials. Over a dozen articles exist describing GVs but only 6 studies reported are randomized controlled trials. None show notable cost results. None show notable cost results.
Study Questions Do GVs for patients with type 2 diabetes lower healthcare utilization and outpatient (OP) charges? Do GVs for patients with type 2 diabetes lower healthcare utilization and outpatient (OP) charges? Is the lack of statistically significant findings in previous studies due to endogeneity of the GV variable in cost models? Is the lack of statistically significant findings in previous studies due to endogeneity of the GV variable in cost models?
Design and Sampling Randomized controlled clinical trial Randomized controlled clinical trial 506 uncontrolled type 2 diabetic patients with a HgbA1c > 8.0% invited to participate 506 uncontrolled type 2 diabetic patients with a HgbA1c > 8.0% invited to participate 186 patients enrolled after signing IRB approved consent documents 186 patients enrolled after signing IRB approved consent documents Patients randomly assigned to participate in group visits or traditional one-on-one patient- physician visits Patients randomly assigned to participate in group visits or traditional one-on-one patient- physician visits
Data Collection and Key Variables Charges for OP, ED visits, and IP stays collected at study end Charges for OP, ED visits, and IP stays collected at study end Distance to provider (living within 10, 20, or 30 miles of the provider); Distance to provider (living within 10, 20, or 30 miles of the provider); Charlson scores, ICD-9 based measures of illness severity; Charlson scores, ICD-9 based measures of illness severity; Diagnosis codes; procedure codes. Diagnosis codes; procedure codes.
Statistical Analysis Mann-Whitney tests used for differences in mean charges by service (OP visits, ED visits, and IP stays) Mann-Whitney tests used for differences in mean charges by service (OP visits, ED visits, and IP stays) Separate charge models used for OP visits and ED visits (IP stays had too few nonzero observations to estimate models reliably) Separate charge models used for OP visits and ED visits (IP stays had too few nonzero observations to estimate models reliably) OP visit models include controls for payer, Charlson score, distance to provider, and a binary indicator of GV treatment OP visit models include controls for payer, Charlson score, distance to provider, and a binary indicator of GV treatment With charge models, distance to provider (within 10, 20, or 30 miles) used as binary indicators (>30 miles omitted as those more likely to use OP services from providers whose charges are not captured in our data) With charge models, distance to provider (within 10, 20, or 30 miles) used as binary indicators (>30 miles omitted as those more likely to use OP services from providers whose charges are not captured in our data)
Statistical Analysis (cont.) Estimating the OP charge model using ordinary least squares (OLS), we hypothesized when unobserved patient characteristics correlate with both the intervention (GVs) and the outcome (healthcare costs), endogeneity can bias OLS results Estimating the OP charge model using ordinary least squares (OLS), we hypothesized when unobserved patient characteristics correlate with both the intervention (GVs) and the outcome (healthcare costs), endogeneity can bias OLS results In many RCTs, treatment assignment, assessment, termination, and dosage are controlled for by the researcher (exogenous), a necessary condition for statistical tests of between-group differences in patient outcomes to be unbiased. In many RCTs, treatment assignment, assessment, termination, and dosage are controlled for by the researcher (exogenous), a necessary condition for statistical tests of between-group differences in patient outcomes to be unbiased.
Statistical Analysis (cont.) Study patients, randomized to treatment modality (GVs or usual care), largely determined the intensity (dosage) of their own treatment by choosing not to attend all GVs or not staying for the entirety of any given GV Study patients, randomized to treatment modality (GVs or usual care), largely determined the intensity (dosage) of their own treatment by choosing not to attend all GVs or not staying for the entirety of any given GV This choice, likely affected by the treatment arm to which the patient was randomized, created endogeneity of the GV variable in the charge model This choice, likely affected by the treatment arm to which the patient was randomized, created endogeneity of the GV variable in the charge model Endogeneity can be controlled for by estimating a treatment effect model based on Heckman control function Endogeneity can be controlled for by estimating a treatment effect model based on Heckman control function
Statistical Analysis (cont.) Re-estimating the OP charge model using a treatment effect model and simultaneously estimating likelihood of GV participation and OP charges controlled for potential endogeneity. Re-estimating the OP charge model using a treatment effect model and simultaneously estimating likelihood of GV participation and OP charges controlled for potential endogeneity. Initial probit model found factors likely to affect GV participation to be included in the GV participation likelihood equation in the treatment effect model. Initial probit model found factors likely to affect GV participation to be included in the GV participation likelihood equation in the treatment effect model. OLS for the OP charge model revealed the same for the charge equation in the treatment effect model using drive time as a continuous measure in the GV participation part of the model and including payer, Charlson score, and binary indicators for distance to provider in the OP charge models. OLS for the OP charge model revealed the same for the charge equation in the treatment effect model using drive time as a continuous measure in the GV participation part of the model and including payer, Charlson score, and binary indicators for distance to provider in the OP charge models.
Results Mann-Whitney test results show that GV patients had 34.7% higher OP expenditures, 49.1% lower ED expenditures, and 30.2% lower total expenditures compared with those in control group (P <.05 for all). Mann-Whitney test results show that GV patients had 34.7% higher OP expenditures, 49.1% lower ED expenditures, and 30.2% lower total expenditures compared with those in control group (P <.05 for all). Initial OLS with robust standard errors estimates of the unadjusted OP charge model revealed statistically significant effects of GV treatment on OP care initially estimating that GV treatment increased OP costs by $699.52/patient/year Initial OLS with robust standard errors estimates of the unadjusted OP charge model revealed statistically significant effects of GV treatment on OP care initially estimating that GV treatment increased OP costs by $699.52/patient/year
Results (cont.) Though there was a statistically significant and marginally positive effect on GVs in the OP cost model that did not correct for endogeneity, the treatment effect model showed a statistically significant marginally negative effect of GV treatment on OP charges of $ Though there was a statistically significant and marginally positive effect on GVs in the OP cost model that did not correct for endogeneity, the treatment effect model showed a statistically significant marginally negative effect of GV treatment on OP charges of $ To understand how GV treatment reduced OP charges, we estimated a treatment effect model of specialty care visits and found that GV treatment leads to a reduction of 4.15 specialty care visits/GV patient/year. To understand how GV treatment reduced OP charges, we estimated a treatment effect model of specialty care visits and found that GV treatment leads to a reduction of 4.15 specialty care visits/GV patient/year.
Summary In our study of GVs for inadequately insured patients with type 2 DM, we demonstrate that, after controlling for endogeneity of the GV variable, GV treatment statistically significantly lowers outpatient charges by decreasing specialty care visits In our study of GVs for inadequately insured patients with type 2 DM, we demonstrate that, after controlling for endogeneity of the GV variable, GV treatment statistically significantly lowers outpatient charges by decreasing specialty care visits