Diabetes mellitus is associated with reduced life expectancy and affects approximately 25% of patients receiving care in the Veterans Health Administration (VA). A number of primary care interventions have been shown to result in a substantial reduction in mortality. There are large differences in the provision of these life-prolonging treatments and also a two-fold variation in facility mortality rates. These differences exist despite great effort on the part of the VHA to improve and standardize care.
Members of our team and studies of others have shown that the way care is structured affects the way care is delivered. Known structural factors of importance include, but are not limited to, access to care, integration of care, and adequacy of resources. Few studies, however, have examined the relationship of the structure of care to long-term outcomes, and none have examined mortality. Thus, there is an important gap in the information needed to guide decisions about how best to organize and support primary care.
The VA is uniquely situated to examine this issue. It has more than 250 outpatient primary care clinics, comprehensive computerized databases with which to examine treatments and outcomes, and extensive survey data about the structure of care.
The overall objective was to identify structures of care that affect patient outcomes in order to establish an evidence base for management decisions about how primary care can best be organized and supported. The specific objectives were to determine the association of mortality with access to care; integration of care; adequacy of resources; and diabetes-related initiatives.
We used a combination of patient-level and clinic-level data to build hierarchical logistic regression models. Patient-level independent variables were those identified by our previous work: patient demographics; co-existing conditions; complications of diabetes; and patient-reported physical and mental functioning. For the clinic level variables, we used the 2007 VA Clinical Practice Organizational Survey (CPOS) questions in combination with selected questions from the Survey of Healthcare Experiences of Patient (SHEP) FY 2007 to create indexes with which to assess the degree to which the following Patient Aligned Care Team (PACT) elements were present
a.Integration: measured by how many different kinds of providers are included on the PACT team.
b.Coordination: The degree to which difficulty arises in working with various consultation services.
c.Access: Whether follow up appointments are a bottleneck.
d.Communication: Measured as quality of communication among team members and between team members and administration
e.Quality: Emphasis on addressing patient complaints, tracking of problems, systematic approach to solving problems, etc.
f.Resources: Adequacy of space, clinic personnel, support personnel, IT support.
We used for the analysis diabetic patients from the SHEP cohort FY 2007 from age 18 through 85 that were seeing in VA clinics surveyed in CPOS. Our analytic sample was 15,507 diabetic patients that received care across 224 VA clinics. The mean age was 65. The patients were most likely to be white (78.4%), male (96.9%), married (61.4%) and retired (50.5%) or unable to work (29.7%). The mean Charlson comorbidity index was 2.25. The mean baseline physical (PCS) and mental (MCS) health scores were 35.4 and 46.7, respectively. There were significant variations of PACT indexes across clinics.
We applied a previously developed model for mortality measured at 24 months. Adjusters included patient demographics, Charlson index, diabetes severity and PCS and MCS from the VR-12. The performance of the model was good as reflected by a c-statistics of 0.80. There were significant variations of mortality across the clinics. The 2-year mortality rate was 6.1%. There was a large variation across clinics regarding mortality rates, ranging from 1% to 16%.
We used hierarchical linear regressions to examine the relationship between the PACT indexes and risk adjusted 2-year mortality rates. Among the 10 PACT indexes, we found a statistically significant inverse relationship between quality of communication and risk-adjusted 2-year mortality (p = 0.02). However, this did not hold up when:
-- We repeated the analysis 14 times using sampling with replacement (2 out of 14 repetitions not significant)
-- We examined the sickest 1/3 of patients as judged by VR-12 scores.
-- We added variables to the model: rural vs. urban clinic, Medicare penetration.
We then went on to examine the relationship of HbA1c, prescription of lipid lowering agents, and prescription of ACEI/ARBs and the PACT indexes. We found paradoxical relationships--i.e. higher (better) index scores correlated with worse HbA1c and prescribing practices, though none were statistically significant.
We should highlight several important findings. First, there were wide variations of the PACT indexes across VA clinics. Second, there were large variations of mortality rates across VA clinics. Third, there was a very weak association between PACT indexes and mortality.
None at this time.
Health Systems, Diabetes and Related Disorders
Quality Indicators, Quality of Care