Preventable Diabetes

Evaluation of Preventable Diabetes Hospitalizations in the VHA

Co-Investigator: Brad Doebbeling, MD, MSc

Project Number IIR 02-079 funded by the VA Health Services Research and Development

November 1, 2003-October 31, 2006 

 Rationale and Objectives: Ambulatory care sensitive conditions (ACSCs) are those for which related hospitalizations may be avoidable by providing appropriate ambulatory care services. However, the relationship between appropriate, timely ambulatory care and risk for specific hospitalizations has never been demonstrated at the individual level. Diabetes is an ideal disease state in which to study these associations since nationally accepted performance measures for ambulatory care are widely adopted. We will evaluate hospitalizations for uncontrolled diabetes, diabetic ketoacidosis, and lower extremity amputations. These are three  of the fourteen adult  Prevention Quality Indicators (PQIs) developed and proposed for national implementation by  the Agency for Healthcare Research and Quality: 

Aim 1. To operationalize outcome measures: selected ACSC hospitalizations related to diabetes that may be affected by ambulatory processes of care.

Aim 2. To investigate associations between patient level factors and ambulatory diabetes care and between patient level factors and the ACSC hospitalizations in Aim 1.

Aim 3. To evaluate ambulatory processes of care as the primary exposure and determine the extent to which these processes are multivariately associated with the ACSC hospitalizations defined in Aim 1.

Aim 4. To investigate the contribution of institutional and community level factors on the relationship between ambulatory processes of care and ACSC hospitalizations

Research Design:

The project is  a retrospective analysis of all veteran clinical users with diabetes, dually eligible for Medicare in FY97/FY98, who were alive on 9/30/98. Ambulatory care and hospitalizations will be followed for FY99-FY01. The database is derived from multiple sources:  VA Healthcare Analysis and Information Group,  Veterans Support Center (NCPD);  CMS MEDPAR  A and MEDPAR B claims data. It is supplemented by the Large Veterans Health Survey.

Analyses:

Because outcomes are categorical, we will use logistic regression for the first two aims. We will determine entry order of variables a priori (hierarchical regression approach) and will select our best performing model based on magnitudes, signs, and significance levels of regression parameters; C-statistic; and goodness-of-fit of models.

For specific aim 3, to assess the association between explanatory variables (including individual-level, cluster-level factors) and outcomes  while accommodating the correlation of the data, we plan to use marginal models as the primary modeling approach, but will compare our results to the random effect approach. The former gives an effect that is averaged across facilities and the latter gives a net effect when the facility effect is held constant; in addition, the fact that the latter effect estimate appears to serve as an upper bound for the effect of ambulatory processes of care (conditioned on variables included in the models) is clinically appealing.  In both modeling approaches, we will safeguard the consistency of variance estimates by using sandwich variance estimates.

Anticipated Impact: 

This study, with its rich patient level data and totality of utilization and events, will permit us to demonstrate whether or not a convincing  association exists between ambulatory care practices and selected diabetes hospitalizations. These results will be of importance to the VA because of its policy emphasis upon footcare, and will also influence the national debate over the use of PQIs.