TennCare Diabetes Program Evaluation

1
TennCare Diabetes Program Evaluation
An Individually-Matched Control Group Evaluation
of a Disease Management Program to Improve
Quality and Control Costs in a Diabetic Medicaid
Population
Presentation to AcademyHealth Kenton Johnston,
MPH, MS, MA June 4, 2007
2
Overview
TennCare Diabetes Program Evaluation

• Research Objective
• Evaluate diabetes disease management program for
  state Medicaid (TennCare) population
• Outcomes of interest diabetic quality of care
  and medical cost savings (Inpatient
  Prof/Outpatient)
• Outline
• Program Description
• Study Design
• Findings
• Limitations, Conclusions, Implications

3
Program Description
TennCare Diabetes Program Evaluation

• Outcome of diabetes treatment highly dependent on
  self-care
• Non-adherence to recommended regimens an obstacle
  to improved health status
• Medicaid population tends to exhibit higher
  utilization costs, as well as poorer health
  outcomes
• CareSmart Diabetes Disease Management (DM)
  Program developed internally by BCBST for
  TennCare population
• For Type 1 and Type 2 diabetics
• Program behavior change health education,
  self-management, personalized telephone coaching,
  compliance with ADA clinical practice guidelines,
  and PCP support
• Member consent obtained for enrollment in program

4
Study Design Individual Matching With
Propensity Model
TennCare Diabetes Program Evaluation

• Methodological toolbox for DM program
  evaluation
• Randomized controlled trials
• Population based pre-post methodology
• Predictive modeling
• Control group matching (individual, group)
• Problem finding a good control group not easy
• Solution Individually-matched controls using
  propensity scores (matched pairs cohort study)
• Propensity score is continuous number that
  represents individual probability of being in
  study group
• Propensity score reduces entire set of covariates
  to one score for easy individual matching
• This approach allows for smaller n

5
Study Design Population Methods
TennCare Diabetes Program Evaluation

• Study and control group member criteria
• Continuously enrolled in TennCare 24-months of
  2004-05
• Diagnosed with Type 1 or 2 diabetes in 2004 or
  earlier
• Not dually eligible Medicaid only
• 126 study members enrolled in CareSmart Diabetes
  Program for at least 6 months in 2005 were
  individually matched to 126 diabetic controls not
  enrolled in program in 2004 or 2005
• Propensity model covariates demographics,
  diseases comorbidities, quality of care,
  medical utilization, costs
• Baseline Period Jan - Dec 2004 for matching
  control study
• Intervention Period Jan - Dec 2005

6
Study Design Dependent Variables
TennCare Diabetes Program Evaluation

• Diabetic quality of care operationally defined
  according to recommended preventive services
  outlined by ADA
• Screening for kidney disease
• First annual HbA1c screening
• Second annual HbA1c screening
• Retinopathy screening
• LDL cholesterol screening
• Medical services utilization and cost
• Reported as totals (not specific only to
  diabetes)
• Inpatient admissions, inpatient days, inpatient
  
• ER encounters, office visit encounters,
  Prof/Outpatient
• Total
• RX utilization cost data unavailable

7
Findings Baseline Results
TennCare Diabetes Program Evaluation
8
Findings Intervention Quality Results
TennCare Diabetes Program Evaluation

• Statistically significant positive difference on
  4 of 5 measures on overall score
• Improvement in both study control groups from
  2004
• Propensity matched control group enables us to
  rule out secular trend as sole cause

9
Findings Intervention Utilization Results
TennCare Diabetes Program Evaluation

• Statistically significant difference on office
  visits study members had higher utilization
• Office visit finding not surprising given this is
  the setting for quality measures
• Inpatient admissions days lower for study
  members not statistically significant
• ER encounters higher for study members not
  statistically significant

10
Findings Intervention Cost Results
TennCare Diabetes Program Evaluation

• None of the cost findings were statistically
  significant
• Inpatient total costs trending in downward
  direction for study group
• Prof/Outpatient costs higher for study group
• Financial analysisusing control group to
  calculate expected costsshows program savings
  impact for study group

11
Limitations
TennCare Diabetes Program Evaluation

• Unable to analyze RX data
• Psychological or sociological variables not
  included/available for propensity model potential
  source of confounding
• Non-Participation Bias
• Study members agreed to participate in the
  program
• Controls either could not be contacted by
  telephone or refused to participate
• We did not control for practice patterns of
  member providers (data not available for all
  members)
• Lab values unavailable on gt 50 of study and
  control population so we were not able to control
  for these
• Available HbA1c and LDL values showed HbA1c close
  to stat sig (.09) difference in baseline period

12
Conclusions Implications
TennCare Diabetes Program Evaluation

• Conclusions
• Improvement in quality in study group was not due
  solely to general secular trend towards quality,
  but was also positively impacted by the diabetes
  program intervention itself
• Mixed findings for utilization cost, but may be
  showing trend in right direction
• Implications
• DM programs can be successful in improving
  quality of care in chronically diseased state
  Medicaid populations
• A matched-pairs cohort study using propensity
  scores is a valuable tool for evaluating program
  outcomes in small to medium sized populations

13
Thank You
TennCare Diabetes Program Evaluation
Presentation to AcademyHealth Kenton Johnston,
MPH, MS, MA June 4, 2007 E-mail
Kenton_Johnston_at_BCBST.com