New Approaches Focusing on Dynamic Variables Related to Changes in Member’s Health Status: Diabetic HbA1c Predictive Model Brenton B. Fargnoli Blue Cross.

New Approaches Focusing on Dynamic Variables Related to Changes in

Member’s Health Status:

Diabetic HbA1c Predictive Model

Brenton B. Fargnoli

Blue Cross & Blue Shield of Rhode Island

2

Outline

• Background

• Predictive Rules

• Validity

• Applications

Background

3

The Diabetic Epidemic

• Prevalent– 23.6 million people (7.8% of population)

• Expensive– Medical Expenditures: $116 Billion

National Diabetes Statistics, 2007

American Diabetes Association, 2007

• National Diabetes Statistics, 2007

4

Lab Data Gap

Clinical and Economic Effectiveness:• HbA1c<7%: (6, 4.5)• HbA1c>9%: (6, 4.5)• Annual HbA1c Screening: (1,1)

• Thus, it is the lab values, not the presence of screenings which are significant.

de Brantes et al., Am J Managed Care, 2008

5

Variables Associated with HbA1c Level

Association• Age• Drug Adherence• Drug Therapy • Co-Morbidities• Physician Visits• Ethnicity

Shectman et al., Diabetes Care, 2002

No Association• Gender• Income• A1c screenings

6

Predictive Rules

7

HbA1c’s Continuous Risk Gradient

• 1% HbA1c Reduction Associated with Decreases:– 43% Amputations– 36% Nephropathy, Neuropathy, Retinopathy– 30% Depression– 24% ESRD– 14.5% Cataracts– 14% MI– 12.5% Stroke

IMPACT Product

8

Applied HbA1c-Comorbidity RelationshipRetinopathy Example:

A1C %: 9.4 8.4 7.4 6.4 5.4

Retinopathy Prevalence: 0.5566 0.3563 0.228 0.1459 0.0934

(1-Prevalence) 0.4433 0.6438 0.772 0.8541 0.9066

P (0 Co-Morbidities) 0.1151 0.2892 0.4236 0.5307 0.6123

P(Only Retinopathy) 0.1446 0.1601 0.1251 0.0907 0.0631

P(Ret&Neur Only) 0.0601 0.0371 0.0175 0.0077 0.0033

P(Ret + 1) 0.1844 0.1435 0.0823 0.0465 0.0264

P(R, Neur, Dep Only) 0.0057 0.0027 0.0009 0.0004 0.0002

9

Performed for 156 combinations of 9 Co-Morbidities

Predicted A1c from # of Co-Morbidities

10

9.4 8.4 7.4 6.4 5.4 Predicted A1c

P(0) 0.1152 0.2894 0.4236 0.5307 0.61228 6.7732

P(1 Only) 0.2915 0.4195 0.4038 0.3630 0.31888 7.4010

P(2 Only) 0.2544 0.2270 0.1460 0.0943 0.06284 8.0573

P(3 Only) 0.2934 0.2530 0.1659 0.0872 0.04873 8.1723

Polynomial Extrapolation

11

Drug Intensity-Disease Intensity Relationship

12

• High Intensity (+0.75)– Type II Insulin use– ≥ 3 oral anti-diabetics

• Low Intensity (-0.75):– No pharmaceuticals needed

Adapted and Modified from Shectman et al., Diabetes Care, 2002

Drug Adherence

• Reflects:– Self-Management– Drug Effectiveness

• Calculated with Avg. Days Supply Method

• (% Adherence – 82%) x (-1.5)

Adapted and Modified from Shectman et al., Diabetes Care, 2002

13

Rules Summary

• Co-Morbidities:• 0: 6.77• 1: 7.40• 2: 8.06• 3: 8.17• 4: 10.11• 5: 11.81• 6: 13.80• 7: 16.10• 8: 18.70• 9: 21.59• No PCP nor Eye Appts for full

year: (+0.75)

• Pharmacy• Insulin: (+0.75)• ≥ 3 oral anti-diabetics: (+0.75)• None (-0.75)• (% Adherent – 82%) x (-1.5)

14

Predicted HbA1c=(Co-Morbidity Index + Pharmacy Index)/2

Note: All adjustments are from 7.40

Validity

15

Paired T-Test All Inclusive Excluding Physician Visit Outliers

  Actual Predicted

Mean 7.116470588 7.216149433

Variance 1.131392157 0.431441838

Observations 85 85Pearson Correlation 0.289856571Hypothesized Mean Difference 0

df 84

t Stat -0.854070714

P(T<=t) two-tail 0.395494943

t Critical two-tail 1.988610165  

16

  Predicted Actual

Mean 7.388 7.31215

Variance 2.275006 0.437331

Observations 100 100Pearson Correlation 0.338633Hypothesized Mean Difference 0

df 99

t Stat 0.531475

P(T<=t) two-tail 0.59628

t Critical two-tail 1.984217  

Predictions compared with 2005-2007 BCBSRI HEDIS Data

Predictive Power

Method 1 Method 2

Deviation from Mean -0.07585 +0.09968

Avg. Absolute Deviation 0.89341 0.75371

1.0 Deviation Confidence 77% 80%

17

Limitations

• Variance

• Patients skipping full year of appointments

• Variables limited to data fields within pharmacy and insurance claims

18

Applications

19

Disease Management

Patient-Level

• Identify Actionable Members

• Measure Intervention Effectiveness

20

Marketing

Population-Level

• Track and report group’s year over year changes in predicted mean HbA1c

21

References

• NIH. National Diabetes Statistics 2007. http://diabetes.niddk.nih.gov/dm/pubs/statistics/

• American Diabetes Association. Direct and Indirect Costs of Diabetes in the United States. http://www.diabetes.org/diabetes-statistics/cost-of-diabetes-in-us.jsp

• de Brantes F, Wickland P, Williams J:The Value of Ambulatory Care Measures: A Review of Clinical and Financial Impact from an Employer/Payer Perspective. Am J of Managed Care 14: 360-368, 2008

• IMPACT Product: Meta-analysis of case-controlled, longitudinal studies• Schectman J, Nadkarni M, Voss J: The Association Between Diabetes

Metabolic Control and Drug Adherence in an Indigent Population. Diabetes Care 25: 1017-1021,2002

22

Questions

23