Estimating the Prevalence of Diabetes in Wisconsin Through an Innovative Data Exchange
Between a Department of Family Medicine and Public Health
Brian Arndt, MD Assistant Professor
Department of Family MedicineUW School of Medicine & Public Health
WREN ConferenceSeptember 15, 2011
Background• Diabetes is a prevalent chronic disease
affecting over 475,000 adults in Wisconsin• Wisconsin Behavioral Risk Factor Surveillance
System (WI BRFSS) data provide annual statewide diabetes prevalence estimates – Data not useful for estimating prevalence at
smaller geographic areas– Unable to track quality performance
indicators (e.g. A1c levels)
Alternative Surveillance Data• Electronic Health Record (EHR) data from UW
Department of Family Medicine (DFM) Clinics to identify a population with diabetes at a census block level– Geographic analyses and maps may lead to the
identification and surveillance of Wisconsin patients with diabetes at the neighborhood level
• Contains parameters for quality evaluation (A1c, BP, Cholesterol, Kidney health, etc.)
Project Goals
• Can EHR data improve diabetes prevalence estimates over telephone survey data?
• How do diabetes prevalence estimates based on DFM clinic data and BRFSS compare?
• Evaluate Risk, Control, & Co-morbidities• Link EHR data to community indicators
(Median Income, Economic Hardship Index)
BRFSS Diabetes Definition
• Have you ever been told by a doctor that you have diabetes? – Gestational diabetes and pre-diabetes excluded
• Does not distinguish between Type 1 and Type 2
UW MED-PHINEXType 2 Diabetes Definition
• Problem list AND Encounter diagnosis• Problem list OR Encounter Dx, AND
– Fasting glucose ≥ 126 mg/dL– 2 hour GTT glucose ≥ 200 mg/dL– Random glucose ≥ 200 mg/dL– A1c ≥ 6.5%or– Anti-diabetic medication Rx ≥ 1
> 2
UW DFM EHRType 2 Diabetes Prevalence
2007-2009 Criteria Count Prevalence
Problem 8,975 4.7%
Encounter 9,673 5.0%Problem or
Encounter 10,605 5.5%Problem/
Encounter and Labs/Meds 9,034 4.7%
2007-2009 Adult Type 2 DiabetesWI BRFSS Data UW DFM Clinic Data
*NPrevalence
(95% CI) NPrevalence
(95% CI)Overall 2,007 7.2(6.8-7.7) 9,023 6.0(5.9-6.1)
SexFemale 1109 7.0 (6.4-7.7) 4,329 5.2 (5.1-5.4)
Male 898 7.5 (6.8-8.2) 4,694 6.9 (6.7-7.1)
Age Group18-34 34 1.2(0.5-1.8) 366 0.7 (0.6-0.8)
35-64 959 7.0 (6.4-7.7) 5,589 6.8 (6.6-7.0)
65+ 991 18.3 (16.7-19.8) 3,068 17.4 (16.8-18.0)
2007-2009 Adult Type 2 DiabetesWI BRFSS Data UW DFM Clinic Data
*NPrevalence
(95% CI) NPrevalence
(95% CI)Race/Ethnicity
White (Non-Hispanic) 1,617 6.9 (6.4-7.4) 7,676 5.9 (5.8-6.0)
Black (Non-Hispanic) 210 11.7 (8.5- 14.9) 514 11.1 (10.2-12.0)
Other (Non-Hispanic) 124 10.5 (6.6-14.3) 281 6.2 (5.5-6.9)
Hispanic 31 5.5 (2.8-8.1) 352 7.0 (6.3-7.8)
2007-2009 Adult Type 2 DiabetesWI BRFSS Data UW DFM Clinic Data
*NPrevalence
(95% CI) NPrevalence
(95% CI)BMI
Normal or Underweight (<25.0) 249 2.7 (2.2-3.2) 513 1.6 (1.5-1.8)
Overweight (25.0 - <30.0) 613 6.3 (5.5-7.1) 1,458 4.4 (4.2-4.7)
Obese (30.0 - <40.0) 775 12.6 (11.4-13.9) 3,178 11.2 (10.9-11.6)
Morbidly Obese (≥40.0) 233 26.7 (21.5-31.9) 1,440 22.3 (21.3-23.3)
2007-2009 Adult Type 2 DiabetesWI BRFSS Data UW DFM Clinic Data
*NPrevalence
(95% CI) NPrevalence
(95% CI)Smoking
Never 865 5.9 (5.2-6.5) 3,619 5.1 (5.0-5.3)
Former 845 11.2 (10.1-12.3) 3,377 10.2 (9.8-10.5)
Current 294 5.8 (4.7-6.8) 1,326 5.2 (5.0-5.5)
Passive NA - 105 6.7 (5.4-7.9)
Multivariate Logistic Regression of Type 2 Diabetes Risk in Adults
• Good agreement with BRFSS• Each factor is a significant predictor in
direction expected:– Age, Gender, Race / Ethnicity, Smoking, BMI,
Median Income• Insurance Status & Economic Hardship also
predict risk• DFM data volume 4x greater (or more)
compared to BRFSS – provides greater precision and resolution
Economic Hardship Index• Census data from the Census Block Group
level• Index from 1 to 100 (No → Very Hard)• Variables include:
– Crowded housing– Federal poverty level– Unemployment– Less than high school – Dependency (% under 18 or over 64)– Median income per capita
Wisconsin Economic Hardship Index
Madison Economic Hardship Index
MilwaukeeEconomic Hardship Index
Diabetes Co-MorbiditiesOdds Ratio = P(Disease | Diabetes)
P(Disease | No Diabetes)
Co-Morbidity Prevalence OR 95% CI
Depression 25.1% 1.7 1.7-1.8
Asthma 11.0% 1.5 1.4-1.6
COPD 8.4% 4.2 3.8-4.5
CKD 26.1% 9.6 9.1-10.2
Among 9,023 Adult Patients with Type 2 Diabetes
Diabetes Co-MorbiditiesOdds Ratio = P(Disease | Diabetes)
P(Disease | No Diabetes)
Co-Morbidity Prevalence OR 95% CI
IVD- Cardiac 16.2% 7.9 7.4-8.4IVD – Cerebral 4.4% 5.7 5.0-6.4
CHF 9.1% 9.2 8.4-10.1
Among 9,023 Adult Patients with Type 2 Diabetes
Diabetes Co-MorbiditiesOdds Ratio = P(Disease | Diabetes)
P(Disease | No Diabetes)
Co-Morbidity Prevalence OR 95% CI
MI 2.1% 6.4 5.4-7.7PTCA 1.8% 6.9 5.8-8.4
Dementia 3.2% 3.7 3.3-4.3
Among 9,023 Adult Patients with Type 2 Diabetes
Diabetes Co-morbiditiesConclusions
• Each risk is significant • Higher complexity likely leads to higher
utilization & cost• Next Steps – data mining
– What predicts co-morbidity?– Which co-morbidities group together?– What predicts clusters ?
Predictors of HbA1c Control in Patients with Type 2 Diabetes
Kristin GallagherUniversity of Wisconsin
Department of Population Health Sciences M.S. ThesisJune 2011
Methods
• Adult Type 2 Diabetes Definition• Current A1c Value / Binary at 7%• Logistic Regression• Predictors of Poor A1c Control (>7%)
– Age, Gender, Race / Ethnicity, Economic Hardship Index, BMI, Depression
Regression Results Poor A1c Control
Characteristic OR 95% CI P-value
Age Group 0.0033
18-24 0.92 [0.52 - 1.60]
25-34 1.26 [0.98 - 1.62]
35-44 1.26 [1.08 - 1.46]
45-54 1.23 [1.09 -1.39]
55-64 1.00
Race/Ethnicity <.0001
White (Non-Hispanic) 1.00
Black (Non-Hispanic) 1.48 [1.20 - 1.83]
Other (Non-Hispanic) 1.45 [1.09 - 1.93]
Hispanic/Latino 2.08 [1.60 - 2.71]
Regression Results Poor A1c Control
Characteristic OR 95% CI P-value
Sex 0.0031
Male 1.00
Female 0.85 [0.76 - 0.95]Economic Hardship Index 0.0011
EHI <20 1.00
EHI 20 to <30 1.56 [1.18 - 2.05]
EHI >30 1.74 [1.28 - 2.37]
BMI <.0001
Normal or Underweight 1.00
Overweight 1.09 [0.83 - 1.44]
Obese 1.59 [1.23 - 2.06]
Morbidly Obese 1.76 [1.34 - 2.32]
Conclusions
• Socio-demographic factors:– Middle age groups, black, Hispanic, and other
race/ethnicities, obese, and morbidly obese BMI were all significantly associated with having higher odds of being in poor control
– Patients living in areas with increased economic hardship index (20-30; >30) have higher odds of being in poor control – this was significant
• Health factors:– Those without depression were found to have
significantly higher odds of being in poor control
Diabetes Next Steps• Evaluate comorbidity predictors • HEDIS performance definitions & analysis
(PCP & clinic level; P4P)• Measures of utilization in population x status• Data mining & modeling community factors• Expand variables exchanged
Diabetes Next Steps – GIS / Spatial Analysis
Diabetes Next Steps – GIS / Spatial Analysis
Collaborative Effort – Thank you!
• Brian Arndt-UW DFM• Amy Bittrich-DPH• Bill Buckingham-UW APL• Jenny Camponeschi-DPH• Michael Coen-UW Biostats• Tim Chang-UW Biostats• Dan Davenport-UW Health• Kristin Gallager-UW Pop
Health• Theresa Guilbert (PI)-UW
Peds
• Larry Hanrahan-DPH• Lynn Hrabik-DPH• Angela Nimsgern-DPH• David Page-UW Biostats• Mary Beth Plane-UW DFM• David Simmons-UW DFM• Aman Tandias-SLH• Jon Temte-UW DFM• Kevin Thao-UW DFM• Carrie Tomasallo-DPH