A Tool Kit for Prediabetes Surveillance

1
A Tool Kit forPrediabetes Surveillance

• NHANES Analysis

The Surveillance Focus Area Diabetes Primary
Prevention Initiatives February 7, 2007 Seattle,
Washington
2
Outline

• Project objectives
• Prediabetes surveillance at the state level
• Prediabetes surveillance tool kit
• Process
• Summary of results
• Application examples
• Limitations, next steps, and continuing
  discussions

3
Project Objectives

• Develop a tool kit to help
• Estimate prevalence of prediabetes
• At the state level
• Using existing data such as BRFSS
• Monitor behaviors and risk factors
• Improve state diabetes surveillance system
• Identify populations for intervention

4
Prediabetes SurveillanceIn California (age 20-74)

• APrediabetes (3.6), 2005 CA BRFSS
• BAt risk for diabertes (8.8), 2005 CA BRFSS
• CIFG (23), NHANES III
• DIFG or Undx diabetes (26), NHANES III

5
Prediabetes SurveillanceIn California (age 40-74)

• APrediabetes (4.5), 2005 CA BRFSS
• BAt risk for diabetes (8.6), 2005 CA BRFSS
• CIFG or IGT (43), NHANES III
• DIFG, IGT, or Undx diabetes (47), NHANES III

6
Prediabetes Surveillance Tool KitAlgorithms

• Mathematical formulas for calculating the
  probability of having prediabetes
• Use self-reported measures that are available for
  most of the state BRFSS
• Validated with NHANES data
• Classify people for prediabetes with a chosen
  cut-off point

7
Prediabetes Surveillance Tool Kit Recommendations

• Data sources
• List of applicable variables in the existing
  state data
• Additional BRFSS questions may be added
• Applications
• Estimate prevalence of prediabetes
• Monitor behaviors and risk factors in a
  population
• Assist with program and intervention evaluations
• Implications
• Enhance state diabetes surveillance system
  capacity
• Improve diabetes primary prevention strategies
• Limitations

8
Process

• Set objectives
• Identify data sources
• Identify statistical methods
• Run model prediction and validation
• Create the algorithms
• Draw conclusions
• Incorporate expert reviews
• Make recommendations
• tool kit

9
Summary of Results

• Prepared NHANES III data
• Narrowed down to 13 self-reported measures in
  NHANES based on literature review, 11 of those
  are also in BRFSS
• Determined 4 prediabetes-related responses
• Prediabetes IFG (age 20-74)
• IFG or undiagnosed diabetes with FPG (age 20-74)
• Prediabetes IFG or IGT (age 40-74)
• IFG, IGT, or undiagnosed diabetes with FPG or
  OGTT (age 40-74)
• Investigated 16 models
• NHANES vs BRFSS
• Completed cases vs Imputed cases
• FPG vs FPG or OGGT
• Prediabetes vs Prediabetes or undiagnosed
  diabetes
• Completed multiple imputations
• Completed 16 model selection with Akaike
  Information Criterion
• Completed predictability validation with NHANES
  data
• All cases as well as leave-one-out
  cross-validation
• Statistics such as sensitivity, specificity, ROC
  curves, and area under curve
• Completed parameter estimation

10
Summary of Results The Selected Models
Models are based on imputed cases using NHANES
self-reported variables list 1IFG or
Undiagnosed diabetes with FPG 2IFG 3IFG, IGT
or Undiagnosed Diabetes with FPG or OGTT 4IFG or
IGT with FPG or OGTT
11
Summary of Results Prediabetes Prevalence
12
Summary of Results Health Conditions in
Prediabetes Population
13
ApplicationPrevalence of Prediabetes in
California
14
Application Health Conditions in California
Prediabetes Populations
15
Limitations

• Models validated with the national data
• Lab data limited by age groups
• Gaps in BRFSS self-reported measures
• High false positive rate
• Debates on areas of applications
• Population vs individual
• Surveillance vs screening
• Implications on programs and policies

16
Next Steps

• Summarize all the results
• Complete manuscripts for publication
• Complete expert reviews
• Finalize the tool kit
• Disseminate the tool kit

17
Continuing Discussions

• Is this tool useful for the state prediabetes
  surveillance?
• How to translate the results into applications?
• Can this tool be used for prediabetes screening?
• Can this tool help target populations for
  diabetes prevention?
• Can this tool be used in program evaluation?
• Should undiagnosed diabetes or abnormal blood
  glucose be included in the algorithms?