Presented at The 129th Annual Meeting of the American Public Health Association Atlanta, GA, October

1
Forecasts of Diabetes in the U.S. Results from
a Markov Model

• Presented at The 129th Annual Meeting of the
  American Public Health Association Atlanta, GA,
  October 2125, 2001
• Presented by Kristine R. Broglio Thomas J.
  Hoerger James P. Boyle Amanda A.
  HoneycuttLinda S. Geiss Theodore J. Thompson

P.O. Box 12194 3040 Cornwallis Road Research
Triangle Park, NC 27709Phone [phone]
Fax [phone] honeycutt_at_rti.org
www.rti.org
2
Background

• In 1997, 10.3 million people in the U.S. were
  estimated to have diagnosed diabetes
• 35 higher than the estimated number in 1992
• Diabetes prevalence is highest among the oldest
  age groups and among certain racial/ethnic
  minorities
• Future changes in diabetes prevalence may be
  expected given projected changes in the size and
  age/racial composition of the U.S. population and
  life expectancies

3
Study Objectives

• Generate forecasts of diabetes prevalence and the
  number of people with diabetes through 2050 by
  age, sex, and race/ethnicity
• Use a dynamic modeling approach to predict the
  number of people with diabetes in each year based
  on flows into and out of the diabetes state

4
Overview of Previous Studies

• Forecasts of the number in the U.S. with diabetes
• Helms (1992) 12 million in 2050
• King, Aubert, and Herman (KAH 1998) 22 million
  in 2025 (diagnosed and undiagnosed)
• Boyle et al. (2001) 29 million in 2050
• Limitations of previous studies
• Assume constant prevalence (or linear trend)
• Prevalence is not determined based on flows into
  (incidence) or out of (death rates) the diabetes
  state
• Most do not account for prevalence differences
  across race/ethnicity (e.g., Helms and KAH)
• Generate forecasts for age groups (e.g., 0 to 44
  years), rather than individual ages

5
Methods The Markov Modeling Approach

• Markov models are used to simulate the
  progression of individuals through mutually
  exclusive disease states
• 3 states of the diabetes forecasting model no
  diabetes, diabetes, and death
• Age, race/ethnicity, and sex-specific subgroups
  transition between states in each 1-year cycle of
  model
• age (0 to 100 years, plus 101)
• sex (male, female)
• race/ethnicity (non-Hispanic White, non-Hispanic
  Black, Hispanic, and Other)
• Total number of subgroups 816

6
Markov Model of Diabetes Disease States and
Transition Probabilities
7
Data for Model Inputs (I)

• Base Year (1999) Diabetes Prevalence
• Estimated using NHIS for 19801998
• Used to determine initial distribution of
  individuals across the Diabetes and No Diabetes
  states
• 1999 Diabetes Incidence
• Estimated using NHIS for 19841998
• Age-, race/ethnicity, and sex-specific incidence
  rates used to forecast the flows of individuals
  into the Diabetes state
• Base model assumes no temporal changes in
  incidence

8
Data for Model Inputs (II)

• Baseline U.S. Population
• Census Bureau middle series of population
  projections
• Used to determine initial distribution of
  individuals across Diabetes and No Diabetes
  states
• Forecasts of the Number of Live Births
• Census Bureau forecasts of number of children
  under one year of age through 2050
• Added to No Diabetes state in each model cycle
• Forecasts of Net Migration
• Census Bureau middle series of population
  projections through 2050
• Added to Diabetes and No Diabetes states based on
  age, sex, and race/ethnicity-specific 1999
  prevalence

9
Data for Model Inputs (III)

• Forecasts of the U.S. Population Mortality Rate
• Census Bureau mortality rate projections
  through 2050
• Used to estimate Diabetes and No Diabetes
  mortality rates (in conjunction with 1999
  prevalence and estimated relative risk of death
  from diabetes)
• Estimates of the Relative Risk of Mortality from
  Diabetes
• From Tierney et al. (2000)

10
Results Forecasts of the Number of People with
Diabetes, 19992050
11
Results Forecasts of Diabetes Prevalence,
19992050
12
Results Forecasts of Diabetes Prevalence, by
Race/Ethnicity, 19992050
13
Results Impact of Changes in Incidence/Relative
Risk of Death from Diabetes, 19992050
14
Interpreting Results (I)

• Projected growth in the population would lead to
  forecasts of 15.8 M in 2050 v. baseline forecast
  of 35.5 M

15
Interpreting Results (II)

• Projected aging of the U.S. population
  population growth would lead to 19.9 M with
  diabetes in 2050 as compared to baseline
  forecasts of 35.5 M

16
Interpreting Results (III)

• Projected changes in the racial composition age
  composition population growth would lead to
  22.4 M with diabetes in 2050 v. baseline forecast
  of 35.5 M

17
Interpreting Results (IV)

• What accounts for the rest of the growth in the
  number with diabetes from 1999 to 2050?
• Dynamic modeling approach
• Prevalence is determined within the model not
  assumed to remain constant over time
• Projected mortality rates
• Increases in life expectancy mean an increase in
  the number of people at risk of developing
  diabetes in the oldest highest incidence age
  groups

18
Sensitivity Analyses

• Impact of using age-, race/ethnicity-, and
  sex-specific incidence rates that are 20 lower
  than baseline rates
• Projected number with diabetes in 2050 is 29.4
  million
• 2050 forecasts are 2.7 times greater than the
  number with diabetes in 1999
• Impact of using age- and sex-specific relative
  risks of death from diabetes that are 20 higher
  than baseline rates
• Projected number with diabetes in 2050 is 33.99
  million
• 2050 forecasts are more than 3 times higher than
  the number with diabetes in 1999

19
Comparison to Previous Studies

• Our forecasts suggest that 35.5 million people
  will have diagnosed diabetes in 2050, as compared
  to previous forecasts of
• 12 million (Helms, 1992)
• 29 million (Boyle et al., 2001)
• Why are our forecasts higher than previous
  estimates?
• Use 1999 prevalence and incidence data
• Forecasts are by age (0100 years and 101),
  race/ethnicity (NHW, NHB, Hispanic, Other)
• Diabetes prevalence is determined within a
  dynamic model not assumed to remain constant or
  follow a linear trend

20
Study Limitations

• Relative risk of death from diabetes
• Available only for adults
• Not available by race/ethnicity
• Uncertainty about the future
• Population growth
• Diabetes incidence (assumed to remain constant)
• Relative risk of death for people with diabetes
  (assumed to remain constant)

21
Summary and Conclusions

• Forecasts suggest that the number of people with
  diabetes will more than triple between 1999 and
  2050
• Rising from 10.9 million in 1999 to 35.5 million
  in 2050
• Diabetes prevalence is projected to grow from 4
  to 9 from 1999 to 2050
• Race/ethnicity-specific Highest projected
  prevalence among non-Hispanic blacks increases
  from 5.6 in 1999 to 11.9 in 2050
• Age-specific Largest increase for 75 years
  from 12.8 in 1999 to over 23 in 2050
• Findings suggest an increased demand for health
  care resources in the next 50 years to treat
  people with diabetes