Professional Documents
Culture Documents
O R I G I N A L A R T I C L E
T
he high cost of caring for individuals alter these trends in costs. Policymakers are diagnosed and added to the popula-
with chronic diseases is one of the already are keenly interested in develop- tion; contemporaneously, other individu-
most pressing issues in health care ing and pursuing policies that can prevent als with existing diabetes die and leave
in the U.S. today (1). The baby boom the expected rise in disease burden and this subpopulation. With the balance of
generation is aging, and advanced age is head off expensive public commitments these two processes, the prevalence of di-
accompanied by costly chronic ill- to care for the chronically ill. abetes in the total population changes on
nesses. As a result, Medicare and other The forecasting effort presented in an annual basis. The pace of change dif-
health-related governmental programs this article speaks directly to this concern fers over time depending on factors such
will face demographic and epidemio- by improving the rigor of the estimates of as the rate of obesity and age of those at
logical forces that will challenge their health outcomes and health care spend- risk. For instance, the aging of the large
financial viability. ing associated with future trends in the baby boom generation will bring large
In light of the sheer magnitude of incidence, prevalence, and progression numbers of new people into age catego-
costs associated with diabetes, policy- toward complications. We constructed a ries that are at higher risk of developing
makers and the public need to under- model of diabetes costs that accounts for the disease.
stand how these costs will change over the the trends in risk factors for diabetes, the Second, costs associated with diabetes
next decades and how new policies may natural history of disease, and the effects tend to follow a natural progression over
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
time. Complications take time to develop
From the 1Department of Medicine, University of Chicago, Chicago, Illinois; the 2National Opinion Research and inflict damage to the eyes, kidneys, and
Center at the University of Chicago, Chicago, Illinois; and 3Civic Enterprises, Washington, DC.
Corresponding author: Elbert S. Huang, ehuang@medicine.bsd.uchicago.edu. circulatory and nervous systems. Therefore,
Received 9 March 2009 and accepted 12 August 2009. robust projection models must include es-
DOI: 10.2337/dc09-0459 timates of the expected natural history of
© 2009 by the American Diabetes Association. Readers may use this article as long as the work is properly the disease based on alternative levels of dis-
cited, the use is educational and not for profit, and the work is not altered. See http://creativecommons.
org/licenses/by-nc-nd/3.0/ for details.
ease management.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby In developing our forecasting model,
marked “advertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. we account for two types of cohorts—a
care.diabetesjournals.org DIABETES CARE, VOLUME 32, NUMBER 12, DECEMBER 2009 2225
Projecting future diabetes costs
2226 DIABETES CARE, VOLUME 32, NUMBER 12, DECEMBER 2009 care.diabetesjournals.org
Huang and Associates
vary by sex, race, and ethnicity and UKPDS control arm to a functional form studies in the literature (please see the on-
smoothed over ages 24 – 85 years. Esti- used in the original National Institutes of line appendix Table for details).
mates of r are separately smoothed for Health model (11). For the transition be- For this analysis, we used the compli-
age-groups ⬍45, 45– 64, and ⬎64 years tween proteinuria to end-stage renal dis- cation model to predict the average an-
due to substantial heterogeneity across ease, we used probabilities from an nual costs of living with diabetes by
these age ranges. observational study (12). different ages, sexes, racial groups, and
Age-specific annual hazard of pro- For background mortality rates, we major durations of diabetes. A total of
gression to diabetes for people without used race/ethnicity- and sex-specific 10,000 Monte-Carlo iterations (each iter-
diabetes for different sexes and BMI cate- background mortality rates reported in ation representing a patient life) were
gories are calculated based on observed U.S. life table statistics from 1999 (13). To used to generate average estimates. All
incidence of people with diagnosed dia- calculate background mortality rates for costs are expressed in 2007 USD. In esti-
betes and current screening rates. The individuals with diabetes, we subtracted mating costs for future years, we applied
progression hazards increase monotoni- cardiovascular mortality rates for the gen- the cost growth assumptions used by the
cally with age in all categories and are eral population from the overall mortality Congressional Budget Office.
highest for the obese category followed by rates found in life tables. We multiplied
overweight and normal at all ages. these rates by 2.75 as previously done to RESULTS — The results of our model
reflect higher background mortality rates regarding overall population changes in
Lifetime simulation model of for patients with diabetes (11). When pa- obesity, future population size, and
diabetes complications tients developed specific complications, health care spending have been briefly de-
Within a 1-year cycle, patients move from one such as coronary heart disease, stroke, scribed in a related publication (23). We
disease state to another or stay in the current end-stage renal disease, and amputation, expand on those results and describe fore-
disease state until death or age 95 years. we assumed that patients had higher mor- casts for the Medicare population.
Online appendix Fig. S2 displays the tality rates attributable to these complica-
design of the model of diabetes complica- tions (14,15).
tions. This figure presents the structure of Within the model, we accounted for Changes in obesity
the decision analytic model. Hypothetical the effect of individual medications. The Because our model predicts the progression
patients move through the model from benefits of ACE inhibitors were based on from non-diabetes to diabetes, we estimate
left to right for each cycle length (1 year). the findings from the Heart Outcomes changes in percentage of obese, overweight,
Based on initial patient clinical character- Prevention Evaluation (HOPE) Study and normal-weight individuals in the pop-
istics, patients are subject to the risk of (16). Aspirin was assumed to reduce the ulation living without diabetes. Overall obe-
various complications related to diabetes probability of coronary heart disease but sity distribution in the non-diabetes
as well as mortality. Patients who survive to increase the probability of gastrointes- population remains fairly stable over time,
a given year repeat the cycle until death. tinal bleed (17). We assumed that the with ⬃65% of the population being over-
Data on demographic characteristics joint effect of aspirin and an ACE inhibi- weight or obese. The percentage catego-
(sex and race/ethnicity) as well as relevant tor on cardiovascular effects was multipli- rized as overweight in the non-diabetes
clinical characteristics (blood pressure cative. We did not assume that simply the population is expected to remain steady at
levels, cholesterol levels, GHb levels, and processes of care such as foot examination 35% over the time period. The percent cat-
duration of diabetes) are obtained from or routine laboratory tests independently egorized as obese is expected to drop
NHANES and used as data inputs for the produced clinical benefits (18). slightly from 30% in 2009 to 27% in 2033.
simulation models. For each clinical risk This same leveling of the obesity trend is
factor, we use age-, sex-, and race/ Health service utilization and cost found in projections produced by the Cen-
ethnicity-specific distributions of these inputs ters for Disease Control and Prevention for
factors within the models. We assumed that the use of medications the U.S. population (24).
The diabetes complication models in reflects the current distribution of use of
this analysis are derived from U.K. Pro- insulin, oral agents, insulin plus oral Future population size for the U.S.
spective Diabetes Study (UKPDS) results agents, and diet therapy as observed in We found that in 2009, there will be 19.5
(6). Prediction models for all major national studies of diabetes care (19). Dis- million diagnosed and 4.25 million undi-
diabetes-related complications have been tribution of use of different oral glucose- agnosed diabetes cases in the population
developed by the UKPDS study group lowering agents was assumed to be the ages 24 – 85 years. Over the next 12 years,
(7,8). These models have been internally observed distribution in national studies the overall population with diabetes is ex-
and externally validated with cardiovas- (20). Use of ACE inhibitors and aspirin pected to rise (Fig. 2). Among this popu-
cular trial data (9). The UKPDS model therapy was based on recent national re- lation, the distribution of diagnosed and
does not include glucose control as a pre- ports of diabetes care (21). Frequency of undiagnosed individuals will be shaped
dictor, making it unsuitable for evaluating office visits and laboratory tests was as- by the rate of arrival of new cases and
the impact of improved diabetes care on sumed to be that observed in a recent na- continued screening for diabetes by the
end-stage renal disease. Instead, we mod- tional study (22). medical system. The combined effect is
eled the development of microalbumin- We estimated drug costs based on the that the cohort of established diagnosed
uria and proteinuria, which are linked to average type and frequency of drug pre- diabetes will grow, while the cohort of
the intensity of glucose control (10). We scriptions, dosage of medications, and undiagnosed diabetes steadily declines
used prediction models for these interme- wholesale drug prices. Annual costs of and stabilizes at around 3.7 million by
diate complications using optimization microvascular and cardiovascular com- 2020. After 2020, the size of the cohort of
procedures to fit observations from the plications were obtained from recent people with undiagnosed diabetes is esti-
care.diabetesjournals.org DIABETES CARE, VOLUME 32, NUMBER 12, DECEMBER 2009 2227
Projecting future diabetes costs
2228 DIABETES CARE, VOLUME 32, NUMBER 12, DECEMBER 2009 care.diabetesjournals.org
Huang and Associates
example of the type of forecasting model findings, and preparation of the manuscript. pendent diabetes mellitus: a population-
that can be used by policymakers when M.O. and J.C.C. contributed to the overall study based study in Rochester, Minnesota. Ann
considering policies for other chronic dis- conception and design, formulation of analysis Intern Med 1989;111:788 –796
eases. Such models are appropriate when plan, interpretation of findings, and critical revi- 13. Andersen R, DeTurk P. National Vital Sta-
sion of the manuscript. E.S.H. had full access to tistics Report. Hyattesville, MD, National
abundant epidemiological data are avail- all the data in the study and takes responsibility Center for Health Statistics, 2002
able to forecast the natural history of dis- for the integrity of the data and the accuracy of 14. Petty GW, Brown RD Jr, Whisnant JP, et
ease incidence and progression, as is the the data analysis. al. Survival and recurrence after first cere-
case with type 2 diabetes. No other potential conflicts of interest rele- bral infarction: a population-based study
The study has several limitations. First, vant to this article were reported. in Rochester, Minnesota, 1975–1989.
attempts to forecast future costs and utiliza- Neurology 1998;50:208 –216
tions are conditional on current rates of uti- 15. U.S. Renal Data System. USRDS 1994 An-
lization. For example, we have used the References nual Data Report: Appendix D17. Bethesda,
most current estimates of screening rates for 1. DeVol R, Bedroussian A. An Unhealthy MD, National Institute of Health, Na-
diabetes from NHANES. However, rates America: The Economic Burden of Chronic tional Institute of Diabetes and Digestive
Disease. Santa Monica, California, Milken and Kidney Diseases, 1994
change over time, and future changes may Institute, 2007 16. Heart Outcomes Prevention Evaluation
influence our results. Our model also does 2. Helms RB. Implications of population Study Investigators. Effects of ramipril on
not account for individuals under 24 years growth on prevalence of diabetes: a look cardiovascular and microvascular outcomes
of age who enter the population. This limi- at the future. Diabetes Care 1992;15 in people with diabetes mellitus: results of
tation may be particularly relevant for accu- (Suppl. 1):6 –9 the HOPE study and MICRO-HOPE sub-
rately incorporating diabetes prevalence 3. King H, Aubert RE, Herman WH. Global study. Lancet 1999;355:253–259
and incidence in the immigrant population, burden of diabetes, 1995–2025: preva- 17. Hansson L, Zanchetti A, Carruthers SG, et
who may experience heterogeneous rates of lence, numerical estimates, and projec- al. Effects of intensive blood-pressure
developing the disease (25). Lastly, during tions. Diabetes Care 1998;21:1414 –1431 lowering and low-dose aspirin in patients
4. Boyle JP, Honeycutt AA, Narayan KM, et with hypertension: principal results of the
our analysis of transitions across BMI cate-
al. Projection of diabetes burden through Hypertension Optimal Treatment (HOT)
gories, we grouped all individuals who had 2050: impact of changing demography randomised trial: HOT Study Group.
BMIs ⱖ30 kg/m2. We did this because of a and disease prevalence in the U.S. Diabe- Lancet 1998;351:1755–1762
lack of available Medical Expenditure Panel tes Care 2001;24:1936 –1940 18. Singh N, Armstrong DG, Lipsky BA.
Survey data to model transitions across finer 5. Incidence of diabetes in the population aged Preventing foot ulcers in patients with
BMI categories. Many of these limitations 18–79 years [Internet], Atlanta, Georgia, diabetes. J Am Med Assoc 2005;293:
may lead to more conservative estimates of Centers for Disease Control and Prevention. 217–228
the future size of the diabetes populations Available from http://www.cdc.gov/diabetes/ 19. Saydah SH, Fradkin JE, Cowie CC. Poor
and their costs. statistics/incidence_national.htm. Accessed control of risk factors for vascular disease
Despite these limitations, our study 20 April 2008 among adults with previously diagnosed
6. Clarke PM, Gray AM, Briggs A, et al. A diabetes. JAMA 2004;291:335–342
strongly suggests that diabetes will grow in
model to estimate the lifetime health out- 20. Cohen FJ, Neslusan CA, Conklin JE, et al.
the coming decades, both in population size comes of patients with type 2 diabetes: the Recent antihyperglycemic prescribing
and costs, and will have significant impacts United Kingdom Prospective Diabetes trends for U.S. privately insured patients
on the lives of Americans and the financial Study (UKPDS) Outcomes Model with type 2 diabetes. Diabetes Care 2003;
viability of programs like Medicare. Fore- (UKPDS no. 68). Diabetologia 2004;47: 26:1847–1851
casting models like this can help policy- 1747-1759 21. Saaddine JB, Cadwell B, Gregg EW, et al.
makers anticipate future burdens of chronic 7. Stevens RJ, Kothari V, Adler AI, et al. The Improvements in diabetes processes of
diseases and design targeted policies that UKPDS risk engine: a model for the risk of care and intermediate outcomes: United
fight these diseases in the most effective coronary heart disease in type II diabetes States, 1988 –2002. Ann Intern Med
ways possible, both in terms of clinical ef- (UKPDS 56). Clin Sci (Lond) 2001;101: 2006;144:465– 474
671– 679 22. Chin MH, Drum ML, Guillen M, et al.
fectiveness and cost-effectiveness.
8. Kothari V, Stevens RJ, Adler AI, et al. UKPDS Improving and sustaining diabetes care in
60: risk of stroke in type 2 diabetes estimated community health centers with the health
by the UK Prospective Diabetes Study risk en- disparities collaboratives. Med Care 2007;
Acknowledgments — E.S.H. and A.B. were gine. Stroke 2002;33:1776–1781 45:1135–1143
supported by the Diabetes Research and 9. The Mount Hood 4 Modeling Group. 23. Huang ES, Basu A, Capretta JC, et al. Fed-
Training Center at the University of Chicago Computer modeling of diabetes and its eral cost estimates: a look at how a clinically-
(P60 DK20595). The investigators received complications. Diabetes Care 2007;30: driven projection model for diabetes could
support for this project as consultants to the 1638 –1646 inform budget estimates and decisions.
National Changing Diabetes Program of Novo 10. Huang ES, Zhang Q, Brown SE, et al. The Health Affairs 2009;28:w978 –w990
Nordisk. The funder did not participate in the cost-effectiveness of improving diabetes 24. Homer J, Milstein B, Dietz W, et al. Obe-
design and conduct of the study or collection, care in U.S. federally qualified commu- sity population dynamics: exploring
management, analysis, and interpretation of nity health centers. Health Serv Res 2007; historical growth and plausible futures
the data. The funder also did not participate in 42:2174 –2193 in the U.S. In 24th International Confer-
preparation of this manuscript. Novo Nordisk 11. Eastman RC, Javitt JC, Herman WH, et al. ence of the System Dynamics Society. Nijme-
participated in the final approval of the manu- Model of complications of NIDDM. gen, the Netherlands, 2006
script before submission. Model construction and assumptions. Di- 25. Davidson MB. The disproportionate bur-
E.S.H. and A.B. participated in the conception abetes Care 1997;20:725–734 den of diabetes in African-American and
and design of the overall study, formulation of 12. Humphrey LL, Ballard DJ, Frohnert PP, et Hispanic populations. Ethn Dis 2001;11:
analysis plan, analysis of data, interpretation of al. Chronic renal failure in non-insulin-de- 148 –151
care.diabetesjournals.org DIABETES CARE, VOLUME 32, NUMBER 12, DECEMBER 2009 2229