Trends in the Epidemiology of IDDM During 1970-2020 in Fyn County, Denmark ANDERS GREEN, DRMEDSCI ANNE K. SJOLIE, DRMEDSCI
Ol.E ESH0J, MD
OBJECTIVE — To present an epidemiological model applicable to diabetes based on which prevalence rates are estimated from assumed rates of incidence and mortality of diabetes. Furthermore, this study illustrates the model by analyzing epidemiological data on IDDM in a Danish population during 1970-1990, with predictions of prevalence rates for 1990-2020. RESEARCH DESIGN A N D METHODS— The epidemiological model assumes known prevalence rates as well as incidence rates and mortality at a given time. Under assumed rates of incidence and mortality of IDDM and its complications, the prevalence rate is the dependent variable, estimated as a function of calendar time. We use epidemiological data on IDDM (operationally defined as insulin-treated diabetes with onset before age 30 years), blindness, and nephropathy, as well as mortality, as reported for 1973 and 1987 in Fyn County, Denmark. RESULTS— During 1970-1990, the prevalence of IDDM increased steadily because of increasing incidence and decreasing risk of complications and mortality. The relative prevalence of patients with nephropathy increased, whereas that of blind patients decreased considerably. Under specified assumptions regarding the future levels of incidence of disease, complications, and mortality, it is estimated that the prevalence rate of IDDM in the year 2020 will be 45-60% higher than the level in the year 1990. The relative prevalence of patients with nephropathy will increase further, whereas the relative prevalence of blind patients will remain constant at a low level. CONCLUSIONS — IDDM will represent an increasing public health problem in Denmark during the next decades, with increasing overall prevalence rates and a rising proportion of patients with nephropathy. The major determinant of this trend is increasing incidence, combined with declining mortality and declining risk of complications. It is strongly recommended that epidemiological models like the one presented here be further developed and implemented at regional levels to provide data for the dimensioning of the current and future diabetes care systems.
I
DDM develops predominantly in childhood and young adulthood. In many populations, IDDM represents a major health problem for patients and society because of the disabling late complications and risk of premature death. Currently, the epidemiology of IDDM is changing. In many European populations, a steady increase in the incidence has been observed during the last decades
(1). In parallel, the prognosis in IDDM has improved considerably, although the disease is still associated with a relative mortality two to four times that of the general population (2). The net result is that the prevalence of IDDM is increasing at a rate that, in Denmark, has been estimated at ~ l - 2 % per year in the early 1970s (3). Future changes are likely to occur
I;rom the Genetic Epidemiology Research Unit (A.G., O.E.), the Department of Ophthalmology (A.K.S.), and the Department of Internal Medicine and Endocrinology (O.E.), Odense University Hospital, Odense, Denmark. Address correspondence and reprint requests to Anders Green, DrMedSci, Genetic Epidemiology Rest-arch Unit, Winslowparken 15 St., DK-5000 Odense C, Denmark. E-mail:
[email protected]. Received for publication 16 October 1995 and accepted in revised form 7 March 1996.
DIABETES CARE, VOLUME 19,
NUMBER 8,
AUGUST
1996
because of beneficial effects of intensified insulin treatment (4) and possible future prevention of the disease itself. Modeling techniques based on epidemiological methods may assist in describing current and future epidemiological characteristics of IDDM, thereby providing information needed for appropriate planning and priority-setting in diabetes care. We present here an epidemiological modeling system that is applicable to a chronic disease like IDDM and that may be used for predicting trends in the epidemiology of IDDM. The model is illustrated with data on IDDM and associated nephropathy and blindness in a Danish population. Our approach supplements recent models of the benefits of systematical screening for and treatment of early renal disease in a simulated cohort o\i IDDM patients (5) and a recent forecasting model based on assumed incidence and life expectancies lor diabetes in general (6).
RESEARCH DESIGN AND METHODS Epidemiological principles We consider a population that at a given time can be mutually and exhaustively divided into 1) disease-free individuals, 2^ patients without a specific complication, and 3) patients with a specific complication. The relative sizes of these subsets of the population provide the basis for estimation of the relevant prevalence rates at the time specified (Fig. 1). The turnover in the study population is determined by the transition rates between the various compartments (Fig. 1), estimated as the annual number of transitions between two compartments divided by the number of person-years at risk of leaving the candidate compartment. For a given calendar year, the number of person-years at risk is estimated by the size of that compartment multiplied by the time unit for the observations; here, the time unit is 1 year. The model assumes that all prevalence rates are known at one specified time and that the incidence and mortality
801
Trends in the epidemiology oflDDM
Table 1—Basic observations in 1973 and 1987
Population base Age 0-29 years Total population Total patient population Number of prevalent patients Annual number of new cases Annual number of deaths Patients with nephropathy Number of prevalent patients Annual number of new cases Annual number of deaths Patients with blindness Number of prevalent patients Annual number of new cases Annual number of deaths
ljuly 1973
ljuly 1987
199,886 439,689
184,082 456,483
727 25 15
950 28 13
50 9
82 9 4
26 5 4
19 2 2
Data arc n.
rates are known for the time units around this point. For subsequent successive time units, the number of individuals changing status between compartments is estimated from assumed incidence and mortality rates. The prevalence rate represents the dependent parameter that is updated after each time unit, as further
Of the complications to IDDM, we have selected persistent proteinuria as an indicator of severe diabetic nephropathy and blindness as an indicator of severe diabetic retinopathy. Persistent proteinuria is denned as a urinary protein concentration of >0.5 g/l in each of at least two consecutive examinations, with a
minimum interval of 3 months between examinations. Blindness is defined as a visual acuity ^6/60 in the best eye. Epidemiological characteristics during 1970-1990. The prevalence of IDDM (as denned above) in the Funen County has been assessed as of 1 July 1973 (8) and 1 July 1987 (9). These studies also provided estimates of the incidence and mortality of IDDM in the same study population in the years before and immediately after the respective prevalence dates (3,8,10; O.E., unpublished observations). From a clinical follow-up study of the 1973 population of IDDM patients (11,12) and from a cross-sectional investigation of the 1987 population (O.E., unpublished observations), it has been possible to estimate the annual transitions and complicationspecific mortalities in the years surrounding 1973 and 1987, respectively. The characterization of the study population in terms of size and annual turnovers with reference to 1973 and 1987 is shown in Table 1. Table 1 illustrates that in 1973, as well as in 1987, the population of IDDM patients was increasing considerably (as indicated by the larger input as compared with the total output). The incidence of
described in APPENDIX.
Application Study population. We have selected Fyn County, Denmark, as the study population for the present analysis. The time span of the trend analysis has been defined as the period 1970-2020, since the region has been under survey for IDDM during the last 20 years. Data on the population size and composition are available on an annual basis from official population statistics. We have used 1 calendar year as the time unit for the incidence rates and for updating the prevalence rates. Definitions of disease and complications studied. Because of the lack of wellestablished definitions of IDDM for use in large-scaled epidemiological surveys, we have, for the present purpose, denned IDDM as insulin-treated diabetes with onset before age 30 years. Although we hereby neglect individuals with true IDDM developing at higher ages, a previous study in the Fyn County has demonstrated that this definition leads to a patient group comprised of almost entirely truly insulin-dependent diabetic subjects (7). 802
DEATHS AMONG PATIENTS WITHOUT COMPLICATION
+ IDDM - COMPLICATION
+ IDDM + COMPLICATION NEW CASES OF COMPLICATIONS
DEATHS AMONG PATIENTS WITH COMPLICATION
NEW CASES OF IDDM
Figure 1—Graphic overview of the basic epidemiological model.
DIABETES CARE, VOLUME 19,
NUMBER 8,
AUGUST
1996
Green and Associates
30n
Incidence (per 105)
25Model 1 20Model 2
1510-
sumed future incidence rates are presented for each model in Fig. 2 (dotted lines). Concerning the future incidence rates of complications and the future mortality rates, we have assumed that the trends modeled for the last part of the 1970-1990 period continue during the rest of the period of prediction. The modeled rates are illustrated in Fig. 3A (incidence rates of nephropathy and blindness, respectively) and 3B (mortality rates according to complication status as well as overall mortality). RESULTS
Model 3
5-
Trends in prevalence rates Figure 4A shows the estimated trends in the overall prevalence of IDDM during the periods of 1970-1990 (solid line) and 1970 1980 2010 2020 1990 2000 1990-2020 (dotted lines according to the incidence models described above). The Calendar time prevalence rate increased from 1.6 per Figure 2—Modeled incidence rates (per 100,000 people aged 0-29years). For specification ojmodels, 1,000 people in 1970 to 2.1 in 1990. Unsec text. der model 1 (continuously increasing inIDDM has increased with an average of Modeling the trends for the years 1990- cidence rates), the prevalence rate will in1.2% annually (Fig. 2, solid line), and the 2020. Predicting prevalence rates beyond crease exponentially to 3.4 in 2020. The overall mortality rate has decreased as a 1990 requires assumptions regarding the increase will become slightly less under consequence of a dramatic reduction in future incidence rates of disease and com- model 2 (constant incidence rates from the mortality among patients with com- plications and of mortality rates. Because year 1995). Under model 3 (annual replications. The mortality among patients the incidence of disease is a major deter- duction in incidence rate by 5% from year without complications has remained al- minant of future prevalence, we have de- 2000), the prevalence rate will increase to most constant and that among those with veloped three different scenarios assum- a maximum at 2.6 in year 2013, after both blindness and nephropathy has in- ing different incidence trends. The first which it will decrease very slowly. creased somewhat (data not shown). Fur- (model 1) assumes that the incidence of thermore, whereas the number of patients IDDM continues to rise with an annual Trends in relative prevalences with nephropathy has been increasing, increase of 1.2%, corresponding with the according to complication status the number of patients with blindness has annual increment estimated for the pe- Figure 4B shows the estimated trends in decreased. The change in the profile of the riod of 1970-1990. The second scenario the relative prevalences (percentage of papatient population from 1973 to 1987 (model 2) assumes that during the period tients) according to complication status. can only be accounted for by a drastic of 1990-1995, the increase in annual in- The relative prevalence of nephropathy reduction in the incidence of complica- cidence gradually levels off to 0, corre- increased from 6.7% in 1970 to 9.5% in tions, combined with changes in the mor- sponding with a constant incidence rate 1990. This increasing trend will continue tality levels, as described above. Using the from 1995 onward. The third scenario for all models, most markedly for model 1 profiles and annual turnover rates in 1973 (model 3) assumes that from the year (continuously increasing incidence). In and 1983 as checkpoints, we have fitted 1995, it will be possible to partially pre- contrast, the relative prevalence of blindthe annual changes in the turnover rates vent the development of IDDM so that the ness decreased from 3.2% in 1970 to by visual smoothing. The most simplistic incidence during the period of 1995- ~ 2 % in 1990, and this level will remain functions are shown in Fig. 3. To fit the 2000 is gradually reduced from an annual fairly constant, regardless of model. observations, it has been necessary to as- increase of 0% in 1995 to an annual desume that the reduction in the incidence crease of 5% in 2000, which is assumed to CONCLUSIONS — Fpidemiological of complications was most pronounced in be maintained throughout the remaining modeling is sensitive to underlying obserthe early 1970s. period of prediction. The selection of an vations and assumptions. As for the It should be noted that the inci- annual reduction in the incidence of 5% is 1970-1990 period, we have valid crossdence rates of IDDM have been expressed based on the assumptions that 10% of all sectional data for the years 1973 and with reference to the 0- to 29-year age- new future cases of IDDM develop in sub- 1987, together with reasonable estimates group, whereas the prevalence rates refer jects with a positive family history of of incidence and mortality during the surto the total population, since the preva- IDDM and that 50% of these cases will be rounding years. Furthermore, many of prevented by some measure(s). The as- the features for this period are confirmed lent patients represent all age-groups. DIABETKS CARE, VOLUME 19,
NUMBER 8,
AUGUST
1996
803
Trends in the epidemiology oflDDM
2.0n
Incidence rate (per 100 patient-years)
1.5\ \
1.0-
Nephropathy \
0.5-
Blindness 0.0 1970
1980
1990
2000
2010
2020
Calendar time
B 20-i
Mortality rates (per 100 patient-years)
15-
10Blind patients
5-
Patients with nephropathy All patients
1970
1980
1990
2000
2010
2020
Calendar time Figure 3—Modeled rates of complications (A) and mortality (B) in patients with IDDM. from other studies. The increasing incidence of IDDM in Denmark has been well characterized, with an annual increase of — 1% (13). Previous studies have demonstrated a decrease in the overall mortality of IDDM in Denmark (2,14) as well as a decreasing incidence of persistent proteinuria (15). Our estimated complication-specific mortality rates in the 1970s confirm the results of other Danish studies (16) showing that patients with persistent proteinuria have a mortality rate that is several times higher than the rate for patients without this complication. Our analysis also demonstrates that the mortality in patients with complications has decreased considerably. The dramatic decline in the risk of blindness correlates with the establishment of services for ophthalmological screening and aggressive laser treatment during the 1970s in
804
the study region. Therefore, we believe that our epidemiological characterization of the population of IDDM patients during 1970-1990 is reliable. This part of our analysis clearly shows that blindness among IDDM patients has declined considerably by the year 1990, whereas, in contrast, the proportion of patients with nephropathy has been increasing. As for the estimated future prevalence rates, we predict that IDDM will remain a significant and increasing health problem in the decades to come. We consider model 3 (partial prevention of IDDM from year 2000) to be less realistic as compared with models 1 and 2. Accordingly, the total prevalence rate will increase by 45-60% as compared with the level in 1990. The proportion of patients with persistent proteinuria will continue its increase. The major determi-
nants of this trend are the future incidence rates combined with the expected slowly decreasing mortality rates. If further reduction in mortality should take place, the future prevalence will become even higher. Measures for earlier detection and more aggressive treatment in patients at risk of severe complications have already been implemented in Danish health care for patients with IDDM, as reflected in the decreasing incidence rates of diabetic complications. We assume in our projections that these trends will continue. The cost, however, is the maintenance of a health care system that covers all IDDM patients in terms of frequent routine examinations to commence preventive treatment when needed. It is evident that the demand for resources for education, self-control, and frequent screenings for complications will increase considerably during the next decades in Denmark. We acknowledge that the assumptions made for our predictions may be questioned. It is, however, easy to obtain alternative models from alternative sets of assumptions. This may be of particular relevance for populations with incidence and mortality rates different from those of the Danish. For the same reason, modeling like the kind presented here should be done in regional wellcharacterized populations only. Minimal requirements for the model comprise initial estimates of incidence, prevalence, and mortality. If these indexes are estimated continuously, i.e., on an annual basis, they may act as checkpoints and references for more qualified trend projections. Further refinements, including subclassification of the patients according to age and sex, are needed but may require mathematical modeling for the purpose of smoothing, as has been done before on retrograde estimation of incidence data from cross-sectional data (17). However, the greatest obstacle in this respect is that with an increasing degree of details in the model, more detailed epidemiological data will be needed. We conclude that in Denmark and comparable populations, IDDM represents an increasing public health problem due to substantial increases in the prevalence rate and in the proportion of patients with nephropathy. Because of the chronicity of the disease, possible measures of prevention of IDDM and/or per-
DIABETES CARE, VOLUME 19,
NUMBER 8,
AUC;UST
1996
Green and Associates
Prevalence rate 4.0n (per 1000) Model 1
3.5-1
'.- Model 2
3.0-i
Model 3
2.5-i
2.0-i 1.5-i 1.0-j 0.5-: 0.0
1970
1980
1990
2000
2010
2020
Calendar time
B 20-i % of patients
Nephropathy (model 3)
15-
*"*mm.
Nephropathy (model 2)
-""""
Nephropathy (model 1)
10-
5Blindness (all models)
1970
1980
1990
2010
2000
2020
Calendar time Figure 4—Projected trends in the prevalence oj IDDM (A, prevalence rates oj IDDM) and in the relative prevalences oj complications (B, blindness and nephropathy in IDDM patients) according to incidence models. For specification oj incidence models, see Fig. 2 and text.
sistent proteinuria will change this trend to a modest degree only. We recommend that epidemiological models for diabetes like the one presented here should be further developed and applied to various scenarios of future trends in incidence of disease and complications.
APPENDIX: DESCRIPTION OF THE BASIC EPIDEMIOLOGICAL MODEL — We outline the principles and estimation procedures in the basic epidemiological model (Fig. 1) of IDDM with two possible patient states, i.e., presenting with or without a specified complication, respectively. The model assumes that we consider a geographically well-defined popuDIABETES CARE, VOLUME 19,
NUMBER 8,
AUGUST
lation, with net migration equal to 0 over the time span of the projection. The transitions between the various compartments of the model are determined by the incidence and mortality rates that are assumed as part of the various scenarios entertained. From these transitions, the sizes of the compartments determining the prevalences are updated after each time unit. In the following notation, (t) refers to the time point t for prevalence estimates. For incidence and mortality, (0 refers to the time unit immediately following time point i. Input The number of new cases, d(t), of the disease developed per time unit after time 1996
point t is estimated as d(t) = N(fl X ID(t), where N(t) is the general population at risk at time point t and ID(f) is the incidence rate of IDDM during the time unit concerned. Similarly, the number of complication-free patients, c(t), developing the complication per time unit after time point t is estimated as c(t) = D(0 X IC(0, where D(t) is the number of patients without complications at time point t and IC\t) is the incidence rate of the complication during the time unit concerned. Output The withdrawal from the compartments is determined by the mortality rates applicable for each compartment. For the patients without complications, D(t), output also comprises transitions to the compartment of patients with complication, C{t). From the population of patients without complication, the number of deaths, x((), per time unit after time point t is estimated as D(t) X (mortality rate for patients without complication). Further output is represented by transitions to the compartment of patients with complications, as described above. From the population of patients with complications, the number of deaths, ?(t), per time unit after time point t is estimated as C(t) X (mortality rate for patients with the complication). Updating the prevalences At time point t + 1, the absolute number of patients without complication is estimated as D(0 + d(t) - c(t) - x(t). Similarly, at time point t + 1, the absolute number of patients with complications is estimated as C(t) + c(t) - z(t). Thus, the overall prevalence proportion of patients is [D(t + 1) + C(t + 1)]/N(t + 1), and the relative prevalence of patients with complications becomes C{t + l)/[D(t + 1) + Getting started For time point t = 0, it is assumed that the absolute numbers of subjects in the various compartments are known and that the inputs and withdrawals from the compartments also are known in absolute numbers. These numbers provide for the estimation of the various incidence and mortality rates, as well as prevalence proportions. For the subsequent time units, the incidence and mortality rates are modeled under the scenarios entertained, and the population is updated per time 805
Trends in the epidemiology oflDDM
Br Med J 306:1722-1723, 1993 Ruwaard D, Hoogenveen RT, Verkleij H, Kromhout D, Casparie AF, van der Veen EA: Forecasting the number of diabetic patients in the Netherlands in 2005. Am] References Public Health 83:989-995, 1993 1. Bingley PJ, Gale EAM: Rising incidence of Green A, Hougaard P: Epidemiological IDDM in Europe. Diabetes Care 12:289studies of diabetes mellitus in Denmark. 295,1989 IV. Clinical characteristics of insulin2. Green A, Borch-Johnsen K, Andersen PK, treated diabetic patients. Diabetologia 25: Hougaard P, Keiding N, Kreiner S, Deck231-234, 1983 ertT: Relative mortality of type 1 (insulinGreen A, Hauge M, Holm NV, Rasch LL: dependent) diabetes in Denmark, 1933Epidemiological studies of diabetes melli1981. Diabetologia 28:339-342, 1985 tus in Denmark. II. A prevalence study 3. Green A: Epidemiologic and public health based on insulin prescriptions. Diabetoloaspects of insulin-treated diabetes in Dengia 20:468-470, 1981 mark. Diabetes Care 8 (Suppl. 1):77-81, 9. Eshoj O, Green A, Borch-Johnsen K, 1985 Feldt-Rasmussen B, Beck-Nielsen H: In4. The Diabetes Control and Complications creased prevalence of insulin-treated diaTrial Research Group: The effect of intenbetes mellitus in Funen County, Densive treatment of diabetes on the developmark. JInt Med 235:405-410, 1994 ment and progression of long-term complications in insulin-dependent diabetes 10. Green A, Hougaard P: Epidemiological studies of diabetes mellitus in Denmark. mellitus. N Engl J Med 329:977-986, V. Mortality and causes of death among 1993 insulin-treated diabetic patients. Diabeto5. Borch-Johnsen K, Wenzel H, Viberti GC, logia 26:190-194, 1984 Mogensen CE: Is screening and intervention for microalbuminuria worthwhile in 11. Sjolie AK: Ocular complications in insulin treated diabetes mellitus: an epidemiologpatients with insulin dependent diabetes?
unit with respect to the resulting prevalence rates.
806
12.
13.
14.
15.
16.
17.
ical study. Ada Ophthalmol Suppl. 172:1— 76,1985 Sjolie AK, Green A: Blindness in insulintreated diabetic patients with age at onset