The Minnesota Department of Health (MDH) in concert with the Centers for Disease Control (CDC) conducted population-based studies of diabetes mellitus in ...
I
ncidence, Prevalence, and Mortality of Diabetes Mellitus in Wadena, Marshall, and Grand Rapids, Minnesota: The Three-City Study
A. P. BENDER, DVM, PhD, J. M. SPRAFKA, PhD, MPH, H. G. JAGGER, RN, MPH, K. H. MUCKALA, MD, C. P. MARTIN, MD, AND T. R. EDWARDS, MD
The Minnesota Department of Health (MDH) in concert with the Centers for Disease Control (CDC) conducted population-based studies of diabetes mellitus in three Minnesota communities. The use of hospital and clinic records alone for case ascertainment purposes would have missed 16% of studyeligible diabetic individuals. Thus, studies that use only hospital or clinic records may present a biased view of the natural history of diabetes. Physician-defined diabetes without additional diagnostic review yielded a prevalence of 1.6% (age standardized to the 1970 US Caucasian population). Diabetes incidence was 117 per 100,000 (age standardized to the same population). These results compare well with those of a Mayo Clinic population-based study in Rochester, Minnesota, and indicate the utility of physiciandefined diabetes as an epidemiologic case definition. Despite differences in population size and structure as well as different medical care systems, diabetes incidence, prevalence, and mortality in these communities was remarkably similar, DIABETES CARE 1986; 9:343-50.
E
stimates of diabetes prevalence and incidence vary considerably, due in part to population differences but also due to interstudy differences in diagnostic criteria and case ascertainment.1"5 In the last 40 yr, postprandial, random, and fasting blood glucose, glucose tolerance tests, glycosuria, history and recall of physician diagnosis, use of insulin or oral agents, hospital discharge and/ or medical record review in population-based or quasi-population-based settings have been used to define diabetes melPrevalence and incidence have also been reported for a wide variety of age ranges, adding to the complexity of nonstandardized population comparisons.4 It is widely believed, for example, that the prevalence and incidence of diabetes in the United States has been increasing steadily since the 1930s.11'12 Much of the evidence for this assumption rests on the above-referenced community studies. However, when data from these studies are age standardized, many of the rates are remarkably similar13 and evidence for a temporal increase in age-standardized rates of diabetes must be viewed with caution.1415 Due to the differences in case-ascertainment methods, diagnostic confirmation techniques, age structure of the population studied, and other personal characteristics, such as obesity,4 it is difficult to draw the needed inferences about
prevalence of diabetes mellitus in the United States required for programmatic evaluation. The Minnesota Department of Health (MDH) and the Centers for Disease Control (CDC) recognized the importance of good baseline data to define programs for control of diabetes and its complications.16 As part of the CDC Diabetes Control Project the MDH conducted population-based studies in Wadena, Marshall, and Grand Rapids, Minnesota. The first phase of the study reported here was the determination of the incidence, prevalence, and mortality of individuals with diabetes residing within the three cities. METHODS
Rationale and methods for the conduct of population-based studies in the Three-City Study have been previously described.17 Briefly, there were three criteria for a community to be included in the study: J) population between 5000 and 15,000—large enough to ensure an adequate number of cases, yet small enough to facilitate case ascertainment and record abstraction; 2) self-contained medical system sufficiently distant from major medical centers to limit amount of care sought outside the community, and 3) interested and cooperative medical community and civic leaders. Three cities, Wadena,
DIABETES CARE, VOL. 9 NO. 4, JULY-AUGUST 1986
343
THE THREE-CITY STUDY/A. P. BENDER AND ASSOCIATES
TABLE 1 Size and medical characteristics of participating communities in the Minnesota Three-City Study
1980 population Median ages Men Women Total Number of: Hospitals Medical clinics Physicians General practice Internal medicine Surgery Ophthalmology Other Pharmacies Nursing homes
Wadena
Marshall
Grand Rapids
4699
11,131
7934
29.9 36.5 33.2
24.9 26.7 25.6
29.2 33.4 31.3
1 1 7 6
1 3 15 8
1
3 1
1
3 1
2 4 1
2 4
1
1
4 23 13 5 2
1
Marshall, and Grand Rapids, satisfied these criteria and case ascertainment began in Wadena on May 18, 1981. Table 1 contains several demographic and medical characteristics of these communities. The study population in each community consisted of all individuals with physician-diagnosed diabetes living within the city limits anytime between January 1, 1979, and the date when case identification was initiated in that community. For Wadena this time period was 1/1/79-5/18/81; for Marshall, 1/1/79-11/1/81; and for Grand Rapids, 1/1/795/31/82. A certified medical records technician conducted case identification and record abstraction in all three communities, with assistance from a second abstractor for Marshall. An initial list of diabetic individuals was generated by reviewing local hospital discharge diagnoses and clinic records (primary medical sources). In Wadena and Grand Rapids, review of clinic records was facilitated by computerized billing records, which included ICD-9 coded reason(s) for medical encounters. It was recognized that use of primary medical sources alone for ascertainment would miss the following diabetic individuals: 1) those not seeking care anytime during the study period; 2) those seeking care outside the community; 3) those with an address change during the study period (i.e., current address ineligible but past address eligible); 4) those with no diabetes diagnosis code on discharge or clinic records for medical encounters during study period. After ascertaining known diabetic patients from primary medical sources, all secondary sources serving these communities were contacted and asked to provide a roster of their diabetic clientele or to allow the medical records technician access to their files for chart review. Secondary sources included pharmacies, public health nursing, nursing homes,
344
diabetes teaching teams, networking through local American Diabetes Association (ADA) chapters, schools, and other civic organizations. In conjunction with the medical communities, a variety of publicity techniques (e.g., TV, radio, newspaper articles) were implemented to enhance community awareness and stimulate diabetic case ascertainment. A survey to identify individuals with diabetes was conducted in Wadena and Grand Rapids by enclosing a return-addressed post card in each resident's electric bill. Due to differences in billing practices, a survey of this type was not feasible in Marshall and the identification effort was replaced by two survey instruments in the local newspaper. The roster of diabetic individuals generated from secondary sources and the survey were cross-checked with the initial roster obtained from review of primary sources and all duplications were omitted. A master list of all individuals from primary and secondary sources in the communities was compiled. After the eligibility determination (physician diagnosis of diabetes in medical record and living within the city limits anytime during the study period) all hospital, clinic, and secondary source records for study subjects were abstracted. A detailed medical history from the time diabetes was diagnosed until the end of the study period, the subject's death, or their moving from the city was compiled from the abstracted data. To ensure consistency, one staff epidemiologist (J.M.S.) coded all abstracted data. Also, a 5% sample of original records was reabstracted blindly, and a 10% sample of all abstracts was recoded blindly in the first city (Wadena) to ensure consistency and accuracy of the data-gathering process. Follow-up was accomplished primarily through medical records, but post offices, relatives, neighbors, and civic organizations were also used. Death certificates for subjects who died within the study period were obtained from the MDH Division of Vital Statistics and the causes of death, both underlying and contributing, were coded and tabulated.
RESULTS
A total of 2625 individuals were identified as potential candidates for study. Of these, 2030 (77.3%) were ineligible, 2025 lived outside the city limits, and 5 could not be confirmed through medical records or physician inquiry as having diabetes. The remaining 595 met all study criteria and their medical profiles were constructed from hospital and physician records. An estimate of the number of individuals with diabetes that would have satisfied study criteria yet were not identified can be approximated from the survey results.18 Overall, the ascertainment was estimated to be ^ 9 5 % complete. One hundred fifty-one diabetic individuals were identified in Wadena, 188 in Marshall, and 256 in Grand Rapids. Table 2 contains the distribution of record sources in which the 595 diabetic patients were initially located. Records from hospital and clinics/physician offices identified 83.9% of the subjects during the study period, and secondary sources accounted for
DIABETES CARE, VOL. 9 NO. 4, JULY-AUGUST 1986
THE THREE-CITY STUDY/A. P. BENDER AND ASSOCIATES
TABLE 2 Distribution of sources through which diabetic individuals were initially located for inclusion in the Minnesota Three-City Study
Primary sources Clinic or physician offices Hospital Secondary sources Pharmacies Self-identification survey Diabetes teaching team Nursing home Public health nursing Other
Total (N = 595)'
Wadena (N = 151)
404 (67.9) 95 (16.0)
126 (83.4)t
22 (3.7) 21 (3.5) 12 (2.0)
1 (0.7) 7 (4.6)
27 (4.5) 7 (1.2) 7 (1.2)
2 (1.3)
3 9 3 0
(N = 188)
Grand Rapids (N = 256)
70 (37.2) 78 (41.5)t
208 (81.2)t 15 (5.9)
Marshall
11 4 7 9 2 7
(2.0) (6.0) (2.0) (0.0)
10 10 2 9 2 0
(5.9) (2.1) (3.7) (4.8) (1.1) (3.7)
(3.9) (3.9) (0.8) (3.5) (0.8) (0.0)
'Number in parentheses is percent of total number, t Indicates first source used for case ascertainment.
16.1%. The combined total of postdiagnosis experience of the 595 subjects was 4201 yr. Crude prevalence ranged from 1.7 to 3.2%, with an overall crude prevalence of 2.5%. The age distribution by gender as of January 1, 1979, is given in Table 3 and emphasizes the significant contribution of individuals >70 yr to the diabetic community. Age-specific prevalence per 1000 population stratified by gender and city is given in Table 4. Prevalence increases dramatically with age in both sexes for all communities, the highest prevalence occurring in those >70 yr. Overall, male prevalence exceeded female prevalence in the younger age categories, female prevalence exceeded male prevalence in the middle-age categories, and no gender differences were apparent in older age categories. This trend, however, demonstrated considerable variability between communities. Prevalence estimates standardized to the 1970 US Caucasian population by city and gender are given in Table 5. Overall standardized prevalence was 1.60%, ranging from 1.38% in Marshall to 1.78% in Grand Rapids. Female prevalence was greater than male prevalence (1.75 vs. 1.44%, respectively), and this finding was consistent in all three communities. Male prevalence ranged from 1.23 to 1.69%, whereas female prevalence ranged from 1.50 to 1.92%. An attempt was made to estimate prevalence by diabetes type with a simplified interpretation of the classification scheme proposed by the National Diabetes Data Group.19 Insulindependent (type I) diabetes was defined as those individuals labeled as ketosis prone by their physician or ketoacidotic at diagnosis and using insulin for the duration of their disease. Non-insulin-dependent (type II) diabetes was defined as diabetic individuals who did not meet the criteria for the type I definition. Of 595 eligible study subjects, 55 (9.2%) were defined as type I, with an average age at diagnosis of 16.3 yr (ranging from 1.6 to 80.8 yr). The majority (80%) were 120% of ideal at diagnosis. Type I and type II prevalence estimates by gender and city standardized to the 1970 US Caucasian population are given in Table 5. Overall prevalence of type I diabetes appeared greater in men than in women (0.20 vs. 0.16%, respectively), where prevalence of type II diabetes was greater in women than in men (1.59 vs. 1.24%, respectively). The majority of diabetes in these communities (~91%) was the type II variety. Indirectly standardized prevalence by city and gender are shown in Table 6. Expected numbers were generated with data from Palumbo et al.10 and were compared with observed numbers in each city. Confidence intervals for observed-toexpected ratios were calculated with the method of Bailar and Ederer.20 Observed cases were defined as those prevalent at the midpoint of the study period (September 1980). Significant differences between observed and expected cases were found for the number of women in the communities of WaTABLE 3 Age distribution of diabetic individuals living in Wadena, Marshall, or Grand Rapids as of January 1, 1979, in the Minnesota Three-City Study
Age category (yr)
N
