The Belgrade childhood diabetes study: a multivariate analysis of risk determinants for diabetes (original) (raw)

Abstract

Background: The aim of this study was to evaluate some hypotheses about factors related to the development of type 1 diabetes mellitus. Methods: A case–control study was conducted in Belgrade during the period 1994–1997. A total of 105 recently onset diabetic and 210 control children, individually matched by age (±1 year), sex and place of residence, were included in the study. Results: According to multivariate regression analysis, the following factors were related to type 1 diabetes: stressful events and symptoms of psychological dysfunction during the 12 months preceding the onset of the disease [odds ratio (OR) 3.48, 95% confidence interval (CI) 2.15–5.65; and OR 2.15, 95% CI 1.33–3.48], irregular vaccination (OR 16.98, 95% CI 1.38–208.92), infection during 6 months preceding the onset of the disease (OR 4.23, 95% CI 1.95–9.17), higher education of father (OR 1.50, 95% CI 1.05–2.14), mother's consumption of nitrosoamines-rich food during pregnancy (OR 4.33, 95% CI 1.95–9.61), alcohol consumption by father (OR 3.80, 95% CI 1.64–8.78), insulin-dependent and non-insulin-dependent diabetes mellitus in three generations of children's relatives (OR 20.04, 95% CI 4.73–84.81; and OR 5.52, 95% CI 2.45–12.46), and use of ultrasound diagnostic techniques during pregnancy (OR 0.42, 95% CI 0.17–1.00). Conclusions: Among non-genetic factors, those affecting the child during pregnancy are especially important because of their preventability.

Type 1 diabetes mellitus is believed to develop as an autoimmune disease in genetically susceptible subjects following exposure to environmental factors.1 The evidence of a genetic component to type 1 diabetes is based on the high concordance rate in monozygotic twins as compared with dizygotic twins, the higher recurrence risk among relatives of patients with type 1 diabetes as compared with general population risk and the well-established associations with genetic markers.2 The evidence of a non-genetic component to type 1 diabetes is based on the following facts: that the concordance rate in monozygotic twins is <50%,3 the great variability in the incidence rate of type 1 diabetes both between genetically very similar neighbouring populations and within genetically homogenous populations, the rapidly increasing incidence of type 1 diabetes in many countries, its seasonal pattern and its occurrence in epidemic form similar to epidemics of infectious diseases.4,5 On the basis of the existing data it is postulated that out of environmental factors, fetal infections, maternal–fetal blood group incompatibility, cow's milk protein and nitrosamine products may initiate, through different mechanisms, β-cell damage and thus starts the autoimmune process. On the other hand, environmental factors such as cold weather, different postnatal infections, high growth rate and psychological stress could act as accelerating risk factors through increasing the workload of β-cells.5

The aim of the present study was to evaluate some of the hypotheses about factors related to the development of type 1 diabetes.

Materials and methods

A case–control study was conducted in Belgrade (∼320 000 inhabitants aged 0–16 years old) during the period 1994–1997. The population of Belgrade is multiethnic but the majority of Belgrade citizens, almost 90%, are Serbs.6

The case group comprised children ≤16 years old who were hospitalized for the first time because of type 1 diabetes mellitus in two hospitals in Belgrade, to which all children with suspected diabetes up to the age of 16 years have been referred. Diagnosis of type 1 diabetes was made according to WHO criteria.7 Out of 116 children diagnosed as type 1 diabetes cases during the 1994–1997 period, 105 (90.5%) were included in the study. Five children subsequently changed their place of residence and could not be traced, and parents of six children refused participation in the study.

For each case, two outpatient controls were chosen among children treated for skin diseases (erytema solarae, urticaria, verrucae vulgares, verrucae plantaris, herpes simplex, impetigo contagiosa streptococcica, impetigo staphylococcica, impetigo vulgaris, ecthyma, strophulus infantum, eczema, acne vulgaris, angulus infekciosus, onychomycosis, pediculosis capitis and scabies). Parents of all control children accepted participation in the study.

Cases and controls were individually matched by age (±1 year), sex and place of residence (all were from Belgrade). In both groups there were 50.5% boys and 49.5% girls. There was only slight difference in age between cases and controls: 18.1% versus 17.1% were in the age group 0–4 years, 31.4% versus 31.9% were in age group 5–9 years, 41.9% versus 41.0% were in age group 10–14 years and 5.8% versus 7.7% were in age group 15–16 years.

The questionnaire covered a variety of questions concerning characteristics of children's parents and family characteristics, and questions concerning neonatal factors, infant feeding, nursery and kindergarten attendance, infectious and some non-infectious diseases, compulsory vaccination in personal histories (through vaccination certificate), presence of stressful events and symptoms of psychological dysfunction during 12 months preceding the onset of the disease, as well as questions about the presence of different diseases among three generations of children's relatives. Information on stressful events and symptoms of psychological dysfunction were collected by the use of the questionnaire applied in a study conducted by Siemiatycki et al.8

Data about study participants, cases and controls, were obtained from their parents, most frequently from their mothers, within 6–12 weeks after diagnosis of diabetes was established. A trained interviewer conducted all interviews.

In the statistical analysis, univariate and multivariate conditional logistic regression methods were applied. For all calculations, SPSS software was used.

Results

Tables 1–4 present variables which were, according to univariate logistic regression analysis, associated with type 1 diabetes at a level _P_≤0.1.

The following family characteristics (table 1), were more frequently reported in diabetic children, in comparison with their controls: higher education of father; mother's and father's specific occupational exposure to radiation, low temperature, petroleum and its derivates, metallic dust, asbestos, steel, organic solvents, dyes and lacquers; more than five family members; and day nursery attendance. There were no statistically significant differences between cases and controls in mother's education, occupation of father and mother, marital status of parents, <6 m2 of living space per family member and kindergarten attendance (_P_≤0.10).

As presented in table 2, many factors present during pregnancy (infections, complications, medication, hormonal therapy, consumption of nitrosoamines-rich food, consumption of coffee, coca cola and alcohol, and cigarette smoking), as well as gestational age >41 weeks, need for postnatal intensive care, short duration of breast feeding and early introduction of supplementary milk, were also related to type 1 diabetes. Fathers of diabetic children more frequently consumed alcohol. Use of ultrasound diagnostic techniques and tea consumption during pregnancy were more frequently reported among controls. There were no statistically significant differences between cases and controls in mother's weight gain during pregnancy >15 kg, father's smoking habits, parental age at child birth, birth order, birth weight and length of child, and mode of delivery (_P_>0.10).

In the personal histories of type 1 diabetes cases (table 3), various infectious and non-infectious diseases (whooping cough, rubella, frequent respiratory, urinary gastrointestinal and some other infections, febrile convulsions, and allergic diseases) were more frequently present. Infections during the 6 months preceding the onset of the disease (common cold, sore throat, bronchitis, otitis media, gastro-enteritis, urinary tract infections, rubella and chickenpox) were also more frequent among cases than among control children. Both cases and controls had received vaccines that are compulsory in Yugoslavia, but in 6.7% of cases and 1.0% of controls some vaccinations were not administered at the prescribed times, but were postponed for various reasons. Postvaccinal complications were more often recorded in diabetic than in control children. Stressful events and symptoms of psychological dysfunction were more frequent in diabetic than in control children. Although the questionnaire provided more detailed data about different types of stressful events and about different symptoms of psychological dysfunction, in this paper only total frequency of these two indicators of stress is presented. There was also some evidence that dose–response risk for type 1 diabetes increased with greater number of both stressful events and symptoms of psychological dysfunction. There were no statistically significant differences between cases and controls in the frequency of the following diseases: measles, chickenpox, mumps, exanthema subitum, scarlet fever, mononucleosis, otitis and/or sinusitis and coeliac disease (coeliac disease was present in the history of only one diabetic child).

As can be seen from table 4, many diseases, including type 1 and type 2 diabetes, and thyroid, rheumatic, gastrointestinal, allergic and other diseases, were more frequent among relatives of diabetic children than among relatives of control children.

Multivariate logistic regression analysis was performed in two steps. First, we used five separate models including all variables that, according to univariate logistic regression analysis, were related to type 1 diabetes at a significance level _P_≤0.10. Model 1 included family characteristics; model 2, neonatal characteristics; model 3, infant diet; model 4, personal history; and model 5, family history. Variables significantly related to type 1 diabetes according to these five multivariate regression analyses were put together in model 6. These variables were as follows: higher education of father, more than five family members, day nursery attendance, infections and complications during pregnancy, consumption of nitrosoamines-rich food and coffee by mother during pregnancy, gestational age >41 weeks, alcohol consumption by father, less frequent use of ultrasound diagnostic techniques during pregnancy, early introduction of supplementary milk, rubella and frequent respiratory and other infections during life-time preceding the onset of the disease, infections during 6 months preceding the onset of the disease, irregular vaccination, stressful events and symptoms of psychological dysfunction during the 12 months preceding the onset of the disease, and type 1 and type 2 diabetes, allergic and other diseases among relatives.

According to the results obtained by the use of multivariate regression analysis with model 6 and presented in table 5, the following factors were significantly related to type 1 diabetes: stressful events and symptoms of psychological dysfunction during the 12 months preceding the onset of the disease, irregular vaccination, infection during the 6 months preceding the onset of disease, higher education of father, mother's consumption of nitrosoamines-rich food during pregnancy, alcohol consumption by father, and type 1 and type 2 diabetes in three generations of children's relatives. Use of ultrasound diagnostic techniques during pregnancy was independently related to insulin-dependent diabetes mellitus at significance level of _P_=0.0505.

Discussion

The results of the present study are in agreement with the well-established view that the cause of type 1 diabetes mellitus is complex, involving genetic susceptibility as well as non-genetic determinants.

The indication of a genetic component to type 1 diabetes in this study is based on the finding that a positive family history of type 1 and type 2 diabetes was a risk factor for the occurrence of type 1 diabetes in children.

The tendency to clustering of type 1 diabetes within families is well established. Several studies have found consistent estimates of type 1 diabetes risk in the range of 5–10% among siblings and children of type 1 diabetes patients.9 In a study from Sweden,10 the odds ratio for type 1 diabetes in children with reported type 1 diabetes in three generations of relatives was 5.52. In our study, when analysing three generations of relatives (grandparents, parents and parental siblings, cousins and siblings), the odds ratio for type 1 diabetes in children, when a relative had type 1 diabetes, was 20.04 (95% confidence interval 4.73–84.81). As indicated by large confidence intervals, in the present study the estimation of type 1 diabetes risk in those having relatives with the same disease was not precise enough. Since the incidence of type 1 diabetes in Belgrade is low to medium,11 for more precise estimates of familial risk a larger sample is needed.

The significantly increased type 1 diabetes risk when type 2 diabetes was reported in relatives found in the present study has also been found also in some other investigations,10 although not in all of them.12 The relationship between type 1 and type 2 diabetes may suggest a genetic association between these two diseases. On the other hand, similar environmental risk determinants for type 1 and type 2 diabetes may also explain this association.

Among environmental factors that, in the present study, were independently related to type 1 diabetes in children, some of them are considered as accelerating risk factors for type 1 diabetes (infections during the 6 months preceding the disease, irregular vaccination, stressful events and psychological dysfunction).

In many studies, including the present one, recent infections, during the 3 months, 6 months or 1 year preceding the onset of diabetes, increased the risk for type 1 diabetes.13 In the Swedish study,13 the total frequency of infectious diseases during the last year before diagnosis showed a dose–response relationship in the risk of developing type 1 diabetes. The most probable explanation of this finding is that infections precipitate the onset of type 1 diabetes by increasing the peripheral insulin need.

The possible association between viral diseases and type 1 diabetes aroused an interest in vaccination as a possible risk factor for diabetes; thus is it was postulated that vaccination may precipitate an autoimmune response towards β-cells.13 However, a protective effect of measles vaccine was found in one study,13 and a protective effect of immunization (against tuberculosis, diphtheria, tetanus, whooping cough, polio and measles) was demonstrated in the study by Telahun et al.14 Summarizing all recently performed investigations Hiltunen et al.15 concluded that there was no clear evidence that any currently used vaccine can prevent or induce diabetes in humans. The most probable explanation of the preventive effect of regular vaccination (against tuberculosis, diphtheria, whooping cough, tetanus, poliomyelitis, measles, mumps and rubella) found in the present study is that due to immunization, children evade infections that could be related to type 1 diabetes in one of the postulated mechanisms. Of the diseases against which vaccination is compulsory in Yugoslavia, whooping cough and rubella were more frequently reported by diabetic children than by controls.

According to the results in the present study, stress appeared to be a risk factor for type 1 diabetes, and the risk of type 1 diabetes increased both with increasing numbers of stressful events and with increasing numbers of social and psychological dysfunctions. These findings are in line with the majority of other studies, particularly with a case–control study conducted by Siemiatycki et al.8 in Montreal, Canada. Similar results were obtained in other studies. For example, Robinson and Fuller16 found that young adults with type 1 diabetes reported more stressful events than did their siblings in the period preceding onset of the disease.

It is probable that stress (social, psychological, physical), by causing excessive release of cortisol and catecholamine, increases insulin requirements and in that way, by inducing an increased workload on the β-cells, precipitates the onset of disease.5 The other possible mechanism is that stress could accelerate the autoimmune destruction of β-cells.17 There is experimental evidence that stress affects the functioning of the immune system.18

The positive relationship of higher education level of the father with diabetes in children in the present study is difficult to explain. In studies in which association between type 1 diabetes and education level was found, it was between type 1 diabetes and lower education level of parents, especially of the mother.19 This was explained by correlation of mother's education and preventive health behaviour of the family.20 It is possible that education of the father is related to type 1 diabetes through socioeconomic status. The association of type 1 diabetes with socioeconomic status has been found in many studies, but the direction of this association is not consistent. In some studies, type 1 diabetes incidence has been found to be greater in lower socioeconomic classes,21 and in some other studies the frequency of type 1 diabetes was greater in higher socioeconomic classes.22 It has been postulated that different effects of socioeconomic status on type 1 diabetes occurrence can result from the fact that health-related behaviour and food consumption patterns differ by socioeconomic status in different countries, possibly in different ways.23 For example, while in some countries the diet of lower social strata is characterized by nitrosamines-rich food, in some other countries the opposite is the case. Concerning the results obtained in this study, it is necessary to mention that under contemporary conditions in Yugoslavia, characterized by a prolonged political and economic crisis, education level is not related to the living standard in a logical way as is usually the case in many other countries.

Increased frequency of use of ultrasound diagnostic techniques among mothers of control children might indicate that they are more concerned about their own health and the health of their children. An active or passive attitude towards health could in many ways be related to the occurrence of type 1 diabetes.

The association between the development of type 1 diabetes and the intake of nitrosamines-rich food by mothers during pregnancy, found in the present study, was for the first time indicated by a descriptive epidemiological study from Iceland24 and later proved in a detailed dietary study performed in Finland.25 A cause–effect relationship between nitrosamine compound and β-cell destruction is supported by findings in experimental animals showing that nitrosamine compounds are toxic to the β-cell.26

Significantly more frequent alcohol consumption among fathers of diabetic children in the present study could have two possible explanations. Either alcohol could exert its harmful effect on germ cells, or it could be related to type 1 diabetes development affecting some components of family socioeconomic status.

There are two major potential sources of bias in case–control studies: inappropriate control group and recall bias. In the present investigation, children and adolescents with skin disease were chosen as controls, although in previous case–control studies of type 1 diabetes most frequently healthy individuals were used as controls. In case–control studies, controls can be chosen among healthy people or among people with some disease (other than the disease under investigation). The controls must be selected to represents not the entire non-diseased population (without the disease under investigation), but the population of individuals who would have been identified and included as cases had they also developed the disease. Each type of control has its advantages and disadvantages. Selection of controls among diseased people in a case–control study decreases the likelihood of non-response and recall bias.27 During the period 1994–1997, when the study was conducted, it was preferable to use as controls people who visited physicians to seek medical advice and treatment, rather than to use healthy controls. At the beginning, during the design stage of the study, it was planned to use schoolmates as controls, but in a pilot study it was realized that the majority of parents of healthy children were not willing to participate. On the other hand, subjects who came to seek physician's advice could not refuse participation in the study. According to data in the literature,28 transmissible skin diseases are more frequent in children from families with low socioeconomic conditions, the household density being the main or the only socioeconomic factor significantly related to skin diseases. In contrast, eczema was more frequent among children from high social classes and it was postulated that it could be attributed to the early exposure of children to artificial milk in infancy.29 It is indeed difficult to find any disease, that is not, in one way or another, associated with socioeconomic conditions. It is also true that controls chosen among patients will differ, in one way or another, from controls chosen among healthy individuals. In the present study, cases and controls did not differ in household density, which is, according to the literature, related to skin diseases. At the same time, since skin diseases are most frequently transmitted among schoolmates, the composition of such control group would be very similar to controls taken among healthy children from the school or neighbourhood of a type 1 diabetes child.

In a case–control study there is always concern about reliability of data collected, since these data are retrospective, and recall of past exposure can be incomplete, incorrect and, what is worse, can differ between cases and controls. Specifically, it is possible that parents of diabetic children, looking for a cause that made their child ill, may remember past events better than parents of healthy children or parents of a child with a less serious disease, such as skin disease. However, we do not expect important differences in recall between parents of cases and parents of controls. Among factors found to be independent risk factors for insulin-dependent diabetes mellitus, data about stress and infections were asked for a recent period, 1 year or 6 months preceding the onset of the disease, and they were asked separately for each stressful event (hospitalization, accident, death of close relative, divorce of parents and so on), each of seven psychological dysfunctions (sleeping problems, nightmares, disliking school, learning problems, low grades at school, having no close friends, problems with peers), and specific and non-specific infections. We also do not expect recall bias in family histories. Since diabetes is a serious disease it would be quite unusual for people not to know or not to remember the presence of such a disease among their relatives. Concerning other variables, since parents were not aware of the hypotheses tested, one could expect that the proportion of wrongly classified subjects on account of incomplete or incorrect recall would be similar in both case and control groups.

In conclusion, the majority of environmental risk factors for type 1 diabetes identified so far in the present study were found to be independently related to the development of type 1 diabetes in children. The exception is early supplementation with cow's milk, which, after adjustment for confounding factors, was not independently associated with type 1 diabetes. However, findings about cow's milk proteins as a risk factor for type 1 diabetes are not consistent.30

Table 1

Distribution of diabetic children (_n_=105) and their controls (_n_=210) according to family characteristics, household crowding and attendance of preschool institutions

a

According to conditional univariate logistic regression analysis.

Table 1

Distribution of diabetic children (_n_=105) and their controls (_n_=210) according to family characteristics, household crowding and attendance of preschool institutions

a

According to conditional univariate logistic regression analysis.

Table 2

Distribution of diabetic cases (_n_=105) and their controls (_n_=210) according to neonatal factors and infant feeding

a

According to conditional univariate logistic regression analysis.

Table 2

Distribution of diabetic cases (_n_=105) and their controls (_n_=210) according to neonatal factors and infant feeding

a

According to conditional univariate logistic regression analysis.

Table 3

Personal histories of diabetic cases (_n_=105) and their controls (_n_=210)

a

According to conditional univariate logistic regression analysis.

b

Three or more infections per year.

c

Herpes zoster, stomatitis angularis, pseudocroup, impetigo, pulmonary tuberculosis.

d

Rhinitis, asthma, nutritive and medicaments allergy.

Table 3

Personal histories of diabetic cases (_n_=105) and their controls (_n_=210)

a

According to conditional univariate logistic regression analysis.

b

Three or more infections per year.

c

Herpes zoster, stomatitis angularis, pseudocroup, impetigo, pulmonary tuberculosis.

d

Rhinitis, asthma, nutritive and medicaments allergy.

Table 4

Positive family history of different diseases in three generations of relatives of diabetic cases (_n_=105) and their controls (_n_=210)

a

Adrenal disease was present in only one relative of one diabetic child.

b

According to conditional univariate logistic regression analysis.

c

Rhinitis, asthma, nutritive and medicaments allergy.

d

Trombocitopenia, alopecia areata, psoriasis, uveitis chronica, epilepsy, anemia perniciosa, psychotic and malignant diseases.

Table 4

Positive family history of different diseases in three generations of relatives of diabetic cases (_n_=105) and their controls (_n_=210)

a

Adrenal disease was present in only one relative of one diabetic child.

b

According to conditional univariate logistic regression analysis.

c

Rhinitis, asthma, nutritive and medicaments allergy.

d

Trombocitopenia, alopecia areata, psoriasis, uveitis chronica, epilepsy, anemia perniciosa, psychotic and malignant diseases.

Table 5

Factors related to type 1 diabetes according to multivariate logistic regression analysis

OR: odds ratio; CI: confidence interval.

Table 5

Factors related to type 1 diabetes according to multivariate logistic regression analysis

OR: odds ratio; CI: confidence interval.

• The aim of this study was to evaluate some hypotheses about factors related to the development of type 1 diabetes mellitus.
• Stress, infections, irregular vaccination, some prenatal factors and diabetes mellitus in three generations of children's relatives were related with type 1 diabetes.
• Among non-genetic factors, those affecting child during pregnancy are especially important because of their preventability.

This work was supported by the Ministry for Science and Technology of Serbia, through contract No. 8774/1991–1995 and 1996–2000.

References

1

Nerup J, Lernmark A. Autoimmunity in insulin-dependent diabetes mellitus.

Am J Med

1981

;

70

:

135

–41.

2

Green A. Prevention of IDDM: the genetic epidemiologic perspective.

Diabetes Res Clin Pract

1996

;

34

(Suppl):

101

–6.

3

Olmos P, Hearn R, Heaton DA, et al. The significance of concordance rate of type 1 (insulin-dependent) diabetes mellitus in identical twines.

Diabetologia

1988

;

31

:

747

–50.

4

WHO DIAMOND Project Group on Epidemics. Childhood diabetes, epidemics, and epidemiology: an approach for controlling diabetes.

Am J Epidemiol

1992

;

135

:

803

–16.

5

Dahlquist G. Non-genetic risk determinants of type 1 diabetes.

Diabetes Metab

1994

;

20

:

251

–7.

6

Statistical year book for Belgrade. Belgrade: Institute for Informatics and Statistics,

1993

.

7

WHO Study Group of Diabetes Mellitus. WHO technical report series No. 727. Geneva: WHO,

1985

.

8

Siemiatycki J, Colle E, Campbell S, et al. Case–control study of IDDM.

Diabetes Care

1989

;

12

:

209

–16.

9

Tillil H, Kobberling J. Age-corrected empirical genetic risk estimates for first-degree relatives of IDDM patients.

Diabetes

1987

;

36

:

93

–9.

10

Dahlquist G, Blom L, Holmgren G, et al. The epidemiology of diabetes in Swedish children 0–14 years – a six-year prospective study.

Diabetologia

1985

;

28

:

802

–8.

11

Vlajinac HD, Bojović BM, Šipetić SB, et al. Insulin dependent diabetes mellitus: incidence in childhood in Belgrade 1982–92.

J Epidemiol Community Health

1995

;

49

:

107

–8.

12

Altobelli E, Chiarelli F, Valenti M, et al. Family history and risk of insulin-dependent diabetes mellitus: a population-based case–control study.

Acta Diabetol

1998

;

35

:

57

–60.

13

Blom L, Nystrom L, Dahlquist G. The Swedish childhood diabetes study. Vaccinations and infections as risk determinants for diabetes in childhood.

Diabetologia

1991

;

34

:

176

–81.

14

Telahun M, Abdulkadir J, Kebede E. The relation of early nutrition, infections and socio-economic factors to the development of childhood diabetes.

Ethiop Med J

1994

;

32

:

239

–43.

15

Hiltunen M, Lonnrot M, Hyoty H. Immunization and Type 1 diabetes mellitus: is there a link?

Drug Saf

1999

;

20

:

207

–12.

16

Robinson N, Fuller JH. Severe life events and their relationship to the aetiology of insulin-dependent (type 1) diabetes mellitus.

Pediatr Adolesc Endocrinol

1986

;

15

:

129

–33.

17

Bateman A, Singh A, Kral T, Solomon S. The immune–hypothalamic–pituitary adrenal axis.

Endocrinol Rev

1989

;

10

:

92

–112.

18

Laudenslager ML, Ryan SM, Drugan RC, et al. Coping and immunosuppression: inescapable but not escapable shock suppresses lymphocyte proliferation.

Science

1983

;

221

:

568

–70.

19

Blom L, Dahlquist G, Nystrom L, et al. The Swedish childhood diabetes study – social and perinatal determinants for diabetes in childhood.

Diabetologia

1989

;

32

:

7

–13.

20

Steel JL, McBroom WH. Conceptual and empirical dimensions of health behaviour.

J Health Soc Behav

1972

;

13

:

382

–92.

21

Lawler Heavner J, Cruickshanks KJ, Hay WW, et al. Birth size and risk of insulin-dependent diabetes mellitus (IDDM).

Diabetes Res Clin Pract

1994

;

24

:

153

–9.

22

Tarn AC, Gorsuch AN, Spencer KM, et al. Diabetes and social class.

Lancet

1983

;

2

:

631

–2.

23

Rasanen L, Ahola M, Kara R, Uhari M. Atherosclerosis precursors in Finnish children and adolescents. VIII. Food consumption and nutrient intakes.

Acta Paediatr Scand Suppl

1985

;

318

:

135

–53.

24

Helgason T, Jonasson MR. Evidence for a food additive as a cause of ketosis-prone diabetes.

Lancet

1981

;

2

:

716

–20.

25

Virtanen SM, Jaakkola L, Rasanen L, et al. Nitrate and nitrite intake and the risk for type 1 diabetes mellitus in Finnish children.

Diabet Med

1994

;

11

:

656

–62.

26

Helgason T, Ewen SW, Rose JS, Stowers JM. Diabetes produced in mice by smoked-cured mutton.

Lancet

1982

;

2

:

1017

–22.

27

Rothman KJ, Greenland S. Modern Epidemiology. Philedelphia, P.A: Lippincott-Raven,

1998

.

28

Gibbs S. Skin disease and socioeconomic conditions in rural Africa: Tanzania.

Int J Dermatol

1996

;

35

:

633

–9.

29

Odueko OM, Onayemi O, Oyedeji GA. A prevalence survey of skin diseases in Nigerian children.

Niger J Med

2001

;

10

:

64

–7.

30

Meloni T, Marinaro AM, Mannazzu MC, et al. IDDM and early infant feeding. Sardinian case control study.

Diabetes Care

1997

;

20

:

340

–2.

© The Author 2005. Published by Oxford University Press on behalf of the European Public Health Association. All rights reserved.