Absence of breast-feeding is associated with the risk of type 1 diabetes: a case–control study in a population with rapidly increasing incidence
- First Online:
- Cite this article as:
- Malcova, H., Sumnik, Z., Drevinek, P. et al. Eur J Pediatr (2006) 165: 114. doi:10.1007/s00431-005-0008-9
- 259 Views
There are indications that the effect of environmental factors on the risk of type 1 diabetes mellitus (T1DM) is increasing over time. This can be documented by the rapid increase of T1DM incidence in genetically stable populations. Our aim was to study an association of T1DM with the variable factors of the perinatal period and of early infancy, using data from children born over a period of changing exposure to some of the studied factors. A case–control dataset was analysed, consisting of 868 diabetic children and 1,466 anonymous controls, mostly schoolmates of the children with T1DM. The data were collected using structured questionnaires completed by parents. After performing univariate analyses, the associations were analysed using multiple logistic regression adjusted for potential confounders, including the year of birth. The risk of T1DM decreased with increasing duration of breast-feeding, while no breast-feeding was associated with an increased T1DM risk, OR=1.93 [95% CI: 1.33–2.80], breast-feeding for more than 12 months was protective, OR=0.42 [95% CI: 0.22–0.81], both being relative to the reference category of breast-feeding for 1–3 months. A short duration of day-care attendance (none or less than 1 year) was weakly associated with the risk of T1DM, OR=1.65 [95% CI: 1.05–2.62]. No association was detected between T1DM and signs of prenatal infections, perinatal stress factors, birth size and weight, indicators of crowding or the presence of a domestic pet in the household. Short breast-feeding period and short attendance to day care is associated with the risk of T1DM in Czech children.
KeywordsDiabetes mellitusType 1EpidemiologyRisk factorsInfant nutrition
- 95% CI
95% confidence interval
Type 1 diabetes mellitus
Type 1 diabetes mellitus (T1DM) is a consequence of immune-mediated destruction of pancreatic β-cells in a genetically predisposed individual. Controversy still exists on which environmental factors—and to what extent—are responsible for the disease’s susceptibility. The effect of environmental factors is probably increasing; most European populations have experienced a continuous increase in T1DM incidence despite their genetic stability [1, 4]. This increase has been most pronounced in the populations of Central and Eastern Europe. In these populations, the putative environmental factors started to act later or with different strengths compared to the Western societies. The delay makes these populations interesting for investigating environmental factors in T1DM pathogenesis. In a previous paper, we addressed the T1DM risk associated with maternal age at delivery and birth order under the conditions of a rapid fall in natality and a swift increase of average maternal age .
In the present work, we aimed to investigate the associations of T1DM with several parameters of pregnancy, of the perinatal period and of early infancy, namely with signs of infection in pregnancy, with term and mode of delivery, birth weight and length, with infant feeding, attendance to day‐care, size of the community and the presence of domestic pet animals in the household.
Subjects and methods
The study design was case–control. Retrospectively obtained questionnaire data on the putative T1DM risk factors were compared between children with T1DM (cases) and their non-diabetic peers (control subjects). The study was approved by the Ethics Committee at the Motol University Hospital, Charles University, Prague, Czech Republic.
Logistics of the study
The study was conducted by the network of the population-based Czech Childhood Diabetes Register. Overall, 51 local pediatric diabetology centres took part in the study, with no geographical predominance compared with those who did not take part. The last 1,000 diabetic patients recorded by the Diabetes Register, aged less than 15 years at the onset of the disease, and under the age of 18 years at the time of the start of the study (March 2000), were invited to participate by their pediatric diabetologists. In addition, the diabetologists were encouraged to include children who manifested with T1DM since the last annual report to the Diabetes Register. The control subjects (schoolmates of diabetic children, unrelated to the case, non-diabetic and of approximately the same age) were chosen and their parents were approached directly by the parents of the patients. According to the criteria explained in letters to the parents, optimally, three control subjects should have been contacted.
The data were obtained using structured questionnaires identical for the children with T1DM and for the non-diabetic control subjects. The questionnaire was designed to evaluate the following variables: signs of infections during pregnancy (fever above 38°C, diarrhoea and rash, reported separately for each trimester), perinatal parameters (term of delivery, mode of delivery, birth weight and length, neonatal jaundice), infant feeding (duration of breast-feeding and time of introduction of formula or other supplementary feeding) and vitamin D supplementation. In children who already completed their fifth year of life, we also evaluated other putative factors of the pre-school period related to diabetes risk: day-care attendance, size of the community and the presence of pet animals in the household. Most questions had the option of “I do not remember/I do not want to tell,” which was recorded as “unknown” in the analysis.
The distribution of the investigated factors was first evaluated by univariate analyses and expressed as odds ratios (OR) with their 95% confidence intervals, and the continuous variables were analysed using the analysis of variance. Logistic regression models were built for mutually related exposures or for exposures that change over time, with adjustment to the calendar year. Discriminant functional analysis was used for the identification of the more important regressor among the correlated nutrition variables of breast-feeding and the introduction of supplementary feeding.
Characteristics of the study group
Children with T1DM
Number of distributed questionnaires
Number of returned questionnaires
Age at study
Median (25th; 75th percentile)
13 (10; 16)
12 (9; 15)
Distribution of age at T1DM onset
Distribution of studied parameters in children with T1DM and control patients
OR [95% CI]
Signs attributable to infections during pregnancy
Term of delivery
Preterm (<week 38)
Full term (week 38–42)
Postdate (>week 42)
Mode of delivery
Birth length and weight, median (25th; 75th percentile)
50 (49; 52)
50 (49; 52)
3,350 (3,028; 3,700)
3,350 (3,045; 3,650)
Unadjusted OR [95% CI]a
Adjusted OR [95% CI]b
Total duration of breast-feeding in completed months
No breast feeding
Time at introduction of formula or other supplementary feeding
Unadjusted OR [95% CI]c
Adjusted OR [95% CI]b
Vitamin D supplementation during the first year of lifed
OR [95% CI]
None of the term of delivery, mode of delivery, birth weight and length, or neonatal jaundice was associated with a subsequent risk of T1DM. The proportion of caesarean sections substantially increased over the period when the studied children were born, from 5.9% deliveries in the first quartile to 11.4% in the last quartile of the period. The association was, therefore, investigated also with an adjustment for the calendar year of birth, birth weight and length; none of the variables was significantly associated with T1DM (data not shown).
Univariate analysis of infant feeding showed that the risk of T1DM proportionally decreases both with increasing duration of breast-feeding and with the age when formula or other supplementary feeding was introduced. In a logistic regression model with the two variables and adjustment for the calendar year of birth, the duration of breast-feeding turned out as the more important regressor, while the association of T1DM with the time of introduction of supplementary feeding was found secondary to that of breast-feeding. Further discriminant functional analysis confirmed these results.
Analysis of putative risk factors operating in the pre-school age. Only children five years of age or older were analysed
Unadjusted OR [95% CI]
Adjusted OR [95% CI]a
Domestic pet in the household
Size of the community (number of inhabitants)
Less than 1,000
Total duration of day-care attendance in completed yearsb
Unadjusted OR [95% CI]
Adjusted OR [95% CI]c
None or <1 year
We analysed whether the variables associated with T1DM influences also the age at which it is diagnosed. The association was investigated using a model with dichotomised age at diabetes onset (below and over median) as the outcome, and the factors previously associated in the univariate analyses as the predictors. Neither of the variables influenced significantly the age at diabetes onset (data not shown).
We performed a large case–control study on type 1 diabetes mellitus (T1DM) environmental risk factors, finding significant association with parameters of infant nutrition and with attendance to day‐care.
The present study shows an association between the lack of breast-feeding and T1DM risk, while breast-feeding over 12 months is associated with protection. The parallel effect of the early introduction of supplementary feeding is secondary to the effect of breast-feeding. Such associations have been known for a long time; nevertheless, the question on which of the exposures is primary has not yet been fully answered (reviewed in ). The situation is further complicated by the putative effect of exposure to cereals, as indicated by results from the DAISY birth cohort . However, such a variable cannot be tested in a retrospective case–control study without the risk of a substantial recall bias. Although case–control data like ours may confer important pieces of the whole picture of nutritional risk factors, the final answer has to be provided by the ongoing prospective interventional studies as in the international TRIGR project .
Prenatal and perinatal variables
Delivery by caesarean section has been reported to significantly increase the risk of T1DM in some [8, 12], but not other, studies [3, 9, 18]. In the present study, the proportion of caesarean sections doubled over the study period. This change was parallel in children with T1DM and in controls, indicating that even the newly introduced indications for caesarean section probably do not include factors primarily associated with T1DM.
The present study, as did other case–control studies (e.g. [2, 13]), failed to show a difference in the birth size of diabetic children compared to the non-diabetic population. However, the lack of association is not universal; in a large multicentre EURODIAB study , low birth weight and short birth length were protective, and a Norwegian-population-based record linkage study showed an increase in T1DM risk with higher birth weight .
Limitations of the retrospective approach are seen in the analysis of neonatal jaundice. Although we had collected data on the occurrence of neonatal jaundice, most parents cannot recollect whether phototherapy was instituted or not. The same is true about vitamin D supplementation: most parents remember having given this supplementation but only 40% can recollect for how long it was given.
Day-care attendance and other proxy measures of exposure to infectious agents
The concept of the hygiene hypothesis suggests that an early encounter of a young immune system with common infectious agents may protect against the development of autoimmunity and atopy. The early exposure to infections may be indicated by several proxy measures. One is day-care attendance, which has been reported to be negatively associated with T1DM (reviewed in ). In our study, the effect of none or very short day-care attendance held borderline significance—its epidemiological relevance seems low, given that only few children attend day‐care for less than a year and that there is no decreasing trend in T1DM risk with longer day-care attendance.
The size of the community in which the family lived is another proxy measure of exposure to infectious agents. This parameter reflects the population density, and also, to a certain extent, the differences in socioeconomic status between urban and rural areas. In previous studies, a higher population density was associated with lower diabetes risk [11, 15]. Further, we investigated whether the families kept a domestic pet animal, since we hypothesized that the exposure to infections and allergens may also come from pets in the household, independently of the child's social mixing. Conceivably, the size of the community is strongly negatively correlated with the proportion of households owning a domestic pet—in control patients, a domestic pet was present in 84% of households in villages under 1,000 inhabitants, but only 42% households in cities with 100,000 or more inhabitants. Negative association between contact with pets and childhood diabetes has been recently found in a study from UK , but not in other studies [14, 16]. In the present work, neither the presence of a domestic pet, nor the size of the community was associated with the risk of T1DM. This may, amongst other factors, be attributed to the shallow socioeconomic stratification in Czechoslovakia within the period when the children were born.
The present study is among one of the largest European case–control studies on the T1DM-related environmental factors, and it is the first report on this topic from our population. Although a prospective observation on genetically preselected individuals (as is currently being used by several studies) is obviously superior to our approach, retrospective case–control design is the only way to investigate the T1DM associations within certain time periods of interest. In our case, it was a rapid rise in T1DM incidence in a genetically stable population. Swift changes in exposure to some of the candidate factors are rather easy to see; over the period in which the studied children were born, the proportion of caesarean sections doubled, the length of day-care attendance shortened and the duration of breast-feeding lengthened. However, as also seen from our data, these changes cannot explain the rise of T1DM incidence in our population—there are undoubtedly other reasons, as yet unrecognised.
The present study shows a negative association of T1DM with breast-feeding, a weak positive association with none or short attendance to day-care and a lack of association with other factors, namely, the introduction of supplementary feeding and selected perinatal parameters.
The members of the Czech Childhood Diabetes Registry Study Group are acknowledged for an excellent collaboration and we also thank the participating families for their continuing support. We thank especially: S. Kolouskova, M. Snajderova, L. Brazdova, I. Titzova, M. Sykora, M. Svojsik, M. Machalkova, J. Bartosova, L. Tyce, H. Safarikova, J. Maly, P. Skala, V. Janstova, J. Kopriva, R. Machytka, E. Novakova, V. Kratochvil, A. Martiskova, L. Kocinova, P. Mikyska, H. Nemcova, M. Hromadkova, H. Strmenova, J. Soucek, P. Eichl, K. Fiklik, A. Benesova, A. Filakova, E. Hladakova, Z. Jezova, S. Pokorny, B. Cervickova, I. Velesikova, H. Vavrova, J. Fromel, M. Honkova, V. Prokop, J. Havlicek, Y. Ptoszkova, J. Klabochova, R. Zemanova, J. Berankova, J. Gayer, L. Zlamalova, J. Brozova, M. Imramovska, S. Hyckova, J. Skvor, M. Gregora, E. Senkyrova, V. Jamelska, P. Mokros, J. Matejkova, J. Bazant and M. Schubert. The study was supported by the Ministry of Education of the Czech Republic (grant MSM 0021620814) and a grant from the Czech Diabetes Society.