Study population

All live-born children in Denmark from 1 January 1997 through 31 December 2010 (n = 912,797) were identified in the Medical Birth Registry. We excluded 38,218 children from multiple pregnancies (n = 37,895) and pregnancies with errors in the personal identification number (n = 323). Furthermore, we excluded 16,378 children with events before their two years birthday due to either death (n = 3,412), emigration (n = 12,790) or diagnosis of type 1 diabetes (n = 176). The final population included 858,201 live-born singleton children born to 527,927 mothers.

The children were followed from age two until their fifteenth birthday (n = 60,934), the date of type 1 diabetes diagnosis (n = 1,503), death (n = 618), first emigration (n = 21,787) or end of follow-up by December 2012 (n = 773,359), whichever came first (Fig 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Inclusion, exclusion and data censoring.

Shown are the number of children who met the exclusion criteria, the number of events and the distribution of antibiotics redemption and diagnosis of childhood type 1 diabetes.

https://doi.org/10.1371/journal.pone.0161654.g001

Exposure

Outpatient redemptions of antibiotic prescriptions for the child during the first two years of life were classified into either: any type of antibiotics (yes or no), narrow-spectrum antibiotics (yes or no) or broad-spectrum antibiotics (yes or no), classified in accordance with the Danish Integrated Antimicrobial Monitoring and Research Program 2013 (S2 Table).[21] Consequently, children exposed to both narrow-and broad spectrum antibiotics, would be classified as exposed in all analyses evaluating effects of any type, narrow-spectrum or broad-spectrum antibiotics, respectively, whereas children treated with only broad-spectrum antibiotics, would be included in the exposed group in analyses evaluating effects of any type and broad-spectrum antibiotics but in the un-exposed group, when evaluating the effect of narrow-spectrum antibiotics. The classification into the three different exposure groups were used in separate analyses.

Outcome variable

We followed children during 13 years from age two to fourteen, both inclusive, contributing 5,906,069 person-years at risk and an average duration of follow-up of 6.9 years. We defined childhood type 1 diabetes in children who before the age of fifteen fulfilled one of two of the following criteria or both. Either they were discharged from a hospital with a diabetes diagnosis (type 1 or unspecified) or they had redeemed at least two prescriptions on insulin/insulin analogues within one year. Children who additionally redeemed two prescriptions on oral anti-diabetics within one year were considered having type 2 diabetes. The date of onset of type 1 diabetes was defined as the first of either the date of the first diabetes-related hospital admission or the date for the second redemption of insulin/insulin analogues (S1 Table).

Confounders and other covariates

We based selection of possible confounders on knowledge from previous studies on risk factors for type 1 diabetes, as well as on theoretical considerations regarding factors known to be associated with both the use of antibiotics and risk of type 1 diabetes.[22] Confounders comprised birth year (1997–2010), sex (male or female), parity (primiparous or multiparous) and mode of delivery (vaginal delivery, intrapartum cesarean section or prelabor cesarean section) (S1 Table).

Included parental confounders were genetic predisposition measured as maternal or paternal diabetes status at childbirth, as well as maternal and paternal educational level and age (<25, 25–30 or >30 years) at childbirth. Maternal and paternal type 1 diabetes was defined in parents discharged since 1977 with a diabetes diagnosis (type 1 or unspecific). Parents with two prescriptions on oral antidiabetics within one year were interpreted as having type 2 diabetes. The onset date of type 1 diabetes was defined as the date for the first diabetes-related hospital admission. The variable was classified into two categories (type 1 diabetes diagnosed before childbirth: yes or no). Education was recorded as the highest level of ongoing or finished education and classified into three categories (elementary school/high school, short education/skilled worker or medium/long education). Indicators (yes or no) on maternal redemption of antibiotics during pregnancy (Trimester 1, 2 or 3) and during the first six months after delivery were derived for any, narrow-spectrum or broad-spectrum antibiotics.

Birth weight (<2,500, 2,500–4,000 or >4,000 grams) and gestational age (<34, 34–36, 37–40 or >40 weeks gestation) were available.

Data analysis

We estimated childhood type 1 diabetes hazard ratios (HRs) for redemption of antibiotics using a Cox proportional hazards model, using the survival package (version 2.37–7) in R (version 3.0.2). The proportional hazards assumption of the Cox regression model implies that the association with antibiotics does not change with age. To examine the validity of this assumption, we fitted models where we allowed the association to change in three-year intervals. These models validated that the proportional hazards assumption was met.

Robust (sandwich) standard errors were used to adjust for multiple births from the same woman. To examine the impact of different confounders we conducted partially adjusted models as well as fully adjusted models including all confounders (year of birth, mode of delivery, maternal- and paternal type 1 diabetes status at childbirth, maternal- and paternal age at childbirth, maternal- and paternal educational level at childbirth, offspring sex, parity). Missing covariates were handled by complete case analysis.

Additional models evaluated the potential impact of gestational age and birth weight. We explored the potential interaction between antibiotic redemption and delivery mode, parental type 1 diabetes status at childbirth, parity and maternal redemption of antibiotics during pregnancy and during the first 6 months after delivery by adding an interaction term to the model. To illustrate the effect of mode of delivery on the number of children, who on average should be exposed to broad-spectrum antibiotics during the first two years of life to cause childhood type 1 diabetes in one child (diagnosed from two to fifteen years of age), we calculated the number needed to harm for different subgroups of mode of delivery (all deliveries, intrapartum cesarean section and prelabour cesarean section, respectively).[23]

Finally, potential effects of specific categories of antibiotics were evaluated in analyses of children exposed to the following antibiotic sub-categories based on ATC-codes: J01CA (Extended-spectrum penicillin), J01CR (Combined penicillins, including beta-lactamase inhibitors), J01CE (Beta-lactamase sensitive penicillins), and J01FA (Macrolides).

Other studies

Associations between use of antibiotics and subsequent development of childhood type 1 diabetes have been evaluated in three other human observational studies. A case-control study from Finland including 437 children diagnosed with type 1 diabetes and 1,748 matched controls, found that overall, there was no association between antibiotic use before pregnancy, during pregnancy or during childhood and the risk of childhood type 1 diabetes. However, when evaluating effects of subgroups of antibiotics as well as several potential sub-comparisons combining antibiotic exposure before, during and after pregnancy, they found an increased risk of childhood type 1 diabetes in children from mother-child pairs, where macrolides were used both by the mother before pregnancy and by the child compared to pairs where neither used macrolides (OR 1.76, 95% CI 1.05–2.94). Furthermore, children born to mothers treated with phenoxymethyl penicilins (OR 1.70, 95% CI 1.08 to 2.68) or quinolone antimicrobials (OR 2.43, 95% CI 1.16 to 5.10) one year prior to pregnancy had increased risk of type 1 diabetes. The use of more than seven purchases of antibiotics during childhood was associated with increased risk of type 1 diabetes in the child (OR 1.66, 95% CI 1.24 to 2.24), but no dose-response effect was observed in the antimicrobial sub-groups. The study did not account or correct for the number of included comparisons. [11] A Dutch case-control study including 925 individuals with type 1 diabetes and 3,591 controls found that patients in the type 1 diabetes group received more antibacterials than controls (49.8 vs. 40.0%, P<0.001), and that the number of anti-infective prescriptions were higher in the type 1 diabetes group from eight years before until four years after the diagnosis.[10] A previous Danish cohort study including all children born in Denmark 1995–2003, hereof 454 children diagnosed with type 1 diabetes, found no statistically significant association between overall use of antibiotics during childhood and risk of type 1 diabetes (rate ratio 1.16, 95% CI 0,91 to 1,50). No specific class of antibiotics (extended-spectrum penicillin, beta-lactamase sensitive penicillins, macrolides or other systemic antibiotics) was associated with type 1 diabetes, no specific age of use was associated with type 1 diabetes and no specific age of onset was associated with antibiotic use. Hospitalization with serious bacterial infections was not associated with increased risk of type 1 diabetes. The study did not explore potential effects of mode of delivery.[9]

Strengths and limitations

The Danish registers provide unique data as they are nationwide, almost complete and with high validity.[16–20] The prospectively collected data limits the risk of selection and information bias. It is a strength of the study that antibiotics are only sold on prescription and redemptions are automatically transferred to the Register of Medicinal Product Statistics.[19]

No other studies have explored the potentially modifying effect of delivery mode, parity and genetic predisposition or focused on the increasing use of broad-spectrum antibiotics and the present study is the largest with the longest follow-up time. Despite the rising incidence childhood type 1 diabetes is still a rare disease, but it is associated with an increased morbidity.[24] Studies have shown that up to 50% of all antibiotic prescriptions could have been omitted,[25] and that the cesarean section rates are much higher than medically indicated in many countries.[26] Therefore a number needed to harm of around 500 in children delivered by cesarean section is of clinical importance.

Antibiotics can be categorized as either narrow- or broad-spectrum antibiotics according to the spectrum of activity, but there is no universal algorithm for this categorization,[27] and the categorization is complicated as the spectrum of activity changes due to changes in the resistance patterns over time. Traditionally, bacteria are categorized using the Gram stain. Antibiotics only active either against Gram-positive or Gram-negative bacteria have a narrow spectrum whilst antibiotics active against both groups have a broad spectrum.[21;28] We chose the Danish antibiotics categorization which is well defined and founded on considerations of the national resistance patterns,[21] but other categorizations have been suggested.[29–31] We found an increased risk of childhood type 1 diabetes in children exposed to extended-spectrum penicillins and macrolides, of which the first is considered broad-spectrum following the DANMAP categorization but narrow-spectrum following other categorizations and the latter the opposite way around. This finding underlines the potential association between antibiotic treatment during childhood and later childhood type 1 diabetes, but it also stresses the challenges associated with the categorization into broad-and narrow-spectrum antibiotics.[21;29–31] We focused the analyses on the potential association between treatment with antibiotics within the first two years of life, after which the microbiota has been found to be indistinguishable from the adult microbiota.[2] We furthermore focused on broad-spectrum antibiotics, well-knowing that 31% of the children had redeemed both narrow- and broad-spectrum antibiotics. No dose-response effects were found.

Unfortunately, we did not have access to information on the underlying infection or antibiotic treatment during hospitalization. However, missing information regarding patient compliance and antibiotic treatment during hospitalization dilute the effect and underestimate the true effect of antibiotics, and a previous Danish study including information regarding hospitalization with serious bacterial infection did not find an effect of this exposure on type 1 diabetes risk.[9] It is a further limitation, that we did not have access to information regarding ethnic origin, infant diet or breastfeeding.

The predictive validity of type 1 diabetes diagnoses based on discharge codes from the National Patient Registry or on redeemed insulin prescriptions was in 1996 estimated to be 96% in the general population and in 2007 to be 94% to 97% in children.[32;33] There are several potential pathogenic explanations to the findings of an association between redemption of broad-spectrum antibiotics within the first two years of life and increased risk of childhood type 1 diabetes only in children delivered by cesarean section.

Firstly, it may be a causal association with broad-spectrum antibiotics. The pathogenesis of type 1 diabetes is supposed to be multifactorial, with a strong genetic predisposition influenced by several intrauterine, perinatal and postnatal triggers. Children delivery by cesarean section has a less diverse and more pathogenic microbiota,[34] which combined with a prolonged recovery phase following treatment with broad-spectrum antibiotics may result in eradication of harmless and beneficial types of bacteria and overgrowth of more pathogenic bacteria, introducing an imbalance in the microbiota and disturbance of the immunological maturation. This “hygiene theory” has been proposed to be a potential etiology for increased autoimmune activity.[1;2]

Secondly, it has been proposed that delivery by cesarean section may in itself weaken the baby´s immune system through a reduced stress-response. Combined with a more pathogenic gut-microbiota caused by broad-spectrum antibiotics this may induce autoimmune activity or accelerate autoimmune processes.[2;35]

Thirdly, it may not be a causal association with broad-spectrum antibiotics but rather an association with the underlying infection which may trigger autoimmunity. Studies show inconsistent results.[1;36;37] The findings of the association only among children delivered by cesarean section indicate a “two-hit” pathogenesis, that these children are vulnerable to harmful effects of either infections or side-effects from broad-spectrum antibiotics.

Finally, the association may reflect that patients with diabetes have more bacterial infections and therefore redemption of broad-spectrum antibiotics before onset of diabetes could act as a proxy for sub-clinical diabetes.[38] However, then it would be expected that similar associations were found irrespective of mode of delivery.

We found a comparable association with broad-spectrum antibiotics in children delivered by prelabor and intrapartum section. This could indicate that the coding of cesarean section is invalid, that the microbiota is equally affected from intrapartum and prelabor cesarean section or that the interaction with cesarean section is related to shared factors like anesthesia, operative procedure, a reduced fetal stress response or antibiotics given in relation to the operation. Additionally, intrapartum antibiotics would have been given to a proportion of women having intrapartum cesarean section, which could in theory disturb the gut microbiota of these children more than the disturbance attributed to cesarean section per se. Unfortunately, we did not have access to information regarding intrapartum antibiotic treatment.

The finding that paternal and maternal type 1 diabetes are strong predictors of type 1 diabetes is known from other studies. We found, that children of mothers with a short education had a higher risk of type 1 diabetes. Other studies have found diverging associations between maternal educational level and the risk of type 1 diabetes in the children. The finding of a higher risk of type 1 diabetes in firstborn children may be supportive of the “hygiene hypothesis” as they would not have been exposed to microbes from siblings.