S1-5 Tables (supplementary results) 3. S1-2 Figures (supplementary results).. 8. Details of the literature review... 10

Transcription

1 Infant feeding and risk of type 1 diabetes in two large Scandinavian birth cohorts Authors: Nicolai A. Lund-Blix, Stine Dydensborg Sander, Ketil Størdal, Anne-Marie Nybo Andersen, Kjersti S. Rønningen, Geir Joner, Torild Skrivarhaug, Pål R. Njølstad, Steffen Husby, Lars C. Stene Content: Supplementary Materials 2 S1-5 Tables (supplementary results) 3 S1-2 Figures (supplementary results).. 8 Details of the literature review S6-7 Tables (Overview of previous studies).. 11 Supplementary Material References

2 Supplementary Materials Registries The primary data on type 1 diabetes in the Norwegian Mother and Child Cohort Study (MoBa) was obtained from the Norwegian Childhood Diabetes Registry and secondarily the Norwegian Patient Registry. The Norwegian Childhood Diabetes Registry includes all new cases of childhood-onset diabetes, reported from all pediatric departments in Norway. Cases are included in the Norwegian Childhood Diabetes Registry based on a clinical diagnosis of type 1 diabetes, using the first insulin injection as the date of diagnosis.(1) The few cases of type 2- or monogenic diabetes were excluded from the endpoint definition.(2) In addition, the Norwegian Patient Registry contains data from all Norwegian government owned hospitals and outpatient clinics in addition to private practices receiving governmental reimbursements. Individual level data are available from 2008 onwards. We identified and included four cases with at least two type 1 diabetes registrations in the Norwegian Patient Registry that were not registered in the Norwegian Childhood Diabetes Registry. The Danish National Birth Cohort (DNBC) was linked to the Danish Childhood Diabetes Registry. The Danish Childhood Diabetes Registry contains data on children diagnosed with type 1 diabetes from all over Denmark. Cases are included in the Danish Childhood Diabetes Registry based on a clinical diagnosis of type 1 diabetes, using the date of first insulin injection as the date of diagnosis.(3) Use of formula in the hospital. Norway and Denmark comply with the WHO / UNICEF Baby Friendly Hospitals Initiative. However, if breastfeeding is contraindicated or impossible, infants may receive formula or banked human milk after maternal consent. In MoBa, use of formula in hospitals defined the end of full breastfeeding and age of introduction of cow s milk based formula. In DNBC, because of an imprecision in the answer alternatives for hospital feeding, we could not differentiate between formula and other types of feeding given at the neonatal ward. This was reported in a total of 0.36% of children, and we decided for simplicity to ignore this minor misclassification because it is too rare to influence any of the results. Finally, the use of hypoallergenic (hydrolysed) formula is rarely used in Norway and Denmark (reported used, only in the neonatal ward, for 2.3% of DNBC babies). It is therefore unlikely that differentiating hypoallergenic and conventional formula would influence our results and we were unable to meaningfully study the association of hypoallergenic formula with risk of type 1 diabetes. Multiple imputation of missing data As a sensitivity analysis, we imputed missing data using multiple imputation with chained equations.(4-6) The analyses were done in Stata, version 14 as described below. In the Norwegian Mother and Child Cohort Study (MoBa), information obtained on duration of any breastfeeding at age 6 months were used to impute data at 18 months of age, which was missing in 17.9% who were lost to follow-up between age 6 and 18 months. Similarly, missing data for the main covariates were imputed (between 0.02 and 3.7% had missing values for the covariates used for adjustment). After imputation, 88,870 participants in MoBa were included in the analysis. In the Danish National Birth Cohort (DNBC), 20.4% of participants in the 6 months interview were lost to follow-up and did not provide infant feeding data at 18 months. Data on adjustment covariates was missing for 0.008% to 27.0%, and these were similarly imputed. We decided not to impute missing data for full breastfeeding, because this was due to a change in questionnaire used in the interviews during the recruitment period, and was missing in a large proportion of participants merely because participants were not asked to provide the relevant information. Multiple imputation of full breastfeeding in MoBa showed very similar results (data not shown). For the imputation of any breastfeeding duration, we used the pmm option (predictive mean matching) with the following additional predictor variables: type 1 diabetes during follow-up, maternal age at delivery, delivery by caesarean section, birth weight, sex, parental type 1 diabetes, parity (complete data available). Additional covariates used in the prediction with a limited amount of missing data were also imputed: gestational age at delivery, maternal smoking during pregnancy, and maternal body mass index before pregnancy (covariates were imputed with linear or logistic models, as appropriate). In both cohorts (MoBa and DNBC), we imputed m=20 data sets. In the multiple imputation analysis, the variance of the resulting estimates takes the uncertainty in the imputation procedure into account. The estimated hazard ratios with 95% confidence intervals resulting from the multiple imputation analysis are shown in Figure S2.

3 Supplementary Table S1. Characteristics among included and excluded participants in the Norwegian Mother and Child Cohort Study (MoBa) and in the Danish National Birth Cohort (DNBC)*. * Percentage (%) is column percent among those with non-missing data for each variable MoBa reported type 1 diabetes in both mothers and fathers. DNBC reported maternal diabetes only (diabetes of any type). DNBC did not have detailed data on maternal education level after high school

4 Supplementary Table S2. Characteristics among those who were breastfed and never breastfed in the Norwegian Mother and Child Cohort Study (MoBa) and in the Danish National Birth Cohort (DNBC)*. * Percentage (%) is column percent among those with non-missing data for each variable MoBa reported type 1 diabetes in both mothers and fathers. DNBC reported maternal diabetes only (diabetes of any type). DNBC did not have detailed data on maternal education level after high school

5 Supplementary Table S3. Breastfeeding exposure characteristics and risk of type 1 diabetes (T1D) in the Norwegian Mother and Child Cohort Study (MoBa) and the Danish National Birth Cohort (DNBC). The main results of this table are displayed in Figure 2. PY: Person-years; HR: Hazard ratio.

6 * Adjusted for gender, birth weight, gestational age, caesarean section, parity, maternal smoking in pregnancy, maternal age, maternal BMI, and parental diabetes. Any breastfeeding duration longer than 18 months was set to 18.5 months in trend analysis using continuous months. Full breastfeeding duration longer than 6 months was set to 6.5 months. There was no significant heterogeneity between the studies for any of the comparisons (all p-values >0.10). Post hoc analysis because of apparent non-linearity (threshold effect) for those not breastfed at all. Because this was a small group, estimates are essentially unchanged. To obtain model based HR estimates and confidence intervals for differences per x months, compute exp(ln(estimate)*x).

7 Supplementary Table S4. Sensitivity analysis: excluding those with parental diabetes: Breastfeeding exposure characteristics and risk of type 1 diabetes after excluding those with parental diabetes. HR: Hazard ratio. MoBa: The Norwegian Mother and Child Cohort Study (MoBa). DNBC: The Danish National Birth Cohort. * Adjusted for gender, birth weight, gestational age, caesarean section, parity, maternal smoking in pregnancy, maternal age, maternal BMI, and parental diabetes. Adjusted for all the above covariates, except parental diabetes which was excluded from this analysis. Parental diabetes: maternal or paternal type 1 diabetes in MoBa; maternal diabetes of any type in DNBC. These are the same estimates as in S3 Table, shown here for comparison.

8 Supplementary Table S5. Sensitivity analysis of the impact of adjusting for parental native language on the association between duration of any breastfeeding and risk of type 1 diabetes (T1D) in the Norwegian Mother and Child Cohort Study (MoBa). PY: Person-years; HR: Hazard ratio. * Adjusted for gender, birth weight, gestational age, caesarean section, parity, maternal smoking in pregnancy, maternal age, maternal BMI, and parental diabetes. Additionally adjusted for parental native language. For 90.4 percent of included children, both parents reported Norwegian as their native language, 8.2 percent reported another native language for either the mother or the father, and 1.4 percent reported another native language for both parents. Any breastfeeding duration longer than 18 months was set to 18.5 months in trend analysis using continuous months. P-trend = 0.97

9 Supplementary Figure S1. Risk of type 1 diabetes among participants who were included in the main analysis versus those who were excluded because of incomplete information on infant feeding, in a. The Norwegian Mother and Child Cohort Study (MoBa) and b. The Danish National Birth Cohort (DNBC). The hazard ratio for type 1 diabetes comparing those with and without complete infant diet information was 0.89 (95 % CI ) in MoBa and 0.99 (95 % CI ) in DNBC. a

10 b

11 Supplementary Figure S2. Results after multiple imputation of missing data. a: Comparison of complete case analysis versus multiple imputation of duration of any breastfeeding and covariates in the Norwegian Mother and Child Cohort Study (MoBa), and b: the Danish National Birth Cohort (DNBC). c: Pooled analysis of MoBa and DNBC after multiple imputation of missing data in both studies. T1D: Type 1 diabetes. HR: Hazard ratio. All HRs were adjusted for the following covariates: parental type 1 diabetes, gender, mode of delivery, birth weight, gestational age at delivery, and the mother s parity, age at delivery, education, smoking in pregnancy, and pre-pregnancy body mass index. Details of the methods used for multiple imputation are described in the supplemental methods section above.

12 Details of the literature review Search strategy: The latest meta-analysis on the field included studies up to April 31, 2011,(7) and was in turn based on a meta-analysis conducted in 1996,(8) we searched PubMed for subsequent publications. PubMed August 28, 2015: ((((iddm[mesh Terms] OR insulin dependent diabetes mellitus[mesh Terms] OR type 1 diabetes)) AND ((((breastfeeding[mesh Terms]) OR infant nutrition[mesh Terms]) OR early nutrition[mesh Terms]) OR infant feeding)) We searched for studies with type 1 diabetes as the outcome and breastfeeding/infant feeding as the exposure variable. The search was based on a previous pooled analysis of observational studies conducted in 2012 (including studies published up to May, 2011),(7) with three cohort studies included. The search resulted in 51 studies (340 studies when no date limitations were added to the search). Studies were screened for relevant information and identification of prospective cohort studies. We further screened reference lists in the identified articles as well as the author s personal data bases based on informal searches. We identified three additional recently published cohort studies in the search, as well as one older cohort study identified from references lists or personal databases. The latter study (Virtanen et al. 1998) was a cohort of siblings of patients with type 1 diabetes, but had a very long recall period for infant feeding (see comment in S6 and S7 Tables). Only the most recent publication from each cohort is included in the table in cases where similar analyses of the same cohort have been published in several papers (S6 Table). Selected studies with islet autoimmunity as the endpoint are in addition briefly summarized in a separate table (S7 Table), although our a priori decision was that such results should not be emphasized in the discussion of our results (or introduction) of the main text of our paper.

13 Supplementary Table S6. Overview of previously published cohort studies of infant feeding and type 1 diabetes (T1D). The most recent publication from each cohort is included. FDR: First degree relative.

14 Supplementary Table S7. Overview of selected cohort studies of infant feeding and islet autoimmunity as the endpoint (not type 1 diabetes). The most recent publication from each cohort is included* * IA: Islet autoimmunity. AAb: Autoantibodies. IAA: Insulin Autoantibodies. GADA: Glutamic Acid Decarboxylase (GAD) Autoantibodies. IA-2A: IA-2 Autoantibodies. ZnT8A: Zinc transporter 8 Autoantibodies. ICA: Islet cell autoantibodies. Quest: Questionnaire. FDR: First degree relative. T1D: Type 1 diabetes. To avoid false positive islet autoantibodies, and to assure a strong relation with future type 1 diabetes (useful surrogate endpoint), islet autoantibodies of at least two types should be repeatedly positive and cut-offs for positivity should be greater than the 99th percentile of normal controls and validated in external quality assurance schemes.(20-22) Some studies presented results for two or more alternative definitions such as >=1 AAb and >=2 AAb. T1D: Number with islet autoimmunity reported to have progressed to T1D.

15 Supplementary Material References 1. Skrivarhaug T, Stene LC, Drivvoll AK, Strøm H, Joner G. Incidence of type 1 diabetes in Norway among children aged 0-14 years between 1989 and 2012: has the incidence stopped rising? Results from the Norwegian Childhood Diabetes Registry. Diabetologia. 2014;57(1): Irgens HU, Molnes J, Johansson BB, Ringdal M, Skrivarhaug T, Undlien DE, et al. Prevalence of monogenic diabetes in the population-based Norwegian Childhood Diabetes Registry. Diabetologia. 2013;56(7): Thorsen SU, Eising S, Mortensen HB, Skogstrand K, Pociot F, Johannesen J, et al. Systemic levels of CCL2, CCL3, CCL4 and CXCL8 differ according to age, time period and season among children newly diagnosed with type 1 diabetes and their healthy siblings. Scandinavian journal of immunology. 2014;80(6): White IR, Royston P, Wood AM. Multiple imputation using chained equations: Issues and guidance for practice. Stat Med. 2011;30(4): Sterne JA, White IR, Carlin JB, Spratt M, Royston P, Kenward MG, et al. Multiple imputation for missing data in epidemiological and clinical research: potential and pitfalls. BMJ. 2009;338:b Cummings P. Missing data and multiple imputation. JAMA Pediatr. 2013;167(7): Cardwell CR, Stene LC, Ludvigsson J, Rosenbauer J, Cinek O, Svensson J, et al. Breast-feeding and childhood-onset type 1 diabetes: a pooled analysis of individual participant data from 43 observational studies. Diabetes care. 2012;35(11): Norris JM, Scott FW. A meta-analysis of infant diet and insulin-dependent diabetes mellitus: do biases play a role? Epidemiology (Cambridge, Mass). 1996;7(1): Lund-Blix NA, Stene LC, Rasmussen T, Torjesen PA, Andersen LF, Rønningen KS. Infant Feeding in Relation to Islet Autoimmunity and Type 1 Diabetes in Genetically Susceptible Children: The MIDIA Study. Diabetes care. 2015;38(2): Chmiel R, Beyerlein A, Knopff A, Hummel S, Ziegler AG, Winkler C. Early infant feeding and risk of developing islet autoimmunity and type 1 diabetes. Acta diabetologica Frederiksen B, Kroehl M, Lamb MM, Seifert J, Barriga K, Eisenbarth GS, et al. Infant exposures and development of type 1 diabetes mellitus: The Diabetes Autoimmunity Study in the Young (DAISY). JAMA pediatrics. 2013;167(9): Virtanen SM, Hyppönen E, Läärä E, Vähäsalo P, Kulmala P, Savola K, et al. Cow's milk consumption, disease-associated autoantibodies and type 1 diabetes mellitus: a follow up study in siblings of diabetes children. Diabet Med. 1998;15: Ponsonby AL, Pezic A, Cochrane J, Cameron FJ, Pascoe M, Kemp A, et al. Infant anthropometry, early life infection, and subsequent risk of type 1 diabetes mellitus: a prospective birth cohort study. Pediatric diabetes. 2011;12(4 Pt 1): Savilahti E, Saarinen KM. Early infant feeding and type 1 diabetes. Eur J Nutr. 2009;48(4): Ievins R, Roberts SE, Goldacre MJ. Perinatal factors associated with subsequent diabetes mellitus in the child: record linkage study. Diabetic medicine : a journal of the British Diabetic Association. 2007;24(6): Norris JM, Barriga K, Klingensmith G, Hoffman M, Eisenbarth GS, Erlich HA, et al. Timing of initial cereal exposure in infancy and risk of islet autoimmunity. JAMA. 2003;290(13): Virtanen SM, Takkinen HM, Nevalainen J, Kronberg-Kippila C, Salmenhaara M, Uusitalo L, et al. Early introduction of root vegetables in infancy associated with advanced -cell autoimmunity in young children with human leukocyte antigen-conferred susceptibility to Type 1 diabetes. Diabet Med. 2011;28(8): Couper JJ, Beresford S, Hirte C, Baghurst PA, Pollard A, Tait BD, et al. Weight gain in early life predicts risk of islet autoimmunity in children with a first-degree relative with type 1 diabetes. Diabetes Care. 2009;32(1): Holmberg H, Wahlberg J, Vaarala O, Ludvigsson J. Short duration of breast-feeding as a risk-factor for -cell autoantibodies in 5-year-old children from the general population. Br J Nutr. 2007;97(1): Barker JM, Barriga KJ, Yu L, Miao D, Erlich HA, Norris JM, et al. Prediction of autoantibody positivity and progression to type 1 diabetes: Diabetes Autoimmunity Study in the Young (DAISY). J Clin Endocrinol Metab. 2004;89(8): Ziegler AG, Rewers M, Simell O, Simell T, Lempainen J, Steck A, et al. Seroconversion to multiple islet autoantibodies and risk of progression to diabetes in children. JAMA. 2013;309(23): Steck AK, Vehik K, Bonifacio E, Lernmark A, Ziegler AG, Hagopian WA, et al. Predictors of progression from the appearance of islet autoantibodies to early childhood diabetes: The Environmental Determinants of Diabetes in the Young (TEDDY). Diabetes Care. 2015;38(5):