Gluten intake and risk of celiac disease: Long-term follow-up of an at-risk birth cohort

Abstract

Objective:

To determine the association between the amount of gluten intake in childhood and later celiac disease (CD), for which data are currently scarce.

Methods:

The prospective DAISY cohort includes 1875 at-risk children with annual estimates of gluten intake (grams/day) from age one year. From 1993 through January 2017, 161 children were, via repeated tissue transglutaminase (tTGA) screening, identified with CD autoimmunity (CDA, persistent tTGA positivity); of these, 85 children fulfilled CD criteria of biopsy-verified histopathology or persistently high tTGA levels. Cox regression, modeling gluten intake between ages one and two years (i.e., in one-year-olds), and joint modeling of cumulative gluten intake throughout childhood, were used to estimate hazard ratios adjusted for confounders (aHR).

Results:

Children in the highest third of gluten intake between the ages of one and two years had a 2-fold greater hazard of CDA (aHR 2.17; 95%CI, 1.22–3.88; P-value=0.01) and CD (aHR 1.96; 95%CI, 0.90–4.24; P-value=0.09) than those in the lowest third. The risk of developing CDA increased by 5% per daily gram increase in gluten intake (aHR 1.05; 95%CI, 1.00–1.09; P-value=0.04) in one-year-olds. The association between gluten intake in one-year-olds and later CDA or CD did not differ by the child’s HLA genotype. The incidence of CD increased with increased cumulative gluten intake throughout childhood (e.g., aHR 1.15 per standard deviation increase in cumulative gluten intake at age six; 95%CI, 1.00–1.32; P-value=0.04).

Conclusions:

Gluten intake in one-year-olds is associated with the future onset of CDA and CD in children at risk for the disease.

INTRODUCTION

Celiac disease (CD) is an autoimmune disorder present in 1–3% of the U.S. population. (1, 2) Over the life-course, CD causes reduced life-expectancy and significant social and economic costs to individuals and society at large.(3, 4) In CD, gluten intake fuels a T-cell mediated small-intestinal inflammation and flattening of the gut mucosa. The etiologies of CD are likely to involve a multifactorial interplay between genetic and environmental factors.(5)

Since the 1950s, gluten has been known as the driver of the autoimmune process of CD. Even so, few studies have examined whether the amount of gluten intake predicts later CD. Two studies have suggested that Swedish children with a high gluten intake by age two years are at an increased risk of CD.(6, 7) Other European studies have not found an association between gluten intake in early life and the development of CD.(8, 9) To our knowledge, there are no similar examinations from the U.S. Given the short follow-up in previous studies (average <5 years) it is also unknown if gluten intake in later childhood influences the risk of CD development. Finally, no previous study has tested whether or not gluten intake relates to the onset of CD autoimmunity (CDA), a preclinical manifestation of the disease.

Using prospectively collected, long-term follow-up data from a U.S. birth cohort we tested the hypothesis that the amount of gluten intake in early childhood as well as throughout childhood increases the risk of later developing CD and CDA.

METHODS

The Diabetes Autoimmunity Study in the Young (DAISY), conducted in Denver, CO, follows children genetically at risk for type 1 diabetes and CD. Children were screened to detect CD and CDA. Gluten intake from age one year throughout childhood was assessed via repeated food-frequency questionnaires (FFQ). This study has been approved by the Colorado Multiple Institutional Review Board. Parental consent was obtained for all participants. Details on the formation of the DAISY cohort and its follow-up have previously been reported;(10, 11) an overview of its design is shown in Supplementary Figure 1.

Study sample

Out of 2547 DAISY enrollees, 1875 participants had dietary data for the current analysis (Figure 1, flowchart). Children were recruited from two populations:

1. General-population newborns that through mass screening for CD-permissive human leukocyte antigen (HLA) genotypes (defined below) were identified to have an increased risk for the disease. Children born in 1993–2009 at the St. Joseph’s hospital in Denver, CO, and without major neonatal morbidities were eligible for screening; 31,766 newborns were screened (87% of eligible children) out of whom 1043 with CD-associated HLA genotypes were enrolled in this study.

2. Unaffected children with a first-degree relative with type 1 diabetes in the Denver metropolitan area. Since 1994, 832 siblings and offspring of type 1 diabetes patients were included in DAISY (median age at study entry two years). These children were recruited irrespective of their HLA genotype.

Figure 1.

Flowchart illustrating the formation of the study sample.

CDA, celiac disease autoimmunity; DAISY, Diabetes Autoimmunity Study in the Young; HLA, human leukocyte antigen.

Outcomes: Screening-detected celiac disease (CD) and CD autoimmunity

Participants were screened using CD-specific tissue transglutaminase autoantibodies (tTGA) at 9, 15 and 24 months of age, and yearly thereafter, as previously described.(12, 13) Once positive (tTGA index ≥0.05), the test was repeated after three to six months and participants with two positive tTGA tests were referred for workup of CD.

The primary outcome of this study was CDA as defined by two consecutive positive tTGA tests. Our secondary outcome was CD, which we defined as a positive tTGA test and evidence of small-intestinal inflammation or villous atrophy (Marsh(14) grade two or higher). Children who did not undergo biopsy, but who had repeated tTGA tests above an index of 0.5 (i.e. ten times the upper limit of normal), were also considered to have CD for the purposes of this study. A tTGA index above 0.5 has previously been shown to have a positive predictive value of 96% for biopsy-confirmed CD.(15) All participants with CD met criteria for CDA.

Exposure: Gluten intake in childhood

Semi-quantitative FFQs were administered annually to the parents starting when the child was two years of age. The questionnaire covered the child’s usual diet in the previous year. Therefore, the FFQ administered at age two years reflected the intake of one-year-olds. Beginning at age 10 years, the participants filled out the age-appropriate Youth/Adolescent Questionnaire covering usual dietary intake in the previous year. The pediatric FFQ has been validated in DAISY using biomarkers and 24-hour dietary recalls.(16, 17) The Youth/Adolescent Questionnaire has been validated against 24-hour dietary recalls and shown to be reproducible in this age group.(18, 19) However, the accuracy of gluten intake has not specifically been evaluated. Both the pediatric and youth/adolescent questionnaires assess the average intake of common food items, including composite dishes such as pizza, hamburgers and pasta dishes, cereals, bakery products, breads, crackers, cookies and candy. For each food item one portion was described and a range of consumption frequencies were listed; available responses ranged from never or less than once a month to six or more times a day. We have previously shown that data from the pediatric and the youth/adolescent questionnaires in DAISY can be used in the same analysis when the analysis includes an indicator variable of questionnaire type.(20) More information on the questionnaires is located at https://regepi.bwh.harvard.edu/health/.

Gluten intake (grams/day) was derived from the protein content of wheat-, barley- and rye-containing foods as defined by standardized food recipes and popular online U.S. recipe tools (details are provided in the Supplementary Material). We used a conversion factor of 0.75 to estimate the protein gluten content.(21, 22) As in previous research,(6, 23) the same conversion factor was used for wheat, barley and rye. Oats was not considered as a gluten-containing grain since the immunogenicity of oats strains are typically low and safe to eat for patients with CD.(24) We did not consider trace amounts of gluten from condiments or gluten cross-contamination during production and handling,(25) as the contribution of these sources to the total amount of gluten intake would be negligible.(25, 26) To reduce the impact of erroneously collected data, 62 FFQ records (0.4% of total FFQs) with an unrealistically high gluten intake (>35 grams/day) were excluded. All participants were on a gluten-containing diet at the time of the first dietary assessment at age two years. During the subsequent follow-up, 43 out of 1875 participants (2%) reported at least once to be on a gluten-free diet (GFD); two out of the 43 participants developed CDA/CD. In our analyses we coded these 43 participants to consume zero gram of gluten when they were on a GFD; in a sensitivity analysis we examined the impact of that approach by excluding these 43 participants from our data.

Genotyping

To account for CD-associated genetic susceptibility we genotyped and categorized individuals in one of the following groups based on HLA genotype: DQ2/DQ2; DQ2/DQ8; DQ2/X, DQ8/DQ8, DQ8/X, and X/X where X is neither DQ2 nor DQ8. Details on the genotyping procedure have been published.(11)

Other variables

We preselected adjustment variables of maternal and child characteristics that may be associated with gluten intake and CD development. Information on child’s sex, first-degree relatives with CD, parent-reported race-ethnicity and maternal age at the time of delivery were retrieved from enrollment questionnaires. We also included data on the age at gluten introduction, breastfeeding duration, regardless of whether it was exclusive or partial, type of FFQ and the child’s reported total energy intake.(27)

By study design, participants were screened for the onset of islet autoimmunity, a preclinical stage of type 1 diabetes. We considered islet autoimmunity a potential confounder since its appearance may affect study retention, be associated with CD and trigger dietary changes of the child. As described in previous literature,(28, 29) the timing of islet autoimmunity was defined by the appearance of at least one autoantibody against insulin, tyrosine phosphatase-like protein IA-2, zinc transporter 8 or glutamic acid decarboxylase twice or more in succession.

Statistical methods

We used SAS (V9.4, SAS Institute, NC) for the statistical analyses. We decided to examine the effect of gluten intake in one-year-olds (as this is the earliest time in which the amount of gluten intake was collected) as well as throughout childhood by using two statistical models:

1. Cox proportional-hazards models examining if the development of CDA and CD depends on gluten intake in one-year-olds. Gluten intake was treated as a fixed (time invariant) exposure. Motivated by the results of a previous study,(6) we decided in advance (a priori) to model gluten intake as a continuous variable (grams/day) and as categorical based on tertiles. We found no violation of the proportional-hazards assumption. For these analyses, hazard ratios (HRs) and 95% confidence intervals (CIs) were estimated accounting for right-censored data. The Cox model included random effects (frailties) treating siblings from the same family as clusters. Efron’s approximation was used to handle ties in failure times.

2. Joint longitudinal exposure-survival models examining whether the gluten intake throughout childhood predicted the risk of CDA and CD. Briefly, joint models account for potential correlation between the trajectory of longitudinal exposure (gluten) measurements and the time-to-event data (i.e., an interrelationship between gluten intake and CDA/CD development).(30) In this approach the longitudinal and the survival data are linked through shared random effects. The joint modeling approach efficiently accounts for measurement errors of repeated exposure assessments. In this analysis, we simultaneously considered all data on gluten intake (grams/day), treated as a time-varying exposure. The analysis did not assume a certain window of susceptibility for high gluten intake. We modeled gluten intake using a linear mixed-effects model, fitting a simple linear form of intake over time. A piecewise exponential distribution with six internal knots was found to be the best fit for our time-to-event submodel using AIC. The optimal number of internal knots for the baseline hazard function was determined by comparing the goodness of fit for models using different number of internal knots. The longitudinal and survival submodels were linked using a cumulative association structure.(31) This allowed estimation of the association between the tendency (high vs. low random subject effect) in gluten intake over time and the subsequent risk for developing CDA or CD. We also used a current-value association structure to estimate the risk of CDA and CD at observed event times associated with gluten intake at those times.

In both the Cox and joint modeling approaches, participants were followed until the appearance of CDA (or CD), defined by the timing of first tTGA-positive test, or else until the time of the last tTGA-negative measurement as of January 19, 2017.

Adjustment variables

Both the Cox and joint model included as covariates the child’s sex, family history of CD, parent-reported race-ethnicity, maternal age at the time of delivery, HLA genotype, breastfeeding duration, timing of infant gluten introduction, total energy intake and the timing of islet autoimmunity. In the Cox model, total energy intake and islet autoimmunity were treated as fixed covariates as of age two years. In the longitudinal joint model, total energy intake, timing of islet autoimmunity, as well as survey type were included as time-varying covariates.

Subanalysis

Motivated by the results of a previous study,(6) we tested whether any association found between gluten intake in one-year-olds and later development of CDA/CD differed by the child’s HLA genotype or family history of CD. Finally, our Cox and joint modeling analyses were rerun after excluding 43 participants who at least once during follow-up were on a GFD.

RESULTS

Out of 1875 children in the study, 902 (48%) were girls. Most of the children (n=1414 [76%]) were non-Hispanic white (Table 1). On January 19, 2017, over 22,443 person-years of follow-up, 161 participants had developed CDA; of these, 85 fulfilled CD-criteria. The incidence of CDA and CD was highest among those with the HLA-DQ2/2 genotype (Supplementary Tables 1–2). In total, we had 16,538 FFQ records with an average of nine records per child.

Table 1.

Characteristics of study participants

	All n=1875	CDA n=161	CD n=85
Female, n (%)	902 (48)	93 (58)	44 (52)
Age at end of follow-up (years), median (IQR)A	13 (6–17)	5 (4–8)	5 (4–7)
Year of birth, median (range)	1997 (1986–2009)	1998 (1986–2009)	1998 (1987–2007)
HLA genotype, n (%) B
DQ2/DQ2	128 (7)	26 (16)	16 (19)
DQ2/DQ8	397 (21)	49 (31)	29 (34)
DQ2/X	289 (15)	41 (26)	26 (31)
DQ8/DQ8	181 (10)	16 (10)	6 (7)
DQ8/X	601 (32)	24 (15)	6 (7)
X/X	277 (15)	4 (3)	2 (2)
Race-ethnic group, n (%)
Non-Hispanic white	1414 (76)	133 (83)	72 (86)
Other	453 (24)	25 (17)	12 (14)
Family history of CD, n (%)	82 (4)	25 (16)	19 (22)
Developed islet autoimmunity, n (%)	199 (11)	21 (13)	11 (13)
Maternal age at delivery (years), median (IQR)	30 (27–34)	32 (29–35)	33 (30–35)
Duration of breastfeeding (months), median (IQR)	6 (2–11)	6 (2–12)	6 (3–13)
Age at gluten introduction, n (%)
< 4 months	147 (8)	12 (8)	7 (8)
4–5.9 months	614 (34)	48 (30)	21 (25)
≥ 6 months	1056 (58)	99 (62)	56 (67)

^AAge at time of first positive tissue transglutaminase autoantibody (tTGA) test or age of censoring at last tTGA-negative measurement by January 19, 2017.

^BHuman leukocyte antigen (HLA) haplotype where X is neither DQ2 nor DQ8.

Two children lacked HLA data, eight had missing race-ethnicity data, 20 lacked data on maternal age at delivery, 36 lacked data on breastfeeding duration and 58 lacked data on age at gluten introduction. CD, celiac disease; CDA, celiac disease autoimmunity; IQR, interquartile range.

Early gluten intake and risk of later CDA and CD

At the first dietary assessment (reflecting the diet in the second year of life) the average daily gluten intake was 6.4 (standard deviation [SD], 1.7) grams among participants in the lowest third, 10.9 (1.2) grams in the middle third and 18.1 (4.2) grams for participants in the highest third.

Compared with one-year-olds in the lowest third of gluten intake, those in the highest third had a 2-fold increased risk of CDA (adjusted HR [aHR] 2.17; 95%CI, 1.22–3.88; P-value=0.01; Figure 2). Cumulative incidence curves for CDA according to thirds of gluten intake between the ages of one and two years are presented in Figure 3. Examining gluten intake as a continuous variable yielded an aHR of 1.05 (95%CI, 1.00–1.09; P-value=0.04) for each one gram increase in daily intake or an aHR of 1.20 (95%CI, 1.01–1.43; P-value=0.04) for the four grams of gluten provided in a slice of bread (60 grams).

Figure 2.

Hazard ratios for celiac disease autoimmunity (CDA) and celiac disease (CD) according to gluten intake between one and two years of age.

Diamonds and horizontal bars represent adjusted hazard ratios (HRs) and 95% confidence intervals (CIs), respectively. The adjusted model accounts for child’s sex, family history of CD, parent-reported race-ethnicity, maternal age at the time of delivery, human leukocyte antigen genotype, breastfeeding duration, timing of infant gluten introduction, total energy intake and the timing of islet autoimmunity. For CDA there were overall significant differences across tertiles of gluten intake in both unadjusted (P-value=0.05) and adjusted analyses (P-value=0.01) (df=2). For the outcome CD, the corresponding P-values were 0.13 (unadjusted) and 0.17 (adjusted).

Figure 3.

Probability (cumulative incidence) of celiac disease autoimmunity (CDA) according to thirds of gluten intake between the ages of one and two years.

Curves display failure distributions for CDA adjusted for covariates as described by Snapinn et al.(41) and Storer et al(42). The solid line (“-“) indicates the lowest third, the dashed line (“- -“) the middle third and the dashed-dot line (“- .”) the highest third of gluten intake.

Similar association patterns as above were found with the development of CD, when compared with one-year-olds in the lowest third of gluten intake, those in the highest third had almost a 2-fold increased risk of CD (aHR 1.96; 95%CI, 0.90–4.24; P-value=0.09; Figure 2). The association between gluten intake in one-year-olds and future CDA or CD did not differ by the child’s HLA genotype or family history of CD (interaction P-values>0.15).

Gluten intake throughout childhood and risk of CDA and CD

Characteristics of gluten intake are illustrated in Supplementary Figure 2. The incidence of CD increased with the cumulative gluten intake throughout childhood (per one gram, aHR 1.01; 95% CI 1.00–1.01; P-value=0.04) (Supplementary Table 3). At age six (mean cumulative gluten intake 22,262 [SD, 8291] grams) this corresponded to an aHR of 1.15 (95% CI, 1.00–1.32) per SD of cumulative gluten intake. The cumulative gluten intake was not associated with CDA. Gluten intake at observed event times (“current-value”) was not associated with the onset of CDA or CD at that time.

Subanalyses

Excluding 43 participants who at least once during follow-up were on a GFD yielded essentially unchanged aHRs for CDA/CD compared with our main analyses (Supplementary Tables 4–5).

DISCUSSION

In this cohort of at-risk children, gluten intake between the ages of one and two years and, to a lesser extent, the cumulative intake throughout childhood, were associated with the appearance of CDA/CD.

Strengths and limitations

Strengths of this study include its prospective design where validated dietary assessments were done before the onset of CDA, which minimizes the risk of recall bias and reverse causality, i.e., a changing diet as a result of tTGA-positivity. It is unlikely that the awareness of earlier normal (negative) tTGA tests affects gluten intake levels. Repeated tTGA-screening should ensure a high sensitivity for CD.(15) The power of this study was also increased by its long follow-up period and it is the first study to examine the association of gluten intake in later stages of childhood with the development of CDA or CD. The detailed data on potential confounders and predictors for CD, including genetic susceptibility, are additional strengths. Some analyses became significant with adjustment suggesting that the preselected adjustment variables were negative confounding variables.

A limitation of this study, as well as of previous studies in this field,(6–9) is that there is no established method for exact assessment of gluten intake. Similar to others,(6, 8, 21, 32, 33) we used self-reported data to derive the gluten amount from the protein content of wheat, barley and rye as defined by standardized or often-used food recipes. Although this approach will most likely, on average, correctly derive gluten quantities, we cannot rule out misclassification of gluten intake related to erroneous recall. Still, because dietary assessments were done before the onset of CDA, any potential misclassification is unlikely to be differential with respect to later CD risk, but may have resulted in attenuated HRs. Data from FFQs are likely to primarily reflect the usual diet (i.e., food habits) rather than current diet. While it may be difficult to estimate absolute food intake in a growing child, in whom feeding habits and choices may change significantly, semi-quantitative FFQs have proved useful for ranking participants based on quantified dietary intake for the study of associations.(34) The gluten intake in this study was largely similar to that reported for European children and young adults,(8, 32, 33) despite differences in terms of dietary instruments used and age distributions under study. To our knowledge, there are no previous estimates on gluten intake in pediatric populations in the U.S..

While we included data on the timing of gluten introduction during infancy, we lacked data on the quantity or pattern of gluten consumption at weaning. Another potential limitation is that dietary data were collected once per year. It is therefore unknown if sudden changes in diet may have an effect on CD development or relates to fluctuating tTGA-levels while on regular diet. Related to this limitation is our inability to examine how gluten intake, in parallel with intermittent exposures, e.g. infections, may shape the likelihood of developing CD. Yet, few previous studies in this field have such detailed long-term follow-up data on diet as the current study. Finally, this analysis was restricted to a subset of the DAISY cohort with available dietary data (Figure 1). Hence, missing data are a potential, but unverifiable, source of bias.

Interpretation of findings and previous literature

In this cohort, gluten intake in childhood predicted the later development of CDA/CD. This view is supported by observations linking geographic variations in CD prevalence with differences in gluten consumption.(35–37) A first individual-level analysis, published no less than fifteen years ago, revealed in retrospectively collected data that Swedish infants with a high intake of gluten-containing foods were at increased risk of CD diagnosed before, but not after, age two years.(7) Recently, a prospective analysis by Aronsson et al. of 132 Swedish children with CD found that those in the highest third of gluten intake in the first two years of life had a 2–3 fold increased CD risk compared with children who consumed less (odds ratio per increased gram of gluten 1.05 [95% CI 1.01–1.10]).(6) In that study, as in ours, the association of gluten intake and CDA/CD did not differ by the child’s HLA genotype. In other European studies gluten intake at one to three years of age has not been significantly related to the development of CD.(8, 9)

To our knowledge, this is the first study to link the amount of gluten intake in early childhood with the later development of CDA and CD in a U.S. population. Our results are consistent with those reported by Aronsson et al, but we contribute by demonstrating an equally strong association pattern with CDA as with CD. Another novel finding is the indication that gluten intake in one-year-olds was more strongly associated than cumulative intake throughout childhood, in development of CDA/CD.

The mechanism for the association between early gluten intake and CD remains to be elucidated. Speculatively, in the genetically susceptible child, a high gluten intake in early childhood may act as a primer for other environmental perturbations that initiate immune reactions leading to CD. Wheat intake may mediate or activate gut-permeating activities, a key feature early in CD pathogenesis and also a characteristic for prodromal stages of the disease.(38) Hence we speculate that at least one other time-varying factor must be present for CDA to develop, which would explain why high gluten intake between the ages of one and two years was more strongly associated than the longer-term cumulative intake. For example, early childhood is associated with a higher frequency of infections and immature immune functions, which in concert with gluten intake, might modify the risk of CD.(39) More research is needed to corroborate our findings and to determine if there is an optimal amount of gluten intake in early childhood.

Our conclusions are subject to the caveat that this study was restricted to at-risk children from the Denver metropolitan area. We do not know the extent to which these results can be generalized to other populations. Further, although the association between gluten intake in one-year-olds and later CDA was adjusted for multiple covariates, the observational nature of our data prevents us from excluding the possibility of unmeasured confounding. This study also lacks analyses of dietary patterns that may be associated with gluten intake level and the risk of CD. (40) Therefore, pending corroborative evidence, we do not recommend a change in pediatric feeding practices.

Conclusions

In this cohort of at-risk children, gluten intake between the ages of one and two years and, to a lesser extent, the cumulative intake throughout childhood, was associated with the appearance of CDA/CD.

STUDY HIGHLIGHTS.

What is current knowledge:

• Celiac disease (CD) is an autoimmune disorder in which genetically susceptible individuals develop an immune reaction to gluten

• Environmental factors are thought to be conducive to this loss of tolerance to gluten.

• Data are scarce and inconsistent whether the amount of gluten intake influences the risk of CD.

What is new here:

• In this at-risk birth cohort, one-year-olds with the highest gluten intake had a 2-fold greater hazard of CD.

• The incidence of CD also increased with increased cumulative gluten intake throughout childhood.

Abbreviations:

aHR

adjusted hazard ratio

CD

celiac disease

CDA

celiac disease autoimmunity

CI

confidence interval

DAISY

Diabetes Autoimmunity Study in the Young

FFQ

food frequency questionnaire

GFD

gluten-free diet

HLA

human leukocyte antigen

SD

standard deviation

tTGA

tissue transglutaminase antibody