Antibiotic Exposure and the Development of Celiac Disease
Antibiotic Exposure and the Development of Celiac Disease
This is the first study to find a positive association between antibiotic use and subsequent CD. Antibiotic exposure was also linked to small-intestinal inflammation and to normal mucosa with positive CD serology, both of which may represent early CD. The consistent association between the multiple groups, the slightly stronger association between repeated use of antibiotics compared with no use as well as the association with use of certain antibiotics (e.g., metronidazole) and CD may suggest that antibiotic exposure, possibly through a changed gut microbiota, plays a pathogenic role in early CD development. However, given the lack of time-response effect, within the limited time window studied, we cannot rule out non-causal explanations for our findings.
Observational studies on drugs are particularly susceptible to the concerns of reverse causation and confounding-by-indication. Reverse causation defines the causality bias if the exposure is a response to manifestations of the undiagnosed disease. In CD it is difficult to date the true onset of disease and thereby to establish whether antibiotic use truly preceded CD or whether the antibiotic was given for the symptoms of as yet undiagnosed CD. Several studies have shown a mean diagnostic delay of 5–11 years from onset of CD symptoms until diagnosis, a time associated with an increased number of consultation visits and possibly an increased likelihood of receiving antibiotic prescriptions. To reduce the risk of reverse causation and the effect of comorbidity, which may act as a confounder by increasing the possibility of receiving antibiotic prescriptions, we performed two post-hoc analyses restricted to individuals exposed to antibiotics in the first six months of life or individuals without hospital admission. Although these post-hoc analyses revealed largely unchanged ORs, they do not rule out residual comorbidity or reverse causation.
Observational studies on drugs may also be subject to confounding-by-indication in which the indication for treatment and not the treatment per se is associated with the outcome. Individuals with undiagnosed CD have an increased risk of several diseases that may, in concert, increase their likelihood to receive antibiotics. For example, because antibiotics are frequently misused in viral infections, confounding may be introduced when antibiotics are erroneously used to combat adenovirus or rotavirus infections, both proposed as risk factors for CD development. However, the Swedish Medical Products Agency do not recommend antibiotic treatment in diarrhoeal illnesses, except for cases of severe bacterial gastroenteritis. Further, just as for diagnosed CD, undiagnosed CD may be associated with bacterial infections, which may have also influenced our results. Finally, the fact that all three cohorts were similarly associated with antibiotic use raises the possibility that an external factor, i.e. gastrointestinal symptoms such as diarrhoea, increases the "risk" of both antibiotic use and the performance of a small bowel biopsy.
It is well-established that the intestinal microbiota influences the maturation of the intestinal immune system. Meanwhile several studies have found an imbalanced composition of the intestinal microbiota in those with CD.In vitro studies suggest that intestinal dysbiosis may, in the presence of gliadin, increase intestinal epithelial permeability and enable epithelial translocation of gliadin peptides potentially triggering CD. Other data suggest that the distinct intestinal microbiota in CD may have pro-inflammatory properties that affect the immune response elicited by gluten. Although this study lacks conclusive evidence for a causal association between antibiotic use and subsequent CD, our results do not refute the hypothesis that the intestinal microbiota affects CD development. A causal association may also be supported by the slightly stronger association to subsequent CD and certain antibiotics (e.g., metronidazole) that have a major impact on the anaerobic bacteria of the colon. Consequently, today's prevalent use of antibiotics and their potential public heath impact on CD development warrant attention in future research.
Antibiotic use has been associated with the development of several immunological diseases, including inflammatory bowel disease and asthma. More importantly with regard to CD, most but not all studies, have failed to find an association between antibiotic use and subsequent type 1 diabetes, a disease that otherwise shares many aetiological traits with CD.
A major strength of this study is our use of multiple groups on the CD spectrum (CD, small-intestinal inflammation and normal mucosa with positive CD serology). With this study design, we were able to examine the association of antibiotic treatment by the degree of mucosal abnormality. Multiple groups also improved our evaluation of potential causality. Another strength is the use of prospectively recorded exposure and outcome data, which eliminate the risk of recall bias. Furthermore, this study provided detailed information on antibiotic use, including time and age of exposure, type of antibiotics and number of courses.
The use of biopsy data enabled us to identify a representative population with CD. In Sweden, more than 95% of gastroenterologists obtain a small-intestinal biopsy before CD diagnosis, implying that biopsy records have a high sensitivity for diagnosed CD. We regard the risk of misclassification in CD as low. In an earlier validation study 108 (95%) of 114 individuals with villous atrophy had CD. Misclassification could be more of a concern in inflammation because villous atrophy may be patchy and not all inflammation is related to CD or to a pre-coeliac state. Furthermore, any potential misclassification of histopathology should be non-differential regarding antibiotic use and therefore should not lead to spurious associations, but to an underestimation of the true effect.
Our third cohort included individuals with normal small-intestinal mucosa, but positive CD serology. Most of these individuals had a single positive AGA serology with a lower specificity for CD than TTG or EMA. Thus, it may be argued that this condition does not represent a pre-coeliac state. However, when Hill et al. reviewed 26 studies of CD serology, they observed a median AGA specificity of 93%.
Antibiotic exposure was determined by the Swedish Prescribed Drug Register, which includes nationwide high-quality data on all dispensed prescribed medications. Self-medication, i.e. obtaining an antibiotic without prescription, is very rare in Sweden, estimated to be 0.3% of all antibiotics used. A limitation of our study is the recent start of the Swedish Prescribed Drug Register (established in July 2005) and the left truncation of exposure data in which individuals diagnosed with CD early in the study period (and their matched controls) will have little chance of being classified as antibiotic users because of lack of antibiotic data before July 2005. However, this loss of prior antibiotic data should not be differentially related to future CD status, and therefore only bias our results toward the null.
Discussion
This is the first study to find a positive association between antibiotic use and subsequent CD. Antibiotic exposure was also linked to small-intestinal inflammation and to normal mucosa with positive CD serology, both of which may represent early CD. The consistent association between the multiple groups, the slightly stronger association between repeated use of antibiotics compared with no use as well as the association with use of certain antibiotics (e.g., metronidazole) and CD may suggest that antibiotic exposure, possibly through a changed gut microbiota, plays a pathogenic role in early CD development. However, given the lack of time-response effect, within the limited time window studied, we cannot rule out non-causal explanations for our findings.
Observational studies on drugs are particularly susceptible to the concerns of reverse causation and confounding-by-indication. Reverse causation defines the causality bias if the exposure is a response to manifestations of the undiagnosed disease. In CD it is difficult to date the true onset of disease and thereby to establish whether antibiotic use truly preceded CD or whether the antibiotic was given for the symptoms of as yet undiagnosed CD. Several studies have shown a mean diagnostic delay of 5–11 years from onset of CD symptoms until diagnosis, a time associated with an increased number of consultation visits and possibly an increased likelihood of receiving antibiotic prescriptions. To reduce the risk of reverse causation and the effect of comorbidity, which may act as a confounder by increasing the possibility of receiving antibiotic prescriptions, we performed two post-hoc analyses restricted to individuals exposed to antibiotics in the first six months of life or individuals without hospital admission. Although these post-hoc analyses revealed largely unchanged ORs, they do not rule out residual comorbidity or reverse causation.
Observational studies on drugs may also be subject to confounding-by-indication in which the indication for treatment and not the treatment per se is associated with the outcome. Individuals with undiagnosed CD have an increased risk of several diseases that may, in concert, increase their likelihood to receive antibiotics. For example, because antibiotics are frequently misused in viral infections, confounding may be introduced when antibiotics are erroneously used to combat adenovirus or rotavirus infections, both proposed as risk factors for CD development. However, the Swedish Medical Products Agency do not recommend antibiotic treatment in diarrhoeal illnesses, except for cases of severe bacterial gastroenteritis. Further, just as for diagnosed CD, undiagnosed CD may be associated with bacterial infections, which may have also influenced our results. Finally, the fact that all three cohorts were similarly associated with antibiotic use raises the possibility that an external factor, i.e. gastrointestinal symptoms such as diarrhoea, increases the "risk" of both antibiotic use and the performance of a small bowel biopsy.
It is well-established that the intestinal microbiota influences the maturation of the intestinal immune system. Meanwhile several studies have found an imbalanced composition of the intestinal microbiota in those with CD.In vitro studies suggest that intestinal dysbiosis may, in the presence of gliadin, increase intestinal epithelial permeability and enable epithelial translocation of gliadin peptides potentially triggering CD. Other data suggest that the distinct intestinal microbiota in CD may have pro-inflammatory properties that affect the immune response elicited by gluten. Although this study lacks conclusive evidence for a causal association between antibiotic use and subsequent CD, our results do not refute the hypothesis that the intestinal microbiota affects CD development. A causal association may also be supported by the slightly stronger association to subsequent CD and certain antibiotics (e.g., metronidazole) that have a major impact on the anaerobic bacteria of the colon. Consequently, today's prevalent use of antibiotics and their potential public heath impact on CD development warrant attention in future research.
Antibiotic use has been associated with the development of several immunological diseases, including inflammatory bowel disease and asthma. More importantly with regard to CD, most but not all studies, have failed to find an association between antibiotic use and subsequent type 1 diabetes, a disease that otherwise shares many aetiological traits with CD.
A major strength of this study is our use of multiple groups on the CD spectrum (CD, small-intestinal inflammation and normal mucosa with positive CD serology). With this study design, we were able to examine the association of antibiotic treatment by the degree of mucosal abnormality. Multiple groups also improved our evaluation of potential causality. Another strength is the use of prospectively recorded exposure and outcome data, which eliminate the risk of recall bias. Furthermore, this study provided detailed information on antibiotic use, including time and age of exposure, type of antibiotics and number of courses.
The use of biopsy data enabled us to identify a representative population with CD. In Sweden, more than 95% of gastroenterologists obtain a small-intestinal biopsy before CD diagnosis, implying that biopsy records have a high sensitivity for diagnosed CD. We regard the risk of misclassification in CD as low. In an earlier validation study 108 (95%) of 114 individuals with villous atrophy had CD. Misclassification could be more of a concern in inflammation because villous atrophy may be patchy and not all inflammation is related to CD or to a pre-coeliac state. Furthermore, any potential misclassification of histopathology should be non-differential regarding antibiotic use and therefore should not lead to spurious associations, but to an underestimation of the true effect.
Our third cohort included individuals with normal small-intestinal mucosa, but positive CD serology. Most of these individuals had a single positive AGA serology with a lower specificity for CD than TTG or EMA. Thus, it may be argued that this condition does not represent a pre-coeliac state. However, when Hill et al. reviewed 26 studies of CD serology, they observed a median AGA specificity of 93%.
Antibiotic exposure was determined by the Swedish Prescribed Drug Register, which includes nationwide high-quality data on all dispensed prescribed medications. Self-medication, i.e. obtaining an antibiotic without prescription, is very rare in Sweden, estimated to be 0.3% of all antibiotics used. A limitation of our study is the recent start of the Swedish Prescribed Drug Register (established in July 2005) and the left truncation of exposure data in which individuals diagnosed with CD early in the study period (and their matched controls) will have little chance of being classified as antibiotic users because of lack of antibiotic data before July 2005. However, this loss of prior antibiotic data should not be differentially related to future CD status, and therefore only bias our results toward the null.