This study shows considerable changes in CD risk over time in Sweden that are dependent on age and year of birth. Over the 36 years of follow-up we have observed an overall increase in CD risk in the entire child population except for a temporary decline from 1995 to 1997. At the same time, a substantial drop in CD incidence has occurred in the subgroup of children below 2 years of age, with almost zero CD cases diagnosed within the first year of life. The changing CD pattern is also illustrated by a continued increase in the median age at diagnosis. Our findings from cumulative incidence analysis suggest that year of birth may determine specific CD risks, exemplified by excess risk in the birth cohorts of 1985–1995 and 2000–2002. The observed fluctuations in CD cannot be explained by genetic changes, and therefore strongly suggest the role of cyclic environmental and lifestyle risk factors in CD etiology.
The main strength of this study is the identification of CD cases through the Swedish National Childhood CD Register, with a prospective reporting since 1998 from all pediatric departments in the country. Differences in population coverage may not have introduced any major biases, since our findings for the period from 1973 to 1990, with 15% coverage of the total child population are comparable to findings from an earlier national wide study conducted during the same period . We have also previously shown that 40% coverage was sufficient for estimating the national CD incidence .
In this study CD diagnosis was based on villous atrophy in order to avoid overestimation of the CD risk. We excluded all cases without villous atrophy even if they had elevated serological markers, minor enteropathy and symptoms typical for CD. The observed overall increase cannot by due to mass screening for the general population since this has not been implemented in Sweden. Additionally, serological testing in genetically susceptible high-risk populations is encouraged and conducted liberally. However, this cannot be solely responsible for the observed increase. The latest ESPGHAN guidelines where CD diagnosis may be based on serological markers were introduced in 2012  and therefore did not influenced our findings since our follow-up ended in 2009.
The limitation of this study is that the CD register primarily captures cases detected clinically. However, it is well established that most CD cases both worldwide and in Sweden remain undiagnosed, and thus most CD studies and knowledge of CD are limited to clinical cases [15, 16]. In fact, our recent CD screening study revealed two thirds of all cases were only identified through screening .
We found an overall increase in CD incidence in children mainly due to a rising incidence in children aged 2–14.9 years. An increase in CD occurrence [18–23] along with other immune-mediated disorders such as diabetes, asthma, and inflammatory bowel disease, is not unique to Sweden and has been reported elsewhere [24–26]. Increased awareness may be one of the contributing factors to the rising incidence observed in children aged 2–14.9 years. However, this is insufficient in explaining variations seen in the youngest children in whom there has been a sharp increase from 1985 to 1994, a sudden decline from 1995 to 1997 and yet again a rapid increase from 2000 to 2002 followed by a persistent decline.
The median age at diagnosis has increased throughout this follow-up; between 1973 and 1994 it was 1.2 years, and rose to 6.7 years from 2004 to 2009. The upward shift in age at diagnosis can partly be explained by the fact that the epidemic birth cohorts of 1985–1995 are getting older and are carrying along an excess risk. However we also notice a similar pattern in the post epidemic birth cohorts thus suggesting additional exposures that are probably related to modern life-styles and are yet to be identified. Increased awareness of CD symptoms that are typical in older children, as well as screening of siblings, could have contributed to the shift towards an older age at diagnosis, but the role of additional environmental factors should not be overlooked.
The variations in incidence among children below 2 years of age strongly suggest a role of environmental and/or lifestyle exposures that change over time. We have previously shown that infant feeding practices have played a crucial role in the onset and the end of a 10-year Swedish CD epidemic during the mid 1980s up until 1995 [6, 7]. The continuing drop in incidence among children below 2 years of age began in 1995, coinciding with a one-third decrease in the average daily consumption of gluten-containing flour in children younger than 2 years of age, and introduction of the new national infant feeding recommendation that encouraged introduction of gluten in smaller amounts and preferably from 4 months of age with continued breastfeeding . To date this recommendation remains in effect and most likely contributes to the continued CD decline in the youngest age group.
We previously showed a gap in the cumulative incidence between the epidemic and post-epidemic cohorts, but as both cohorts grow older, the gap continues to narrow. This is true regardless of whether cases were clinically detected or detected through screening [12, 16]. In fact, during this follow-up, the cumulative incidence of the post-epidemic cohorts of 2000–2002 has surpassed that of the epidemic birth cohort (1985–1995) at certain ages. It is not yet clear as to why the 2000–2002 birth cohorts have a higher risk compared to other post-epidemic cohorts, since infant feeding guidelines have remained the same to date, thus stressing the role of other yet unidentified environmental risk factors.
Other environmental and lifestyle factors such as an increase in gluten consumption may explain the observed changes including the gradual increase seen over time among children 2–14.9 years, but can hardly explain the shorter and sudden changes that have been observed, especially in children below 2 years of age. The variations in the youngest children seemed to have an epidemic pattern both in period from 1985 to 1995 and the period from 2000 to 2002 and therefore are more likely to be associated with periodical changes such as episodes of infections [8, 27, 28]. Interestingly, during the Swedish CD epidemic, rod-shaped bacteria were often found in the small intestinal mucosa of children with CD, but not in controls, and rarely in biopsies from later periods . Changes in the intestinal microbiota have also been associated with CD susceptibility and might play a role in explaining our observations [30–32].
Another crucial factor that needs to be examined is whether the observed CD increase is due to increasing immigration. Sweden, just like other European countries, has experienced a positive trend of net migration from abroad. Nevertheless, it is unlikely that the demonstrated CD increase is due to increasing immigration, because the majority of the immigrant population comes from low CD-prevalent countries. During the 1980s, immigration was mainly from Iran, Lebanon, Poland and Turkey. In the 1990s immigration was mainly from the former Yugoslavia, and in the 2000s from Iraq, Somalia, Afghanistan and Ethiopia . Moreover, a recent study of worldwide immigration in Sweden showed non-European ethnicities to have a much lower CD susceptibility compared to the native population .