A systematic review was conducted for IBD disorders. Studies were initially included from 1970 to 2019. To improve transparency in methodology, the study protocol for this review was published as “Study Protocol Epidemiology of Inflammatory Bowel Disease in Childhood and Adolescence: a Systematic Review” [21].

This systematic literature search was performed in the PubMed and Embase databases via Ovid. In addition, a manual search was performed in bibliographies of previously published and relevant systematic reviews. For detailed methodology and screening of this systematic review, we refer to the published study protocol [21].

For this study, we updated the previous systematic review from 2019 and 2020 to 2022. In this update, which was carried out until August 2022, we used the same search term as before but also included studies published up to December 31, 2021, which covered the observation period from 1970 to 2019. The complete search strategy can be viewed in Multimedia Appendix 1. The inclusion and exclusion criteria shown in Table 1 were defined for this study.

The title and abstract screening and the full-text screening were carried out independently by 2 project participants. The extraction of the data and the corresponding consistency checks were also carried out by 2 project participants. In case of disagreement, a third project participant was consulted for mediation. All included studies were evaluated for study quality. The critical appraisal tools Critical Assessment of Structure Prediction and Scottish Intercollegiate Guidelines Network were used for this purpose. In addition, a risk of bias analysis, following the procedure described in the Cochrane Handbook [22], was performed (see Multimedia Appendix 2 [5-7,23-92]). Studies of poor quality were not excluded from the quantitative synthesis to avoid loss of information.

An ethics vote was not required for this systematic review because patients were not directly involved in this study.

All included studies were independently screened for incidence rates and study characteristics using a standardized table summary of findings. In case of missing data, contact with the authors was made. The data were exported to a database and processed for statistical analysis. For studies by 1 author that reported multiple incidence rates for children and adolescents, the mean values of incidence rates and study sizes were calculated for the respective observation period.

For the planned meta-regression, we classified possible moderators of heterogeneity into 2 dimensions: geographic and environmental factors on the one hand, and socioeconomic factors on the other. Longitude and latitude as well as exposure to UV sunlight (UV radiation index [UVI]) were assigned to the first dimension of geographical and environmental factors. The geographic data were extracted from Geoplaner (version 3.1; Martin Nathensen). When studies are nationwide or involve multiple centers within a country, the mean latitude value applied to the corresponding country or area was considered. In addition, we used the mean latitude to calculate the absolute distance to the equator irrespective of the northern or southern location [15]. We extracted the UVI from the United Nations Sustainable Development Goals data from the World Health Organization database [93].

The second dimension of possible moderators included socioeconomic factors. For this purpose, we used the percentage of gross domestic product (GDP) spent on health, which we extracted from the Organisation for Economic Co-operation and Development database “Health expenditure and financing” [94]. The Human Development Index (HDI) was included in the analysis as another possible moderator. The HDI assesses a country’s developmental state and combines life expectancy at birth, expected years of schooling, and gross national income per capita [95]. The values relevant to this study were extracted from the United Nations Development Programme’s Human Development Reports [95] from 1990 onward and averaged for statistical analysis. In addition, data on the GDP of the respective included countries from the GENESIS database of the Federal Statistical Office were used for further moderator analysis [96]. Furthermore, the universal health coverage (UHC) service coverage index Sustainable Development Goals 3.8.1 was extracted from the World Health Organization database [93]. UHC quantifies coverage of essential health services and is defined as the average coverage based on tracer interventions that include reproductive, maternal, newborn, child health, infectious diseases, noncommunicable diseases, service capacity, and access among the population [97].

We performed random-effects meta-analyses and meta-regressions to assess the variability of incidence rates. Analysis was performed with R (version 4.2.1.; R Foundation for Statistical Computing) software using the metafor package version 3.8-1 [98]. Meta-analysis was performed on a log scale (log incidence rates) using the general inverse variance method. Random effects and the extent of heterogeneity were estimated using the restricted maximum likelihood estimator. For the meta-regression, a multivariate model was constructed to identify further moderators of heterogeneity in incidence rates. The pooled incidence rates for each observation period formed the dependent variable. The observation period for each study was averaged and assigned as the starting time of the given study. The absolute distance of the included countries from the equator, UVI, HDI, health expenditure as a percentage of GDP, GDP, and UHC were included as additional independent variables in the regression model. In addition to the estimate of τ², the Q-test for heterogeneity and the I² statistic are reported. The I² value was interpreted according to Higgins and Thompson [99] as follows: 0% to 40%, possibly insignificant; 30% to 60%, moderate heterogeneity; 50% to 90%, substantial heterogeneity; and 75% to 100%, considerable heterogeneity [99]. The influence of the moderators was evaluated using the R² statistic as a measure of the explained heterogeneity. An a priori significance level of 0.05 (5%) was set for all statistical methods. To control the risk of publication bias, statistical methods such as the Egger regression test and the rank correlation test were applied to quantitatively assess the risk of publication bias. In addition, we applied the trim-and-fill analysis and the fail-safe N-analysis (Rosenberg method) to consider and control the potential risk of publication bias.

A total of 3153 studies were found from the previous systematic search conducted in 2019. The update of the systematic literature search yielded another 83 records. After removing duplicates, 77 studies were screened in the systematic literature search update. Another 5 studies from the update were included in the qualitative and quantitative synthesis. In total, the systematic literature research resulted in 81 findings from 29 countries with the search terms CD, UC, and indeterminate colitis. CD was the subject of a total of 74 studies, which were included in the qualitative synthesis of this work. Further, 2 studies had to be excluded retrospectively due to the lack of population reference. The meta-analysis included 72 studies from 26 countries and the meta-regression included 52 studies (Figure 1 [100]). The references of the included studies and a table summary of included studies can be viewed in Multimedia Appendices 3 [6,7,23-75,77-92] and 4 [7,23-75,77-92,101].

In the 72 studies eligible for meta-analysis, the incidence rates on the linear scale varied from 0.14/10⁵ to 11.40/10⁵. Performing a random-effects meta-analysis revealed that the overall mean incidence rate was 2.64/10⁵ (95% CI 2.09 to 3.34; on log-scale –10.54, 95% CI –10.78 to –10.31), whereas the I² value of 97.88% suggests that the substantial heterogeneity of this study’s results cannot be explained by a sampling effect alone (Multimedia Appendix 5 [5-7,23-92]). In the following, we set out to identify factors that can account for the substantial dispersion in study results. Interestingly, the individual weights for each study were largely dominated by the contribution of the between-study variance while the study-specific variance (ie, the sampling effect) had a smaller effect. Consequently, the studies in the random-effects meta-analysis have rather similar relative weights.

We included studies from a 50-year observation period, from 1970 to 2019. To assess whether the time point of this study influenced the CD incidence, we performed a meta-regression in which time is considered the continuous variable, whereas incidence rates are the dependent variable. Figure 2 confirms that the moderator time has no significant effect on incidence rates for CD. Moreover, time as a moderator cannot explain the heterogeneity, so the remaining heterogeneity remains substantial (test of moderators P=.39; I²=97.85%; R²=0.00; see also Multimedia Appendix 6-11 [5-7,23-92] subgroup analysis incidence CD in 10-year steps). These results suggest that there must be other moderators to explain the observed heterogeneity. Figure 2 also displays a slight negative trend with a simultaneous increase in heterogeneity. Some of the studies with low incidence values (depicted in the lower right corner) are from Taiwan, Finland, Saudi Arabia, Mexico, and Argentina, reinforcing the impression of greater geographical division. In the next step, we specifically examine the influence of the geographic component on CD incidence rates.

It is interesting to see that the highest mean incidence rates per 100,000 children and adolescents during the observation period from 1970 to 2019 were observed in Australia (11.12 new cases/10⁵), Finland (6.31/10⁵), Canada (7.12/10⁵), Germany (6.15/10⁵), and New Zealand (6.07/10⁵). The lowest incidence rates were reported in studies from countries in Asia and South America. Strikingly, an incidence of CD almost twice as high was reported in Australia compared to the other included countries (Figure 3 and forest plot geographic variation in incidence rates of CD in Multimedia Appendix 12). These data suggest geographic heterogeneity, which we first consider at the continental level.

Indeed, a meta-regression with the variable continents indicated that 41.34% of the heterogeneity can be explained. The test for moderators is significant (P<.001). Comparing this regression model with a more complex model in which we added the factor time to the moderator continents, an ANOVA showed no significant model improvement, confirming the notion that time does not act as a major moderator. In a further step, we examined the development of the incidence of CD over time for each continent individually. The results of this analysis suggest that the incidence of CD has developed differently in relation to the continents. Increasing trends were noted for North America, Europe, and Asia. For South America and Australia or the Pacific region, we found no confirmation of increasing incidence because of too few data points (see Multimedia Appendices 13 and 14).

The results also suggest that distance from the equator may affect the incidence of CD. A corresponding meta-regression, which included absolute distance from the equator as a moderator, showed that CD incidence increased significantly with increasing distance from the equator (Figure 4). Extrapolated to 1000 km, the incidence rate increased by 0.36%. The test for moderators yielded a significant result (P<.001). However, given the considerable heterogeneity in study results, distance from the equator formally contributed only moderately to better explain this variance (R²=29.14%; Table 1 and Figure 4). We found similar results when we recalculated the analysis for the country-specific UVI instead of the absolute distance from the equator. The results show that incidence rates decrease with increasing UV exposure. The results with this factor were significant in the test of moderators (P<.001) and 18.57% heterogeneity was resolved (Table 1 and Figure 4). Given the correlation between the moderators of absolute distance from the equator and UVI (r=–0.87, P<.001), we refrained from a joint regression model to avoid problems of collinearity and unreliable coefficient estimates.

In the next step, we investigated the extent to which socioeconomic factors could be considered moderators of heterogeneity. The results of the corresponding meta-regression showed that the HDI, health expenditure as a percent of GDP, and the UHC index acted as moderators. Accordingly, the frequency of CD increases with increasing values of each moderator (Table 2 and Figure 5). To avoid issues with collinearity and unreliable coefficient estimates resulting from the correlations between the socioeconomic factors, we decided not to use a joint regression model.

In our analysis, we identified different widely independent factors study timing, absolute distance from the equator, and HDI or UHC as univariate moderators of CD incidence rates. To explain the high degree of heterogeneity between studies that we observed during the analyzed study period, we used a multifactorial meta-regression model that accounted for these complementary moderators as the final step of our investigation. As a result, the corresponding model showed a joint R² of 62.5%, indicating that almost two-thirds of the heterogeneity can be explained by these 3 moderators. The test for moderators was significant at P<.001 (see Tables 3 and 4).

This study is limited by the use of 2 databases for systematic literature search. Another limitation of this study is the exclusion of studies that were not published in English, Spanish, French, or German. Given that the included studies were mainly from countries with good access to the health care system (UHC >70), the underreporting of countries with poorer access to the health care system should be discussed. We controlled the risk of publication bias using the Eggers regression test, rank correlation test, trim and fill analysis, and fail-safe N analysis (Rosenberg method). Although these methods did not statistically indicate a bias due to publication bias, a small bias cannot be completely ruled out.

Based on available study data from 1970 to 2019, we could not identify a global, temporal trend toward increasing CD incidence rates, although these effects are reported for individual countries or regions. Instead, we could demonstrate that a substantial part of the observed heterogeneity between the published study results can be explained by geographic location and socioeconomic factors. Our study can be used to provide quantitative estimates of these trends for CD in childhood and adolescence. However, to establish causal relationships regarding potential risk factors, further studies are necessary, including those conducted in countries with lower levels of development. Nevertheless, our analysis provides valuable information to drive future research and health policies aiming to reduce the incidence of CD among children and adolescents. This needs continuous global monitoring of the incidence of IBD in childhood and adolescence to fully understand the trends in IBD incidence [102]. To this end, the implementation of internationally standardized and interoperable registries, coupled with the dissemination of health data via federated networks grounded on a CDM, such as the OMOP CDM, is deemed advantageous. The OMOP CDM aligns most closely with the requisites conducive to expediting data exchange within longitudinal studies [112,113]. The usage of such registries and data networks holds the potential to streamline the exhaustive and standardized accumulation as well as dissemination of data. This, in turn, would enhance our comprehension of CD and foster evidence-based approaches for preventive and interventional strategies.