Dementia is currently a major public health problem, affecting approximately 50 million people worldwide. It is predicted that there will be 152 million cases in 2050 (World Alzheimer report 2018), with 50%–70% being Alzheimer’s dementia (AD) and 25% being vascular dementia (VD)^[1,2]. As a heterogeneous neurodegenerative disease with a complex pathogenic interplay between genetic and modifiable risk factors^[3-5], the etiology of dementia remains unclear. Therefore, exploration of the novel pathogenesis of dementia is urgently required.

The brain–gut axis has received increasing attention in the field of dementia etiology. Inflammatory bowel disease (IBD), two types of ulcerative colitis (UC) and Crohn’s disease (CD), is a chronic disease with pathological characteristics of intestinal inflammation^[6]. Previous evidence has indicated that IBD could increase the risk of developing an autoimmune disease^[7], and can cause neuroinflammation^[8]. Clinical studies have shown an association between IBD and the risk of dementia. A longitudinal study among Taiwanese showed the hazard ratio (HR) of developing dementia in IBD patients was 2.54 (95% confidence interval [CI]=1.91-3.37), and the greatest impact was on AD by IBD (HR=6.19, 95% CI=3.31-11.57)^[9]. However, subsequent studies did not find such large association^[10,11]. Two other studies based on the Swedish National Patient Register database and the UK Biobank did not find any significant associations between IBD and dementia risk^[12,13]. Previous studies have yielded conflicting findings regarding the relationship between IBD and dementia. Since the possible association between IBD and dementia found in these observational studies may be subject to unmeasured confounding factors, reverse causality, and selection bias, the causal association between IBD and dementia remains unclear.

The Mendelian randomization (MR) approach has been widely deployed in genetic epidemiology using observational data to examine potential causal associations^[14,15]. Although three MR studies have explored the causal association between IBD and AD^[16-18], the findings have been inconsistent. The MR approach is still in the development stage and faces the challenge of pleiotropy bias. A recently developed method known as causal analysis using summary effect estimates (CAUSE) can effectively identify and address the issues of correlated and uncorrelated pleiotropic effects^[19]. Additionally, an increasing number of studies have used genetic correlation^[20,21] or co-localization analysis^[22,23] to assist MR analysis in elucidating causal associations.

Therefore, this study aimed to update the current evidence on the association between IBD and dementia based on observational studies. We then explored the stability of the results using serial M methods, especially CAUSE, genetic correlation, and co-localization analyses. Our study aimed to summarize the associations between IBD and dementia, and provide evidence of triangulation from observational findings and genetic associations.

In the first part, a meta-analysis was conducted and reported following the statement of preferred reporting items for systematic reviews and meta-analyses (PRISMA) released in 2020 (Supplemental Appendix 1, available in www.besjournal.com)^[24]. Four databases, PubMed, Web of Science, Embase, and the Cochrane Library, were searched from inception to February 2024 to identify eligible studies (the search strategies were shown in Supplemental Appendix 2, available in www.besjournal.com). In addition, the references of the relevant systematic reviews were scanned to identify eligible studies. A flowchart is presented in Figure 1.

Figure 1.  Flow diagram of the literature selection process.

Two independent investigators (MC and WC) retrieved and abstracted the full text of potentially eligible articles. Observational studies concentrated in cohorts, case-controls, and nested case-controls were considered eligible for inclusion if they investigated the associations of IBD, including UC and CD, with dementia, including AD and VD. Studies that were duplicates, unrelated to study design, systematic reviews or meta-analyses, comments or conference abstracts, published in languages other than English, and publications based on the same database, were further excluded.

For each eligible study, we extracted information related to author, year, country, sample size, adjustment model, and risk estimates, such as hazard ratio [HR], risk ratio [RR], odds ratio [OR], or standardized incidence ratios, together with their 95% confidence interval [CI] and the definition of exposures and outcomes. We recorded the risk estimates after adjusting for confounding factors. For studies based on the same database and in the same country, we extracted information from the study with the largest sample size. We applied the Newcastle-Ottawa Scale (NOS) to evaluate the quality of the included articles. Two investigators (MC and WC) independently extracted and analyzed the data, and a third investigator (DL) resolved any disagreements.

We used random-effects models with the Mantel-Hanszel method to combine OR and RR, given that heterogeneity is common in observational studies. Owing to the low incidence of dementia, the OR can approximate the RR in observational studies. I² statistics and the Cochrane Q test were used to test for heterogeneity. Additionally, a leave-one-out sensitivity analysis, omitting one study in turn, was used to explore the studies that could potentially affect the association. Egger’s test and funnel plots were used to detect potential publication bias.

A P <0.05 was considered statistically significant. All the data analyses were performed by “meta” package in R version 4.1.3.

We used the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) to evaluate the certainty of the evidence^[25]. The potential certainty of the overall evidence was rated as extremely low, low, moderate, or high. The hierarchy of evidence from observational studies was deemed low at first and was downgraded with violation of the five domains: risk of bias, inconsistency, indirectness, publication bias, and imprecision, while it was upgraded if the effect size was large, the study showed a dose-response correlation, or the effect size was underestimated due to negative bias.

In the second part, a two-sample MR analysis was performed to investigate the potential causal relationship between IBD and the risk of dementia. In addition, genome-wide LDSC was used to assess the genetic association between IBD and dementia, and co-localization analysis was used to investigate the local genetic structure shared between IBD and dementia and to assess potential pleiotropy. An overview of the design and analysis process is shown in Figure 2. Because our analyses were based on summary-level genome-wide association study (GWAS) data, there was no need for ethical approval or consent to participate.

Figure 2.  An overview of the MR study design. LD, linkage disequilibrium; MAF, minor allele frequency; MR, Mendelian randomization.

The MR study was conducted according to the Strengthening the Reporting of Observational Studies in Epidemiology using Mendelian randomization (STROBE-MR; Supplemental Appendix 3, available in www.besjournal.com)^[26]. It is well known that MR requires three main assumptions^[27]: (1) the instrumental variables (IVs)–genetic variants are strongly correlated with IBD; (2) the genetic variants are independent of the confounders of the IBD-dementia relationship; and (3) the genetic variants are independent of dementia, except IBD. As in all MR studies, the pleiotropic bias of the second assumption is difficult to avoid. The MR study process is shown in Figure S1 (available in www.besjournal.com).

Figure S1.  Mendelian randomization model.

The summarized GWAS data for IBD and the two main forms of IBD, including UC and CD, were reported in a published paper^[28], which was based on a meta-analysis of three GWASs from the UK low coverage whole genome sequencing IBD study, UK HumanCoreExome genotyped IBD study, and IIBDGC genotyped IBD study, and included 25,042 patients with IBD and 34,915 healthy controls, 12,194 with CD and 28,072 healthy controls, and 12,366 with UC and 33,609 healthy controls. Definitions of the IBD are shown in Supplementary Table S2 (available in www.besjournal.com).

The summarized GWAS data for dementia, including AD and VD, were based on FinnGen biobank analysis. The summarized data included more than 393,024 participants. Summarized GWAS data are available at https://storage.googleapis.com/finngen-public-data-r10/summary_stats/. In addition, we used GWAS-summarized data for AD, based on a meta-analysis of four large GWASs²⁹ (named AD-meta). The GWAS summarized data can be downloaded at https://ctg.cncr.nl/software/summary_statistics/. Additional information is provided in Supplementary Table S1 (available in www.besjournal.com).

Genetic variants of single nucleotide polymorphisms (SNPs) associated with IBD/UC/CD (P < 5 × 10^-8) were selected as candidate IVs, and linkage disequilibrium [LD] clumps (with the lowest P value having LD r² < 0.001) were used to further select the final IVs based on the 1,000 genomes of European samples. Genetic variants known to have pleiotropic effects were excluded from analyses using the Open Targets Genetics Tool.

The inverse-variance weighted method was used to investigate the overall causal correlation between IBD and risk, the inverse-variance weighted (IVW) dementia. The IVW Q test was performed to assess the heterogeneity of the selected IVs. Fixed-effects IVW models were used if there was no heterogeneity (P > 0.05); otherwise, random-effects IVW models were used. We also conducted sensitivity analyses, including weighted median (WM), penalized weighted median (PWM), Mendelian Randomization-Egger, MR-Pleiotropy Residual Sum and Outlier (PRESSO), MR-Robust Adjusted Profile Score (RAPS) and CAUSE^[19]. MR-Egger analysis was used to test pleiotropy based on the intercept.

The R² [R² = 2 × EAF × (1-EAF) × Beta²] of each SNP was estimated, and then summed up to assess the overall R². The F-statistics and power were calculated using https://shiny. cnsgenomic. org/mRnd/. Higher R², F-statistic, and power indicated a lower risk of weak IVs bias.

A key assumption of MR is that the IVs are not associated with any confounders of IBD or dementia (Supplementary Figure S1 available in www.besjournal.com). Therefore, we assessed the effect of potential pleiotropy on causal estimates using three analytical approaches. First, an MR-Egger analysis was performed to test for pleiotropy. Second, we used the CAUSE method to elucidate the correlated and uncorrelated horizontal pleiotropic effects^[19]. Third, we combined genetic evidence from genetic correlations and Bayesian co-localization analysis.

Genome-wide LDSC^[30] been used to assess genetic associations between IBD and dementia. If the proportion of heritability explained by genome-wide significant SNPs, as determined by the MR approach, is low, the accuracy of the MR results may be lower than that of LDSC. LDSC used all SNPs, including those that did not achieve genome-wide significance. The results of the LDSC analysis are presented as genetic correlation (r_g) with standard error (SE).

Co-localization analysis was used to assess whether the two associated traits were consistent with shared causal variants according to the included IVs and to evaluate potential pleiotropy caused by the MR approach. The involved hypotheses have been reported^[31], and we also used the posterior probability hypothesis 4 (PPH₄) to quantify the support of the hypothesis. The regions with 500 kb windows upstream and downstream of each instrumental variable in the MR were selected for analysis, and the average value of PPH₄ across all regions was taken as the final co-localization result.

The analyses were two-sided, with Ps of < 0.05 regarded as significant associations. PPH₄ level, which was higher than 75%, was considered suggestive of evidence for causal genetic variant for both traits. Data analyses were performed using “TwoSampleMR” and “CAUSE” packages in R version 4.3.1, LDSC v1.0.1, “coloc” package in R version 4.3.1.

A total of 2,072 articles were identified from the databases. After careful examination of the titles and abstracts, 2,040 articles were excluded. Ten articles met our criteria and were included in the final meta-analysis after careful full-text evaluation (Figure 1). All ten 10 articles were cohort studies^{[9-13,18,32-35]}. There were two articles^{12, 34} that involved the use of the same database, and both were included in the analysis because of the different reported exposures.

The characteristics of the studies are presented in Supplementary Table S2. These studies mainly reported results from individuals of European and Asian ancestry. Ten observational studies involving 80,565,688 participants were included in this meta-analysis. All cohort studies were followed up for more than 5 years. The definitions of exposures and outcomes varied across the studies (Supplementary Appendix 4, available in www.besjournal.com). The estimates for the individual studies are presented in Figure 3.

Figure 3.  Forest plot for the pooled estimates of the association between inflammatory bowel disease and dementia.AD, Alzheimer’s disease; CD, Crohn’s disease; CI, confidence interval; IBD, inflammatory bowel disease; RR, risk ratio; UC, ulcerative colitis; VD, vascular dementia.

According to the NOS tool, the included studies had scores between 6 and 7. Most of the included studies were based on electronic medical records or Medicare databases. The absence of a direct assessment of exposure and outcome validity, which might induce misclassification bias, was assessed as low quality. In addition, half of the included studies did not match confounders between the IBD and non-IBD groups (Supplementary Table S2).

The risk estimates of IBD for dementia, AD, and VD were calculated. For dementia, the combined RR of IBD was 1.36 (95% CI = 1.04-1.78) based on random effects model; the five studies have shown significant heterogeneity (I² = 84.8%; Figure 3). Meta-analysis based on types of UC and CD in the six studies provided the RR of 1.26 (95% CI = 0.97-1.63; I² = 83.4%) and 1.21 (95% CI = 1.11-1.32; I² = 54.9%). For AD, the combined RR of IBD in the four studies was 2.00 (95% CI = 0.96-4.13; I² = 99.8%; Figure 3). Meta-analysis based on types of UC and CD in the five studies that provided the RR of 1.84 (95% CI= 0.93-3.60; I² = 99.6%) and 1.35 (95% CI = 0.83-2.20; I² = 78.7%). Only one study reported an association between IBD and VD. The RR of IBD, UC and CD individually was 2.60 (95% CI = 1.18-5.70), 4.39 (95% CI = 1.64-11.80) and 1.10 (95% CI = 0.24-5.15), respectively (Figure 3).

Considering the large heterogeneity across studies, we used a leave-one-out sensitivity analysis to explore which study affected the effect size and caused heterogeneity. As shown in Table 1, when Zhang's study was excluded, the heterogeneity was much lower in most analyses (Table 1). The combined risk of dementia was 1.22 (95% CI = 1.20-1.23; I² = 0) in IBD patients, 1.07 (95% CI = 1.05-1.10; I² = 14.0%) in UC patients, 1.18 (95% CI = 1.10-1.26; I² = 0) in CD patients when Zhang's study was excluded. Combined risk of AD was 1.45 (95% CI = 0.93-2.27; I² = 99.6%) in IBD patients, 1.44 (95% CI = 0.83-2.51; I² = 99.7%) in UC patients, 1.09 (95% CI 0.99-1.19; I² = 45.0%) in CD patients when Zhang's study was excluded.

The Egger’s test show insignificance (P = 0.782), and the funnel plot did not perform obvious asymmetry (Supplementary Figure S2), indicating no significant publication bias.

Figure S2.  Funnel plot for assessment of publication bias.

Overall, the quality of evidence related to the association between IBD and dementia was low based on the GRADE framework (Supplementary Table S3). The main consideration for downgrading was the high heterogeneity across studies.

All variances of IBD, UC, and CD explained by IVs were greater than 0.5, and the F-statistics were greater than 10, indicating that the included SNPs satisfied the strong relevance assumption. Detailed information on the IVs is presented in Supplementary Table S4.

As shown in Table 2, the conventional MR analyses showed insignificance of genetically instrumented IBD with dementia (OR=1.01, 95% CI=0.98-1.03). In addition, UC and CD were also not associated with the increased risk of dementia (OR=1.00, 95% CI=0.98-1.03; OR=0.99, 95% CI=0.97-1.02). The associations of IBD, UC or CD with dementia were robust in all sensitivity analyses (all Ps > 0.05), except for that the WM and PWM analyses showed significant associations between CD and dementia (OR=0.96, 95% CI=0.93-0.99; OR=0.95, 95% CI=0.92-0.98; Supplementary Table S4). We used the CAUSE method to elucidate the correlated and uncorrelated horizontal pleiotropic effects, avoiding false positives induced by correlated horizontal pleiotropy. We did not find any significant causal associations between genetically instrumented IBD, UC, or CD and dementia (Table 2), indicating that the significant associations of genetically instrumented IBD and CD with dementia identified by the WM and PWM analyses might be caused by horizontal pleiotropy.

The conventional MR analyses showed insignificance of genetically instrumented IBD, UC or CD with AD (OR=0.98, 95% CI = 0.95-1.01; OR = 1.00, 95% CI = 0.96-1.05; OR=0.99, 95% CI=0.96-1.02). The robust findings were presented in the database of AD-meta (OR = 1.00, 95% CI = 0.99-1.00; OR = 1.01, 95% CI = 1.00-1.01; OR=1.00, 95% CI = 1.00-1.01; Table 2). The associations between IBD, UC, CD, and AD were robust in all sensitivity analyses (all Ps > 0.05), except for the MR-Egger and MR-RAPS analyses, which showed significant associations of genetically instrumented UC with AD in the AD-meta database (OR = 1.02, 95% CI = 1.003-1.03; OR=1.004, 95% CI=1.00-1.01; Supplementary Table S6). We did not find any significant causal associations between genetically instrumented IBD, UC, or CD, and AD using the CAUSE method, indicating that the significant association between genetically instrumented UC and AD in the AD-meta database might be caused by horizontal pleiotropy.

The conventional MR analyses showed insignificance of genetically instrumented IBD, UC, or CD with VD (OR = 1.02, 95% CI = 0.97–1.07; OR = 1.00, 95% CI = 0.94–1.05; OR = 0.97, 95% CI = 0.92–1.01; Table 2). Sensitivity analyses using various statistical models yielded robust results (Supplementary Table S7). The CAUSE method confirmed the insignificant causal association between genetically instrumented IBD, UC, CD, and AD.

The genetic correlation between IBD and dementia ranged from 0.001 to 0.043 (Table 3). There were no significant genetic correlations among IBD, CD, UC, or dementia (all Ps > 0.05).

There was no shared causal variant to prove the association between IBD and dementia using co-localization analysis based on the average value of PPH₄ across all regions (PPH₄ < 75%; Table 3). These co-localization results suggest that there may be no common biological mechanism between IBD and dementia.

Our meta-analysis indicated that IBD was associated with the risk of dementia but not with the risk of AD. However, considerable heterogeneity has been observed among these studies. Our study did not provide genetic evidence of a causal association between IBD and the risk of all-cause dementia. Exposure to IBD may not independently contribute to the risk of dementia, and the increased risk of dementia observed in observational studies may be due to unobserved confounding factors or detection biases.

Five systematic reviews and meta-analyses have summarized the relationship between IBD and dementia; however, previous findings were inconsistent^36-40. Additionally, previous meta-analyses did not consider all exposures (including IBD, UC, and CD outcomes) or outcomes (including all-cause dementia, Alzheimer's dementia, and vascular dementia). Our updated meta-analysis includes all exposures and outcomes. Considering the possibility of duplication of publications based on the same database, we screened and included the largest sample. One study based on the UK Biobank did not support a significant association between IBD and dementia¹³, while another study based on the same database showed that IBD was associated with early-onset dementia⁴¹. These findings suggest that different confounding factors and dementia types may have affected our results. Our meta-analysis suggested that great heterogeneity was mainly induced by Zhang et al. 's study, which might be due to differences in individual characteristics (i.e., symptom severity, treatment, inclusion criteria, race, ethnicity, and adjusting factors).

Considering unobserved confounding factors, we performed genetic analyses to further clarify the association between IBD and dementia. Combining genetic evidence from MR, genetic correlation, and co-localization analyses, there is a lack of evidence to support the causality of IBD with dementia risk. Our findings are somewhat inconsistent with those of previous MR studies. A recent MR study showed that genetically instrumented IBD is associated with a decreased AD risk^[16], whereas two other studies did not provide evidence for this association^[17,18]. Considering that pleiotropy poses a challenge in interpreting MR results, we used the CAUSE method, which can correct for correlated and uncorrelated horizontal pleiotropic effects^[19]. The CAUSE study showed no significant causal associations of genetically instrumented IBD, UC, or CD with dementia or AD, indicating that the recent MR study and significant associations of genetically instrumented IBD, UC, or CD with dementia or AD by sensitivity analyses might be caused by horizontal pleiotropy. In addition, we combined genetic evidence from MR with genetic correlation and co-localization analyses to simultaneously address the limitations of MR, such as its limited power due to the small number of SNPs and potential pleiotropy.

Evidence triangulation in our study suggests that the association between IBD and dementia found in some observational studies may not be causal. Some reasons for this may be as follows: First, IBD and dementia have many shared risk factors such as an unhealthy lifestyle, which is equivalent to residual confounding factors. If IBD occurs secondary to a pre-dementia state or subclinical dementia, reverse causation could also occur, inducing false associations. In addition, the possible association between IBD and dementia based on observational studies may be confounded by factors that are difficult to adjust for, such as social class, institutionalization, and medical comorbidities associated with IBD treatment. Furthermore, a greater number of medical examinations in patients with IBD than in the control population induces detection bias and results in a false association between IBD and dementia, as most observational studies are based on electronic medical record databases. Finally, the difference is likely due to the use of an elderly population, where selection bias due to recruitment on surviving exposure and the competing risk of dementia may be more severe.

Several factors may explain the association between IBD and dementia. Environmental and behavioral factors play a significant role in the development of IBD and dementia. The combination of lifestyle and drugs for the treatment of IBD is a novel measure that has been shown to alleviate cognition in dementia and may represent the most promising way to prevent and treat dementia. More and more evidence supports the mutual connection between the gut and central nervous system in diseases, known as the “gut-brain axes”^[42-44]. Intestinal homeostasis is associated with many psychiatric and neurological syndromes through the gut-brain axis, which describes the signal transduction between the microbiome, gut, and central nervous system^{[42,43,45-50]}. The gut microbiome plays an important role in the relationship between IBD and dementia^[50,51]. Previous studies have shown that smoking increases the risk of IBD and dementia, whereas physical activity decreases the risk of IBD and AD^[4,52]. In addition, a higher intake of ultra-processed foods is correlated with the risk of IBD and AD^[53,54]. Because the association between IBD and dementia is unlikely to be causal, further studies are warranted to investigate the factors shared by the comorbidities of IBD and dementia.

Evidence triangulation from a meta-analysis of observational studies and genetic associations provided a deeper understanding of the association between IBD and dementia. However, this study had some limitations. First, the heterogeneity of our meta-analysis was high and largely unexplainable. While this may be attributed to the differences in individual characteristics (i.e., race, ethnicity, symptom severity, medical comorbidities associated with IBD treatment, and diagnostic criteria), study duration, adjusting factors, sample size, and the limited number of studies prevented us from exploring heterogeneity sources by subgroup analysis or meta-regression. Furthermore, we did not obtain individual-level data and were unable to control for confounding factors. In addition, different controls were not used to explore the effects of confounding factors. However, MR studies may still be affected by weak instruments and pleiotropic biases. While leveraging GWAS data with large sample sizes, our study was limited by its ability to detect small effects. We combined the genetic correlation and MR methods and conducted multiple supplementary and sensitivity analyses to validate the robustness of our findings. Finally, we used GWAS data from individuals of European ancestry, even though the clinical and genetic characteristics of IBD and dementia may differ between cultural and ethnic groups. Therefore, our findings may not be generalizable to other racially or ethnically diverse populations.

Overall, our meta-analysis suggests that IBD is associated with the risk of dementia but not with an increased risk of AD, with considerable heterogeneity among the studies. However, genetic evidence suggests no causal association between IBD and dementia. Exposure to IBD may not independently contribute to the risk of dementia, and the increased risk of dementia observed in observational studies may be due to unobserved confounding factors or detection biases. Further exploration of the shared factors underlying the comorbidity of IBD and dementia may help identify potential targets for the gut-brain axis comorbidity of IBD-dementia prevention.

Study conception: DL and ZY; Study design: DL, ZY and FS; Data extraction and analysis: DL, MC, WC and YJ; Manuscript drafting: DL, YJ, FL, and TL. All co-authors have revised the manuscript and approved the submitted version.

Our study did not include individual data from the human participants. The analyses for this study were based on publicly available summary datasets, and no additional ethical approval or consent to participate was required.