We conducted a systematic review and meta-analysis in accordance with Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (27). The predefined protocol was registered at Open Science Framework (OSF; identification number 9xvgz; date of registration: February 28th, 2020).

We predefined the following inclusion criteria: (1) original studies without restriction regarding publication status (i.e., peer-reviewed full-text journal articles, non-peer-reviewed full-text manuscripts, and conference abstracts) to avoid data limitation, (2) reporting on or including data allowing the calculation of: point (≤4 weeks), 12-month, lifetime prevalence, or 12-month incidence of PTSD [i.e., diagnosed PTSD through structured interviews or the assessment based on PTSD according to DSM-5 (4) or ICD-10 (28) or prior DSM/ICD versions] and/or PTSS (i.e., assessment providing an indication for elevated PTSS but is not based on classification system) in people with (3) CD [defined following the definition of Kampling et al. (29) who specified a list of ICD-10 diagnoses to meet the criteria of CD]. Studies had to be available in (4) English or German language. If no full text was available even after contacting the authors, this led to exclusion.

The literature search (including backward searches) was performed using the electronic databases MEDLINE, Embase, and PsycINFO from 13.03.2020 to 24.07.2020. Publications from their inception to June 2020 were considered. The search string combines terms related to CD/CD events with those related to PTSD/PTSS. Proximity operators were used and MeSH (Medical Subject Headings) terms were applied where appropriate. The search strategy is provided in Supplementary Table 1 and was validated using the PRESS guideline (30).

Two reviewers (FL and PG) independently screened titles and abstracts of all studies that resulted from the search, reviewed full-text articles of all potentially relevant articles, and extracted data from eligible full-text articles. Discrepancies were resolved by discussion with a third reviewer (LS). Data collected included author, publication year, country, sex, age, CD diagnosis, PTSD/PTSS, index trauma, type of PTSD/PTSS measurement instruments, time of PTSD/PTSS measurement after CD diagnosis or initial treatment, sample size, and estimates. If results were based on the same study sample, the most comprehensive and recent publication was considered. For intervention studies with multiple measurement time points, baseline assessment data were included to evaluate cohorts prior to study interventions. To account for the sample dependence of multiple estimates within a study, the most recent eligible PTSD or PTSS assessment date for the prevalence or incidence estimate and the most commonly used measurement tool to collect PTSD or PTSS were extracted. If information or data were missing, the corresponding authors of the study were contacted and asked for further information.

Two researchers (FL and PG) independently assessed each included study for quality by using a risk of bias (RoB) appraisal instrument based on the Newcastle-Ottawa Scale (31) as well as the recommendations of the Cochrane Collaboration for the assessment of the risk of bias for non-comparative studies (32). This combined instrument addresses the quality of (1) sample representativeness, (2) prospective scheduling, (3) transparent, non-selected reporting of sample characteristics and outcomes, (4) sample size, (5) assessment, (6) data quality, and (7) comparability against a control group (full details regarding scoring provided in Supplementary Appendix 2). Studies were classified as having a high (1 to 3 points), moderate (4 to 6 points), or low (7 to 9 points) risk of bias. Discrepancies were resolved by discussion with a third reviewer (LS).

Prevalence and incidence of PTSD or PTSS in people with CD were calculated by pooling the study-specific estimates by using a generalized linear mixed model (GLMM). GLMM is an elaborate approach advocated especially for proportions (33). Since the effect sizes are based on continuous outcome data, the maximum-likelihood was used to estimate the GLMM (34). Prevalences and incidences are reported as percentages with corresponding 95% confidence interval (CI). Binary data were pooled as Odds Ratio (OR) by using the Bakbergenuly Sample Size Method (35) and the Paule-Mandel estimator was used to estimate the between-study variance (34). Bakbergenuly sample size method uses weights based on study-level sample size to estimate the overall effect, leading to a reduction in between-study variance bias than conventional methods (35). Estimates of PTSD or PTSS in people with CD versus without CD are reported as OR with corresponding 95% CI.

Forest plots were created to visually assess heterogeneity. Quantification of heterogeneity between studies was examined using the I² statistic -as the ratio of variance between studies to total variance in the meta-analysis- and was tested for significance by using the Q statistics. A substantial level of heterogeneity was indicated by an I² statistic value of 60% and greater (36). The larger I², the greater the heterogeneity within the meta-analysis, because in it the scatter between the studies proportionally outweighs the random scatter within the studies (36). Records were defined as outliers if their CI did not match the CI of the pooled effect and differed significantly. This is because records with a high sampling error are likely to derivate significantly from the pooled effect (37). Potential moderators of prevalence and incidence estimates were calculated using random effects meta-regressions (36). Following the recommendation of Schwarzer et al. (38) and Borenstein et al. (39), meta-regression analysis was conducted in case of ≥10 studies per outcome (subgroup). From a number of ≥2 events per subcategory of the variable, the results of the meta-regressions are described descriptively (40, 41). The influence of each study on pooled estimate results was calculated using sensitivity analysis (leave one out analysis). Publication bias was investigated by funnel plot and Egger’s test (42).

All analyses were performed using the packages tidyverse, meta and metafor in the R software version 4.0.2 (43). Statistical tests were two-sided and used a significance threshold of P < 0.05. Changes to the a priori study registration were that in addition to narrative reviews, case reports and lack of full text availability, secondary literature such as meta-analyses, systematic reviews were excluded. However, the literature from the secondary literature was used for the additional manual search. In the case of substantial heterogeneity (>60%), data were pooled and presented with reports on heterogeneity and its interpretation. Possible reasons for the statistical heterogeneity were explored. In order to use a consistent method, meta-regressions were performed when there was substantial heterogeneity in the population (I² > 60). Meta-regression analyses were performed in the case of ≥ 10 studies per outcome. The changes are shown in Supplementary Table 3.

A total of 6,103 references were identified through the electronic database searches and 26 additional studies through backward searches. After removal of duplicates, a total of 4,616 studies were reviewed; of these, 589 full-text articles were reviewed, and 64 were included in the systematic review. Reasons for exclusion of full texts were wrong population, wrong study design, language, intensive care unit, no CD or not on CD list, not only CD, no prevalence/incidence calculable, no PTSD/PTSS was measured, same population sample and no availability of full text and others (for PRISMA flow diagram see Figure 1).

Included studies were conducted in 17 different countries, predominantly in the USA (31 studies; 48%). 134 records from 60 references (n = 21,213) reported on the point prevalence of PTSD or PTSS; three records from three references (n = 913) reported on the 12-month prevalence of PTSD, and nine records of six references (n = 826) reported on the lifetime prevalence of PTSD. No references were found for the 12-month incidence of PTSD or PTSS in people with CD. The most commonly used questionnaires were Impact of Event Scale (Revised), PTSD Checklist and Post-traumatic Stress Diagnostic Scale. Structured Clinical Interview for DSM was the most frequently used structured interview. Nine different categorized CD were extracted (Tables 1, 2). Sample sizes of studies varied between 12 and 6,542 individuals. The mean age of individuals across studies was 41.14 years (SD = 8.72); 55.5% were female and 84.2% Caucasian. CD was diagnosed at a mean age of 21.52 years (SD = 15.64), from then on lasting a mean of 8.03 (SD = 4.58) years. With respect to study quality, four (6.3%) studies were at low, 53 (82.8%) at moderate, and seven (10.9%) at high risk of bias. A summary of selected study characteristics is presented in Supplementary Tables 4–6. See Supplementary Appendix 7 for references of included studies.

The most common measurement was self-report questionnaires (70.7% for PTSD, and 91.9% for PTSS); the remaining point prevalence estimates were collected by means of structured interviews. The duration of CD was 9.69 (SD = 6.08) years regarding PTSD and 5.36 (SD = 4.44) years regarding PTSS. The pooled estimate for the point prevalence of PTSD in people with CD [number of included studies (k) = 41] was 12.7% (95% CI, 8.6 to 18.4%), with substantial heterogeneity (Q = 1233.82, P < 0.001, τ² = 1.36, I² = 96.0%) (Figure 2). After eliminating 18 outliers, heterogeneity was I² = 82.2% (Q = 130.57, P < 0.001, τ² = 0.25) and pooled estimate was 11.1% (95% CI, 8.8% to 14.0%). Sensitivity analysis (k = 41) demonstrated that no individual study affected the overall prevalence estimate of PTSD for more than 1% (Supplementary Figure 8).

The pooled estimate for the point prevalence of PTSS in people with CD (k = 24) was 19.6% (95% CI, 13.2 to 28.1%), with substantial heterogeneity (Q = 635.55, P < 0.001, τ² = 1.14, I² = 95.8%) (Figure 3). After eliminating ten outliers, heterogeneity was I² = 70.2% (Q = 46.84, P < 0.001, τ² = 0.39) and pooled estimate was 17.6% (95% CI, 14.3% to 21.5%). Sensitivity analysis (k = 24) demonstrated that no individual study affected the overall prevalence estimate of PTSS for more than 2% (Supplementary Figure 9).

Meta-regressions for point prevalence of categorized CDs comorbid to PTSD (P = 0.593) or PTSS (P = 0.656) did not yield significant results. The point prevalences for comorbid PTSD in CD subcategories range from 23.6% (95% CI, 16.8 to 31.9%) for cerebrovascular disorders (k = 4) to 6.6% (95% CI, 1.9 to 20.9%) for cardiovascular diseases (k = 5), with substantial heterogeneity (Q = 1112.62, P < 0.001, τ² = 1.83, I² = 98.1%) (Table 1). The point prevalences for comorbid PTSS in CD subcategories range from 21.7% (95% CI, 11.6% to 36.8%) for malignant neoplasm (k = 13) to 13.1% (95% CI, 2.4% to 48.2%) for metabolic disorders (k = 3), with substantial heterogeneity (Q = 627.14, P < 0.001, τ² = 1.29, I² = 97.5%). Descriptively, for comorbid PTSS, the highest prevalence of 50.3% was reported for musculosketal disorders (k = 1) and the lowest prevalence of 9.4% was reported for infectious disease (k = 1) (Table 2). Meta-regressions for point prevalence of PTSS in people with CD yielded a significant result for type of PTSS instrument [structured interview (k = 2) 2.7% (95% CI, 1.0 to 6.9%) vs. questionnaire (k = 22) 22.3% (95% CI, 15.7 to 30.7%; P = 0.005)], with substantial heterogeneity (Q = 601.85, P < 0.001, τ² = 0.93, I² = 96.9%), for RoB sum score in PTSS studies [k = 24; g = −0.53 (95% CI, −0.89 to −0.16; P = 0.004)], with substantial heterogeneity (QM = 8.05, τ² = 0.96, I² = 96.4%), and for duration of CD in PTSS studies [k = 7; g = 0.10 (95% CI, 0.04 to 0.17; P = 0.002)], with substantial heterogeneity (QM = 9.55, τ² = 0.09, I² = 95.3%). These potential moderators were not a moderating factor for point prevalence in studies on PTSD, nor were the variables time of assessment of PTSD or PTSS in years after CD diagnosis or initial CD treatment, timepoint of CD diagnosis, stage of CD, age, setting, gender, ethnicity, and publication year, for both PTSD and PTSS (Table 1 for PTSD and Table 2 for PTSS).

All included studies assessed 12-month PTSD prevalence using a structured interview; studies assessing PTSS were not available. The pooled estimate for the 12-month prevalence in people with CD (k = 3) was 8.75% (95% CI, 5.5% to 13.5%), with substantial heterogeneity (Q = 14.20, P < 0.001, τ² = 0.38, I² = 86.1%) (Figure 4). No outliers were detected. The number of studies was too small (≤ 10 studies) to calculate meta-regressions. Descriptively, an estimate for specific CDs of 14.3% for nervous system diseases (k = 1) and a weighted mean of 6.5% for malignant neoplasms (k = 2) can be reported.

All studies assessed lifetime PTSD prevalence using a structured interview; studies assessing PTSS were not available. The pooled estimate for lifetime prevalence of PTSD in people with CD (k = 6) was 12.1% (95% CI, 7.6 to 18.5%), with substantial heterogeneity (Q = 35.02, P < 0.001, τ² = 0.30, I² = 84.4%) (Figure 5). After eliminating one outlier, the heterogeneity was I² = 72.3% (Q = 13.21, P < 0.01, τ² = 0.45), and the pooled estimate was 10.2% (95% CI, 6.4 to 15.7%). Sensitivity analysis (k = 6) demonstrated that no individual study affected the prevalence estimate of PTSD for more than 2% (Supplementary Figure 10). The number of studies was too small (≤ 10 studies) to calculate meta-regressions. Descriptively, weighted means for comorbid PTSD and specific CDs showed a lifetime prevalence of 11.0% for malignant neoplasms (k = 2), 9.9% for musculoskeletal disorders (k = 2), and 15.5% for nervous system diseases (k = 2).

All studies on the comparison between PTSD or PTSS in people with versus those without CD assessed PTSD, with 87.5% using self-report questionnaires; studies assessing PTSS were not available. Meta-analytic pooling (k = 8) yielded an OR of 7.09 (95% CI, 2.49 to 20.17) (Supplementary Figure 11), with substantial heterogeneity I² = 88.4% (Q = 60.51, P < 0.001, τ² = 10.05). After eliminating five outliers (k = 3) the heterogeneity decreased to a moderate I² = 36.6% (Q = 6.31, P = 0.178, τ² = 0.71) and the estimate yielded an OR of 9.96 (95% CI, 2.55 to 38.94) (Figure 6). The number of studies was too small (≥ 10 studies) to calculate meta-regressions. Descriptively, people with versus those without specific CD showed increased risk for PTSD of OR = 2.51 and OR = 3.23 for malignant neoplasm (k = 2), OR = 26.45 for musculoskeletal disorders (k = 1), and OR = 2.08 to 61.91 for nervous system diseases (k = 5).

The Egger’s test for point prevalence estimates of PTSD (Intercept = 0.91 (95% CI, −1.4% to −3.2%, P = 0.45) (Supplementary Figure 12) and PTSS (Intercept = 0.78 (95% CI, −2.7% to −4.3%, P = 0.67) (Supplementary Figure 13) in people with CD does not indicate the presence of funnel plot asymmetry and thus publication bias. The number of studies of 12-month and lifetime prevalence of PTSD in people with CD were too small to test for small-study effects with Egger’s test assessing publication bias.

When interpreting the results some limitation of our systematic review and meta-analysis should be considered. Random error is a general problem despite the total number of 22,952 subjects on which the analysis was based. Access to the original study data and thus to epidemiologic data was not available. Study populations were found mainly from industrialized nations and only German or English language studies were included, which limits the world-wide generalizability. Although this meta-analysis included different categories of CD, most studies included people with malignant neoplasms, cardiovascular diseases, musculoskeletal disorders or nervous system diseases, so the results are dominated by these CD categories. Causal attributions between CD and PTSD or PTSS cannot be drawn based on the data.