We conducted a thorough literature review following the Preferred Reporting Items for Systematic Reviews and Meta-analyses guidelines (PRISMA guidelines) (14). A computerized search was carried out on PubMed, Embase, Web of Science, and Cochran Library databases to identify relevant studies published from the time of inception to March 2025. The search utilized the following terms to ensure comprehensiveness:(((((((stomach[Title/Abstract]) OR (gastric[Title/Abstract])) OR ((esophagus[Title/Abstract]) OR (esophageal[Title/Abstract]))) OR (((((colon[Title/Abstract]) OR (colonic[Title/Abstract])) OR (colorectal[Title/Abstract])) OR (rectum[Title/Abstract])) OR (rectal[Title/Abstract]))) OR ((((liver[Title/Abstract]) OR (hepatic[Title/Abstract])) OR (hepatocellular[Title/Abstract])) OR (intrahepatic bile duct[Title/Abstract]))) OR ((pancreatic[Title/Abstract]) OR (pancreas[Title/Abstract]))) AND ((((carcinoma[Title/Abstract]) OR (neoplasms[Title/Abstract])) OR (cancer[Title/Abstract])) OR (tumor[Title/Abstract]))) AND ((suicide[Title/Abstract]) OR (suicides[Title/Abstract])). A detailed outline of the process of study selection is presented in Supplementary Table 1.

The eligibility criteria for this review were determined based on the PICOS framework as follows:

Only original studies conforming to the Population, Intervention, Control, and Outcome (PICO) framework were eligible for inclusion. Studies based on case reports and case series were excluded.

The inclusion criteria were as follows: (a) population characteristic: patients with digestive cancers: (b) outcomes: the statistical result is SMR of patients with cancer of the digestive system; (c) study type: all observational research (retrospective or prospective); (d) SMR shows no correlation between cancer patients and general population; (e) studies utilizing appropriate statistical methods for analyses and having sufficient data. The exclusion criteria were the following: (a) The number of patients is less than 100; (b) risk ratios other than SMR/HR; (c) The findings were specific to patients’ suicidal intentions. (d) confidence intervals ≠ 95%.

After removing duplicates, the initially identified articles underwent screening based on their titles and abstracts. If the abstracts seemed to meet the predetermined inclusion and exclusion criteria, a full-text assessment was conducted. Two reviewers (Wang and Yang) independently carried out the literature search and study selection using a standardized protocol. Any discrepancies between reviewers were resolved through consensus.

The following data were extracted from studies selected by two authors: (a) Study characteristics: author name, country, year of publication, study type, and year of inclusion. (b) Patient characteristics: total number of patients, gender, region. (c) Tumor characteristics: tumor type, tumor prognosis, years after diagnosis. Any discrepancies between authors were resolved through consensus.

The Newcastle-Ottawa Standard (NOS) checklist was employed to evaluate the quality of the articles included in Supplementary Table 2. Each study underwent scoring by two independent reviewers, and any disparities were resolved through discussion and consensus with a third author.

As outcome measure, we used SMRs as RR estimates. In the absence of SMR data, we calculated the SMR values by comparing the observed number of suicides to the expected number in each country where the studies were conducted. Subsequently, we aggregated the observed and expected cases within each subgroup to determine the overall SMR. In order to minimize patient overlap, we opted for studies with the widest range of years of coverage and the largest sample size, and avoided overlapping follow-up years. Statistical analysis was conducted utilizing STATA v.12 software and R statistical software, the meta for packages, with a 2-sided 5% significance level. Meta-analytic pooling was executed exclusively when three or more studies were accessible. Meanwhile, we employed subgroup analysis and random-effects meta-regression to assess the potential changes in suicide rates among cancer patients according to cancer site, cancer prognosis, gender, geographic region, diagnostic period and the year of recruitment. Heterogeneity was assessed utilizing the Cochran Q-test and I² statistics. A random-effects model was employed to amalgamate effect indicators in instances where I² ≥ 50% and P < 0.05. Conversely, a fixed-effects model was utilized. Publication bias was examined when the analysis included ten or more studies.

We initially retrieved 2836 records, of which 1720 remained after excluding duplicates. Following title and abstract screening, 80 records were selected. Among these, 60 did not meet the inclusion criteria, resulting in 20 cohort studies for further analysis. Considering the potential overlap of patients in studies using the same database with recruitment periods and tumor types, we excluded eight records based on recruitment years, retaining a core set of 12 independent studies to mitigate the impact of patient overlap on meta-analysis outcomes. The flowchart of the literature selection process is depicted in Figure 1. In these twelve retrospective cohort studies, the results were consistently presented using standardized mortality ratios (SMRs), encompassing a total of 6,763,452 patients diagnosed with digestive system cancers, with follow-up periods ranging from 1960 to 2019. Among the total cohort of 6,763,452 samples of included studies, specifically, five studies incorporated both male and female participants, whereas only two studies were exclusively designed to examine female cohorts, for a total sample size of > 766,474 males and > 452,502 females. Geographically, the included studies spanned across Europe, the United States, and Asia, focusing on five major types of digestive system cancers: esophageal, hepatic, pancreatic, colorectal, and gastric cancers. Collectively, these studies sought to maximize representativeness by covering diverse geographic locations, balanced gender participation, and multiple types of digestive system cancers. The characteristics of the 12 retrospective cohort studies are presented in Table 1. All 12 studies were classified as high quality according to the Newcastle-Ottawa Standard (NOS).

Our comprehensive analysis revealed that the SMR among patients with digestive system cancer was 2.25 times higher than that of the general population (SMR = 2.25, 95% CI = 1.89–2.66, P < 0.000), accompanied by substantial heterogeneity (I² = 97.4%, P < 0.0001) in Figure 2. As the SMRs for patients with esophageal cancer, liver cancer, pancreatic cancer, gastric cancer, and colorectal cancer have been recorded together in certain studies. To obtain an integrated SMR for each cancer, we analyzed the results of these studies separately by cancer type in Supplementary Figure 1. The SMRs were significantly increased for esophageal (SMR = 3.35, 95% CI = 2.37-4.75, P < 0.000) (10b, 9, 21, 23), pancreatic (SMR = 3.87, 95% CI = 2.33-6.41, P < 0.000) (10, 24b, 21, 23), stomach (SMR = 2.102, 95% CI = 1.81-2.43, P < 0.000) (10, 17, 18b, 21–23), liver (SMR = 1.89, 95% CI = 1.38-2.59, P = 0.066) (10, 18b, 21, 23) and colorectum cancers (SMR = 1.57, 95% CI = 1.36-1.81, P < 0.000) (15, 20a, 10, 18b, 16, 19, 21, 23). To pinpoint potential sources of heterogeneity, we conducted further subgroup analyses. And the summary of subgroup analyses of suicide in patients is presented in Figure 3 and Table 2.

The prognosis and time since diagnosis of digestive system tumors significantly influence patients’ emotional states and survival prospects. Given the generally poor treatment outcomes for most digestive system cancers, we categorized these cancers based on 5-year relative survival rates, defining them as having intermediate prognosis (5-year survival rates between 50–90%) or poor prognosis (5-year survival rates <50%) (25). Subgroup analysis by prognosis revealed notable differences in SMRs among patients (I² = 84.7%, P value for heterogeneity = 0.004) in Supplementary Figure 2. Specifically, for cancers with intermediate prognosis, such as colorectal cancer, SMR was reduced (SMR = 1.57, 95% CI = 1.36-1.81, P < 0.000). Conversely, for cancers associated with poor prognosis, including esophageal, liver, pancreatic, and gastric cancers, SMR was higher (SMR = 2.55, 95% CI = 1.89-3.42, P = 0.01), with lower heterogeneity (I² = 73.8%; P value for heterogeneity = 0.01, heterogeneity P < 0.05). Furthermore, significant statistical differences were also identified in the subgroup analyses stratified by time since diagnosis, as illustrated in Supplementary Figure 3. The included studies were categorized into two cohorts: those within the first year following diagnosis and those beyond the one-year post-diagnosis period. The SMR within the first year after diagnosis was significantly higher than more than one year after diagnosis, with high heterogeneity between studies (I² = 93%; P value for heterogeneity < 0.000). A multivariable meta-regression demonstrated that data pertaining to poor-prognosis cancers were redundant to suicide mortality within the first-year post-diagnosis, with poor prognosis emerging as the predominant contributing factor.

The SMRs among male patients (SMR = 2.31, 95% CI = 1.90-2.79, P < 0.000) diagnosed with digestive system cancer exhibited a pronounced elevation compared to that among female patients (SMR = 1.97, 95% CI = 1.56-2.50, P < 0.000) relative to their respective general populations, but the results did not reach statistical significance (P value for difference = 0.316; I² = 87.6%; P value for heterogeneity < <0.000) in Supplementary Figure 4. Furthermore, we observed that suicide mortality among patients with digestive system cancers was higher in the United States (SMR = 2.79, 95% CI = 1.50-5.21, P < 0.000) surpassed those in Europe (SMR = 1.85, 95% CI = 1.56-2.21, P = 0.463) and Asia (SMR = 1.96, 95% CI = 1.58-2.42, P = 0.004) in Supplementary Figure 5. Similar variations in suicide mortality were also identified across cohorts with different years of recruitment, as illustrated in Supplementary Figure 6. Patients recruited before the year 2000 (SMR = 2.89, 95% CI = 1.57-5.33, P < 0.000) exhibited noticeably higher suicide mortality compared to those diagnosed after 2000 (SMR = 1.84, 95% CI = 1.67-2.03, P = 0.196). Likewise, no statistically meaningful variations in SMRs were identified across gender (P value for difference = 0.316) and year of recruitment (P value for difference = 0.153), consistent with the absence of regional disparities.

The risk factors were also further categorized based on severity of suicide mortality (low, moderate, and high), as presented in Table 3.

We performed a sensitivity analysis utilizing a random-effects model on the 12 eligible studies identified in the systematic review, encompassing a cumulative sample size of 6,763,452 cancer patients in Supplementary Figure 7. From methodological, statistical, and clinical perspectives, heterogeneity is primarily associated with patients’ sex, geographic region, and disease prognosis. Notably, prognosis exerts a significant effect on suicide risk: the poorer the prognosis, the greater the physical, psychological, and economic burdens borne by patients, thereby increasing their susceptibility to suicide. The analysis produced coherent and stable findings. While sample overlap is conceivable, notably within SEER studies, the findings remain consistent with our initial analysis. Specifically, the standardized mortality ratio (SMR) was 2.25 (95% CI = 1.89–2.66) in the analysis based on 12 studies, compared to 1.95 (95% CI = 1.74–2.18) based on 9 studies.

A funnel plot displayed slight asymmetry concerning the pooled SMR across all 12 eligible studies, irrespective of potential sample overlap in Supplementary Figure 8. Nevertheless, both Begg’s correlation test and Egger’s regression test showed inadequate evidence to support the presence of publication bias (Begg’s test: P-value = 0.945; Egger’s test: P-value = 0.407).

Through meta-analysis of 12 high-quality cohort studies, we have discovered that the overall SMR among patients with digestive system cancer is significantly higher than that of the general population, reaching more than twice the level. Especially, our study revealed a 3.9-fold elevation in suicide mortality among patients within the first year after cancer diagnosis and for those with a poor prognosis, their SMR also increases by 2.5 times. Previously, meta-analyses specifically focusing on patients with digestive cancers were nearly scarce, or there has been a lack of comprehensiveness, differences in sample size, and statistical capabilities (26). There was only one study on suicide among patients with digestive system cancers, and that was based on data from the SEER study (27). This study yielded valuable insights into the risk of suicide within this population. Nonetheless, this study was limited to patients in the United States and the follow-up period was too short. To comprehensively grasp the multiple factors contributing to suicide, it is imperative to obtain data from more countries and over an extended duration of time. In contradistinction, our meta-analysis transcended the confines of those previous analyses, encompassing a broader array of databases for the literature search. We diligently computed missing SMRs wherever data were accessible, augmenting our analytical rigor. Furthermore, we undertook supplementary analyses, including meta-regression, interaction analyses, and sensitivity analyses, thereby enriching the depth and breadth of our investigation.