Analysis of risk factors for bleeding after gastric endoscopic submucosal dissection with an emphasis on age and sex differences: A nationwide population-based study.

Materials and methods

Data sources and study design
We conducted a population-based, retrospective cohort study utilizing a customized Korean National Health Insurance Service (NHIS) claims database, which covers the entire Korean population. The NHIS database contains a wide range of data, including demographic information, healthcare utilization, diagnostic codes (International Classification of Diseases, 10th Revision [ICD-10]), procedures, prescriptions, and medical expenditures. This extensive coverage makes it one of the most inclusive and robust global resources, enabling comprehensive analyses of health conditions and outcomes for epidemiological research [12]. Additionally, we integrated data from the National Health Screening Database, which is part of the NHIS. The National Health Screening Program is a government-initiated health screening initiative aimed at early detection of health risk factors and diseases. The health check-up data includes results from standardized health examinations, lifestyle questionnaires, and laboratory tests.

Study population
Patients who underwent ESD for gastric cancer or adenoma between November 1, 2011 and December 31, 2022 were identified using the relevant diagnostic and procedural codes in the NHIS database, as previously defined (Supplementary Table 1) [4, 13]. Inclusion criteria encompassed all adult patients (aged 19 years and older) who had their gastric ESD procedure during the study period. Patients were excluded if they had a diagnosis of gastric cancer or adenoma more than one year before the index ESD, were under 19 years old, had incomplete eligibility information, underwent ESD at multiple institutions on the same day, were treated at institutions performing fewer than 10 procedures annually, had no hospitalization records on the procedure day, or lacked health check-up data within two years from the procedure.
The study was exempt from IRB review (No. 2307–014-19479) by the Institutional Review Board of Chung-Ang University Hospital as it involved only de-identified, publicly available data with minimal risk to participants. All data were accessed and analyzed in accordance with ethical standards and institutional guidelines. Patient consent was waived due to the aforementioned reasons.

Study outcomes, variables and definitions
A bleeding event was defined as any occurrence within 30 days following gastric ESD, identified by cross-referencing diagnosis and procedure codes with treatment and admission records (Supplementary Table 2). This operational definition was previously validated [4, 13].
Potentially confounding variables comprehensively included various factors, comprising basic demographic factors (age, sex, socioeconomic status [SES]), tumor characteristics (histology and lesion multiplicity), medical information (comorbidities, Charlson Comorbidity Index [CCI], and medications), and lifestyle behaviors (smoking history, drinking habits, and body mass index [BMI]).
Comorbidities were defined using ICD-10 codes [14] in admission and outpatient diagnosis within 1 year prior to the procedure (listed in Supplementary Table 3). The CCI was derived by adding up the weights assigned to 17 comorbidities according to Quan’s algorithm. The use of medication usage was investigated for aspirin, P2Y12 receptor inhibitors, other antiplatelet agents, warfarin, direct oral anticoagulants, nonsteroidal anti-inflammatory drugs (NSAIDs), cyclooxygenase-2 (COX-2) inhibitors, and steroids. Health-related behaviors, including BMI, smoking status, and drinking habits, were assessed using health check-up data, and classified based on official guidelines [15, 16]. Detailed classification criteria are provided in Supplementary Table 4.

Statistical analysis
Baseline characteristics are presented as number (percentage) for categorical variables, mean ± standard deviation for continuous variables. Mixed-effect logistic regression models with a 2-level (patient and hospital) hierarchical structure were used to estimate odds ratios (ORs) and 95% confidence intervals (CIs) for bleeding events. To estimate the OR for bleeding events for each risk factor, we employed multivariable models adjusted for all other variables. To avoid multicollinearity, we did not include both CCI and its component comorbidities in the same regression model simultaneously. When estimating the adjusted ORs for individual comorbidities included in the CCI calculation, CCI was excluded from the model, and conversely, the individual comorbidities were omitted when assessing the adjusted OR for CCI. To identify risk factors specific to different age groups, we employed models incorporating interaction terms between the age group and each variable. Relative ORs comparing non-elderly (age < 75 years) to elderly patients (age ≥ 75 years) were obtained using interaction terms between each risk factor and the age group. We selected 75 years as the age cutoff to reflect recent trends in the aging population and evolving clinical practice. As the proportion of elderly patients undergoing ESD continues to increase, many recent studies, as well as elderly-specific criteria such as the EL classification in the eCura system, [17] now commonly use 75 years as the threshold to define the elderly patients undergoing gastric ESD. [18, 19] To examine sex differences in risk factors, we employed interaction terms between sex and each risk factor to determine the female-to-male relative ORs. A p-value < 0.05 was considered statistically significant. Statistical analyses were performed using SAS Enterprise Guide 7.1 (SAS Institute Inc., Cary, NC).

Results

Baseline characteristics of the study subjects
During the study period, 164,968 ESD cases were included in the final analysis (Fig. 1). The mean age of the enrolled participants was 64.8 years, and 70.4% were male. The elderly group (age ≥ 75 years) represented 18.6% of all cases. Patient demographics and baseline characteristics are summarized in Table 1. The mean age was 78.8 years in the elderly group and 62.4 years in the non-elderly group. The male predominance was less pronounced in the elderly group. The elderly had a higher mean CCI, with gastric cancer more prevalent than adenoma. Comorbidities, medications associated with bleeding, and underweight status were more common in the elderly, while obesity, smoking, and drinking were less frequent.

Demographic trends of patients undergoing gastric ESD
The number of procedures steadily increased over time across all age groups. Notably, the proportion of patients aged 75 years and older showed a marked increase, from 14.2% in 2012 to 24.7% in 2022 (Fig. 2A, Supplementary Table 5). In contrast, there was no notable change in the sex distribution over the study period (Fig. 2B, Supplementary Table 5).

Incidence of post-procedural bleeding following gastric ESD
The overall incidence of bleeding was 4.74%. In the elderly group, the incidence was significantly higher at 5.36% compared to 4.60% in the non-elderly group, although this significance disappeared after adjustment (Table 2). When considering age and sex simultaneously, the bleeding incidence was significantly lower in both the non-elderly female group (3.15%) and the elderly female group (4.37%) than in the non-elderly male group (5.16%), with the significance remaining after adjustment. Conversely, the elderly male group showed a higher bleeding risk (5.90%), although the significance disappeared after adjustment.

Risk factors for post-procedural bleeding
In the multivariable analysis, risk factors included younger age, male sex, high CCI (≥ 2), gastric cancer, resection of multiple gastric lesions, various comorbidities (hypertension, diabetes, other cancers, angina, liver cirrhosis, renal failure with or without hemodialysis, heart failure, and cerebral infarction), and use of antithrombotic agents (aspirin, P2Y12 inhibitors, warfarin, and DOACs), NSAIDs, and Cox-2 inhibitors (Table 3). Underweight, current smoking, and heavy drinking were also associated with increased bleeding risks, while overweight or obesity were associated with lower risks.
Among the comorbidities, renal failure undergoing hemodialysis showed the highest risk among all factors (aOR: 7.793, 95% CI: 6.138–9.896). Notably, the bleeding risk increased with higher dialysis frequency (p for trend < 0.0001; Supplementary Table 6). Regarding medications, anticoagulants had higher OR values compared to antiplatelet agents, with warfarin showing the highest risk (aOR: 2.771, 95% CI: 2.247–3.416).

Risk of post-procedural bleeding in elderly and non-elderly patients
Next, we analyzed the risk factors for bleeding separately in the elderly and non-elderly groups to assess differences in their patterns. Most risk factors were common to both elderly and non-elderly groups, with some exceptions (Table 4, Supplementary Table 7). Specific risk factors for the elderly group included diabetes mellitus, cerebral infarction, and the use of COX-2 inhibitors and steroids. Conversely, intracerebral hemorrhage, current smoking, and alcohol consumption were specific risk factors in the non-elderly group. In addition, obesity reduced bleeding risk in both groups, while overweight was protective only in the non-elderly group.
Some factors demonstrated significantly different impacts on bleeding risk between age groups. In the elderly group, the risk associated with a high CCI (≥ 5) (ratio of aOR: 1.209, 95% CI: 1.008–1.451) and DOAC (ratio of aOR: 1.399, 95% CI: 1.050–1.863) increased (Table 4). Conversely, male sex (ratio of aOR: 0.825, 95% CI: 0.725–0.939), current smoking (ratio of aOR: 0.812, 95% CI: 0.673–0.979), and heavy drinking (ratio of aOR: 0.794, 95% CI: 0.653–0.965) showed a diminished effect in the elderly group.

Sex differences in the risk of post-procedural bleeding with sensitivity analyses
We further analyzed sex differences in risk factors in elderly and non-elderly groups. Significant sex differences were observed only in the non-elderly group, with gastric cancer (ratio of aOR: 1.163, 95% CI: 1.022–1.325), warfarin use (ratio of aOR: 2.010, 95% CI: 1.145–3.527), and heavy drinking (ratio of aOR: 1.265, 95% CI: 1.010–1.585) showing a greater impact on bleeding risk in females (Fig. 3), while SES (bottom 25% vs. top 25%) (ratio of aOR: 0.769, 95% CI: 0.636–0.929) and obesity (ratio of aOR: 0.744, 95% CI: 0.638–0.867) had a stronger protective effect. These differences were no longer significant in the elderly group. Detailed univariable and multivariable results, including all variables, are provided in Supplementary Table 8.
The pronounced sex differences in the younger cohort prompted further analysis of bleeding risk by lowering the age cutoff in 5-year increments to 60 years. SES (bottom 25% vs. top 25%), gastric cancer, and obesity consistently showed significant differences across all analyses (Table 5). Notably, gastric cancer and obesity exhibited a significantly increasing influence of sex differences in younger patients (p = 0.0446 and 0.0385, respectively), while SES (bottom 25% vs. top 25%) showed a similar trend without reaching statistical significance (p = 0.0671, Fig. 4).

Discussion
In this large-scale study on the risk factors for bleeding following gastric ESD, we utilized comprehensive national health data to identify key risk factors. Our findings revealed not only that risk of post-ESD bleeding varies by age and sex but also that the types and impact of individual risk factors differ significantly across these demographic groups.
The identified risk factors encompassed various clinical, tumor, and lifestyle characteristics. Notably, our study corroborates previous findings that demonstrate an increased bleeding risk associated with younger age and male sex [6, 7]. Although the elderly group exhibited a higher crude incidence of bleeding, this association was attenuated after adjustment, likely due to the greater prevalence of comorbidities and more frequent use of antithrombotic agents in this population rather than age itself. This inverse correlation between age and bleeding risk, consistent with prior research, may relate not only to physiological differences across groups but also to the possibility that certain risk factors exert differential effects by age [7]. These considerations prompted us to investigate these risks further across age and sex subgroups. Tumor-related factors, including gastric cancer and lesion multiplicity, were associated with increased bleeding risk. The higher bleeding risk in gastric cancer compared to adenoma is well-recognized and is likely attributed to increased angiogenesis and hypervascularity in EGC [20]. A high CCI and various comorbidities increased bleeding risk, with renal failure— particularly in patients undergoing hemodialysis—posing the greatest risk, surpassing even anticoagulant use. Chronic renal failure requiring hemodialysis is a well-established risk factor, likely attributable to low albumin levels and dialysis-related anticoagulation and platelet disturbances [21]. Patients with multiple or severe comorbidities were more prone to post-ESD bleeding. Conditions associated with micro- and macrovascular inflammation, such as organ failure or other cancers, may increase delayed bleeding risk due to impaired ulcer healing and hemostasis process [21–23]. Antithrombotic use was another significant risk factor, with warfarin and DOACs carrying the highest risk, followed by P2Y12 inhibitors and aspirin. This finding aligns with current guidelines, which recommend discontinuing anticoagulant therapy while continuing aspirin use [24]. NSAIDs and COX-2 inhibitors increase bleeding risk, albeit modestly. This effect may stem from COX-1 inhibition, impairing platelet function and mucosal healing [25]. Lifestyle factors further contributed to bleeding risk. Intriguingly, being underweight significantly increased bleeding risk, while being overweight or obese appeared protective. Obesity has been associated with thrombosis through mechanisms such as adipocytokine release, increased activity of the coagulation cascade, and decreased activity of the fibrinolytic cascade [26]. Although debated, the obesity paradox has been observed in various diseases [27]. Previously, obesity has been shown to reduce major bleeding risk in anticoagulant users, whereas being underweight increases it [28]. Additionally, smoking and alcohol intake exhibited a dose–response pattern, with current smoking and heavy drinking amplifying bleeding risk. These findings highlight the significant impact of lifestyle factors on bleeding risk.
Elderly patients showed a higher incidence of bleeding, likely due to confounding conditions such as comorbidities and antithrombotic use. This finding indicated the need for age-specific risk assessments. Notably, high CCI and DOAC use exerted a significantly stronger impact in the elderly, whereas the influence of male sex, smoking, and drinking was less pronounced. The greater impact of high CCI in elderly patients may reflect the additive effect of multiple or severe comorbidities on frailty and physiological vulnerabilities. This could impair tissue repair, ulcer healing, and hemostatic processes [29]. Similarly, the amplified risk with DOAC use likely reflects age-related pharmacokinetic and pharmacodynamic changes. Dose adjustment is recommended in elderly patients due to reduced clearance, increased accumulation, and variability in plasma concentration [24, 30]. On the other hand, the effects of smoking and drinking were diminished in elderly patients, which might be due to reduced metabolic response or lower intake of these substances, though this remains unconfirmed. Male sex showed a diminished effect in the elderly, highlighting marked sex-based differences in the non-elderly group, prompting further analysis.
Significant sex differences in bleeding risk, with a more pronounced influence of various factors in females exclusively in the non-elderly group, were observed. This finding suggests that age may modulate sex-based variations. Previous studies indicated age-dependent sex differences in treatment effect, such as increased stroke risk from hormone therapy in women and reduced cardioprotective effect of aspirin in younger women, paralleling our findings [9]. In our study, gastric cancer, warfarin use, and heavy drinking significantly increased bleeding risk in females, while low SES and obesity had stronger protective effects compared to males. Bleeding risk variations between sexes might be associated with sex-related differences in clinicopathological characteristics of EGC, [31–33] drug metabolism, [34–36] health behaviors, and obesity-related pathophysiology. The lower bleeding risk associated with low SES in females suggests that socio-environmental factors may also influence procedural outcomes. Sex differences in bleeding risk for SES, gastric cancer, and obesity showed stronger trends in younger patients. These findings underscore complex sex- and age-related interactions, likely mediated by hormonal factors, warranting further mechanistic research.
This study has several strengths, including the inclusion of diverse variables—such as health behaviors, tumor characteristics, and medication usage—beyond basic demographics. Incorporating tumor characteristics like gastric cancer type and lesion multiplicity helped address common limitations of claims data. Detailed lifestyle data from health screenings enabled us to assess the effects of smoking, alcohol use, and BMI on bleeding risk. The use of a national dataset spanning over 10 years, one of the largest of its kind, along with validated operational definitions, strengthens the reliability of the study results. Moreover, this study highlights the growing proportion of elderly ESD patients, emphasizing the relevance and timeliness of the study.
However, the study has limitations. First, claims data may lack lesion and procedure-specific details, although we tried to address this by incorporating available data on histology and lesion multiplicity. Additionally, unmeasured confounders, such as frailty, menopausal status, and childbirth history, could impact outcomes across age and sex groups. However, we sought to mitigate these shortcomings by including extensive information on comorbidities, BMI, health behaviors, and medications, to adjust for these potential confounders. Second, there could be potential misclassification of outcomes, as outcomes were not directly extracted from individual medical records. Nonetheless, we used validated operational definitions, which offer enhanced accuracy compared to studies relying on unvalidated definitions. Third, although the prescription date of antithrombotic agents is available, the administrative claims data do not capture the exact date of resumption after temporary discontinuation. While we cautiously assume general adherence to guideline recommendations [37]—which advise early resumption of antithrombotic therapy after confirming adequate hemostasis and absence of post-procedural bleeding—this data limitation may have impacted the accuracy of drug exposure assessment and the study results. Lastly, although our analysis demonstrates significant associations between bleeding risk and various clinical factors, the absence of a validated prediction model restricts the clinical applicability of these findings. Future studies should focus on developing a comprehensive risk stratification tool that incorporates relevant clinical variables to enhance the accuracy and predictive value of bleeding risk assessments. Such models could improve personalized patient care, guiding more tailored decision-making in clinical practice.
In conclusion, this nationwide real-world analysis identified diverse risk factors for post-ESD bleeding, highlighting age- and sex-specific effects. These factors should be considered for personalized risk prediction, decision-making, and post-procedural management, particularly in the elderly.

Supplementary Information
Below is the link to the electronic supplementary material.