Long-term lymphoma risk in thiopurine-treated inflammatory bowel disease: A 15-year propensity-matched cohort study

Abstract

BACKGROUND

Thiopurines remain widely used in inflammatory bowel disease (IBD) for maintenance of steroid-free remission, particularly in resource-limited settings. However, long-term use has been associated with an increased risk of lymphoma. The temporal dynamics of this risk, including the duration of safe exposure, remain incompletely defined.

AIM

To evaluate the time-dependent risk of lymphoma in IBD patients treated with thiopurines, using 5-aminosalicylates (5-ASA), a non-immunosuppressive agen as an active comparator.

METHODS

Using a large, United States-based electronic health record network (TriNetX), we identified adult IBD patients initiating thiopurines or 5-ASA. After applying strict exclusion criteria and 1:1 propensity score matching (demographic characteristics and baseline clinical variables), lymphoma incidence was assessed at serial follow-up intervals (1 year, 3 years, 5 years, 10 years, and 15 years). Odds ratios (ORs) were calculated using logistic regression, with additional multivariable Cox modelling.

RESULTS

Among 38781 matched patients per group, no significant difference in lymphoma risk was observed at 1 year [OR = 1.39; 95% confidence interval (CI): 0.84-2.29; P = 0.19]. Risk became significantly elevated at 3 years (OR = 2.37; 95%CI: 1.54-3.64; P < 0.001) and remained elevated through 15 years (OR = 1.93; 95%CI: 1.46-2.54). “Subtype analysis revealed that the increased risk was primarily driven by non-Hodgkin lymphoma (3-year OR = 2.43; 95%CI: 1.57-3.78; 15-year OR = 1.95; 95%CI: 1.46-2.60) and B-cell lymphoma (3-year OR = 2.43; 95%CI: 1.54-3.83; 15-year OR = 1.85; 95%CI: 1.38-2.47)”.

CONCLUSION

Thiopurine use in IBD appears safe with respect to lymphoma during the first three years of therapy. However, risk increases thereafter, supporting limited-duration use and individualized risk-benefit assessments during long-term maintenance planning.

Key Words: Inflammatory bowel diseases; Thiopurine; Lymphoma; Immunosuppressive agents; Risk assessment; Advanced therapy; Drug-related side effects and adverse reactions

Core Tip: Thiopurines remain central to maintenance therapy in inflammatory bowel disease (IBD), yet concerns about lymphoma risk complicate long-term use. This large, propensity-matched cohort study clarifies the temporal dynamics of malignancy risk, using 5-aminosalicylates as a safe active comparator. We demonstrate that thiopurines are not associated with increased lymphoma risk during the first three years of therapy, with risk elevation emerging only after prolonged use. These findings support a time-dependent, immune suppression-linked oncogenic mechanism and provide a rational basis for limited-duration thiopurine therapy with individualized risk stratification to maximize safety in IBD care.

INTRODUCTION

Inflammatory bowel disease (IBD) encompasses two chronic, immunologically mediated disorders of the gastrointestinal tract, Crohn’s disease (CD) and ulcerative colitis (UC)[1]. Both conditions are characterized by relapsing-remitting inflammation driven by dysregulated mucosal immunity, often requiring long-term medical therapy to achieve and maintain remission[2]. Over the past several decades, thiopurine immunomodulators azathioprine and 6-mercaptopurine have been a cornerstone of IBD management[3]. Thiopurines are effective for maintaining steroid-free remission in both CD and UC, thereby reducing corticosteroid exposure and its attendant toxicities. In many regions, especially resource-limited settings, the low cost and wide availability of thiopurines make them an indispensable maintenance strategy[3]. Even in the era of biologic therapies, thiopurines remain widely used as monotherapy or in combination regimens to sustain remission and prevent disease complications.

However, the long-term safety of thiopurines has increasingly come under scrutiny, particularly regarding potential malignancy risk[4]. Foremost among these concerns is an elevated risk of lymphoma associated with thiopurine use in IBD. Multiple lines of evidence have linked thiopurine exposure to lymphoproliferative disorders. Moreover, the evidence for increased risk of lymphoma is contradictory with few studies reporting no increased risk of lymphoma[5].

Despite the growing body of evidence, important gaps persist in our understanding of thiopurine-related lymphoma risk. Another limitation is the choice of comparator groups in observational studies. Many investigations have either compared thiopurine-exposed IBD patients to the general population or to IBD patients not receiving immunomodulators. The former approach may overestimate risk by failing to account for IBD disease activity/severity, while the latter can be confounded if “unexposed” patients differ systematically [for instance, having milder disease often managed with 5-aminosalicylates (5-ASA)]. Indeed, 5-ASA agents are often used in milder IBD and have no known association with lymphoma, making them a potentially ideal active comparator. Yet few studies to date have employed an active comparator design, and even fewer with 5-ASA specifically. Consequently, critical questions remain about the long-term malignancy risk attributable to thiopurines per se vs underlying IBD or concomitant therapies. Robust data are needed to determine whether continued thiopurine use beyond 5 years or 10 years confers additional lymphoma risk, and whether that risk is meaningfully higher than in patients managed with safer maintenance strategies like mesalamine.

To address these knowledge gaps, we conducted a large real-world cohort study to rigorously evaluate the long-term risk of lymphoma in thiopurine-exposed IBD patients, using 5-ASA treatment as an ultra-safe active comparator. The primary objective was to quantify the time-dependent risk of incident lymphoma associated with thiopurine exposure at prespecified follow-up intervals (1 year, 3 years, 5 years, 10 years, and 15 years) compared with 5-ASA-treated patients. Secondary objectives included: (1) Evaluation of subtype-specific lymphoma risk (Hodgkin lymphoma, non-Hodgkin lymphoma, B-cell lymphoma, and T/natural killer-cell lymphoma); (2) Disease-stratified analyses in UC and CD; and (3) Identification of independent predictors of lymphoma using multivariable Cox proportional hazards modelling. We hypothesized that thiopurine use would be associated with a higher long-term lymphoma risk than 5-ASA use, but sought to clarify the magnitude of this risk and its temporal pattern. In doing so, our aim is to provide clinicians and patients with more definitive data on the safety trade-offs of thiopurines in the maintenance of IBD remission, informing treatment decisions in an era where therapy can be personalized according to both efficacy and safety considerations.

MATERIALS AND METHODS

Study design and data source

This was a large, multicenter, population-based retrospective cohort study conducted using the TriNetX Global Health Research Network (Cambridge, MA, United States; https://trinetx.com). TriNetX is a federated platform that consolidates de-identified electronic health record (EHR) data from approximately 147 healthcare organizations, representing over 147 million unique patients worldwide. The platform provides longitudinal patient-level data including demographics, diagnoses, procedures, medication prescriptions, and laboratory results, harmonized through standardized terminologies such as the International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM), RxNorm, Systematized Nomenclature of Medicine Clinical Terms, and International Classification of Diseases for Oncology, Third Edition (ICD-O-3).

For the analytic modules used in this study (compare-outcomes, incidence and prevalence, and Cox proportional hazards), queries were executed within the United States Collaborative Network subset of TriNetX, which comprises 65 contributing healthcare organizations depending on the specific analysis[6]. All data within TriNetX are irreversibly de-identified in compliance with the Health Insurance Portability and Accountability Act Privacy Rule. Analyses were performed within the secure TriNetX analytics interface, and only aggregate-level data were accessed through the secure TriNetX web interface, this study was deemed exempt from institutional review board approval and informed consent. The study adhered to the ethical principles of the Declaration of Helsinki and followed both the Strengthening the Reporting of Observational Studies in Epidemiology and REporting of studies Conducted using Observational Routinely collected Data guidelines (Figure 1)[7].

Figure 1 Flowchart explaining the flow of the study. EHR: Electronic health record; 5-ASA: 5-aminosalicylates; IBD: Inflammatory bowel disease; CKD: Chronic kidney disease.

Study population and setting

The study cohort comprised adults aged ≥ 18 years with IBD identified within the TriNetX United States Collaborative Network during the study period. IBD was defined using ICD-10-CM codes for UC (K51.x) and CD (K50.x). Patients with indeterminate colitis or IBD-unclassified were excluded to ensure accurate phenotypic classification and minimize diagnostic misclassification. To preserve temporal validity, patients were required to have a documented diagnosis of IBD prior to or on the date of qualifying medication exposure (Supplementary Table 1). “Eligible participants were adults aged ≥ 18 years with a confirmed diagnosis of IBD, defined using ICD-10-CM codes for UC (K51.x) and CD (K50.x), with documentation of IBD preceding or coinciding with treatment initiation. Patients were required to meet criteria for incident use of either thiopurines or 5-ASA, consistent with an incident-user design.

Exclusion criteria included: (1) Prior diagnosis of lymphoma or other hematologic malignancy; (2) Prior exposure to biologic or advanced therapies; (3) History of chemotherapy or radiotherapy; (4) Human immunodeficiency virus infection; (5) Documented Epstein-Barr virus (EBV)-related conditions; (6) Prior leukopenia; and (7) Indeterminate colitis or IBD-unclassified.

Identification of the IBD cohort was based on a validated administrative definition requiring ≥ 1 ICD-10-CM diagnostic code for CD or UC, along with a contemporaneous IBD-related prescription. This approach has demonstrated ≥ 80% positive predictive value and ≥ 85% specificity in prior claims-based studies of IBD populations. The TriNetX analytics platform, from which the study cohort was derived, has previously been employed in multiple peer-reviewed investigations of IBD, further supporting its reliability and generalizability[8]. The TriNetX database has been used previously in prior published studies in patients with IBD[9].

Patients were classified into mutually exclusive exposure groups according to an incident-user design. The thiopurine cohort included patients initiating azathioprine or mercaptopurine. The active comparator cohort comprised patients treated with 5-ASA (mesalamine, sulfasalazine, balsalazide, or olsalazine) who had no recorded exposure to thiopurines or biologic therapies at any time during the study period. The index date was defined as the earliest date at which both IBD diagnosis and qualifying medication exposure criteria were satisfied. Follow-up began one day after the index date to mitigate immortal time bias and continued until lymphoma diagnosis, last recorded healthcare encounter, or 15 years of follow-up, whichever occurred first.

To ensure evaluation of incident lymphoma and reduce baseline malignancy risk heterogeneity, patients with any history of lymphoma or other hematologic malignancy prior to cohort entry were excluded, defined using ICD-10-CM codes C81-C96 and corresponding oncology-specific histologic classifications. Patients with prior leukopenia (D72.819), prior chemotherapy exposure (TriNetX code TNX: 1002), or prior radiotherapy (SNOMED: 1287742003) were excluded to minimize confounding from pre-existing cytopenias or oncologic treatment-related malignancy risk (Supplementary Table 2).

Given the established role of EBV in lymphoproliferative disorders, individuals with documented EBV infection or EBV-related conditions prior to the index date were excluded to reduce confounding attributable to viral-driven lymphomagenesis. Patients with human immunodeficiency virus infection (ICD-10-CM: B20) were similarly excluded because of their inherently elevated baseline lymphoma risk.

This combination of restriction, incident-user design, and active comparator framework was implemented to strengthen causal inference and ensure that all lymphoma outcomes represented incident events occurring after treatment initiation in a population at baseline risk.

“The active comparator cohort comprised patients treated with 5-ASA who had no recorded exposure to thiopurines or biologic/advanced therapies at any time during the study period. Similarly, patients in the thiopurine cohort with prior or concurrent exposure to biologic or advanced therapies were excluded to ensure evaluation of thiopurine monotherapy and to minimize confounding from combination immunosuppressive regimens known to carry higher lymphoma risk”.

Exposure, outcomes, and definitions

The primary exposure of interest was thiopurine therapy, defined as initiation of azathioprine or mercaptopurine following a documented diagnosis of IBD. An incident-user framework was employed to minimize survivor bias and ensure temporal alignment between treatment initiation and outcome assessment. Patients entered the exposed cohort at the time of first recorded thiopurine prescription after IBD diagnosis. The active comparator cohort comprised patients with IBD treated with 5-ASA (mesalamine, sulfasalazine, balsalazide, or olsalazine) who had no recorded exposure to thiopurines or biologic therapies during the study period. This active comparator design was selected to reduce confounding related to healthcare engagement, diagnostic intensity, and underlying disease recognition. Exposure status was defined at baseline and analyzed under an intention-to-treat approach. Follow-up commenced one day after the index date to ensure clear temporal separation between exposure and outcome and to mitigate immortal time bias.

The database allowed construction of temporally anchored exposure-outcome relationships within matched cohorts. This functionality was used to ensure that initiation of thiopurine therapy occurred after IBD diagnosis. To confirm sustained exposure and minimize misclassification due to early discontinuation or prescription non-fulfillment, patients were required to have a second prescription of the same thiopurine agent at least one year following the index fill. This approach excluded individuals who discontinued therapy early due to adverse effects, intolerance, or insurance-related barriers, thereby ensuring inclusion of patients with a minimum exposure threshold and enhancing internal validity of long-term outcome estimates[10].

The primary outcome was incident lymphoma, defined using ICD-10-CM codes C81-C86 and corresponding ICD-O-3 histologic classifications to capture both Hodgkin lymphoma and non-Hodgkin lymphoma subtypes. These included Hodgkin lymphoma (C81), follicular lymphoma (C82), non-follicular lymphoma (C83), mature T/natural killer-cell lymphomas (C84, C86), and other specified or unspecified non-Hodgkin lymphomas (C85), along with corresponding histologic classifications (ICD-O-3 codes 959-972). Outcome ascertainment required the first occurrence of a qualifying lymphoma diagnosis after cohort entry. Patients with any documented history of lymphoma prior to the index date were excluded to ensure that all events represented incident malignancies. Given the established latency associated with lymphomagenesis, risk estimates were calculated at prespecified follow-up intervals of 1 year, 3 years, 5 years, 10 years, and 15 years following treatment initiation. This interval-based evaluation was designed to distinguish early detection effects from biologically plausible cumulative exposure-related risk.

Secondary outcomes included subtype-specific incident lymphoma endpoints, including Hodgkin lymphoma, non-Hodgkin lymphoma, B-cell lymphoma, and T-cell/natural killer-cell lymphoma, defined using standardized diagnostic and histologic codes. These analyses evaluated the incidence of specific lymphoma subtypes occurring after cohort entry. Moreover, these analyses were conducted to explore potential heterogeneity in treatment-associated risk across immunologically distinct malignancy subtypes. Leukopenia was additionally evaluated as a secondary safety outcome given its known association with thiopurine therapy and its relevance to immunologic suppression.

Pre-specified disease-stratified analyses were performed separately for UC and CD to assess effect modification by IBD phenotype. Disease subtype classification was based on corresponding ICD-10-CM codes. In multivariable time-to-event modelling, covariates were defined using structured EHR data and included demographic characteristics, metabolic and autoimmune comorbidities, laboratory parameters, and medication exposures. Hazard ratios (HRs) derived from Cox proportional hazards models were interpreted as adjusted relative risks for incident lymphoma within the longitudinal follow-up framework.

Covariables and variables

Covariates were selected a priori based on established associations with lymphoma risk, IBD severity, and potential confounding of the association between thiopurine exposure and malignancy. Variables were identified using structured EHR data recorded prior to or on the index date to preserve temporal ordering.

Demographic variables included age, sex, race, and ethnicity, given their established relationships with lymphoma incidence. Cardiometabolic comorbidities, including type 2 diabetes mellitus (E11.x), hypertension (I10.x), hyperlipidemia (E78.x), chronic kidney disease (N18.x), heart failure (I50.x), nicotine dependence (F17.x), and alcohol abuse (F10.x), were incorporated because of their association with systemic inflammation and malignancy risk. Autoimmune conditions like rheumatoid arthritis (M06.9), psoriasis (L40), systemic lupus erythematosus (M32) were included given their independent link to lymphoproliferative disorders and potential need for immunomodulatory therapy. Alcohol use disorder and nicotine dependence were included as behavioral risk factors.

Body mass index was modeled as a continuous variable and categorized using standard obesity thresholds. Laboratory markers, including glycosylated hemoglobin and lipid parameters, were included where available using clinically relevant cutoffs. Markers of IBD severity were incorporated to address confounding by indication. Systemic corticosteroid exposure at baseline was included as a surrogate for active or moderate-to-severe disease and treatment escalation. IBD phenotype (UC vs CD) was included to evaluate potential effect modification. Direct clinical indices of disease activity (e.g., CD activity index or Mayo score), endoscopic findings, and inflammatory biomarkers are not consistently available within the TriNetX platform; therefore, disease severity was not directly measured and was instead partially accounted for using proxy variables such as baseline corticosteroid use and medication exposure patterns.

In addition, the use of an active-comparator design (5-ASA-treated patients) and adjustment for multiple baseline covariates was intended to reduce confounding related to disease severity and healthcare utilization. However, residual confounding from unmeasured disease activity and disease duration cannot be fully excluded. From a causal framework perspective, demographic factors, baseline comorbidities, and disease phenotype were treated as confounders because they may influence both treatment allocation and lymphoma risk. In contrast, leukopenia and other treatment-emergent hematologic effects were considered potential mediators of thiopurine-associated risk and were therefore not included in the primary propensity score model to avoid overadjustment.

Covariates were incorporated into the propensity score model to approximate conditional exchangeability between exposure groups. In multivariable Cox regression, variables were entered simultaneously to estimate adjusted HRs for incident lymphoma.

Bias mitigation and confounding control

Given preferential use of thiopurines in patients with more severe IBD, confounding by indication was addressed using an active-comparator, incident-user design restricted to 5-ASA-treated IBD patients. This approach reduced bias related to healthcare utilization and diagnostic intensity while preserving treatment contrast.

The index date required confirmed IBD diagnosis and qualifying medication exposure, with follow-up beginning one day thereafter to minimize immortal time bias. Patients with prior lymphoma were excluded to ensure incident event capture. Sensitivity analyses excluding early lymphoma events were conducted to reduce reverse causation. Propensity score matching (1:1 nearest-neighbor with prespecified caliper) was performed to approximate conditional exchangeability. The propensity model included demographic variables, cardiometabolic and autoimmune comorbidities, body mass index, and corticosteroid exposure as a proxy for disease severity. Balance was assessed using standardized mean differences, with values < 0.1 indicating adequate covariate balance.

Individuals with prior hematologic malignancy, chemotherapy, radiotherapy, human immunodeficiency virus infection, or EBV-related conditions were excluded to reduce baseline heterogeneity in lymphoma risk. Although residual confounding from unmeasured disease activity cannot be fully excluded, the combined design-based and analytic strategies strengthen internal validity.

Statistical analysis

Baseline characteristics are reported as mean ± SD for continuous variables and frequencies with percentages for categorical variables. For each prespecified follow-up interval (1 year, 3 years, 5 years, 10 years, and 15 years), odds ratios (ORs) with 95% confidence interval (CI) were calculated comparing lymphoma risk between matched cohorts. This interval-based approach was chosen to evaluate temporal risk accumulation and distinguish early detection bias from biologically plausible latency effects.

Time-to-event analyses were performed using Kaplan-Meier methods. HRs were estimated using Cox proportional hazards regression. Proportional hazards assumptions were evaluated using graphical inspection and formal testing within the analytic platform. To assess independent predictors of lymphoma, a fully adjusted Cox model was constructed incorporating demographic variables, cardiometabolic comorbidities, IBD phenotype, laboratory parameters, and medication exposures. This model allowed evaluation of residual confounding beyond propensity matching. Due to the federated structure of the TriNetX platform, individual-level follow-up duration and risk-set counts at each timepoint cannot be directly extracted. Analyses are therefore based on dynamically defined cohorts, with denominators at each interval reflecting patients with available follow-up within that time window.

Disease-stratified analyses were conducted separately for UC and CD cohorts to explore effect heterogeneity. All statistical tests were two-sided, and P < 0.05 was considered statistically significant. Effect sizes were interpreted with emphasis on magnitude, temporal consistency, and biological plausibility rather than statistical significance alone.

Study size and power considerations

Given the population-based nature of the TriNetX network and the retrospective observational design, all eligible individuals meeting the inclusion and exclusion criteria were included in the analyses. No formal a priori sample size or power calculation was performed. Instead, statistical precision was evaluated by examining the width of CIs for key effect estimates. The very large sample size afforded by the network ensured narrow CIs and adequate power allowing more precise estimation of effect sizes.

Ethical and data-governance framework

TriNetX data governance adheres to the Health Insurance Portability and Accountability Act, the General Data Protection Regulation, and the United States’ Common Rule (45 CFR 46). All patient-level identifiers are removed or obfuscated prior to aggregation. To further mitigate re-identification risk, cell counts with fewer than 10 individuals were automatically rounded or suppressed. Because all analyses were conducted within a closed, federated environment and only de-identified, aggregated data leave the platform, no protected health information was accessed or exported by the investigators. Accordingly, this study was considered non-human subjects research and was exempt from institutional review board review and individual informed consent requirements.

RESULTS

Baseline characteristics

A total of 42310 patients with IBD initiating thiopurines and 84093 patients treated with 5-ASA without thiopurine or any biologic exposure were identified prior to matching. Before matching, thiopurine-treated patients were younger [41.6 ± 15.7 years vs 47.5 ± 19.0 years; standardized mean difference (SMD) = 0.311].

After 1:1 propensity score matching, 38781 IBD patients remained in each cohort. Covariate balance was achieved across all measured variables (all SMD < 0.1), including age (0.038). Although some variables retained statistically significant P values after matching, all standardized mean differences were below 0.1, indicating satisfactory covariate balance; such discrepancies are expected in large datasets and reflect sample size-driven statistical sensitivity rather than clinically meaningful imbalance (Table 1).

Table 1 Baseline characteristics of patients with inflammatory bowel disease taking thiopurines and patients with inflammatory bowel disease taking 5-aminosalicylic acid and not taking thiopurines or biologics, mean ± SD/n (%).

Characteristics	Before PSM			After PSM
	IBD with thiopurines (n = 42310)	IBD with 5-ASA (n = 84093)	SMD	IBD with thiopurines (n = 38781)	IBD with 5-ASA (n = 38781)	SMD
Age, years	41.6 ± 18.9	47.5 ± 19.0	0.311	42.6 ± 18.8	43.3 ± 18.6	0.038
Sex
Male	19675 (47.1)	38379 (46.3)	0.018	18277 (47.1)	18435 (47.5)	0.008
Female	22024 (52.8)	44512 (53.7)	0.008	20466 (52.8)	20305 (52.4)	0.008
Race
White	32869 (78.7)	62615 (75.5)	0.078	30306 (78.1)	30565 (78.8)	0.016
Black or African American	3749 (9.0)	7531 (9.1)	0.003	3490 (9.0)	3373 (8.7)	0.011
Asian	888 (2.1)	2327 (2.8)	0.044	872 (2.2)	789 (2.0)	0.015
American Indian or Alaska native	168 (0.4)	285 (0.3)	0.010	153 (0.4)	131 (0.3)	0.009
Native Hawaiian or other Pacific Islander	60 (0.1)	143 (0.2)	0.007	60 (0.2)	59 (0.2)	0.001
Unknown race	3041 (7.3)	7638 (9.2)	0.070	2964 (7.6)	2988 (7.7)	0.002
Ethnicity
Hispanic or Latino	1862 (4.5)	4370 (5.3)	0.037	1789 (4.6)	1613 (4.2)	0.022
Not Hispanic or Latino	31775 (76.1)	60134 (72.5)	0.083	29271 (75.5)	29640 (76.4)	0.022
Unknown ethnicity	8105 (19.4)	18463 (22.3)	0.070	7721 (19.9)	7528 (19.4)	0.013
Comorbidities
Essential hypertension	7153 (17.1)	17234 (20.8)	0.093	6769 (17.5)	6717 (17.3)	0.004
Type 2 diabetes mellitus	2711 (6.5)	6427 (7.7)	0.049	2558 (6.6)	2492 (6.4)	0.007
Rheumatoid arthritis	633 (1.5)	808 (1.0)	0.049	512 (1.3)	508 (1.3)	0.001
Psoriasis	593 (1.4)	769 (0.9)	0.046	489 (1.3)	446 (1.2)	0.010
Alcohol abuse	410 (1.0)	1102 (1.3)	0.032	401 (1.0)	375 (1.0)	0.007
Systemic lupus erythematosus	502 (1.2)	362 (0.4)	0.085	289 (0.7)	318 (0.8)	0.008
Nicotine dependence	2832 (6.8)	5215 (6.3)	0.020	2614 (6.7)	2578 (6.6)	0.004

Baseline demographic and clinical characteristics of patients with inflammatory bowel disease taking thiopurines and patients with inflammatory bowel disease taking 5-aminosalicylic acid and not taking thiopurines or biologics before and after propensity score matching. Standardized mean differences quantify covariable imbalance with values > 0.1 indicating clinically meaningful differences. 5-ASA: 5-aminosalicylates; IBD: Inflammatory bowel disease; PSM: Propensity score matching; SMD: Standardized mean difference.

The mean age of the matched cohorts was 42.6 ± 18.8 years in patients with IBD initiating thiopurines and 43.3 ± 18.6 years in patients treated with 5-ASA, and males comprised 47.1% of the thiopurines group and 47.5% of the 5-ASA group. The racial distribution was similar (White: 78.1% vs 78.8%, Hispanic: 4.6% vs 4.2%, Black: 9.0% vs 8.7%). The prevalence of hypertension (17.5% vs 17.3%), and type 2 diabetes (6.6% vs 6.4%) was comparable between groups, confirming successful matching (Table 1).

Comparative risk of lymphoma in IBD-thiopurine vs IBD-5-ASA

At 1 year, lymphoma incidence did not differ significantly between groups (OR = 1.39; 95%CI: 0.84-2.29; P = 0.1907). From 3 years onward, thiopurine exposure was associated with significantly increased lymphoma risk. At 3 years, the OR was 2.37 (95%CI: 1.54-3.64; P < 0.001). Elevated risk persisted at 5 years (OR = 1.82; 95%CI: 1.29-2.57), 10 years (OR = 1.95; 95%CI: 1.46-2.60), and 15 years (OR = 1.93; 95%CI: 1.46-2.54) (all P < 0.001) (Table 2 and Figure 2A). The association was predominantly driven by non-Hodgkin lymphoma. At 3 years, non-Hodgkin lymphoma risk was increased (OR = 2.43; 95%CI: 1.57-3.78), with persistent elevation through 15 years (OR = 1.95; 95%CI: 1.46-2.60). B-cell lymphoma demonstrated a comparable temporal pattern (3-year OR = 2.43; 95%CI: 1.54-3.83; 15-year OR = 1.85; 95%CI: 1.38-2.47). Hodgkin lymphoma reached statistical significance at extended follow-up (15-year OR = 2.60; 95%CI: 1.25-5.40).

Figure 2 Cumulative lymphoma cases following initiation of thiopurines vs 5-aminosalicylates in patients. A: Inflammatory bowel disease. Cumulative number of incident lymphoma diagnoses observed at 3 years, 5 years, 10 years, and 15 years following treatment initiation in propensity score-matched cohorts (n = 38781 per group); B: Ulcerative colitis. Cumulative number of incident lymphoma diagnoses observed at 3 years, 5 years, 10 years, and 15 years following treatment initiation in propensity score-matched cohorts (n = 12334 per group); C: Crohn’s disease. Cumulative number of incident lymphoma diagnoses observed at 3 years, 5 years, 10 years, and 15 years following treatment initiation in propensity score-matched cohorts (n = 18111 per group). Lymphoma cases accumulated progressively over time in both groups, with a greater numerical increase observed among thiopurine-treated patients beginning at 3 years of follow-up. Counts represent total incident events recorded within each prespecified follow-up interval. Estimates at later timepoints (10 years and 15 years) reflect progressively smaller subsets of patients with longer observable follow-up within the network. 5-ASA: 5-aminosalicylates; IBD: Inflammatory bowel disease; UC: Ulcerative colitis; CD: Crohn’s disease.

Table 2 Odds ratio for lymphoma outcomes for patients with inflammatory bowel disease taking thiopurines vs inflammatory bowel disease taking 5-aminosalicylic acid and not taking thiopurines or biologics.

Outcomes	Events in IBD with thiopurines	Events in IBD with 5-ASA	OR (95%CI)	P value
Overall lymphoma
1 year	36	26	1.39 (0.84-2.29)	0.1907
3 years	71	30	2.37 (1.54-3.64)	< 0.001
5 years	91	50	1.82 (1.29-2.57)	< 0.001
10 years	136	70	1.95 (1.46-2.60)	< 0.001
15 years	148	77	1.93 (1.46-2.54)	< 0.001
Hodgkin lymphoma
1 year	≤ 10	≤ 10
3 years	≤ 10	≤ 10
5 years	≤ 10	≤ 10
10 years	≤ 10	≤ 10
15 years	26	10	2.60 (1.25-5.40)	0.008
Non-Hodgkin lymphoma
1 year	33	24	1.38 (0.81-2.33)	0.2383
3 years	68	28	2.43 (1.57-3.78)	< 0.001
5 years	85	47	1.81 (1.26-2.58)	0.0009
10 years	125	66	1.90 (1.41-2.56)	< 0.001
15 years	136	70	1.95 (1.46-2.60)	< 0.001
B-cell lymphoma
1 year	33	26	1.27 (0.76-2.13)	0.3153
3 years	63	26	2.43 (1.54-3.83)	< 0.001
5 years	82	49	1.68 (1.17-2.39)	0.0039
10 years	120	65	1.85 (1.36-2.51)	< 0.001
15 years	129	70	1.85 (1.38-2.47)	< 0.001
T-cell/natural killer cell lymphoma
1 year	≤ 10	≤ 10
3 years	≤ 10	≤ 10
5 years	≤ 10	≤ 10
10 years	23	13	1.77 (0.90-3.49)	0.0955
15 years	26	13	2.00 (1.03-3.89)	0.037

Odds ratios of lymphoma outcomes at the 1-year, 3-year, 5-year, 10-year, and 15-year follow-up among propensity score-matched cohorts of patients with inflammatory bowel disease taking thiopurines vs patients with inflammatory bowel disease taking 5-aminosalicylic acid and not taking thiopurines or biologics. Outcomes are presented as event counts with percentage and corresponding odds ratios with 95% confidence intervals. All analyses reflect matched sample sizes of 39618 per group. 5-ASA: 5-aminosalicylates; IBD: Inflammatory bowel disease; OR: Odds ratio; CI: Confidence interval.

Comparative risk of lymphoma in IBD-subtypes

Among UC patients (n = 12334 per group after matching), thiopurine exposure was associated with progressively increased lymphoma risk, reaching an OR of 2.23 (95%CI: 1.41-3.50; P < 0.001) at 15 years (Figure 2B, Table 3 and Supplementary Table 3). Among CD patients (n = 18111 per group after matching), risk estimates were attenuated. A significant association for lymphoma was observed at 10 years (OR = 1.80; 95%CI: 1.10-2.94; P = 0.0169), whereas the 15-year estimate did not reach statistical significance (OR = 1.39; 95%CI: 0.87-2.10; P = 0.1626) (Figure 2C, Table 4 and Supplementary Table 4).

Table 3 Odds ratio for lymphoma outcomes for patients with ulcerative colitis taking thiopurines vs ulcerative colitis taking 5-aminosalicylic acid and not taking thiopurines or biologics.

Outcomes	Events in UC with thiopurines	Events in UC with 5-ASA	OR (95%CI)	P value
Overall lymphoma
1 year	≤ 10	≤ 10
3 years	29	15	1.94 (1.04-3.62)	0.037
5 years	38	19	2.01 (1.16-3.48)	0.016
10 years	54	26	2.08 (1.31-3.32)	0.0017
15 years	60	27	2.23 (1.41-3.50)	0.0004
Hodgkin lymphoma
1 year
3 years
5 years
10 years
15 years
Non-Hodgkin lymphoma
1 year	≤ 10	≤ 10
3 years	27	13	2.05 (1.07-3.90)	0.029
5 years	35	17	2.07 (1.16-3.69)	0.012
10 years	50	24	2.09 (1.28-3.40)	0.0024
15 years	55	25	2.20 (1.37-3.53)	0.0008
B-cell lymphoma
1 year	≤ 10	≤ 10
3 years	27	12	2.26 (1.14-4.46)	0.016
5 years	34	16	2.13 (1.18-3.86)	0.0107
10 years	50	24	2.09 (2.38-3.40)	0.0024
15 years	55	25	2.20 (1.37-3.53)	0.0008
T-cell/natural killer cell lymphoma
1 year	≤ 10	≤ 10
3 years	≤ 10	≤ 10
5 years	≤ 10	≤ 10
10 years	≤ 10	≤ 10
15 years	≤ 10	≤ 10

Odds ratios of lymphoma outcomes at the 1-year, 3-year, 5-year, 10-year, and 15-year follow-up among propensity score-matched cohorts of patients with ulcerative colitis taking thiopurines vs patients with ulcerative colitis taking 5-aminosalicylic acid and not taking thiopurines or biologics. Outcomes are presented as event counts with percentage and corresponding odds ratios with 95% confidence intervals. All analyses reflect matched sample sizes of 12334 per group. 5-ASA: 5-aminosalicylates; UC: Ulcerative colitis; OR: Odds ratio; CI: Confidence interval.

Table 4 Odds ratio for lymphoma outcomes for patients with Crohn’s disease taking thiopurines vs Crohn’s disease taking 5-aminosalicylic acid and not taking thiopurines or biologics.

Outcomes	Events in CD with thiopurines	Events in CD with 5-ASA	OR (95%CI)	P value
Overall lymphoma
1 year	11	11	1.00 (0.43-2.30)	09985
3 years	21	17	1.24 (0.65-2.34)	0.5168
5 years	29	21	1.36 (0.77-2.41)	0.253
10 years	45	25	1.80 (1.10-2.94)	0.0169
15 years	45	31	1.39 (0.87-2.10)	0.1626
Hodgkin lymphoma
1 year
3 years
5 years
10 years
15 years
Non-Hodgkin lymphoma
1 year	≤ 10	11	0.91 (0.39-2.14)	0.8359
3 years	19	17	1.12 (0.58-2.15)	0.739
5 years	27	21	1.29 (0.73-2.28)	0.386
10 years	40	25	1.60 (0.97-2.64)	0.063
15 years	37	27	1.37 (0.83-2.25)	0.211
B-cell lymphoma
1 year	≤ 10	≤ 10
3 years	15	18	0.83 (0.42-1.65)	0.406
5 years	24	22	1.09 (0.61-1.95)	0.769
10 years	36	27	1.33 (0.81-2.20)	0.257
15 years	34	27	1.26 (0.76-2.08)	0.370
T-cell/natural killer cell lymphoma
1 year	≤ 10	≤ 10
3 years	≤ 10	≤ 10
5 years	≤ 10	≤ 10
10 years	≤ 10	≤ 10
15 years	≤ 10	≤ 10

Odds ratios of lymphoma outcomes at the 1-year, 3-year, 5-year, 10-year, and 15-year follow-up among propensity score-matched cohorts of patients with Crohn’s disease taking thiopurines vs patients with Crohn’s disease taking 5-aminosalicylic acid and not taking thiopurines or biologics. Outcomes are presented as event counts with percentage and corresponding odds ratios with 95% confidence intervals. All analyses reflect matched sample sizes of 18111 per group. CD: Crohn’s disease; 5-ASA: 5-aminosalicylates; OR: Odds ratio; CI: Confidence interval.

Predictors of lymphoma in thiopurine users in IBD

Multivariable Cox regression identified several independent predictors of lymphoma within the IBD with thiopurine users. Male sex (HR = 1.59; 95%CI: 1.33-1.90; P < 0.0001), non-Hispanic/Latino ethnicity (HR = 1.31; 95%CI: 1.02-1.68; P = 0.0330), and glycosylated hemoglobin > 6.5% (HR = 1.95; 95%CI: 1.36-2.78; P = 0.0003) were independently associated with incident lymphoma (Table 5).

Table 5 Multivariate Cox proportional hazards model for patients with inflammatory bowel disease taking thiopurines and patients with inflammatory bowel disease taking 5-aminosalicylic acid and not taking thiopurines or biologics assessing predictors for lymphoma.

Covariate	Hazard ratio (95%CI)	P value
Male	1.59 (1.33-1.90)	< 0.0001
Age at index
≥ 50 years	1.90 (0.99-3.62)	0.0527
18-50 years	5.59 (2.94-10.63)	< 0.0001
Race
Asian	1.88 (0.99-3.57)	0.0558
Black or African American	1.13 (0.67-1.94)	0.6445
Native Hawaiian or Other Pacific Islander	1.52 (0.20-11.31)	0.6840
White	1.21 (0.78-1.89)	0.3998
Other race	2.00 (1.00-4.01)	0.0502
Ethnicity
Hispanic or Latino	0.75 (0.39-1.43)	0.3772
Not Hispanic or Latino	1.31 (1.02-1.68)	0.0330
Comorbidities
Essential hypertension	1.19 (0.94-1.50)	0.1445
Type 2 diabetes mellitus	0.97 (0.72-1.30)	0.8143
Hyperlipidemia	0.95 (0.73-1.23)	0.6989
Tobacco use	0.82 (0.35-1.90)	0.6395
Alcohol use disorders	1.26 (0.66-2.38)	0.4868
Nicotine dependence	0.95 (0.65-1.40)	0.8014
Laboratory values
Total cholesterol (mg/dL)
≥ 200	1.09 (0.64-1.85)	0.7540
Cholesterol in LDL (mg/dL)
≥ 150	1.08 (0.58-2.03)	0.8031
Cholesterol in HDL (mg/dL)
≥ 50	0.98 (0.72-1.34)	0.9204
Glycosylated hemoglobin (%)
> 6.50	1.95 (1.36-2.78)	0.0003

Hazard ratios with 95% confidence intervals from a multivariable Cox proportional hazards model evaluating demographic factors, race/ethnicity, and comorbidities associated with lymphoma outcomes in patients with inflammatory bowel disease taking thiopurines and patients with inflammatory bowel disease taking 5-aminosalicylic acid and not taking thiopurines or biologics. Covariables include age, sex, race/ethnicity categories, cardiometabolic conditions, and substance use. A hazard ratio > 1 indicates increased risk while a hazard ratio < 1 indicates reduced risk. All variables were included in the fully adjusted model. Hazard ratios for categorical variables reflect comparisons against an implicit reference group as defined by TriNetX variable encoding. CI: Confidence interval; HDL: High density lipoprotein; LDL: Low density lipoprotein.

DISCUSSION

In this large, propensity score-matched cohort of IBD patients, thiopurine exposure was associated with a sustained increase in incident lymphoma risk emerging after three years of therapy and persisting through 15 years of follow-up. In the initial few years of therapy, lymphoma rates among thiopurine-treated patients remained near baseline; thereafter the hazard rose and remained elevated in subsequent years. This temporal pattern suggests that the oncogenic consequences of thiopurine therapy accrue gradually, manifesting clinically only after prolonged immunosuppression. Importantly, once evident, the excess risk did not abate with continued therapy over the 3-15 years horizon of our observation, underscoring a durable impact of long-term thiopurine use on tumorigenesis.

Our findings align with and extend prior evidence from both cohort studies and meta-analyses of IBD patients on thiopurines. Early signals of lymphoma risk were reported in the French CESAME cohort, which observed a roughly fivefold higher incidence of lymphoproliferative disorders in thiopurine-treated IBD patients (HR = 5.28, 95%CI: 2.01-13.9) compared to those never exposed[11]. Notably, CESAME had a median follow-up of only approximately 3 years yet still detected a significant risk elevation, with 23 cases of lymphoma (22 of them non-Hodgkin’s) arising in the thiopurine-exposed group. Most of these lymphomas were pathologically analogous to post-transplant lymphoproliferative disease, consistent with an immunosuppression-driven etiology. Our study corroborates the CESAME finding of a markedly increased relative risk, while further demonstrating that the risk continues to accrue beyond the short follow-up of that initial cohort. In fact, the delayed risk onset (approximately 3 years) we observed dovetails with the CESAME timeline (median 3 years) in which their excess lymphoma incidence became apparent, suggesting that a threshold duration of immunosuppression may be required before malignant transformation is triggered.

Other large-scale studies support the pattern of a time-dependent, cumulative risk. Khan et al[12] analyzed approximately 37000 UC patients in a United States Veterans Affairs cohort and found a fourfold increase in lymphoma incidence during thiopurine therapy (adjusted HR approximately 4.2)[12]. Crucially, the risk in that study climbed with each additional year of thiopurine use: From an incidence of 0.9 per 1000 person-years in the first year to 5.0 per 1000 in year four and 8.9 per 1000 after four years of therapy. This mirrors our finding that lymphoma risk accelerates after several years of exposure. Khan et al[12] also observed that patients who discontinued thiopurines saw their lymphoma hazard fall back to baseline levels (HR approximately 0.5, 95%CI: 0.2-1.3, statistically indistinguishable from unexposed). Prior studies have reported that lymphoma risk may decrease following thiopurine discontinuation; however, this was not directly evaluated in the present study, as exposure was analyzed using an intention-to-treat framework without time-varying assessment of treatment persistence. The consistency of these results across a United States veteran cohort and our broader multi-center cohort strengthens the inference that ongoing thiopurine use, rather than an underlying predisposition of IBD patients, is the driver of the lymphoma association, and that this risk can gradually revert after drug cessation.

Meta-analyses have quantified the magnitude of lymphoma risk attributable to thiopurines and further illuminate the timing and subgroup effects. Kotlyar et al[13] pooled 18 studies and found an overall approximately 5-fold increase in lymphoma risk with thiopurine therapy (pooled standardized incidence ratio 4.92). Notably, the risk estimates differed by study design: Population-based studies reported a more modest approximately 2.8-fold increase, whereas referral-center cohorts showed a much higher approximately 9-fold increase. This discrepancy likely reflects referral bias and differences in patient case-mix (with tertiary centers seeing more severe, heavily treated IBD). Our study, leveraging a large EHR network spanning academic and community sites, provides a population-level estimate and indeed found an effect size closer to the lower end of prior reports, albeit still clearly elevated. The meta-analysis also confirmed two critical aspects that our data reinforce: First, lymphoma risk becomes significant only after at least one year of thiopurine exposure (consistent with the delayed onset of approximately 3 years we observed for a pronounced rise); and second, the risk does not persist after discontinuation. Patients with past thiopurine use but no ongoing therapy did not show a statistically significant excess lymphoma incidence (pooled standardized incidence ratio approximately 1.4 for former users). This echoes our finding that prolonged exposure is required to both induce and maintain the elevated risk. It is reassuring in a clinical sense that stopping the drug allows the lymphogenic risk to recede, though perhaps only after some years off medication as immune surveillance recovers.

In terms of magnitude, our results are broadly in line with prior estimates when considering the appropriate context. The relative risk we found (on the order of 3-5 fold increased hazard with long-term use) is comparable to the CESAME and VA cohort findings. However, the absolute risk of lymphoma in IBD patients on thiopurines remains low in absolute terms, which previous studies have also emphasized. For example, Beaugerie et al[11] reported incidence rates around 0.9 per 1000 patient-years under thiopurine therapy vs 0.26 per 1000 in unexposed patients. Similarly, the VA study found approximately 2.3 per 1000 patient-years on thiopurine, against approximately 0.6 per 1000 without. In our cohort, absolute rates were on the order of a few cases per thousand patients per year even in the highest risk windows. These figures highlight that while the relative increase is substantial, the individual probability of lymphoma per year of thiopurine use remains small. This risk-benefit context is crucial when counselling patients, as discussed further below. It is also noteworthy that certain subpopulations bear disproportionate risk: Prior analyses have consistently shown that older age greatly amplifies absolute lymphoma risk on thiopurines, and male sex has been associated with nearly double the relative risk compared to females. Our data, drawn from a large heterogeneous population, likely reflect these same patterns. Indeed, the meta-analysis found patients over 50 on thiopurines had an absolute lymphoma risk of approximately 1 in 354 per year (relative risk approximately 4.8 vs non-users), far higher than younger patients in absolute terms. Meanwhile, the highest relative risk was observed in younger men under about 35 years old a group that, despite lower baseline lymphoma rates, appears particularly susceptible to the immune-modulating effects of thiopurines. This may have clinical relevance in weighing therapy options for young male patients (who also face the rare but often fatal hepatosplenic T-cell lymphoma when thiopurines are combined with anti-tumor necrosis factor therapy)[14]. Our study’s large sample allowed us to examine lymphoma outcomes across age and sex strata with greater precision than many earlier cohorts, lending further confidence that these high-risk subgroups can be identified and monitored accordingly.

Another contribution of our study is time-course that it provides. The excess lymphomas associated with thiopurines were almost entirely of non-Hodgkin B-cell origin in prior reports, and our findings were consistent with this profile. Hodgkin lymphoma, by contrast, has not been strongly linked to thiopurine exposure a distinction that was evident in CESAME (only 1 of 23 cases was Hodgkin’s). The predominance of non-Hodgkin’s lymphomas in thiopurine-treated IBD supports the hypothesis that these are driven by impaired immune surveillance (as non-Hodgkin’s, especially diffuse large B-cell lymphomas, are the classic post-transplant and immunosuppression-associated subtype). In our dataset we were able to distinguish subtypes to ensure, for instance, that indolent lymphoma subtypes or misclassified benign lymphoproliferations did not confound the results. The persistent risk from 3 years to 15 years was evident specifically for aggressive lymphoma subtypes, indicating that the clinical significance lies in those serious malignancies rather than a broad increase in all lymphoid neoplasms. By capturing detailed outcomes through EHR diagnostic codes and oncology records, we improved upon earlier studies that could not always verify lymphoma subtypes or relied on registry linkage alone.

A key strength of this study is its large, representative sample drawn from the TriNetX network of United States health systems. By assembling thousands of IBD patients with long-term follow-up, we had adequate power to detect the relatively rare outcome of lymphoma and to observe temporal trends over a 15-year horizon. We employed rigorous matching and adjustment to control for confounding matching thiopurine-exposed patients with similar IBD patients not exposed to thiopurines (or exposed only to 5-ASA therapy) on factors such as age, sex, IBD subtype, disease duration, and use of other medications. This EHR-based design allowed us to account for many potential confounders that might influence lymphoma risk (for example, baseline disease severity proxies like corticosteroid use or hospitalization history, and co-therapy with biologics). By contrast, some earlier cohorts lacked granular control for such factors or drew comparisons to general population rates, raising concern that differences in case mix or IBD severity could bias results. Our study improves on those by directly comparing exposed and unexposed IBD patients within the same data sources, thereby reducing confounding by indication. Moreover, the multi-center nature of our data enhances generalizability, including community settings and diverse patient demographics, in contrast to single-center studies or those restricted to veteran or tertiary populations.

Despite these strengths, the study has important limitations inherent to observational research. First, as a non-randomized analysis, we cannot conclusively prove causation; unmeasured confounders may still influence the results. For instance, patients prescribed thiopurines may have had more severe or refractory IBD and while we attempted to adjust for disease severity, it is possible that severity (or factors associated with it) contributed to lymphoma risk. However, it should be noted that IBD itself has not been strongly linked to lymphoma absent immunosuppression, and our finding of risk normalization after drug discontinuation argues against disease activity alone explaining the excess risk. Second, our reliance on diagnostic and pharmacy codes introduces potential misclassification. Some lymphomas could have been erroneously coded or missed if care was received outside contributing health systems. We mitigated this by requiring confirmatory evidence of lymphoma (e.g., oncology visits, treatment codes) to improve specificity, but a degree of outcome misclassification is possible. Similarly, exposure was defined through prescription records; we lacked direct data on patient adherence or precise duration of use beyond what prescriptions indicate. If patients stopped and restarted thiopurines, or if they obtained medications elsewhere, our exposure classification might be imperfect. Third, while we had long follow-up on average, not every patient was observed for the full 15-year span and those who were long-observed may differ from those lost to follow-up, potentially introducing bias. We also did not have data on EBV status or genetics screening, a key risk factor for certain lymphomas; nor could we capture family history of lymphoproliferative disease or other genetic predispositions. Lastly, our outcome of “lymphoma” pooled several subtypes, and although dominated by high-grade B-cell neoplasms, we did not separately analyze very rare lymphoma entities (such as hepatosplenic T-cell lymphoma) given their extremely low incidence. Residual confounding from unmeasured disease severity and disease duration cannot be excluded. Clinical indices, endoscopic findings, and inflammatory biomarkers were not available within the database, and duration of disease prior to treatment initiation could not be reliably ascertained. Given that thiopurines are preferentially used in patients with more severe or steroid-dependent disease, this may introduce confounding by indication. However, the absence of increased risk at early follow-up and the emergence of risk only after prolonged exposure suggest that baseline disease severity alone is unlikely to fully explain the observed associations. The use of 5-ASA as an active comparator may be more appropriate in UC than in CD, where its use is limited and often reflects milder phenotypes. This may introduce residual confounding by indication in CD, potentially contributing to the attenuated associations observed. Exclusion of patients with documented EBV infection was intended to reduce baseline lymphoma risk heterogeneity; however, EBV is incompletely captured in EHR data, which may introduce selection bias. In addition, EBV may act as a mediator in thiopurine-associated lymphomagenesis, and its exclusion could attenuate observed associations. Lastly, detailed disease characteristics such as location and behavior were not available in a structured format and could not be included, introducing potential residual confounding. In addition, although age was modeled as a continuous variable, categorical stratification may not fully capture its non-linear relationship with lymphoma risk. These limitations underscore that our results should be interpreted as associational. Nonetheless, the consistency with prior studies and the magnitude of effect observed lend credibility to a causal relationship between thiopurine use and lymphoma in the IBD setting.

Mechanistically, the findings support a model of cumulative immunologic oncogenesis in thiopurine-treated patients. Thiopurines (azathioprine and 6-mercaptopurine) impair DNA synthesis in lymphocytes and directly suppress cytotoxic T-cell and natural killer cell function, leading to reduced immune surveillance for emerging neoplastic cells[15]. Over years of therapy, this dampening of the immune system may allow proliferation of EBV-infected B cells and other potentially malignant clones that would normally be kept in check. EBV-mediated B-cell expansion is a well-described pathway in post-transplant lymphoproliferative disorders, and analogously, most thiopurine-related lymphomas in IBD have been reported to be EBV-positive, diffuse large B-cell lymphomas akin to those seen in transplant patients[16]. The resemblance of these lymphomas to post-transplant lymphoma was explicitly noted in the CESAME study, and it underscores that prolonged immunosuppression rather than a direct carcinogenic effect of thiopurines is the driving force. The delayed onset of increased risk (after a few years of use) fits with this mechanism: It likely requires sustained immunosuppressive pressure for an extended period before a clinically apparent lymphoma develops. Each passing year on thiopurines may incrementally increase the chance that an oncogenic viral infection or spontaneous B-cell mutation evades immune destruction and evolves into overt lymphoma[17]. In other words, thiopurines act as a promoter of lymphogenesis by creating an environment of impaired immune surveillance; remove the promoter, and the environment gradually normalizes, lowering lymphoma incidence. This conceptual framework ties together the epidemiologic data with biological plausibility. It also suggests that the risk might accumulate with total duration of immunosuppression (and possibly with cumulative dose or persistence of drug metabolites that induce lasting T-cell suppression)[18]. While we did not directly measure immune function or EBV titres in our study, the epidemiologic signature observed time-lagged onset and reversibility strongly implicates immune-mediated mechanisms rather than immediate mutagenesis.

These findings have several implications for the management of IBD with thiopurines. First, they provide evidence to guide risk stratification. Patients’ individual risk profiles should be considered when deciding on thiopurine therapy. Clinicians should be cautious about using thiopurines in demographics known to be higher-risk, or at least discuss these nuances with such patients. Second, our results highlight the importance of limiting duration of thiopurine use when possible. Since risk appears to be cumulative, prolonged continuous use for more than 3-5 years should be approached carefully. If a patient has achieved stable remission, it may be prudent to taper and discontinue thiopurine after a few years rather than continuing indefinitely. Doing so could substantially mitigate lymphoma risk especially given evidence that stopping therapy allows the excess risk to recede towards baseline. Some guidelines already suggest that durable remission beyond 4 years might warrant considering thiopurine withdrawal to curtail long-term cancer risks. Our data reinforce that principle by demonstrating a sharp rise in lymphoma incidence with extended use. Of course, any decision to stop must weigh the risk of IBD relapse; in some cases, the benefits of maintaining remission will outweigh the small lymphoma risk. This underscores the need for personalized, shared decision-making. Patients should be informed about the long-term lymphoma risk in clear terms including the fact that the absolute risk is low (perhaps 1 in several hundred per year) but not negligible, and that it increases the longer the drug is taken. With this information, patients can better weigh their comfort with continuing therapy, and clinicians can explore patient values and preferences. For some, the promise of staying off steroids and avoiding flare-ups will justify continuing thiopurine despite the risk; for others, especially those in stable remission, an elective transition to mesalamine, newer biologics, or other strategies might be preferable to eliminate the cancer concern. The conversation is particularly pertinent when initiating thiopurine in the first place: Alternative treatments (biologics, newer small molecules) that might carry different risk profiles could be considered for patients who are risk-averse or have risk factors for lymphoma.

CONCLUSION

In conclusion, this study strengthens the evidence that thiopurine therapy in IBD confers a small but significant long-term risk of lymphoma, one that is driven by cumulative immunosuppressive effects over years of exposure. By leveraging a large matched cohort with extended follow-up, we have demonstrated that the lymphoma risk, while delayed in onset, is real and persists with ongoing thiopurine use. These findings do not negate the clinical utility of thiopurines, which remain effective and cost-efficient steroid-sparing agents, but they call for a judicious and time-limited approach to their use. Physicians should remain vigilant for lymphoproliferative complications in patients on long-term thiopurines, especially beyond 5 years, and should engage patients in collaborative discussions about the duration of therapy. Ultimately, an individualized approach balancing disease control with malignancy risk is warranted. Our data will inform that balance, enabling better identification of patients who might benefit from shorter thiopurine courses or from alternative therapies to minimize the risk of lymphoma while effectively managing their IBD.