Safety profile of BCG with or without immunotherapy in non–muscle-invasive bladder cancer (NMIBC): a systematic review and meta-analysis

Abstract

Background

Intravesical bacillus Calmette-Guérin (BCG) is the standard of care for BCG-naïve high-risk non–muscle-invasive bladder cancer (NMIBC). Recent phase III trials showed reduced disease recurrence with the addition of immune checkpoint inhibitors (ICIs) to BCG induction plus maintenance but with increased toxicity. We carried out a systematic review and meta-analysis to assess the safety of this treatment intensification.

Patients and methods

Randomized phase II-III trials comparing BCG plus anti-programmed cell death protein 1/programmed death-ligand 1 ICIs with BCG alone in BCG-naïve high-risk NMIBC were identified through MEDLINE/PubMed, American Society of Clinical Oncology, and European Society for Medical Oncology databases up to November 2025. Outcomes included the incidence and relative risk (RR) of treatment-related adverse events (TRAEs), immune-related adverse events (irAEs), serious adverse events (SAEs), treatment discontinuation, and the impact of the BCG schedule.

Results

Three phase III trials including 1895 patients were analyzed. Compared with BCG alone, BCG plus ICIs significantly increased the risk of any-grade TRAEs [RR 1.25, 95% confidence interval (CI) 1.20-1.31, P < 0.00001], grade ≥3 TRAEs (RR 4.02, 95% CI 3.06-5.30, P < 0.00001), SAEs (RR 4.54, 95% CI 2.09-9.85, P = 0.0001), any-grade irAEs (RR 15.18, 95% CI 4.34-53.14, P < 0.0001), grade ≥3 irAEs (RR 9.99, 95% CI 2.12-47.14, P = 0.004), and treatment discontinuation (RR 2.87, 95% CI 1.50-5.52, P = 0.002). Toxicity was mainly driven by systemic immune-mediated events, whereas local BCG-related urinary adverse events were largely comparable between groups.

Conclusions

Adding ICIs to intravesical BCG in BCG-naïve high-risk NMIBC increases treatment-related toxicity, mainly due to immune-mediated adverse events, without substantially worsening local urinary tolerability. Careful patient selection and multidisciplinary management are essential when adopting this treatment optimization.

Highlights

• •BCG plus ICIs increase treatment-related toxicity in high-risk NMIBC.
• •Toxicity is mainly driven by systemic irAEs.
• •Local urinary BCG-related toxicity is not meaningfully increased by ICIs.
• •Grade ≥3 TRAEs, SAEs, and treatment discontinuation are higher.
• •Multidisciplinary management is key when implementing BCG–ICI combinations.

Introduction

Bladder cancer (BC) is diagnosed in most cases (∼75%) in early stages as non–muscle-invasive BC (NMIBC).¹ Approximately 50% of NMIBC cases have high-risk features, with a 5-year probability of disease recurrence of ≤40%-45%.² Of note, although a univocal definition of high-risk NMIBC is still lacking, the main characteristics include tumors invading the lamina propria (T1), noninvasive flat carcinoma [Tis; also known as carcinoma in situ (CIS)], and noninvasive papillary carcinoma (Ta) of high grade.^1,3,4

Adjuvant immunotherapy with intravesical instillation of bacillus Calmette-Guérin (BCG)—once-weekly induction for 6 weeks and maintenance for ≤3 years^1,5, 6, 7—after transurethral resection of bladder tumor (TURBT) is the guideline-recommended standard of care (SOC), leading to a 32% reduction in the risk of disease recurrence compared with intravesical chemotherapy that is even more pronounced in contemporary phase III trials using adequately delivered BCG induction and maintenance compared with old historical series.⁸ However, 30%-40% of patients experience disease recurrence despite BCG treatment,⁹ requiring radical cystectomy (RC) as a recommended option with high-grade early recurrence or BCG-unresponsive disease.² In recent years, systemic immunotherapy with immune checkpoint inhibitors (ICIs) targeting programmed cell death protein 1 (PD-1)/programmed death-ligand 1 (PD-L1) has been investigated in both BCG-naïve and post-BCG NMIBC settings, with the goal of addressing the need for novel systemic treatment options with higher antitumor activity and more durable disease control, and thus prevent/delay disease recurrence and thereby avoid RC, and avoiding the associated morbidity. Pembrolizumab monotherapy showed activity for patients with BCG-unresponsive CIS who were ineligible for or declined RC (cohort A of the KEYNOTE-057, an open-label, single-arm, multicentre phase II study). The high rate (41%) of complete response achieved at 3 months justified the U.S. Food and Drug Administration approval of pembrolizumab as a nonsurgical treatment option and provided the foundation for future trials of immunotherapy in the NMIBC setting.¹⁰ In contrast, phase III trials evaluating ICIs combined with BCG enrolled a broader BCG-naïve high-risk NMIBC population, including patients with papillary NMIBC regardless of the presence of CIS.11, 12, 13 In patients with BCG-naïve high-risk NMIBC, recent data from two prospective randomized phase III trials, CREST (evaluating sasanlimab in association with BCG) and POTOMAC (evaluating durvalumab in association with BCG) have demonstrated the efficacy of delaying disease recurrence with the use of PD-1/PD-L1 inhibitors in combination with BCG.^11,12 In particular, the CREST study reported a significant advantage in event-free survival (EFS) with the subcutaneous anti-PD-1 sasanlimab for 2 years in association with BCG induction plus maintenance (I+M) for 2 years compared with BCG (I+M) alone.¹¹ Similarly, 1 year of intravenous durvalumab [anti-PD-(L)1] combined with BCG (I+M for 2 years) improved disease-free survival compared with monotherapy with BCG (I+M) in patients with BCG-naïve high-risk NMIBC.¹² No significant overall survival advantage has yet been reported for either trial given that longer follow-up is required. Interestingly, both trials had a third comparison arm investigating whether the addition of an ICI to BCG induction could avoid the BCG maintenance instillations. However, ICIs plus BCG (I) schedule did not demonstrate delaying disease recurrence compared with BCG (I+M) in both studies, reinforcing the positive impact of BCG maintenance. In contrast, the phase III ALBAN trial, evaluating atezolizumab in association with BCG, did not demonstrate an EFS benefit from the addition of 1-year atezolizumab to BCG (I plus 1-year maintenance) therapy compared with BCG (I+M for 1 year), with the caveat of limitations due to the shorter duration of BCG maintenance and differences in the definition of EFS (which also included low-grade recurrence).¹³ Collectively, evidence suggests ICIs combined with BCG (I+M) as a potential novel treatment for patients with BCG-naïve high-risk NMIBC.

A highly debated issue about BCG treatment concerns the adherence to treatment schedules and the proper management of BCG local and systemic toxicity that could be related to both the induction and even more the maintenance phases of treatment.¹⁴ Furthermore, the safety profile of the addition of ICIs to BCG is about patients’ adherence to BCG and whether it would be impacted by ICI concomitant therapy. We have conducted a systematic review and meta-analysis to evaluate whether the addition of ICIs to BCG for NMIBC increases the toxicity of BCG monotherapy as well as to further review the ICI treatment-related adverse events (TRAEs).

Patients and methods

Definition of outcomes

The primary endpoint of this analysis was to assess the incidence and relative risk (RR) of TRAEs in patients treated with anti-PD-1/PD-(L)1 ICIs in association with BCG induction/maintenance compared with BCG monotherapy (SOC) for NMIBC. For each trial, the anti-PD-1/PD-(L)1 treatment associated with the BCG (I+M) schedule was considered as the experimental arm, and BCG (I+M) monotherapy as the control, given the efficacy results limited to this comparison. All-grade and high-grade (≥3) events were considered as the main outcomes and the analysis was conducted in order to identify a significant difference between the two treatment arms. For TRAEs, we also evaluated whether the type of BCG schedule (I versus I+M), when administered in association with ICIs, could affect the safety profile. Moreover, we investigated the incidence and RR of discontinuation of any treatment for TRAEs, the incidence and RR of immune-related adverse events (irAEs; either of any and high grade) and of serious adverse events (SAEs). For irAEs, we evaluated whether the type of ICI [anti-PD-(L)1 versus anti-PD-1] and the duration (1 versus 2 years) could affect the RR of events of both any and high grade. Finally, we analyzed the most frequently reported TRAEs (including dysuria, hematuria, pollakiuria, pyrexia, and urinary tract infections) and the most frequently reported irAEs (including thyroid disorders, hepatitis, rash, colitis, and adrenal insufficiency).

Selection of studies

We reviewed MEDLINE/PubMed, the American Society of Clinical Oncology (ASCO), and the European Society for Medical Oncology (ESMO) libraries for citations until November 2025. For the PubMed search, a predefined article-type filter was applied before screening. All records retained after database filtering were independently screened by two reviewers (CC and DO) and disagreements were resolved by consensus. Search criteria were limited to articles published in the English language and phase II or phase III randomized controlled trials (RCTs) in patients with BCG-naïve NMIBC treated with BCG plus ICIs in the experimental arm and with available data about TRAEs. The following inclusion criteria have been adopted: more than 10 patients enrolled and randomized trials reporting data on treatments with BCG plus ICIs in patients with BCG-naïve NMIBC. Principal exclusion criteria were overlapping publications, lack of relevant safety data, and the presence of an ICI in the treatment control arm.

The Medical Subject Headings terms used for search on PubMed, ASCO, and ESMO libraries were ‘urothelial carcinoma’, ‘NMIBC’, ‘BCG’, ‘immunotherapy’, or the name of the drugs (i.e. atezolizumab, durvalumab, and sasanlimab). If more than one publication was found for the same trial, the most recent, complete, and updated manuscript was considered for the final analysis. The quality of the trials was assessed using Jadad’s five-item scale (considering randomization, double-blinding, and withdrawals), with a final score ranging from 0 to 5,¹⁵ and the Cochrane Risk of Bias 2 (RoB 2) tool. Each study was evaluated across the following domains: bias arising from the randomization process, deviations from intended interventions, missing outcome data, measurement of the outcome, and selection of the reported result.¹⁶ Given the focus of the present meta-analysis, the risk of bias assessment was conducted with respect to safety outcomes. Each domain was deemed as ‘low risk of bias’, ‘some concerns’, or ‘high risk of bias’ according to the RoB 2 guidance. An overall risk-of-bias judgment was assigned to each study on the basis of domain-level assessments. Disagreements between reviewers were resolved by discussion and consensus.

The analysis was registered in PROSPERO—the International Prospective Register of Systematic Reviews with the identification code CRD420251179645.

Data extraction

Data extraction was conducted independently by two authors (CC and DO) according to the Preferred Reporting Items for Systematic Review and Meta-Analysis statement.¹⁷ The two authors (CC and DO) resolved any discrepancies by consensus. Information extracted for each trial included the trial name, phase, number of enrolled patients, key inclusion criteria, treatment type used in the experimental and control arms, number of any-grade and high-grade TRAEs, irAEs, SAEs, and discontinuation of treatment for any TRAEs in both treatment arms.

Statistical methodology

The calculation of incidence was carried out from the data available in each study. The proportion of patients with TRAEs, irAEs, and SAEs; the proportion of patients that discontinued treatment because of AEs; and the derived 95% confidence intervals (CIs) were calculated for each study. We also calculated the RR and CIs of events in patients assigned to BCG plus ICIs compared with the control (BCG monotherapy) in the same study; then we evaluated the RR and CIs of events on the basis of the schedule of BCG administration in the experimental arm (only BCG induction or BCG I+M, both in association with an ICI) compared with the control (BCG I+M). To calculate the 95% CIs, the variance of a log-transformed study-specific RR was derived using the delta method.¹⁸ The Mantel–Haenszel method was used for calculating the RR of TRAEs, SAEs, irAEs, and treatment discontinuation due to adverse events (AEs). Of note, in the detailed analysis of the type of AEs, when the events reported in each trial were rare, we used the Peto method for calculating the odds ratio (OR). Statistical heterogeneity between the trials included in the meta-analysis was assessed using Cochrane’s Q statistic and inconsistency was quantified with an I² statistic {100% × [(Q − df)/Q]}.¹⁹ The assumption of homogeneity was considered to be invalid for P values <0.1. Summary incidence and RRs were calculated using random- or fixed-effects models, depending on the heterogeneity of the included studies. When there was no substantial heterogeneity, the pooled estimate that was calculated on the basis of the fixed-effects model was reported using the inverse variance method. When substantial heterogeneity was observed, the pooled estimate that was calculated on the basis of the random-effects model was reported using the DerSimonian and Laird²⁰ method, which considers both within- and between-study variations.¹⁹ An indirect comparison between the groups was carried out using a chi-square test. A two-tailed P value of <0.05 was considered to be statistically significant. All the data were collected using Microsoft Office Excel 2007 (Microsoft Corporation, Redmond, WA). The statistical analyses were carried out using the RevMan software for meta-analysis (v.5.2.3, The Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, Denmark).²¹

Results

Search results

The electronic search revealed 2036 citations. After screening, 64 full-text articles were reviewed for further assessment and 61 citations were excluded because they did not meet the inclusion criteria. This review process led to the selection of three studies for final analysis based on their adequate quality and relevance for inclusion in the meta-analysis11, 12, 13 (Figure 1).

Figure 1.

Study selection process.Source: Page et al.¹⁷ This work is licensed under CC BY 4.0. To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/.

ASCO, American Society of Clinical Oncology; BCG, bacillus Calmette-Guérin; ESMO, European Society for Medical Oncology.

All included trials were randomized phase III studies, evaluating the role of systemic ICIs plus intravesical BCG compared with BCG alone in patients with BCG-naïve high-risk NMIBC. In terms of inclusion criteria with respect to the definition of high-risk NMIBC disease, ALBAN and CREST identified high-grade Ta, T1, or CIS, whereas POTOMAC enrolled patients with one of the following: T1, a high-grade or histological grade 3; CIS; or multiple or recurrent or large tumors (diameter ≥3 cm). Regarding the BCG therapy, two trials (POTOMAC and CREST) investigated the combination of an ICI and BCG using two different schedules of BCG in the experimental arms: (I+M) or induction-only (I). In these cases, we considered only data on AEs of the I+M cohort. On the contrary, the ALBAN trial had only one control arm with the BCG I+M schedule. Moreover, the duration of the BCG maintenance differed across the three trials: in POTOMAC and CREST, BCG maintenance lasted 2 years, whereas it lasted 1 year in the ALBAN study. In addition, the ICI duration also varied across the three trials: durvalumab and atezolizumab were administered for 1 year in POTOMAC and ALBAN, respectively, whereas sasanlimab was administered for 2 years in the CREST study. The main characteristics of the selected trials are shown in Supplementary Table S1 (available at https://doi.org/10.1016/j.esmoop.2026.108529).

Incidence and RR of TRAEs

For incidence of TRAEs of any grade, three studies were selected for the analysis, with a total of 1895 patients. Among them, 950 were treated in the experimental arms with BCG (I+M) plus ICIs therapy, whereas 945 patients received standard therapy with BCG (I+M) in the control arms. In the overall cohort, TRAEs of any grade were reported in 855 of 950 patients treated with BCG plus ICIs, corresponding to an incidence of 90.0%, compared with 71.8% in the control arms (679 events among 945 patients) (Supplementary Table S2, available at https://doi.org/10.1016/j.esmoop.2026.108529). Treatment with BCG (I+M) plus ICIs significantly increased the risk of TRAEs of any grade compared with controls (fixed-effect; RR 1.25, 95% CI 1.20-1.31, P < 0.00001). No significant heterogeneity was observed in this analysis (χ² = 1.15, P = 0.56, I² = 0%) (Figure 2A).

Figure 2.

Relative risk for TRAEs of (A) any grade and (B) high grade in patients treated with BCG (I+M) plus ICIs or BCG (I+M).

ALBAN, phase III trial of atezolizumab plus BCG versus BCG alone in BCG-naïve high-risk NMIBC; BCG, bacillus Calmette-Guérin; CI, confidence interval; CREST, phase III trial of sasanlimab plus BCG versus BCG alone in BCG-naïve high-risk NMIBC; ICI, immune checkpoint inhibitor; I+M, induction plus maintenance; M-H, Mantel–Haenszel; POTOMAC, phase III trial of durvalumab plus BCG versus BCG alone in BCG-naïve high-risk NMIBC; TRAE, treatment-related adverse event.

The incidence of high-grade TRAEs was 24.3% in the experimental arms (231 events among 950 patients) and 6% in the control arms (57 reported cases in 945 patients). Similarly, treatment with BCG plus ICIs did significantly increase the risk of high-grade toxicity (fixed-effect; RR 4.02, 95% CI 3.06-5.30, P < 0.00001). No substantial heterogeneity was documented in the RR analysis for high-grade TRAEs (χ² = 5.23, P = 0.07, I² = 62%) (Figure 2B).

In order to assess the role of BCG maintenance instillations in the toxicity of the combination of BCG plus ICIs, we compared the RR of TRAEs of any and high grade between BCG (I+M) plus ICIs and BCG (I) plus ICIs. For this I+M versus I comparison, data were available for two of the three selected studies for a total of 688 patients receiving BCG (I+M) plus ICIs and 689 patients treated with BCG (I) + ICIs. BCG (I) plus ICIs was responsible for a lower risk of TRAEs of any grade (fixed-effect; RR 0.90, 95% CI 0.86-0.95, P < 0.0001) and high grade (fixed-effect; RR 0.74, 95% CI 0.60-0.90, P = 0.003), compared with BCG (I+M) plus ICIs.

Incidence and RR of SAEs

For incidence of SAEs, data were available from all three selected studies. In the overall cohort, SAEs were reported in 167 of 950 patients treated with BCG (I+M) plus ICIs, corresponding to an incidence of 17.6%, compared with 4.1% in the control arms (39 events among 945 patients). Treatment with BCG plus ICIs (I+M) significantly increased the risk of SAEs compared with BCG (random-effects; RR 4.54, 95% CI 2.09-9.85, P = 0.0001). Significant heterogeneity was observed in this analysis (χ² = 8.99, P = 0.01, I² = 78%) (Figure 3). No treatment-related deaths were reported in the BCG (I+M) plus ICI or BCG (I+M) arms of the three trials. In the CREST BCG induction-only plus sasanlimab arm, two treatment-related deaths occurred due to bacterial pneumonia and myocarditis.

Figure 3.

Relative risk for SAEs in patients treated with BCG (I+M) plus ICIs compared with BCG (I+M).

ALBAN, phase III trial of atezolizumab plus BCG versus BCG alone in BCG-naïve high-risk NMIBC; BCG, bacillus Calmette-Guérin; CI, confidence interval; CREST, phase III trial of sasanlimab plus BCG versus BCG alone in BCG-naïve high-risk NMIBC; ICI, immune checkpoint inhibitor; I+M, induction plus maintenance; M-H, Mantel–Haenszel; POTOMAC, phase III trial of durvalumab plus BCG versus BCG alone in BCG-naïve high-risk NMIBC; SAE, serious adverse event.

Incidence and RR of irAEs

Immune-related toxicity was reported in all three selected studies. irAEs of any grade were reported in 381 of 950 patients treated with BCG (I+M) plus ICIs, corresponding to an incidence of 40.1%, compared with 32 events in 945 patients treated with BCG (I+M) in the control arms (incidence of 3.4%). Treatment with BCG plus ICIs significantly increased the risk of irAEs of any grade compared with BCG (random-effects; RR 15.18, 95% CI, 4.34-53.14, P < 0.0001). Significant heterogeneity was observed in this analysis (χ² = 17.03, P = 0.0002, I² = 88%) (Figure 4A).

Figure 4.

Relative risk for irAEs of (A) any grade and (B) high grade in patients treated with BCG (I+M) plus ICIs or BCG (I+M).

ALBAN, phase III trial of atezolizumab plus BCG versus BCG alone in BCG-naïve high-risk NMIBC; BCG, bacillus Calmette-Guérin; CI, confidence interval; CREST, phase III trial of sasanlimab plus BCG versus BCG alone in BCG-naïve high-risk NMIBC; ICI, immune checkpoint inhibitor; I+M, induction plus maintenance; irAE, immune-related adverse event; M-H, Mantel–Haenszel; POTOMAC, phase III trial of durvalumab plus BCG versus BCG alone in BCG-naïve high-risk NMIBC.

In terms of high-grade irAEs, BCG (I+M) plus ICIs was responsible for 96 events among 950 patients, corresponding to an incidence of 10.1%, compared with seven events in 945 patients treated with BCG (I+M) in the controls (incidence of 0.7%). Treatment with BCG (I+M) plus ICIs significantly increased the risk of high-grade irAEs compared with BCG (I+M) (random-effects; RR 9.99, 95% CI 2.12-47.14, P = 0.004). Significant heterogeneity was observed in this analysis (χ² = 6.44, P = 0.04, I² = 69%) (Figure 4B).

Treatment with anti-PD-(L)1 for 1 year in addition to BCG was responsible for irAEs of any grade in 38.6%, compared with 4.5% with BCG. The incidence of any-grade irAEs with anti-PD-1 for 2 years plus BCG was 42.6%, compared with 1.4% with BCG. No significant difference was observed between the two groups (P = 0.23) (Supplementary Figure S1A, available at https://doi.org/10.1016/j.esmoop.2026.108529). As for high-grade irAEs according to the type and duration of ICI, treatment with anti-PD-(L)1 for 1 year plus BCG was associated with 6.9% of high-grade irAEs, compared with 1.2% with BCG. The incidence of high-grade irAEs with anti-PD-1 for 2 years plus BCG was 15.6%, compared with no high-grade irAEs with BCG. The type and duration of ICI in association with BCG did not significantly affect the RR of high-grade irAEs (P = 0.05) (Supplementary Figure S1B, available at https://doi.org/10.1016/j.esmoop.2026.108529).

Incidence and RR of treatment discontinuation due to TRAES

In the experimental arms of BCG (I+M) plus ICIs, 302 of 950 patients discontinued any treatment because of AEs, corresponding to an incidence of 31.8%, compared with 108 in 945 patients treated with BCG (I+M) in control arms (incidence of 11.4%). Treatment with BCG (I+M) plus ICIs significantly increased the risk of any treatment discontinuation compared with control (random-effect; RR 2.87, 95% CI 1.50-5.52, P = 0.002). Substantial heterogeneity was documented in this analysis (χ² = 19.16, P < 0.0001; I² = 90%) (Figure 5).

Figure 5.

Relative risk for treatment discontinuation due to TRAEs in patients treated with BCG (I+M) plus ICIs compared with BCG (I+M).

ALBAN, phase III trial of atezolizumab plus BCG versus BCG alone in BCG-naïve high-risk NMIBC; BCG, bacillus Calmette-Guérin; CI, confidence interval; CREST, phase III trial of sasanlimab plus BCG versus BCG alone in BCG-naïve high-risk NMIBC; ICI, immune checkpoint inhibitor; I+M, induction plus maintenance; M-H, Mantel–Haenszel; POTOMAC, phase III trial of durvalumab plus BCG versus BCG alone in BCG-naïve high-risk NMIBC; TRAE, treatment-related adverse event.

Incidence and RR of selected AEs

We evaluated the incidence and RR of the most frequent TRAEs of any and high grade, including dysuria, hematuria, pollakiuria, pyrexia, and urinary tract infections, associated with the BCG (I+M) schedule plus ICIs compared with BCG (I+M) (Supplementary Table S3, available at https://doi.org/10.1016/j.esmoop.2026.108529). None of these AEs were significantly increased with BCG (I+M) plus ICIs compared with BCG (I+M), with the exception of urinary tract infection of high grade (Peto, fixed; OR = 4.22, 95% CI 1.47-12.08, P = 0.007) (Supplementary Table S4, available at https://doi.org/10.1016/j.esmoop.2026.108529).

Finally, we analyzed the incidence and RR of the most frequent systemic irAEs of any and high grade, including thyroid disorders, hepatitis, rash, colitis, and adrenal insufficiency, with regard to the BCG (I+M) schedule plus ICIs compared with BCG (Supplementary Table S3, available at https://doi.org/10.1016/j.esmoop.2026.108529). BCG plus ICIs significantly increased the RR of developing thyroid disorders of any grade (fixed-effect; RR 24.12, 95% CI 11.06-52.60, P < 0.00001), but not of high grade (Peto, fixed; OR = 7.43, 95% CI 1.05-52.84, P = 0.05); increased the RR of hepatitis of both any grade (fixed-effect; RR 19.78, 95% CI 4.80-81.52, P < 0.0001) and high grade (Peto, fixed; OR = 7.53, 95% CI 3.31-17.13, P < 0.00001); increased the RR of rash of both any grade (fixed-effect; RR 40.53, 95% CI 10.00-164.22, P < 0.0001) and high grade (Peto, fixed; OR = 7.55, 95% CI 2.30-24.80, P = 0.0009); increased the RR of colitis of any grade (fixed-effect; RR 7.93, 95% CI 1.82-34.54, P = 0.006), but not of high grade (Peto, fixed; OR = 4.20, 95% CI 0.95-18.55, P = 0.06); and increased the RR of adrenal insufficiency of both any grade (fixed-effect; RR 12.27, 95% CI 2.33-64.53, P = 0.003) and high grade (Peto, fixed; OR = 7.48, 95% CI 2.02-27.74, P = 0.003) (Supplementary Table S4, available at https://doi.org/10.1016/j.esmoop.2026.108529).

Quality of the studies

All the studies were RCTs and they were of good quality according to Jadad’s scale (scores ≥3) (Supplementary Table S2, available at https://doi.org/10.1016/j.esmoop.2026.108529). According to the Cochrane RoB 2 tool, the overall risk of bias was judged as ‘some concerns’ for all trials. The risk of bias in each study is reported in Supplementary Figure S2 (available at https://doi.org/10.1016/j.esmoop.2026.108529).

Discussion

TURBT followed by intravesical BCG has been the current SOC for patients with high-risk NMIBC. As for the optimal schedule and duration of intravesical BCG as SOC, 6 weeks course (induction) followed by further 3-week instillations after routine cystoscopy at 3, 6, 12, 18, 24, 30, and 36 months (3-year maintenance phase), as recommended in the Southwest Oncology Group (SWOG) trial protocol, is superior to induction treatment only and is the single most important factor in improving outcomes in NMIBC.^6,22 However, only a limited number of patients (16%) were able to complete the maintenance because of significant toxicity.⁶ Subsequent attempts failed in identifying alternative schedules (either reducing the doses, the number of instillations, or the treatment duration to 1 year instead of 3 years) to increase patients’ adherence to therapy without affecting the antitumor efficacy.22, 23, 24

In this context, CREST and POTOMAC phase III trials evaluated the addition of ICIs (sasanlimab and durvalumab, respectively) to BCG (I+M) in BCG-naïve high-risk NMIBC, reporting consistently improved oncologic outcomes in terms of reduction in the risk of disease recurrence compared with BCG (I+M) alone, supporting the expectation that combination strategies might soon reshape the treatment algorithm of routine practice.^11,12 A rigorous characterization of treatment-related toxicity is important given that the clinical value of therapy escalation in high-risk NMIBC depends not only on efficacy but also on tolerability, adherence, and capability to complete treatment. Furthermore, ICI and BCG differ not only in the variety of side-effects but also in the potential comorbidities that represent an absolute contraindication to treatment (i.e. autoimmune disease for ICI and immunosuppressive conditions for BCG), requiring a careful selection of patients.

Historically, intravesical BCG has been considered a generally well-tolerated treatment, with BCG-related side-effects occurring during the first year of treatment and largely involving lower urinary tract symptoms of bladder irritability (mainly inflammatory cystitis symptoms). In a relatively low proportion of cases (not exceeding 5%-10%), intravesical BCG can be responsible for severe systemic complications, including fever, arthralgia, malaise, allergic reactions, and, only exceptionally (<1%), BCG sepsis.^14,25, 26, 27, 28 Regarding the BCG-related toxicity, the safety outcomes observed in the BCG control arms of the studies included in our analysis were consistent with the historical experience of intravesical BCG therapy.^22,23,28 In particular, BCG (I+M) was responsible for TRAEs in ∼70% of cases, mainly low-grade and generally involving the lower urinary tract (i.e. dysuria in 30% and pollakiuria, hematuria, and urinary tract infections in ∼15% of cases). On the contrary, intravesical BCG rarely caused SAEs (high-grade TRAEs in only 6% and SAEs in <5% of cases), was anecdotally associated with systemic immune-related toxicity (<1%), and led to treatment discontinuation in a low proportion of patients (∼11%). In summary, these data support the manageability of intravesical BCG (induction plus 2-year maintenance), further improved thanks to the experience of clinicians consolidated over the years.

As expected, the addition of systemic immunotherapy to ICIs against PD-1/PD-L1 introduces expected irAEs as well documented in other disease indications. Immune-mediated side-effects can involve any organ system—with the gastrointestinal tract, endocrine glands, skin, and liver as the most frequently affected—and are generally of low grade and broadly manageable in daily clinical practice, although they may infrequently be severe.^29,30 Some features make AEs related to ICIs unpredictable: they are dose-independent, can emerge anytime (also after the end of the treatment), and need treatment (immunosuppressive therapy) to resolve.^29,30 Generally, single-agent ICI targeting PD-1/PD-L1 has been associated with an incidence of irAEs of ∼65%, even though several factors can affect this rate, including the tumor type and the disease setting.³¹ A meta-analysis of fatal irAEs in patients treated with anti-PD-1/PD-(L)1 reported a rate of fatal events of <0.5%,³² reinforcing the key role of appropriate management of toxicity for a favorable outcome.

To our knowledge, this is the first systematic review and meta-analysis to assess the safety of ICIs combined with BCG in patients with BCG-naïve high-risk NMIBC. The aim of our analysis was to assess whether the combination of BCG (I+M) and ICIs could increase the overall burden of treatment-related side-effects. We did observe a 25% increase in the risk of developing AEs of any grade, which can be either local (with an incidence ranging from 15% to 30%) or systemic (from 2% to 15% of cases). Furthermore, the combination of BCG (I+M) plus ICIs quadrupled the risk of high-grade TRAEs (which can occur in approximately one-quarter of treated patients), nearly 10-fold the risk of immune-mediated AEs of high grade (with an incidence of ∼10%), and tripled the risk of discontinuing any of the treatments (which occurred in more than one-third of patients). Importantly, our analysis did not identify an increased risk of local AEs (with the exception of high-grade urinary tract infection) with the addition of ICIs to BCG. These findings support the concept that systemic immune modulation does not substantially worsen the local tolerability of BCG and that the incremental toxicity burden could primarily be driven by immune-mediated effects rather than by amplification of local urinary symptoms. To further support this hypothesis, the addition of ICIs to BCG significantly increased the risk of developing systemic irAEs, although they remained overall uncommon and particularly rare at high grade (with an incidence ranging from 0.4% to 2.4%), consistent with the established clinical experience of immune checkpoint blockade. This distinction has practical implications because irAEs require physician education, multidisciplinary collaboration, and management by dedicated clinicians.^33,34 Interestingly, when evaluating whether the duration of ICIs (1 versus 2 years) could affect the risk of developing irAEs, we found no significant differences except for a trend toward an increased risk of high-grade irAEs with 2 years of sasanlimab. However, the few numbers of events and the presence of a single study (i.e. CREST) in one of the two subgroups, together with the limitation of indirect comparisons, make this finding inconclusive, and it should thus be interpreted with caution.

Of note, given the reported higher toxicity associated with the maintenance phase of the BCG treatment, CREST and POTOMAC studies explored the possibility of adding ICIs to BCG induction-only treatment (thus sparing the maintenance phase) compared with BCG (I+M), with the hypothetical aim of reducing the overall BCG therapy burden. Although in both studies, BCG (I) plus ICI combinations did not delay disease recurrence compared with BCG (I+M), reinforcing the central role of BCG maintenance as a key component also when BCG is combined with ICIs, we evaluated whether the BCG schedule (I versus I+M) could affect the toxicity of the treatment in terms of TRAEs of any and high grade. Interestingly, BCG (I+M) plus ICIs was associated with a 10% lower risk of any-grade TRAEs and a 26% lower risk of grade ≥3 TRAEs compared with BCG (I) plus ICIs. These results suggest that the maintenance phase of BCG treatment, although essential for the oncological outcomes, in a certain way affects the toxicity and therefore deserves careful management of the patient in all phases of the therapy.

Taken together, these findings suggest that treatment intensification by the addition of ICI to BCG (I+M) represents a potential novel strategy for BCG-naïve high-risk NMIBC that requires additional clinician education for appropriate recognition and management of TRAEs and immune-mediated AEs. We do support the feasibility of this novel combination given the antitumor activity reported but also underscore the need for careful patient selection and multidisciplinary management. Implementation should emphasize baseline risk stratification (frailty, comorbidities, pre-existing autoimmune disease, and chronic immunosuppression), patient education for early recognition of immune-mediated symptoms, and continuous cross-talk between urologists and medical oncologists for toxicity management.

This meta-analysis has several limitations. The lack of access to raw data represents the main limit. Furthermore, the evidence base is restricted to three phase III trials, resulting in a relatively limited sample size. In addition, although all trials enrolled patients with high-risk BCG-naïve NMIBC, they differed in several key aspects, including the ICI agent [anti-PD-1/PD-(L)1] and dosing schedule, ICI treatment duration (1 versus 2 years), the intensity and duration of BCG exposure (maintenance phase of instillation for 1 or 2 years), and length of safety follow-up, which may have contributed to heterogeneity across studies. Particularly, in the CREST trial, sasanlimab was administered for 2 years, which could reflect the higher incidence of irAEs. Finally, it was not possible to assess the contribution of each agent (i.e. BCG treatment, ICI, or both) in determining the rate of treatment discontinuation given the lack of a uniform way to report this information in the selected clinical trials.

Ongoing clinical trials are evaluating multiple strategies (including gene therapy, vaccines, agents aimed at enhancing the T-helper type 1 immune response pathway, intravesical therapies targeting fibroblast growth factor receptor alterations or novel intravesical devices associated with ICIs, and antibody-drug conjugates) to improve outcomes in patients with BCG-naïve NMIBC for whom BCG alone may be insufficient (Supplementary Table S5, available at https://doi.org/10.1016/j.esmoop.2026.108529). It would be interesting to evaluate in the future the impact of potential increased toxicity of these novel agents combined with BCG.

In conclusion, adding ICIs to intravesical BCG (I+M) in BCG-naïve high-risk NMIBC resulted in a higher overall burden of TRAEs, with increased rates of treatment discontinuation, grade ≥3 AEs, and SAEs. These AEs were mainly immune-mediated, whereas the incidence of typical BCG-related local urinary AEs did not appear meaningfully increased. Overall high-grade irAEs remained relatively infrequent, supporting the feasibility of this approach, balancing efficacy and toxicity in carefully selected patients managed within a multidisciplinary framework with standardized irAEs monitoring and prompt treatment algorithms.