Characteristics and clinical significance of cutaneous immune-related adverse events during atezolizumab therapy: An individual participant data meta-analysis

Sources of data and ethical approval

This IPD meta-analysis was conducted according to the PRISMA-IPD statement based on the Vivli database.^11,12 All RCTs on ICIs, including Programmed Death-1/ligand 1(PD-1/L1) and Cytotoxic T-Lymphocyte associated Antigen 4 (CTLA-4) antagonists, for cancer treatment were searched, including both phase II and III clinical trials, until May 2021. The design of this study and the use of data were approved by the Vivli platform after review by each sponsor (ID:00007857). Consent was waived, as we analysed only secondary data, and participant consent for the use of their data in research after the original study was obtained by respective study sponsors.

Synopses of the enrolled RCTs

A total of 27 trials were found, but only 17 trials, sponsored by Daiichi and Roche, provided their IPDs. Besides, two RCTs discontinued treatment owing to severe side effects, and follow-up results after treatment were absent. Finally, 15 RCTs of atezolizumab (a PD-L1 antagonist)^13-27 from Roche, were included in the final analysis. The flow chart for this study has been shown in Supplementary Figure S1, and basic information has been shown in Supplementary Table S1. The study duration of all RCTs lasted from 2013 to 2024. Only patients in the atezolizumab group were enrolled in this study. Seven different advanced-stage cancers were included in this study, with non-small cell lung cancer (NSCLC) accounting for 46.7%. For the atezolizumab dosage, 14 (93.3%) trials used a dosage of 1200 mg every 3 weeks via intravenous infusion.

Classification and definition of cirAEs

The occurrence of cirAEs was monitored from the beginning of the study to 30 days after the last administration of atezolizumab (90 days for severe adverse events). The causality of the cirAEs was assessed by third-party evaluators according to the E2A and E2B criteria for ICH, and the onset and recovery times were recorded. Adverse events were graded according to the National Cancer Institute Common Terminology Criteria for AEs (CTCAE), and adverse events with negative correlations were excluded from this study. The names of the cirAEs were recorded according to the Medical Dictionary for Regulatory Activities (MedDRA).^28,29 The hierarchy of MedDRA includes the System Organ Class (SOC), the High-Level Group Term (HLGT), the High-Level Term (HLT), the Preferred Term (PT), and the Lowest Level Term (LLT). For dermatological diseases, there were 10 HLGTs, 56 HLTs, and 1,552 PTs [See Supplementary Table S2]. This study enrolled all patients with cirAEs by selecting the SOC term “skin and subcutaneous tissue disorders” from the MedDRA for analysis. These are also the standardised terminologies for adverse events reported in all included RCT studies.

Evaluation, follow-up, and outcomes

Atezolizumab was intravenously infused at a dose of 1200 mg every 21 days (in the CO39262 study, the dose was 840 mg every 14 days). Treatment was stopped, but the dose was not reduced if the cancer progressed, intolerable adverse reactions occurred, or if the patient died. Other anticancer medications should be prescribed based on the drug instructions. According to the Response Evaluation Criteria in Solid Tumours (RECIST),³⁰ tumour assessments (enhanced CT, MRI, or PET-CT scan) were performed regularly until disease progression or death. Patients who continued the trial after disease progression continued to undergo tumour assessment every 6 weeks until the protocol was discontinued.

The primary outcome of interest was overall survival (OS), defined as the time from randomisation to death from any cause. The secondary outcome of interest was investigator-assessed progression-free survival (PFS), defined as the time from randomisation to disease progression or death according to the RECIST.

Assessment and grading of PD-L1 expression

PD-L1 expression was centrally assessed by immunohistochemistry using SP142 and SP263. PD-L1 expression was assessed by scoring the percentage of PD-L1-expressing tumour cells (TC0: <1% of tumour cells expressing PD-L1; TC1: ≥1% and <5%; TC2: ≥5% and <50%; TC3: ≥50%).³¹

Statistical analysis

For descriptive analyses of patient characteristics and types of cirAEs, simple pooling was performed, and the cirAEs were analysed at the HLGT and PT levels. In all analyses, studies on the same cancer were considered as one cohort, and both 1-stage and 2-stage approaches were applied.¹² The one-stage approach involved a standard aggregate data meta-analysis of the incidence and timing of cirAEs as well as their influence on the OS and PFS of the enrolled patients. In the 2-stage approach, a random-effects (random intercept) generalised logistic mixed-effects model was used to estimate adjusted hazard ratios (HRs) and 95% confidence intervals (CIs).

All statistical analyses were performed in real-time using the Vivli analytics online platform. Quantitative data are presented as medians with interquartile ranges, and differences were compared using the Mann-Whitney test. Categorical data were presented as numbers and percentages, and differences were compared using the chi-square test. HRs and 95% CIs were calculated using an adjusted time-dependent Cox regression model. The multivariable COX analysis in different models was adjusted by some of the following indexes: age, sex, body mass index (BMI), ethnic region, cancer type, Eastern Cooperative Oncology Group (ECOG) score, treatment-naive status, treatment duration, PDL1 expression, metastatic status, and combination therapy. Survival curves were compared using the Kaplan-Meier test with landmark times of 6, 12, and 18 months.

Participant characteristics

A total of 5,870 patients were enrolled in this study, of which 1617 patients (incidence: 27.5%) experienced cirAEs during treatment. The baseline characteristics of the enrolled patients have been shown in Table 1. Most patients experienced mild-to-moderate AEs, while 6.1% experienced severe AEs. In terms of cancer type, 59.3% were NSCLC patients, 86.5% of whom had metastatic cancer. The median duration of treatment with atezolizumab was 174.3 days. According to the inclusion criteria, all enrolled patients were in good physical condition; therefore, comorbidities were not analysed as confounding factors in this study.

Incidence and timing of cirAEs at HLGT level

The incidence [Figure 1a] and timing [Figure 1b] of the various types of cirAEs at the HLGT level have been shown in Figure 1. From an HLGT perspective, the three most common types of cirAEs are epidermal and dermal conditions (EDCs), skin appendage conditions (SACs), and angioedema/urticaria. The earliest types of cirAEs are skin appendage disorders, angioedema/urticaria, and EDCs.

Close

Figure 1:

The (a) incidence and (b) onset time of each type of cirAEs. Different colours represent different types of cirAEs at HLGT level.

Export to PPT

The meta-analysis of the overall incidence in different cohorts has been shown in Supplementary Figure S2. The 1- and 2-stage methods yielded similar overall incidence rates, which were 28% (26-29%) and 30% (25-34%), respectively. In terms of tumour type, the incidence rates ranged from 22% to 42%, with the highest being 42% (35-48%) in melanoma, followed by 39% in renal cell carcinoma. According to the results of the random-effects model, there was low heterogeneity among the studies (τ2=0, I2=91.5%, H2=11.74, P<0.001); therefore, there was no need for result adjustment.

The individual incidence and timing of cirAE according to the meta-analysis have been shown in Supplementary Table S3. The incidence of cirAEs is 25-34%, with a wide range of time intervals for initial onset, ranging from 18 to 913 days. The spectrum of cirAEs also varied significantly among different tumour types. Among all cancers, patients with melanoma had a significantly greater incidence of pigmentary disorders (3.91% vs. 0.32-1.50%). Treatment for melanoma has also led to relatively high incidence rates of other types of cirAEs. In contrast, patients with small-cell lung cancer (SCLC) had a significantly higher incidence of urticaria/angioedema during treatment (1.52% vs. 0.56-1.10%).

Clinical characteristics of cirAEs at PT level

The specific types and characteristics of cirAEs at the PT level have been shown in Table 2. From the PT perspective, the top five incidence rates were as follows: rash/eruption/exanthem (83.9%), pruritus (62.5%), alopecia (51.6%), dry skin (16.6%), and hand-and-foot syndrome (9.9%). The earliest types to occur were alopecia, hyperhidrosis, and urticaria, with median onset times of 21, 22, and 43 days, respectively. On the other hand, bullous dermatosis, lichenoid dermatosis, and benign lesions had median onset times of 420.5, 334, and 197 days, respectively.

Most types of cirAEs were mild and recovered spontaneously or after treatment. However, it should be noted that bullous dermatosis had the highest proportion of severe AE (25%), recovery with sequelae at trial termination (16.7%), causing drug interruption/withdrawal (66.7%), and prolonged hospitalisation (25%), with all patients undergoing medical treatment. Psoriasis was also noted, with 8.1% SAE incidence and 8.1% severe outcomes. Other detailed characteristics of cirAE have been listed in Table 2.

The effect of cirAEs on survival

The influence of the occurrence of different types of cirAEs on the OS and PFS of patients was analysed using an adjusted Cox regression. As shown in Figures 2a and b, the occurrence of any type of cirAE was an independent protective factor for both OS (HR=0.40-0.72) and PFS (HR=0.52-0.84) in all three models in 1-stage analysis. According to the 2-stage analyses, the results were similar, and the HRs were 0.41 (0.37-0.45) for OS and 0.53 (0.6-0.60) for PFS. Figure 2c shows the influence of the different types of cirAEs on OS and PFS. Urticaria, erythema, alopecia, pruritus, rash/eruption/exanthem, acne, and lichenoid dermatosis were independent protective factors for OS, while dry skin, purpura, dermatitis/eczema, alopecia, pruritus, and photodermatitis were independent protective factors for PFS in the adjusted Cox model.

Close

Figure 2:

Impact of the occurrence of cirAEs on overall survival and progression-free survival. Forest plot of the impact of occurrence of any cirAEs on (a) overall survival and (b) progression-free survival in different cohorts. (c) The impact of different types of cirAEs on overall survival and progression-free survival. Red triangles represent statistically significance. HCC: Hepatocellular carcinoma.

Export to PPT

The influence of different types of cirAEs according to HLGT levels on survival has been shown in Supplementary Table S4. Angioedema and urticaria (AU), cornification and dystrophic skin disorders (CDSDs), EDCs, pigmentation disorders (PD), SAC, and skin and subcutaneous tissue disorders (SST) were protective factors for OS according to the univariate analysis, but only EDC was an independent protective factor in both Model 1 (HR=0.416, 0.371-0.466, P<0.001) and Model 2 (HR=0.703, 0.635-0.778, P<0.001). Similarly, AU, EDC, PD, SST, SAC, and skin vascular abnormalities (SVAs) were found to be protective factors for PFS in univariate analysis, but only EDC was an independent protective factor in Model 1 (HR=0.555, 0.508-0.605; P<0.001) and Model 2 (HR=0.832, 0.771-0.897; P<0.001).

Survival curves categorised by cirAEs

The Kaplan-Meier survival curves of OS and PFS categorised by cirAEs were analysed at 6 and 12 months, as shown in Figure 3. Patients with cirAEs had significantly longer OS than those with no cirAEs at 6 months [Figure 3a, P=0.003] or 12 months [Figure 3b, P=0.045]. For PFS, patients with cirAEs had significantly longer survival than patients with no cirAEs at a landmark time of 6 months [Figure 3c, P=0.004], whereas at a landmark time of 12 months, the two groups were not significantly different [Figure 3d, P=0.055]. Considering that most types of cirAEs occurred within 12 months of the initial ICI treatment, patients with cirAEs had significantly longer PFS than patients with no cirAEs. The OS and PFS curves for the occurrence of different types of cirAEs according to HLGT grade at a landmark time of 6 months have been shown in Supplementary Figures S3 and S4. Only the occurrence of EDC was associated with significantly longer OS (P=0.004) and PFS (P=0.037).

Close

Figure 3:

Survival curves of overall survival and progression-free survival. (a) OS landmark time: 6 months, (b) OS landmark time: 12 months, (c) PFS landmark time: 6 months, (d) PFS landmark time: 12 months.

Export to PPT

The impact of management and outcome of cirAEs on survival

Finally, we analysed the influence of the management or outcomes of cirAEs on the OS and PFS of all patients with cirAEs, as shown in Table 3. CirAE undergoing medical treatment (HR=0.731, 0.622-0.859, P<0.001), prolonged hospitalisation (HR=0.442, 0.209-0.937, P=0.033), and drug cessation (HR=0.657, 0.457-0.943, P=0.023) were independent protective factors for OS, while cirAE not recovered at trial termination (HR=1.191, 1.025-1.384, P=0.022) were independent risk factors for poor OS. For PFS, only cirAE undergoing medical treatment was an independent protective factor (HR=0.785, 0.699-0.881, P<0.001).