The Immune-related adverse event (IRAE) Likelihood Score (ILS) identifies “pure” IRAEs strongly associated with outcome in a phase I-II trial population

Background: Immune-related adverse events (IRAE) pose a significant diagnostic and therapeutic challenge in patients treated with immuno-oncology (IO) drugs. IRAEs have been found to correlate with better outcome, but studies are conflicting on the magnitude and significance of this correlation. Estimating the true incidence of IRAEs is particularly difficult in the early phase 1/2 trial setting, with factors contributing to both over- and under-estimation. A key issue is the lack of IRAE diagnostic criteria, necessary to discriminate “pure” IRAEs from other treatment-related adverse events not sustained by an autoimmune process. We present the definitive analysis of a retrospective study conducted on patients treated with IO drugs within phase 1-2 trials at our institute. Methods: We extensively reviewed clinical characteristics and temporal dynamics of IRAEs and empirically developed an IRAE Likelihood Score (ILS) based on availability of invasive or highly specific tests, response to immune suppression, temporal correlation with IO drug initiation, evidence ruling out alternative cause, known relationship with IO. We defined High Confidence (HC) or Low Confidence (LC) IRAEs by clinical consensus and estimated correlation with survival of treatment-related events by multivariate Cox analysis. To mitigate immortal time-bias, we also analysed data at 2- month landmark and modeling IRAEs as time-dependent covariate. Results: 29.2% of 202 patients developed 1 treatment-related adverse event. ILS 5 discriminated between HC and LC IRAEs with >93% specificity and sensitivity. HC IRAE patients (n¼24) had significantly improved outcome for PFS and OS, irrespective of the model used (landmark, time-dependent or uncorrected, HR for PFS ranging 0.24-0.44, for OS 0.18-0.23, all p values <0.01), whereas LC IRAE patients (n¼35) showed no statistically significant correlation. Conclusions: ILS provides a simple system to identify bona fide IRAEs, pruning for other treatment-related events likely due to different pathophysiology. Applying stringent criteria leads to lower and more reliable esti