Background: PD-L1/PD-1 inhibitors have transformed the treatment (tx) of advanced NSCLC. Evidence suggests that the occurrence of irAEs with these agents may predict improved outcomes in cancers such as NSCLC. Atezolizumab (atezo; anti–PD-L1) has shown efficacy and tolerability in NSCLC and is currently approved in the 1L and 2L+ settings. The Ph 3 IMpower130, IMpower132 and IMpower150 trials evaluated atezo + chemo ± bevacizumab (bev) as 1L tx of NSCLC. We explore the association between irAEs and efficacy in these trials. Methods: Each trial enrolled tx-naive patients (pts) with nonsquamous stage IV NSCLC. Pts were randomized to: carboplatin (carbo) + nab-paclitaxel alone or with atezo in IMpower130; carbo or cisplatin alone or with atezo in IMpower132; atezo (A) + bev (B) + carbo + paclitaxel (CP), ACP or BCP in IMpower150. Data were pooled (data cutoffs: Mar 15 2018 [IMpower130]; May 22 2018 [IMpower132]; Sep 13 2019 [IMpower150]) and analyzed by tx (atezo-containing vs control) and irAE status. A time-dependent Cox model and landmark analyses at 1, 3, 6 and 12 mo were used to control for immortal bias. Study protocols required atezo tx interruption/discontinuation for grade (Gr) ≥3 irAEs. Results: 2503 pts were included in the analysis (atezo, n = 1577; control, n = 926). In both arms, baseline characteristics were generally balanced between pts with irAEs (atezo, n = 753; control, n = 289) and without irAEs (atezo, n = 824; control, n = 637). Any-Gr irAEs occurred in 48% (atezo) and 32% (control) of pts; Gr 3-5 irAEs occurred in 11% (atezo) and 5% (control). The most common irAEs (atezo vs control) were rash (28% vs 18%), hepatitis (lab abnormalities; 15% vs 10%) and hypothyroidism (12% vs 4%). Median time to onset of first irAE was 1.7 (atezo) vs 1.4 mo (control). OS HRs (95% CI) from the time-dependent Cox model between pts with vs without irAEs were 0.69 (0.60, 0.78) in the atezo arm and 0.82 (0.68, 0.99) in the control arm; after excluding rash (perceived as the least specific irAE), OS HRs (95% CI) were 0.75 (0.65, 0.87) and 0.90 (0.71, 1.12), respectively. OS landmark data are in the Table. Conclusions: In this exploratory pooled analysis, pts with irAEs had longer OS vs pts without irAEs in the atezo-containing and control arms per the time-dependent Cox model and landmark analyses; this trend remained for the atezo arm after excluding rash. Landmark analyses suggest that in the atezo arm, pts with Gr 1/2 irAEs had the longest OS and pts with Gr ≥3 irAEs had the shortest OS, potentially due to tx interruption/discontinuation. Clinical trial information: NCT02367781; NCT02657434; NCT02366143