Background: Previous results from the ATTRACTION-3 phase 3 trial demonstrated a significant improvement in overall survival and a favorable safety profile compared with taxane chemotherapy (CT) in previously-treated ESCC patients. To our knowledge, no long-term efficacy and safety data of this immune checkpoint inhibitor has been reported in ESCC. Herein, we report the three-year survival data of Nivo in ESCC. Methods: In ATTRACTION-3, 419 patients with unresectable advanced or recurrent ESCC refractory or intolerant to 1 prior fluoropyrimidine/platinum-based CT were randomized in a 1:1 ratio to receive Nivo (N = 210) or the investigator’s choice of CT (paclitaxel or docetaxel) (N = 209) until disease progression or unacceptable toxicity. The primary endpoint was overall survival (OS). A subgroup analysis of OS according to the best overall response (BOR) was performed. The onset of treatment-related adverse events of special interest over time in the Nivo arm was also evaluated. Results: As of data cut-off on 25 May 2020, 3 years after the last patient was enrolled, the median OS (mOS) was 10.91 months with Nivo versus 8.51 months with CT [hazard ratio (HR) 0.79; 95% confidence interval (CI) 0.64-0.97]. The OS rates of patients with Nivo and CT were 20.2 % and 13.5 % at 24 months, and 15.3% and 8.7% at 36 months, respectively. In the subgroup analysis of OS by BOR, mOS in CR/PR patients were 19.91 and 15.41 months (HR 0.84, 95%CI 0.46-1.54) and that in SD patients were 17.38 and 9.36 months (HR 0.45, 95%CI 0.26-0.78) in the Nivo and CT arm, respectively. Furthermore, mOS in PD patients were 10.91 and 6.18 months (HR 0.56, 95%CI 0.33-0.95) in the Nivo and CT arm, respectively. No new safety signals were detected during the three-year follow-up. Time to onset of the event of special interest was within the range of events previously observed in other indications. Conclusions: At three-year follow up, Nivo continued to show improved OS over CT in pretreated patients with advanced ESCC patients. Nivo showed a longer mOS than CT regardless of BOR. During the three-year follow-up, no new safety signals were observed. Clinical trial information: NCT02569242