Purpose: To assess the contribution of induction chemotherapy (IC) before definitive chemoradiation therapy (dCRT) in patients with esophageal cancer (EC) based on recursive partitioning analysis (RPA). Methods and Materials: A total of 496 eligible patients with EC staged by positron emission tomography (PET) who received dCRT from 1998 to 2015 were included, 162 (32.7%) of whom underwent IC before dCRT. RPA was used to risk-stratify patients on the basis of independent prognostic factors to predict progression-free survival (PFS). Outcomes were compared between treatment groups. Results: The median follow-up time was 49.1 months (range, 7.0-155.9 months) for survivors. Compared with the non-IC group, the IC group had a comparable 5-year PFS rate (21.0% vs 23.4%; P=.726) in the whole cohort. Multivariate analysis identified age, performance status, primary tumor length, baseline PET maximum standard uptake value (SUVmax), and maximum lymph node diameter as independent prognostic factors for PFS. RPA segregated patients into 3 prognostic groups: low-risk group (PET SUVmax <9.7 and tumor length <= 5 cm), intermediate-risk group (PET SUVmax >= 9.7 and age >= 67), and high-risk group (PET SUVmax <9.7 and tumor length >5 cm, or PET SUVmax >= 9.7 and age <67). Significant improvements in PFS (P=.006) and locoregional failureefree survival (P=.028) in the IC group in comparison with the non-IC group were observed in high-risk patients, whereas no differences in survival were found between the 2 treatment groups in low-risk or intermediate-risk patients. After propensity score matching, the high-risk group still demonstrated a significantly improved PFS with IC (P=.009). Conclusions: The RPA prognostic grouping provides a useful method of selecting high-risk EC patients who may benefit from IC before receiving dCRT. Prospective validation is warranted. (C) 2017 Elsevier Inc. All rights reserved.