Second-line chemotherapy followed by high-dose therapy (HDT) with autologous stem cell transplantation (ASCT) cures less than half of the patients with relapsed or refractory diffuse large B-cell lymphoma (DLBCL). Prognostic models capable of predicting outcome are essential. In 3 sequential clinical trials, conducted from January 1993 to August 2000, we treated 150 patients with relapsed or primary refractory DLBCL with ifosfamide, carboplatin, and etoposide (ICE) chemotherapy followed by HDT/ASCT for patients with chemosensitive disease. We evaluated the age-adjusted International Prognostic Index at the initiation of second-line therapy (sAAIPI) as a predictor of progression-free survival (PFS) and overall survival (OS). At a median follow-up of 4 years, the PFS and OS are 28% and 34% by intention to treat and 39% and 45% for only those patients with chemosensitive disease. Three risk groups with different PFS and OS were identified by the sAAIPI: low risk (0 factors), 70% and 74%; intermediate risk (1 factor), 39% and 49%; and high risk (2 or 3 factors), 16% and 18% (P <.001 for both PFS and OS). The sAAIPI also predicts the PFS and OS for patients with ICEchemosensitive disease: low risk, 69% and 83%; intermediate risk, 46% and 55%; and high risk, 25% and 26% (P <.001 PFS and OS). The sAAIPI predicts outcome for patients with relapsed or primary refractory DLBCL in both intent-to-treat and chemosensitive populations. This powerful prognostic instrument should be used to evaluate new treatment approaches and to compare results of different regimens.