Rationale and Objectives. Phosphoethanolamine and phosphocholine, shown to be elevated in tumors and possibly related to apoptotic signaling, have the potential to be prognostic variables of cancer treatment. Materials and Methods. The sum of phosphoethanolamine and phosphocholine normalized by nucleotide-triphosphates was determined in tumors of non-Hodgkin's lymphoma (NHL) patients via in vivo P-31 MR spectroscopy. Results. The normalized sum of phosphoethanolamine and phosphocholine showed significant differences in tumors of patients who had a complete response to treatment against those who did not (t-test: 1.45 +/- 0.15, mean +/- standard error, n = 10 vs. 2.28 +/- 0.15, n = 17, P < .001; Fisher test: P < .04; sensitivity and specificity approximate to70%). This parameter also showed significant differences among treatment responses in the previously untreated and aggressive subgroups and in the low and low-intermediate-risk subgroups determined by the international prognostic index (IPI). Further, distinctly different treatment response cutoffs for the parameter were found in different risk groups. When these risk-dependent cutoffs were used, the Fisher test of the whole group improved (P < .0002, sensitivity 80%, specificity 94%). The normalized sum of phosphoethanolamine and phosphocholine and the IPI were better predictor covariates for time to treatment failure when fitted interactively in a Cox regression (P < .0003) than when fitted independently. When time to treatment failure was used as a surrogate of survival in Kaplan-Meier analysis, the interaction of both covariates segregated the cases significantly (P < .008). There was no significance with each covariate independently. Conclusion. The normalized sum of phosphoethanolamine and phosphocholine measured before treatment successfully predicts long-term response to treatment and time to treatment failure in non-Hodgkin's lymphoma, particularly when combined with the IPI. (C) AUR, 2004.