The purpose of this study was to examine the validity of a predictive model for response to treatment and survival in advanced diffuse histiocytic lymphoma. One hundred twenty-seven consecutive patients with Ann Arbor stage II-IV diffuse histiocytic lymphoma, who completed treatment between 1974 and 1984 in one of four different Memorial Hospital combination chemotherapy protocols, were reviewed. The median follow-up time was 66.9 months for survivors (range, 21-153.1 months). Factors studied included: age; sex; Ann Arbor stage; prior therapy; B symptoms; serum lactic dehydrogenase (LDH); sites of initial disease; and tumor bulk. LDH was grouped accordingly (units/liter): low, less than 225; medium, 225-500; high, greater than 500. Each patient was assigned an overall level of site involvement (LSI) from the following mutually exclusive groups: group I, peripheral lymph node (PLN) (including +/- Waldeyer ring involvement, +/- spleen); group II, extranodal disease (EN) +/- PLN; group III, retroperitoneal lymph node (RLN) +/- PLN; group IV, bulky mediastinal disease (MED) +/- any other disease; group V, EN with RLN +/- PLN. The Ann Arbor staging system failed to dissect patient groups differing significantly in their prognosis. Serum LDH, LSI, and age were the only factors important for predicting response and survival after multivariate logistic regression and a parametric Weibull survival analysis. Using three levels of serum LDH and correlating them with the different LSI, four tentative "stages" differing significantly in their survival at 48 months were defined: stage I, low LDH, any LSI (80% alive); stage II, medium LDH, PLN, and/or EN (50% alive); stage III, high LDH, PLN, and/or EN or medium LDH, RLN +/- PLN +/- EN, and/or MED (35% alive); stage IV, high LDH, RLN +/- PLN +/- EN, and/or MED (15% alive). Identification of prognostic stages on the basis of LDH level and LSI will allow more accurate comparison of clinical trials for patients with advanced diffuse histiocytic lymphoma.