OBJECTIVES: To investigate temporal changes in the risk of non-Hodgkin lymphoma (NHL) and estimate NHL incidence in risk groups and the prognostic role of these risks and current, nadir and time-weighted average CD4 cell count. METHODS: Secular trends in time from HIV seroconversion to an NHL diagnosis was estimated using Cox models from data pooled from the 22 seroconverter cohorts in the CASCADE collaboration for three periods (pre-1997, 1997-1998, 1999-2002), adjusting for age at seroconversion, exposure category and sex. RESULTS: Of 7103 seroconverters, 129 developed NHL. Compared with pre-1997, there was little reduction in NHL risk in 1997-1998 [relative risk (RR), 0.66; 95% confidence interval (CI), 0.37-1.17] then a substantial reduction in 1999-2002 (RR, 0.25, 95% CI, 0.12-0.53). Compared with individuals with CD4 cell count > 350 x 10 cells/l, the RR of NHL increased to 1.9 (95% CI, 1.0-3.6), 1.4 (95% CI, 0.6-3.5) and 11.2 (95% CI, 6.3-20.0) at CD4 cell counts 200-349, 100-199 and < 100 x 10 cells/l, respectively. There was no evidence that nadir (P = 0.41) or time-weighted average CD4 cell count (P = 0.38) contributed further to predicting NHL risk or were better predictors than current CD4 cell count. For individuals with CD4 cell count > 350 x 10 cells/l pre-HAART, an NHL incidence of 1.8 and 0.4/1000 person-years was estimated for those at highest and lowest risk, respectively, when classified by age and exposure category. CONCLUSION: There appears to be no justification for initiating HAART at CD4 cell counts > 100 x 10 cells/l based specifically on concerns over NHL. The risk of NHL is, however, greatly increased at lower CD4 cell counts.