El-Galaly, Tarec Christoffer, Villa, Diego, Michaelsen, Thomas Yssing, Hutchings, Martin, Mikhaeel, Nabegh George, Savage, Kerry J., Sehn, Laurie H., Barrington, Sally, Hansen, Jakob W., Smith, Daniel, Rady, Kirsty, Mylam, Karen J., Larsen, Thomas S., Holmberg, Staffan, Juul, Maja B., Cordua, Sabrina, Clausen, Michael R., Jensen, Kristina B., Johnsen, Hans E., and Seymour, John F.
Purpose Development of secondary central nervous system involvement (SCNS) in patients with diffuse large B-cell lymphoma is associated with poor outcomes. The CNS International Prognostic Index (CNS-IPI) has been proposed for identifying patients at greatest risk, but the optimal model is unknown. Methods We retrospectively analysed patients with diffuse large B-cell lymphoma diagnosed between 2001 and 2013, staged with PET/CT and treated with R-CHOP(-like) regimens. Baseline clinicopathologic characteristics, treatments, and outcome data were collected from clinical databases and medical files. We evaluated the association between candidate prognostic factors and modelled different risk models for predicting SCNS. Results Of 1532 patients, 62 (4%) subsequently developed SCNS. By multivariate analysis, disease stage III/IV, elevated serum LDH, kidney/adrenal and uterine/testicular involvement were independently associated with SCNS. There was a strong correlation between absolute number of extranodal sites and risk of SCNS; the 144 patients (9%) with >2 extranodal sites had a 3-year cumulative incidence of SCNS of 15.2% (95% confidence interval [CI] 9.2–21.2%) compared with 2.6% (95% CI 1.7–3.5) among those with ≤2 sites ( P < 0.001). The 3-year cumulative risks of SCNS for CNS-IPI defined risk groups were 11.2%, 3.1% and 0.4% for high-, intermediate- and low-risk patients, respectively. All risk models analysed had high negative predictive values, but only modest positive predictive values. Conclusions Patients with >2 extranodal sites or high-risk disease according to the CNS-IPI should be considered for baseline CNS staging. Clinical risk prediction models suffer from limited positive predictive ability, highlighting the need for more sensitive biomarkers to identify patients at highest risk of this devastating complication. [ABSTRACT FROM AUTHOR]