Background. Despite the improvement of therapeutic regimens, a relevant proportion of multiple myeloma (MM) patients (pts) experience early relapse (ER) [Majithia, Leukemia 2016] and represents an unmet medical need. It is therefore of high clinical interest to identify baseline factors that may predict ER. Aims Methods Data were obtained from 2326 pts enrolled in 8 multi-center clinical trials: NCT01093196, NCT01346787, NCT01857115, NCT01190787, NCT00551928, NCT01091831, NCT01063179 and 2005-004730-41. Here, we included 14 baseline features (fts): age, creatinine, albumin (alb), b2microglobulin (b2m), bone marrow plasma cell (PCbm) were evaluated as continuous variables; free light chain (FLC, λ vs K), M-component subtype (IgA vs others), Revised International Staging System (R-ISS stage II/III vs I), lactate dehydrogenase levels >/≤ upper limit of normal (LDHULN), presence vs absence of chromosomal abnormalities detected by FISH [del17p, t(4;14), t(14;16), t(11;14)], and presence of plasmacytomas as categorical values. Trials were assigned to training and validation set to have a superimposed median (μe) age and follow-up in the two subsets. From the training set, a univariate analysis (UV) on outcome was performed according to both Chi-square and Kruskal tests, as appropriate. Features with pResults ER≤18 models. 10/14 fts were selected based on UV analysis: age, FLC, PCbm, del17p, t(4/14), t(14/16), alb and b2m, LDHULN and R-ISS stage. Pts with complete data were included in the training set (n=923; μe age =66 years [y]; ER=35%) and the validation set (n=313; μe age=67 y; ER=36%), respectively. In the MV incorporating the R-ISS, the R-ISS II/III vs I (OR=1.75, 95% CI:1.26-2.44) and increased PCbm (OR=1.05, 95% CI:1.02-1.08) increased the risk of ER. When the MV analysis was performed including single fts instead of R-ISS, increased PCbm (OR=1.05, 95% CI:1.02-1.08), λ FLC (OR=1.34, 95% CI:1.01-1.79), LDHULN (OR=1.80, 95% CI:1.16-2.81), presence of both del17q (OR=1.58, 95% CI:1.07-2.33) and t(4/14) (OR=2.01, 95% CI:1.36-3.01) increased the probability of ER; increased alb (OR=0.75, 95% CI:0.61-0.94), reduced the risk of ER (table 1). ER≤24 models. 8/14 fts were selected based on UV analysis: FLC, PCbm, del17p, t(4/14), alb and b2m levels, LDHULN and R-ISS stage. Pts with complete data were included in the training set (n=1009; μe age=67 y; ER=45%) and in the validation set (n=352; μe age=67 y, ER=45%). In MV analysis, including R-ISS, both R-ISS (OR=1.88, 95% CI:1.39-2.55) and PCbm (OR=1.04, 95% CI:1.02-1.06) impacted on outcome. When the MV analysis was performed including single fts, λ FLC (OR=1.31, 95% CI:1.01-1.69), PCbm (OR=1.04, 95% CI:1.02-1.07), del17p (OR=1.86, 95% CI:1.30-2.65) and t(4/14) (OR=2.00, 95% CI:1.38-2.88) retained their impact on outcome (table 1). Validation. Each MV model was tested on the validation set. Among the 4 MV models, the ER≤18 incorporating individual fts resulted in the highest AUC (0.66) and was therefore used to build up a prognostic score. Score. The ER score was calculated as 0.047 × PCbm + 0.589 × LDHULN + 0.459 × del17q (IF present) + 0.705 × t(4/14) (IF present) + 0.293 × FLC (IF FLC =λ) - 0.284 × alb. The ER score was calculated as 3 groups of pts with different risks of ER: L (42% of pts; risk of ER18=23%), M (33% of pts; risk of ER18=39%) and H (26% of pts; risk of ER18 =55%). Discussion. This is the first analysis proposing a score that includes standard baseline fts and aims at identifying pts at high risk of ER in the context of novel agent-based therapy. Based on our score, 26% of pts can be defined at high risk. To improve the clinical applicability, the construction of a simplified model with categorized variables is ongoing.