Background: Pts with mCRC frequently receive ≥1 sequential treatment TL. Approximately 50%-60% of pts receive second-line (L2) and 20%-30% third-line (L3) regimens in routine practice. We investigated the pts clinical/tumor characteristics and their prognostic impact across TL. Methods: Data from 37,560 pts enrolled in 48 randomized trials (34 in first (L1), 9 in L2, and 5 in L3) were analyzed. Candidate variables (VAR) measured at enrollment were sex, age, body mass index, performance status (PS), bilirubin, hemoglobin (Hb), platelets (Pl), derived white blood cells-to-absolute neutrophil counts ratio (WBC/ANC), lactate dehydrogenase (LDH), alkaline phosphatase (ALP), primary tumor location, and number and location of metastatic sites (MS). Missing data were imputed. VAR with significant value at all TL were selected to construct a prognostic score of overall survival (OS) in training set (TS, n=30,050; 80%). For each TL, the score was calculated as the sum on the estimations of the VAR’ coefficients from the common multivariate model; Cox’s model was used to define risk groups. The discrimination capability was assessed using the Harrell’s C-index. External validation was done in the validation set (VS, n=7,510; 20%). Results: A total of 26,974 pts in L1, 7,693 pts in L2 and 2,893 pts in L3 were analyzed. The following characteristics increased continuously over TL: ≥2 MS (57%, 72%, 82%), lung metastases (50%, 74%, 91%), lymph nodes metastases (51%, 61%, 80%), KRAS mutation (37%, 47%, 51%) and elevated ALP (46%, 52%, 61%). BRAF mutation decreased (9%, 7%, 5%). In L1 vs L3 trials, 70% vs 89% of patients had primary tumor resection, 10% vs 80% had at least one metastasectomy and 31% vs 78% had a late metachronous (>12 months) metastasis. 7 independent VAR were retained in the prognostic score (PS, Hb, Pl, WBC/ANC, LDH, ALP, and the number of MS); four pt groups with significantly different prognoses were defined (table). This score remained valid when excluding pts with PS 2. Third-line oral drugs (vs placebo) and subsequent line (L2/L1 or L3/L2) were effective in all prognostic groups. Conclusions: Clinical/tumor pt characteristics significantly varied over subsequent TL in patients included in randomized trials. The same prognostic model using practical clinical and biological variables can be used in all TL.

*median OS [95% CI], months.