Paolo Manca

We aimed to develop clinical–molecular prognostic nomogram for patients with microsatellite-stable (MSS) and BRAF wild-type metastatic colorectal cancer (mCRC) using comprehensive genomic profiling and clinical data, taking advantage from a prospectively maintained dataset at three Italian institutions and three randomized trials. The overall cohort was split into development and validation sets. The prognostic model was built on the development set using prespecified landmark intervals. Genomic and clinical predictors were identified within a stability-selection framework based on an L1-penalized Cox model and subsequently incorporated into a stratified Cox model, with metastasectomy modeled as a time-dependent covariate. The overall cohort included 1,071 patients. Stability selection identified RAS [stability selection threshold (SST): 98%] and SMAD4 (SST: 92%) as the only genomic alterations providing additional prognostic information beyond the prespecified clinical model. In the overall population, patients with RAS/SMAD4 double–wild-type tumors had the most favorable overall survival, whereas RAS mutations, SMAD4 alterations, and RAS/SMAD4 co-alterations were associated with inferior outcomes. The final nomogram integrated RAS/SMAD4 with clinical variables and metastasectomy. In the development cohort, time-dependent AUCs were 0.723 and 0.736 at 1 and 4 years, respectively. In the validation cohort, AUCs were 0.657 and 0.711 at the same time points. These findings suggest that prognostically relevant genomic information in MSS, BRAF wild-type mCRC may be captured by targeted next-generation sequencing panels including RAS and SMAD4. By integrating these alterations with clinical variables and time-updated information on metastasectomy, this clinical–molecular nomogram provides individualized dynamic survival estimates. Significance:SMAD4 alterations, alongside RAS mutations, identify mCRCs with aggressive biology and poor outcomes. By combining readily available clinical and genomic data, the nomogram may support survival estimation and patient counseling. It also provides a reference framework for developing future models as first-line mCRC landscape continues to evolve.

Oncoprint of recurrent genomic alterations from the NGS panel with clinicopathologic annotations. Each column represents one patient and each row a recurrently altered gene; cells are binary-coded as mutated (Mut) vs. wild type (WT). Genes are ordered by decreasing cohort prevalence (percentages shown on right), with APC (85%), TP53 (80%), RAS (51%), PIK3CA (19%), SMAD4 (14%), and FBXW7 (12%) among the most frequent; less common events include ERBB2 amplification or mutation, PTEN mutation or loss, EGFR amplification, and FGFR1 fusion. Clinical annotation tracks indicate sex, primary site (left vs. right), resection of the primary tumor, number of metastatic sites (≥2 vs. 1), and onset of metastases (synchronous vs. metachronous). Columns are arranged to highlight shared and the x-axis shows the patient count scale. Alt, altered.

Kaplan–Meier analysis of OS probability according to RAS and SMAD4 status in patients with pMMR/MSS, BRAF wild-type mCRC. A, Kaplan–Meier OS curves for RAS mutational status (wild type vs. mutated) in the overall cohort (N = 1,071). B, Kaplan–Meier OS curves for SMAD4 alteration status (wild type vs. altered) in the overall cohort (N = 1,071). C, Kaplan–Meier OS curves for combined stratification of RAS and SMAD4 status in the overall cohort (N = 1,071). D, Kaplan–Meier OS curves for combined stratification of RAS and SMAD4 status in the MSK-IMPACT-CRC cohort (N = 2,649). Alt, altered; Mut, mutant; Ref., reference; WT, wild type.

Clinical–molecular nomogram for conditional OS in mCRC. The figure shows the Cox model–derived dynamic nomogram predicting OS probabilities. The nomogram estimates 1-, 2-, 3-, and 4-year OS probability conditional on being alive at a prediction landmark time tLp (months since metastatic diagnosis). Landmark times are defined every 3 months from 0 to 30 months. For each tLp, points are assigned for age, primary tumor site (left vs. right), resection of the primary tumor (yes vs. no), ECOG PS (0 vs. ≥1), number of metastatic sites (1 vs. ≥2), onset of metastases (metachronous vs. synchronous), surgery of metastases (performed vs. not performed), time since surgery (in months; specified only if surgery has been performed), and RAS and SMAD4 status. The total points map to conditional OS probabilities on the lower scales at 1, 2, 3, and 4 years. The effect of surgery of metastases is shown on two lines, “Surgery of metastases: No” and “Surgery of metastases: Yes”. On the “No” line, the horizontal axis represents time since metastatic diagnosis (0–30 months); on the “Yes” line, it represents time since surgery (0–30 months after metastasectomy). How to use. First, select a prediction landmark tLp (months since metastatic diagnosis). (i) No surgery by the landmark. For a patient who has not undergone metastasectomy by tLp (no surgery at all or surgery only after tLp, i.e., at month s > tLp), use “Surgery of metastases: No” line at tLp, read the points for all covariates, sum them on the Total Points axis, and then project vertically to obtain the conditional 1–4-year OS from tLp. (ii) Surgery before the landmark. For a patient who underwent metastasectomy at month s ≤ tLp, the predictions depend on the time since surgery, i.e., Δ = tLp − s. Use “Surgery of metastases: Yes” line at Δ, read the points for all covariates, sum them to Total Points axis, and project to the survival scales to obtain the conditional 1- to 4-year OS from tLp. This construction provides dynamically updated, landmark-conditional predictions that reflect both whether surgery has occurred and the time elapsed since surgery.

Supplementary Figure S1. Study flow diagram.

Supplementary Figure S2. Overall survival Kaplan-Meier curves, stratified for development and validation cohorts. P-value is from a two-sided log-rank test.

Supplementary Figure S3. Univariable Cox model association between overall survival and individual gene alterations from Comprehensive Genomic Profiling after filtering.

Supplementary Figure S4. Diagram detailing patient inclusion for BRAF wild-type MSK-IMPACT-CRC cohort (N=2649).

Supplementary Figure S6. Time-dependent discriminative ability measure (Area Under the Curve [AUC]) at 1-, 2-, 3-, and 4-year horizons across 11 prespecified landmark times (0-30 months, 3-month steps) in the development cohort (panels A-D) and the validation cohort (panels E-H). Each point represents the landmark-specific AUC with the corresponding 95% CI.