Figure 6. Organ-specific metastatic risk scores.
A) A multivariate Cox regression was built using the presence of alteration in the primary tumor and primary tumor location as covariates for risk of metastasis (met) to any site. Death is considered a competing event. Dots mark the hazard ratio (HR) for each pathway alteration, and lines the 95%CI (confidence interval). HR>1 indicates a higher risk of metastasis for the pathway alteration. Q-values>0.1 are shown with grey dots; blue and pink are used for significant HR<1 and HR>1, respectively. DDR denotes DNA damage response. B) Frequency of indicated alterations in primary tumor by number of metastatic sites at time of metastasis diagnosis. Cases with no metastasis had at least 5 years (yrs) of follow-up (FU) to confirm metastases did not develop. C) Primary tumor genetic alterations and risk of metastasis to specific organs. Shown are the coefficients of the multivariate Cox model that uses time to metastasis to that organ as dependent variable and presence of alterations in the tumor primary—along with primary tumor location—as covariates. Death is considered a competing event. Dots represent the HR for each pathway alteration and lines the 95%CI. HR>1 indicates a higher risk of metastasis for that pathway alteration. Grey dots represent a q-value>0.1, while blue and pink are used for significant HR<1 and HR>1, respectively. D) Examples for the combined effect of wild-type (WT) or altered (Alt) pathways and primary tumor locations on time to metastases. Shown are, from left to right, time to liver metastasis depending on WNT pathway alterations and primary tumor location, time to peritoneal metastasis depending on WNT pathway alterations and primary tumor location, and time to bone metastasis depending on WNT pathway alterations and primary tumor location. E) Brain metastasis free survival in patients with metastatic CRC by the presence of lung and bone metastases and tumor TGFβ pathway alterations. D, E) Hazard ratios (HRs), 95% confidence intervals, and p-values were estimated using univariate Cox proportional hazards regression, with the indicated reference group for each comparison.
See also Figures S6, S7 and Tables S3–S6.