Extended Data Fig. 6: Supplementary Cox proportional hazards (CPH) models from Digital Pathology analysis.
From: Tumor evolution metrics predict recurrence beyond 10 years in locally advanced prostate cancer
(A) Time to metastasis using clinical co-variates and Gleason Morisita index (Imaging Cohort, n = 250 participants). Increased Gleason Morisita index is significant associated with shorter time to recurrence (p < 0.05), in line with what is seen in time to recurrence (Fig. 4E). Hazard ratio for Gleason Morisita index represents the increase in hazard between the 5th and 95th percentile values (within the Imaging Cohort). (B) Time to recurrence using clinical co-variates, presence of Invasive Ductal Pattern, and Gleason Morisita index (Imaging Cohort, n = 250 participants). Both Gleason Morisita index and Invasive Ductal Pattern are independently significant predictors of risk of patient recurrence (p < 0.05). Presence of Invasive Ductal Pattern was identified at a per-patient level by Reviewing Pathologist. Hazard ratio for Gleason Morisita index represents the increase in hazard between the 5th and 95th percentile values (within the Imaging Cohort). (C) Time to recurrence, including clinical covariates and Reviewing Pathologist’s grade grouping (Imaging Cohort, n = 250 participants). Model hazard ratios suggest a decreasing risk of recurrence as grade group decreases, with respect to reference group 5, albeit without significance. (D) Time to recurrence, including clinical covariates and grade grouping from automated classifier (Imaging Cohort, n = 250 participants). Grade groups 3 and 4 show significantly lower risk of recurrence (P = 0.0164 and P = 0.0214, respectively) compared to reference group 5. (E) Time to metastasis, including clinical covariates and Reviewing Pathologist’s grade grouping (Imaging Cohort, n = 250 participants). There is no clear trend for grade group. (F) Cox model of time to metastasis, including clinical covariates and grade grouping from automated classifier (Imaging Cohort, n = 250 participants). Model hazard ratios suggest a decreasing risk of metastasis from groups 3–5, albeit without significance. All forest plots show 95% confidence interval of hazard ratios and the covariate P values, derived from a Wald test (*P < 0.05, **P < 0.01, ***P < 0.001).
