Joeky Senders

94 Chapter 5 TABLE 1. Continued Characteristic Definition Training set (n = 16,656) Hold-out test set (n = 4,165) p n % n % Tumor size Median [IQR] 45 [35-55] 45 [35-55] 45 [35-55] 45 [35-55] 0.986 Surgery type Biopsy 3882 23.3 977 23.5 0.894 Sub-total resection 5888 35.4 1456 35.0 Gross-total resection 6886 41.3 1732 41.6 Radiotherapy No 2667 16.0 662 15.9 0.871 Yes 13989 84.0 3503 84.1 Chemotherapy No 3647 21.9 883 21.2 0.341 Yes 13009 78.1 3282 78.8 Abbreviations: IQR=interquartile range; mm=millimeters; n=number; SD=standard deviation Inferential analysis The Schoenfeld residuals test demonstrated that the assumption of proportionality was violated for all variables except sex and ethnicity in the CPHR model (all p<.006 and global test p<.001; Supplementary Table S1). The quantile-quantile plot demonstrated a valid log-logistic distribution assumption for the (AFT) model (Supplementary Figure S1). For these reasons, we present the inferential results of the AFT model. The AFT allows for uncomplicated interpretation, as it provides acceleration factors (γ), which represent the relative survival duration of a strata compared to the reference group. For example, a γ of 1.5 reflects an expected survival duration that is 50% longer compared to the reference group. Multivariable AFT analysis identified older age (γ=0.75 per 10 years increase, p<.001), male sex (γ=0.93, p<.001), uninsured insurance status or insurance by Medicaid (γ=0.87, p<.001), midline tumors (γ=0.79, p=.004), tumors primarily located in the parietal lobe (γ=0.91, p<.001), brain stem (γ=0.44, p<.001), or multiple lobes (γ=0.88, p<.001), tumors extending to the ventricles (γ=0.90, p<.001) or across the midline (γ=0.73, p<.001), and larger sized tumors (γ=0.99 per cm, p<.001) as independent predictors of shorter survival (Figure 1). Asian race (γ=1.14, p=.001), Hispanic ethnicity (γ=1.08, p=.007), married marital status (γ=1.15, p<.001), gross- total resection (γ=1.19, p<.001), radiotherapy (γ=1.27, p<.001), and chemotherapy (γ=1.49, p<.001) were identified as independent predictors of longer survival. The AFT model with interaction terms demonstrated that age interacted with extent of resection (γ>1.03 per 10 years increase, p<.02), as well as radiotherapy (γ=1.04 per 10 years increase, p=.03) (Supplementary Table S2).