Adaptive IGRT is generally performed to account for mid-treatment changes in tumor volume or patient weight. We investigated the pattern of radiomic features changes and their prognostic value in comparison to well-established clinical variables. Using an IRB-approved single institution database of head and neck cancer patients, we retrospectively identified adult OPC patients who underwent definitive adaptive IGRT between 2012 and 2018. We collected patient, disease, and treatment characteristics, as well as outcomes measures of overall survival (OS) and event-free survival (EFS). All patients underwent contrast-enhanced computed tomography scans for initial and mid-treatment simulation. Primary gross tumor volumes were manually contoured. A total of 145 radiomics features were selected from the following categories: shape, intensity, and second-order features (neighborhood intensity difference ‘NID’, grey-level co-occurrence matrix ‘GLCM’, and grey-level run length matrix ‘GLRLM’) and analyzed using digital gene expression panel software. Second-order features were calculated both in 3D and 2.5D fashions. Pre- and mid-RT values of radiomics features were compared using Wilcoxon signed-rank test with Bonferroni’s correction. Correlation of clinical, pre-RT and delta-radiomic ([mid-pre]/pre) features to OS and EFS was determined using Harrell's concordance index (C-Index) with a preset cut-off >0.7. Statistical analysis was performed using R software. 42 patients were included, with primary tumors located in the base of tongue (n = 23) and tonsil (n = 20). 37 (86%) were HPV positive. Of the mid-RT radiomic features that changed significantly (p<0.05); NID 3D coarseness and contrast, NID 2.5D busyness and coarseness, GLRLM 2.5D, and shape spherical disproportion and surface area density features increased compared to their pre-RT values. In contrast, lower values were noted for other shape features (volume, sphericity, convexity, and compactness, among others), intensity kurtosis, and most GLCM 3D features. For both EFS and OS, radiomics features were more highly correlated than clinical features. For EFS, delta-intensity entropy and mean had the highest correlation (C-index: 0.77 and 0.74, respectively), with no clinical variables demonstrating C-index >0.7. For OS, pre-RT NID 3D complexity outperformed the highest correlated clinical variable of performance status (C-index: 0.78 vs 0.76, respectively). Our analysis consistently showed that pre-RT or delta-radiomic features better correlated to OS and EFS when compared to clinical variables. Temporal tumor radiomics trajectories can provide radiation oncologists with a non-invasive tool to quantify dynamic tumor changes.