A Biology-Driven Radiomics Model for Unresectable Intrahepatic Cholangiocarcinoma Treated with Radiotherapy

Abstract

<h3>Purpose/Objective(s)</h3> For unresectable intrahepatic cholangiocarcinoma (ICC), we previously showed that pre-therapy CT volumetric enhancement (VE) associates with biological (e.g., degree of stromal infiltration and necrosis, genetic mutations, and infiltrating lymphocytes) and clinical outcomes. Further, we showed that a change in VE following radiotherapy (RT) is prognostic. Here, we integrated these baseline and post-treatment imaging features to appraise their joint prognostic value. <h3>Materials/Methods</h3> We retrospectively identified 112 patients (54% female, median age 63 years) who received definitive RT (78% photon, 22% proton; median biologically effective [BED₁₀] dose 86 Gy, interquartile range [IQR] 75-98 Gy) with triple phase CT imaging at diagnosis, prior to RT, and following RT. A technology company's multimodality tumor tracking software was used to quantify tumor VE ≥1 standard deviation of that of healthy parenchyma on each scan. Based on prior work, each patient was stratified into (1) low/high VE at baseline (cutpoint: 32% VE) and (2) VE-based response/non-response to RT (cutpoint: 26% decrease in VE). Overall survival (OS) and local failure-free survival (LFFS) following RT were assessed using the Kaplan-Meier method. Cox analyses were performed to include the 2 radiomic markers and traditional covariates of survival. Concordance indices (C-index) were used to evaluate model performance. <h3>Results</h3> At a median follow up of 74 months, 46% failed locally and 85% died. The 2-year OS and LFFS were 40% (95% confidence interval [CI], 30-49%) and 26% (CI, 18-34%), respectively. At baseline, median VE was 25% (IQR 14-45%); VE was categorized as low in 43% and high in 57%. Following RT, median change in VE was +3% (IQR -7% to +176%); change was categorized as non-response for 56% and response in 44%. On multivariable analysis for OS, Stage IV disease (hazard ratio [HR] 2.11, CI 1.35-3.31; <i>P</i>=0.001) was associated with poorer survival. High VE (HR 0.38, CI 0.24-0.59; <i>P</i><0.001) and VE-based response to RT (HR 0.19, CI 0.12-0.32; <i>P</i><0.001) were each independently associated with longer survival (C-index 0.77). For LFFS, Stage IV disease (HR 1.61, CI 1.03-2.52; <i>P</i>=0.036) was associated with shorter latency to local failure/death, while high VE (0.40, 95% CI 0.26-0.62; <i>P</i><0.001) and VE-based response to RT (HR 0.29, CI 0.18-0.46; <i>P</i><0.001) were associated with longer latency (C-index 0.74). Median OS stratified by cohort is shown in Table 1. <h3>Conclusion</h3> For ICC patients treated with definitive RT, VE measured on pre-therapy imaging and pre/post-RT imaging each independently had greater predictive value than conventional patient and disease factors. These metrics, with their biological and clinical associations, may help personalize therapeutic approaches for ICC. External model validation is warranted.