Rapid Fire Best of the Best Poster Oral Clinical Oncology Society of Australia Annual Scientific Meeting 2020

Developing a Predictive Model utilising Real-world Pancreatic Cancer data and CT-based Radiomics for Implementation into Clinical Practice (#263)

Sam Banks 1 2 3 , Alexander Grogan 1 2 3 , Mark Tacey 4 , Benjamin Thomson 5 6 7 , Brett Knowles 5 6 7 , Benjamin Loveday 5 6 7 8 , Michael Michael 9 , Michael Jefford 2 9 , Lara Lipton 9 10 , Hui Li Wong 1 2 9 , Sumitra Ananda 9 10 , Peter Gibbs 1 2 10 , Hyun Soo Ko 3 , Belinda Lee 1 2 9 11
  1. Systems Biology and Personalised Medicine, The Walter & Eliza Hall Institute of Medical Research, Melbourne, VIC, Australia
  2. Faculty of Medicine & Health Sciences, The University of Melbourne, Melbourne, VIC, Australia
  3. Department of Cancer Imaging, Peter MacCallum Cancer Centre, Melbourne , VIC, Australia
  4. Department of Biostatistics, Northern Health, Epping, VIC, Australia
  5. Department of Surgery, Royal Melbourne Hospital, Parkville, VIC, Australia
  6. Department of General Surgical Specialties, The Royal Melbourne Hospital, Parkville, VIC, Australia
  7. Department of Surgery, Peter MacCallum Cancer Centre, Melbourne, VIC, Australia
  8. Department of Surgery, The University of Auckland, Auckland, New Zealand
  9. Department of Medical Oncology, Peter MacCallum Cancer Centre, Melbourne, VIC, Australia
  10. Department of Medical Oncology, Western Health, St Albans, VIC, Australia
  11. Department of Medical Oncology, Northern Health, Epping, VIC, Australia

Background: Radiomics technology enables clinical images to be converted into minable voxel data. Textural and spatial analyses of the relationships between voxels can thus be correlated with tumour heterogeneity and clinical outcomes. The aim of this study was to determine the association between pancreatic cancer (PDAC) radiomic data and survival outcomes.

Methods:  Consecutive pancreatic cancer patients treated in two metropolitan centres from 2016-2020 were identified from the prospective PURPLE Pancreatic Cancer registry. Using the index CT scans, forty textural radiomic features were computed (LIFEx software) on the maximal tumour diameter of the primary lesion as the region of interest. Associations between radiomic data and 1-year overall survival were determined using manual stepwise logistic regression analysis. Statistical analysis was performed using IBM SPSS. 

Results: 135 patients were included with median age 68 (IQR 30-88) years. Disease location was: head n=81 (60%), body n=28 (21%), and tail n=26 (19%); 20% (27/135) had resectable disease at diagnosis, 12% (16/135) were borderline resectable and 68% (92/135) were locally advanced. Nine pre-specified radiomic features were significantly associated with 1-year overall survival. A lower textural score for “non-uniformity of grey-levels of the homogenous zones” (GLZLM_GLNU) was associated with an increased likelihood of survival at 1 year (OR 4.15, 95% CI: 1.71-10.1, p=0.002). Adding this radiomic biomarker to a predictive model combining  ECOG, age, neutrophil-lymphocyte ratio and Ca19-9 provided an increase in the predictive performance for survival at 1-year as measured by the C-statistic (increase from 0.708 to 0.763).

Conclusion: This study demonstrates that radiomics can be applied to real-world data to develop prediction models. This non-invasive clinical decision support tool could complement patient treatment and personalise care. Work is under way on a validation cohort, and integration into the PURPLE Registry platform to enable future implementation of radiomic prediction models into routine practice.