Poster Presentation 35th Lorne Cancer Conference 2023

Artificial Intelligence derived body composition to accurately predict chemotherapy toxicities following oxaliplatin therapy in colon cancer patients and role of AI-derived body composition (#106)

Yasser Arafat 1 , Ke Cao 2 , Cheuk S Choi 2 , Josephine SY Yeung 3 , Kamol Kin 2 , Tarana Lucky 4 , Ian Faragher 1 , Margaret Lee 5 , Steven Chan 6 , Paul N Baird 7 , Justin M Yeung 1
  1. Colo rectal surgery, Western Health Melbourne, University of Melbourne, Melbourne, Victoria, Australia
  2. Western Health Melbourne, University of Melbourne, Docklands, VIC, Australia
  3. Pharmacy, Western Precinct, The University of Melbourne, Melbourne, Australia, Melbourne, Victoria, Australia
  4. Gynaecology, Royal Women's Hospital, University of Melbourne, Melbourne, Victoria, Australia
  5. Oncology, Western Health Melbourne, University of Melbourne, Melbourne, Victoria, Australia
  6. Department of Surgery, Western Precinct, The University of Melbourne, Melbourne, Australia, Melbourne, Victoria, Australia
  7. Opthalmology, University of Melbourne, Melbourne, Victoria, Australia

Introduction:

Body composition (BC) is increasingly studied for predicting chemotherapy toxicities. Lean body mass (LBM) could be better than body surface area (BSA) for tailoring chemotherapy doses and can be an independent determinant of dose-limiting toxicity (DLT) and neuropathy in non-metastatic colon cancer patients. We aimed to investigate the associations between these patients’ body composition, gender variation, DLT and neuropathy and evaluate the correlation between LBM and BSA.

Methodology:

A retrospective study (2012-2021) was performed for non-metastatic colonic cancer patients who received oxaliplatin-based chemotherapy following surgical resection. Artificial intelligence-derived (AI) algorithms were used to calculate LBM and BC variables at lumbar 3 levels on CT scans. Associations between DLT, neuropathy and BC were studied. The absolute individual dose of oxaliplatin was normalized to LBM. A ROC curve analysis was performed to obtain the cut-off point for oxaliplatin dose/LBM to predict neuropathy.

Results:

Among the 129 study participants, 84 (65.2%) experienced DLT. Dose-limiting toxicity significantly varied by gender, with a higher prevalence of DLT in women than men (74% vs 55%, p-value = 0.025). Women also had lower muscle indices and higher adiposity than men (p-value <0.001). The correlation between LBM and BSA was weak (R² = 0.514). While BMI and BSA were equally distributed between DLT and no DLT patients, LBM and other BC variables were notably different between the two groups. After normalizing to LBM, the optimal oxaliplatin cut-off for neuropathy was estimated at 3.41 mg/kg LBM. Additional analyses on early cessation or dose reduction secondary to DLT showed that almost half (44%) of all DLT occurred within the first four cycles. The majority of these (78.4%) were attributed to patients who received a dose equal to or higher than the predicted cut-point (>=3.41 mg/kg, p-value = 0.05). Similar findings were also noted for peripheral neuropathy.

Conclusions: 

This is the first Australian study to demonstrate the use of AI-automated body composition measurement in investigating its association with DLT in colon cancer patients. Our findings further support the hypotheses that body composition (i.e., LBM) may be used for optimal drug dosing and assessing the risk/severity of chemotherapy toxicities.