Detailed Abstract
[Oral Presentation 7 - Liver (Liver Disease/Surgery)]
[OP 7-6] Upfront surgery or neoadjuvant chemotherapy for colorectal liver metastases? A machine-learning decision-tree to identify the best potential policy
Simone FAMULARO*1 , Flavio MILANA1 , Matteo CIMINO1 , Eloisa FRANCHI1 , Guido COSTA1 , Fabio PROCOPIO1 , Luca VIGAN? , Matteo DONADON1 , Guido TORZILLI1
1 Department Of Hepatobiliary And General Surgery, IRCCS Humanitas Research Hospital - Humanitas University, ITALY
Background : How to select patients for neoadjuvant chemotherapy rather than upfront surgery is still controversial in case of colorectal-liver-metastases(CLM). A model to assess the best potential candidates to upfront or neoadjuvant strategies was developed by machine-learning.
Methods : Consecutive CLM patients at first hepatectomy were enrolled and grouped whether they had upfront surgery (UFS) or neoadjuvant therapy then surgery (NEOS). Baseline differences were weighted by Inverse-Probability-Weighting(IPW); survival analyses and risk predictions were estimated. A mortality risk model was built by Random-Forest(RF) to assess the best-potential-treatment(BPT) for each patient. A classification-and-regression-tree(CART) was developed to automatically identify the characteristics of BPT-upfront and BPT-neoadjuvant candidates.
Results : Between 2005 and 2019, 448 patients were enrolled: 95 UFS and 353 NEOS. Baseline differences were weighted by IPW, obtaining two balanced pseudopopulation: UFS=432 and NEOS=440. Mortality risk was not impacted by neoadjuvant therapy (HR 1.44, 95%CI: 0.95-2.17, p=0.07). A mortality prediction model was fit by RF, showing an AUROC of 75.2%. NEOS was the BPT for 245 patients, while UFS for 203. At CART, planning a R1vasc surgery was the main factor determining who were best candidates for NEOS and UFS, followed by primitive tumor localization, number of metastases, age, sex, RAS status and pre-operative CEA. In case of R1vasc, patients with CEA=>10, or those with lower CEA but being female or being male with a tumor located in the right colon, were the best candidate to UFS.
Conclusions : Upfront resection should be carefully reconsidered resulting the BPT in several complex conditions compared to NEOS.
Methods : Consecutive CLM patients at first hepatectomy were enrolled and grouped whether they had upfront surgery (UFS) or neoadjuvant therapy then surgery (NEOS). Baseline differences were weighted by Inverse-Probability-Weighting(IPW); survival analyses and risk predictions were estimated. A mortality risk model was built by Random-Forest(RF) to assess the best-potential-treatment(BPT) for each patient. A classification-and-regression-tree(CART) was developed to automatically identify the characteristics of BPT-upfront and BPT-neoadjuvant candidates.
Results : Between 2005 and 2019, 448 patients were enrolled: 95 UFS and 353 NEOS. Baseline differences were weighted by IPW, obtaining two balanced pseudopopulation: UFS=432 and NEOS=440. Mortality risk was not impacted by neoadjuvant therapy (HR 1.44, 95%CI: 0.95-2.17, p=0.07). A mortality prediction model was fit by RF, showing an AUROC of 75.2%. NEOS was the BPT for 245 patients, while UFS for 203. At CART, planning a R1vasc surgery was the main factor determining who were best candidates for NEOS and UFS, followed by primitive tumor localization, number of metastases, age, sex, RAS status and pre-operative CEA. In case of R1vasc, patients with CEA=>10, or those with lower CEA but being female or being male with a tumor located in the right colon, were the best candidate to UFS.
Conclusions : Upfront resection should be carefully reconsidered resulting the BPT in several complex conditions compared to NEOS.
SESSION
Oral Presentation 7
Room B 3/5/2022 10:10 AM - 11:00 AM