A Machine Learning approach to identifying key decision variables within the Oesophageal Cancer MDT
A Machine Learning approach to identifying key decision variables within the Oesophageal Cancer MDT
Background: the drivers behind the Upper Gastrointestinal (UGI) multidisciplinary team (MDT) treatment decisions are not always explicit. National guidelines steer these decisions however human decision-making is vulnerable to subconscious biases. Machine-Learning (ML) may let us uncover these potential biases using simple, explainable AI solutions. This study is the first application of ML which examines decision-making within the oesophageal cancer (OC) MDT.
Methods: we conducted a retrospective analysis of patients discussed at the OC MDT, who underwent resections between 2010-2020. Twenty possible pre-treatment clinical variables available to the MDT were used to develop a Decision-Tree (DT) model with cross-validation, to predict MDT treatment decisions (Surgery alone (S), Neoadjuvant Chemotherapy (C+S), Neoadjuvant Chemoradiotherapy (CRT+S)).
Results: we identified 399 cases (median age 66.1 years, range: 32-83). Mean Area Under the Curve for our model for CRT+S vs C+S, S vs C+S and S vs CRT+S were 64.3%, 77.4%, 80.8% respectively. Variable Importance analysis revealed 6 variables with the largest influence, accounting for 90.2% of total importance: age (22.0%), histology (17.2%), tumour location (15.5%), cT stage (15.4%), cN stage (11.1%), and performance status (9.0%).
Conclusions: our DT model highlights key factors used by the MDT. While most are intuitive, our model suggests Age to be influential and may represent a potential unconscious bias towards chronological age rather than physiological age in determining operative fitness. The generation of such ML models from retrospective data allows us to both challenge assumptions inherent to current, and potentially inform, future MDT decisions.
e4-e4
Thavanesan, Navamayooran
94f0a216-9131-431b-809c-f5a83146343d
Vigneswaran, Ganesh
4e3865ad-1a15-4a27-b810-55348e7baceb
Underwood, Timothy
8e81bf60-edd2-4b0e-8324-3068c95ea1c6
17 January 2023
Thavanesan, Navamayooran
94f0a216-9131-431b-809c-f5a83146343d
Vigneswaran, Ganesh
4e3865ad-1a15-4a27-b810-55348e7baceb
Underwood, Timothy
8e81bf60-edd2-4b0e-8324-3068c95ea1c6
Thavanesan, Navamayooran, Vigneswaran, Ganesh and Underwood, Timothy
(2023)
A Machine Learning approach to identifying key decision variables within the Oesophageal Cancer MDT.
European Journal of Surgical Oncology, 49 (1), .
(doi:10.1016/j.ejso.2022.11.026).
Record type:
Meeting abstract
Abstract
Background: the drivers behind the Upper Gastrointestinal (UGI) multidisciplinary team (MDT) treatment decisions are not always explicit. National guidelines steer these decisions however human decision-making is vulnerable to subconscious biases. Machine-Learning (ML) may let us uncover these potential biases using simple, explainable AI solutions. This study is the first application of ML which examines decision-making within the oesophageal cancer (OC) MDT.
Methods: we conducted a retrospective analysis of patients discussed at the OC MDT, who underwent resections between 2010-2020. Twenty possible pre-treatment clinical variables available to the MDT were used to develop a Decision-Tree (DT) model with cross-validation, to predict MDT treatment decisions (Surgery alone (S), Neoadjuvant Chemotherapy (C+S), Neoadjuvant Chemoradiotherapy (CRT+S)).
Results: we identified 399 cases (median age 66.1 years, range: 32-83). Mean Area Under the Curve for our model for CRT+S vs C+S, S vs C+S and S vs CRT+S were 64.3%, 77.4%, 80.8% respectively. Variable Importance analysis revealed 6 variables with the largest influence, accounting for 90.2% of total importance: age (22.0%), histology (17.2%), tumour location (15.5%), cT stage (15.4%), cN stage (11.1%), and performance status (9.0%).
Conclusions: our DT model highlights key factors used by the MDT. While most are intuitive, our model suggests Age to be influential and may represent a potential unconscious bias towards chronological age rather than physiological age in determining operative fitness. The generation of such ML models from retrospective data allows us to both challenge assumptions inherent to current, and potentially inform, future MDT decisions.
This record has no associated files available for download.
More information
e-pub ahead of print date: 17 January 2023
Published date: 17 January 2023
Identifiers
Local EPrints ID: 486890
URI: http://eprints.soton.ac.uk/id/eprint/486890
ISSN: 0748-7983
PURE UUID: 82b3aa6e-bd97-4d02-8350-fc09e4961c83
Catalogue record
Date deposited: 08 Feb 2024 17:35
Last modified: 08 Aug 2024 02:09
Export record
Altmetrics
Contributors
Author:
Navamayooran Thavanesan
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics