Back to search

University of British Columbia

Causal machine learning to optimize treatment decisions for patients with endometrial cancer

Abstract

dc:description

Background: In endometrial cancer (EC), histopathological parameters, often unreliably assessed from diagnostic sampling, are used to guide surgical decisions. Surgical decisions, like lymph node (LN) assessment, have implications for post-operative adjuvant therapy decisions, which often vary based on nodal status. The Proactive Molecular Risk Classifier for Endometrial Cancer (ProMisE) subtypes are prognostically distinct, can be assigned on biopsy and have demonstrated exceptional reproducibility. Thus, they may be better suited to guide decision-making pre- and post-hysterectomy. As per the current literature, it is not yet well understood how ProMisE can impact decision-making. Objectives: This thesis explores ProMisE in the context of other pre-operative prognostic features that may influence nodal assessment and adjuvant therapy decision-making, measures the concordance of histopathological and molecular features available pre- and post-operatively, explores how ProMisE subtypes could tailor nodal assessment decisions to minimize the variability in adjuvant therapy decisions, and finally, proposes a model for a dynamic treatment regime (DTR) that simultaneously optimizes pre- and post-operative decision-making in EC. Methods: EC patients who underwent surgery, and have a ProMisE subtype assigned, were identified from two institutions. The dataset was generated using retrospective data from previous studies with additional pre-operative parameters collected via chart reviews. Univariate associations among ProMisE, treatment decisions and prognostic factors were measured. Multivariate analyses were used to assess association of survival with LN testing within ProMisE subgroups. A DTR optimizing nodal assessment and adjuvant therapy was estimated using all the parameters available at the time of each decision. To account for the confounders and mediators in our observational data, a counterfactual framework and dynamic weighted survival modelling was used. Results: 903 patients were included. Pre-operative histopathology, age, body mass index and co-morbidities are associated with ProMisE and nodal assessment. ProMisE assigned pre- and post-operatively demonstrates better concordance than histopathology assigned pre- and post-operatively or at surgery and after expert review. Nodal assessment is significantly associated with longer survival in p53abn ECs, but not in POLEmut or MMRd/NSMP. The methods and data used in the identification of the optimal DTR are insufficient due to the high rate of censoring for an accelerated failure time model.

Degree

thesis:*
Name thesis:degree_name
Master of Science - MSc
Level thesis:degree_level
master's
Discipline thesis:degree_discipline
Reproductive and Developmental Sciences
Grantor dc:publisher
University of British Columbia
Year dc:date
2022

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Johal, Parveen
Contributors dc:contributor
  • Talhouk, Aline

Rights

dc:rights
Statement dc:rights
  • Attribution-NonCommercial-NoDerivatives 4.0 International
Language dc:language
eng

Identifiers

dc:identifier.*
Handle dc:identifier
http://hdl.handle.net/2429/81846

Chain of custody

source
Harvested from
University of British Columbia
Base URL
circle.library.ubc.ca/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Johal, Parveen. Causal machine learning to optimize treatment decisions for patients with endometrial cancer. master's thesis, University of British Columbia, 2022. http://hdl.handle.net/2429/81846