Roughly one-fourth of patients with esophageal and gastroesophageal junction adenocarcinoma will undergo a pathological complete response.
"Trimodality therapy refers to neoadjuvant chemoradiotherapy, followed by radical surgery," noted Kohei Yamashita, PhD, at The University of Texas MD Anderson Cancer Center. At the AACR Annual Meeting 2023, researchers presented data from a study that sought to develop a model for predicting which patients would experience a pathological complete response after trimodality therapy for their esophageal and gastroesophageal junction adenocarcinoma (Abstract 951).
When asked about the reasons for performing this study, Yamashita replied, "Generally, patients who achieved pathological complete response show better prognosis. Predicting treatment response prior to the therapy can be useful in clinical practice."
The ability to predict whether a patient would be likely to achieve a pathological complete response could provide important strategic guidance regarding surgical interventions. Setting this study apart from previous ones was the use of machine learning models; prior studies utilized methodologies including logistic regression models.
When asked why this study focused on trimodality therapy, Yamashita explained, "Most patients are treated with trimodality therapy according to The National Comprehensive Cancer Network guidelines for esophageal and gastroesophageal junction cancers, which is based on results from Phase III clinical trials."
He also explained how pathological complete response was defined in their study. "We used the common definition of pathological complete response (ypT0N0M0). That is the lack of all signs of cancer in tissue samples removed during surgery after trimodality. A pathologist checks the tissue samples under a microscope to see if there are still cancer cells left, and reports no residual cancer cells identified."
Study Details
This study included 569 patients with esophageal and gastroesophageal junction adenocarcinoma treated with trimodality therapy at MD Anderson Cancer Center between 2002 and 2022. The medical records of these patients were mined for different clinical data, including age, sex, histology, baseline TNM (tumor-node-metastasis) staging post-chemoradiotherapy, and maximized standard uptake value in 18-F-fluorodeoxyglucose PET/CT before and after treatment.
The mined clinical data were then used to "train" a number of machine learning models, including BART, logistic regression, LASSO, XGboost, and Random Forest. Once trained, the methods were then validated using a "10-fold cross-validation" process.
Commenting on their important findings, Yamashita noted, "We established the clinical predictive model for pathological complete response in patients with esophageal and gastroesophageal junction adenocarcinoma treated by trimodality therapy using cutting-edge machine learning methods with acceptable predictive ability. The absence of cancer on re-biopsy tissue after neoadjuvant chemoradiotherapy was most highly associated with pathological complete response, which has been well established.
"However, we would like to emphasize that other characteristics, including signet ring cell component, poorly differentiated histological type, and deeper tumor depths were also associated with non-pathological complete response. The combination of various relevant variables is needed to make a good predictive model."
Assessing how this study would affect patient treatment, Yamashita observed, "An accurate model can help implement an organ preservation strategy, meaning we can avoid surgery in some patients based on model predictions."
Regarding future directions for their research, he said, "As we refine our predictive model for publication, we are discussing which clinical information to include and whether tissue-based information can be incorporated to further improve predictive capabilities. We believe that this concept of study will contribute to patient quality of life for organ preservation along with oncological safety."
Richard Simoneaux is a contributing writer.