Gastrointestinal Cancer

PD 05 - GI 2 - Poster Discussion

1040 - Machine Learning Prediction of Early Distant Progression for Oligometastatic and Oligoprogressive Colorectal Cancer (CRC) Patients Treated With Stereotactic Body Radiation Therapy (SBRT)

Monday, October 22
11:09 AM - 11:15 AM
Location: Room 217 A/B

Machine Learning Prediction of Early Distant Progression for Oligometastatic and Oligoprogressive Colorectal Cancer (CRC) Patients Treated With Stereotactic Body Radiation Therapy (SBRT)
P. Lang1, M. Kayvanrad2, R. D. Thompson3, P. Cheung4, E. D. Gennatas5, G. Valdes5, and H. T. Chung4; 1Sunnybrook Health Sciences Centre, Toronto, ON, Canada, 2Rotman Research Institute, University of Toronto, Toronto, ON, Canada, 3Saint John Regional Hospital, Saint John, NB, Canada, 4Department of Radiation Oncology, Sunnybrook Odette Cancer Centre, University of Toronto, Toronto, ON, Canada, 5University of California San Francisco, Department of Radiation Oncology, San Francisco, CA

Purpose/Objective(s): Recent studies show SBRT for oligo-metastases (OM) and oligo-progression (OP) confers good outcomes and low morbidity, but clinicians face significant challenges selecting patients who will benefit from SBRT, due to complex interactions of patient, tumor and treatment factors. This study examines the ability of machine learning (ML) based classifiers to identify patients who develop early distant progression (DP, ≤ 90 days since treatment completion) in CRC patients receiving SBRT.

Materials/Methods: All CRC patients treated with SBRT to any site at a single institution for OM/OP in 2009 - 2016 were retrospectively reviewed. Clinical characteristics included age, gender, pre-SBRT CEA, RAS status, ECOG performance, treatment indication (OM/OP), SBRT location, disease free interval since last treatment (DFI), number of prior lines of systemic of therapy, prior use of ablative local therapy, PTV volume and mean PTV BED. Univariable and multivariable logistic regression was used to identify predictors of DP. Classification methods included: logistic regression (LR) gradient boosting (GBM), adaptive boosting (ADA), and random forest (RF). Data was divided into training (75%) and testing (25%) cohorts with monte carlo cross-validation with 10 trials. Classifier performance was assessed by receiver operating characteristic curves. Area under the curve (AUC) values were compared using a paired t-test with Bonferroni adjustment.

Results: 147 patients with 226 treated lesions were included; 203 treated for OM and 23 OP. 31 (15.2%) of the treated lesions were followed by DP within 90 days. No patients died or were lost to follow-up prior to the 90 days. In univariable analysis, age, CEA, treatment indication, DFI, number of systemic therapy lines and mean PTV BED were significantly associated with early DP (p < 0.05). In multivariable analysis treatment indication, DFI, number of systemic therapy lines, and mean PTV BED and were significant predictors of DP (p < 0.05). Performance of the various classifiers is shown in the table below. All ML classifiers were significantly better at identifying patients with DP compared to the logistic regression model (p < 0.05). There was no statistically significant difference in performance between the various ML classifiers. The top ranked variables by the RF classifier were DFI, PTV mean dose, number of systemic therapy lines, CEA and age. These are consistent with predictors found on univariable and multivariable analysis.
LR GBM ADA RF
Sen 0.739 0.740 0.964 0.927
Spec 0.428 0.429 0.457 0.514
Balanced accuracy 0.584 0.584 0.710 0.721
AUC 0.590 0.744 0.767 0.821

Conclusion: Treatment indication, DFI, number of systemic therapy lines, and mean PTV BED are associated with DP. ML classifiers were significantly better at identifying patients with DP compared to the logistic regression model. The ability to predict patients at risk of DP would assist clinicians in identifying patients who may benefit minimally from SBRT for OM/OP disease.

Author Disclosure: P. Lang: None. M. Kayvanrad: None. R.D. Thompson: None. P. Cheung: Independent Contractor; Ontario Ministry of Health and Long-Term Care. Research Grant; Pfizer, Sanofi Aventis, Abbvie. G. Valdes: None. H.T. Chung: None.

Pencilla Lang, MD, PhD, BEng

Princess Margaret Cancer Centre

Disclosure:
No relationships to disclose.

Presentation(s):

Send Email for Pencilla Lang


Assets

1040 - Machine Learning Prediction of Early Distant Progression for Oligometastatic and Oligoprogressive Colorectal Cancer (CRC) Patients Treated With Stereotactic Body Radiation Therapy (SBRT)



Attendees who have favorited this

Please enter your access key

The asset you are trying to access is locked. Please enter your access key to unlock.

Send Email for Machine Learning Prediction of Early Distant Progression for Oligometastatic and Oligoprogressive Colorectal Cancer (CRC) Patients Treated With Stereotactic Body Radiation Therapy (SBRT)