Radiation and Cancer Physics
SS 01 - Physics 1 -Best of Physics
12 - Modeling of Locoregional Control in Hepatocellular Carcinoma After Stereotactic Body Radiation Therapy by Integrating Clinical and Immune Cell Profiles
Sunday, October 21
2:15 PM - 2:25 PM
Location: Room 214 A/B
Issam El Naqa, PhD, MS
University of Michigan
University of Michigan: Employee
Endectra, LLC: Advisory board
Modeling of Locoregional Control in Hepatocellular Carcinoma After Stereotactic Body Radiation Therapy by Integrating Clinical and Immune Cell Profiles
I. El Naqa, D. Owen, K. C. Cuneo, C. Mayo, T. S. Lawrence, and R. K. Ten Haken; Department of Radiation Oncology, University of Michigan, Ann Arbor, MI
Locoregional control (LRC) is associated with long-term outcomes in patients with hepatocellular carcinoma (HCC) who receive liver SBRT. The purpose of the current study is to develop new models for predicting LRC and to evaluate the role of clinical factors and circulating immune cells in predicting post-SBRT response.
Data from 146 HCC patients who received SBRT from 2005-14 were analyzed retrospectively. Tumor doses (median prescribed = 49.8 Gy, delivered in 3 or 5 fractions) were bio-corrected to 2 Gy equivalents (EQD2) using the LQ-L model. Circulating immune cell (lymphocytes, neutrophils, platelets) profiles (ICPs), red blood cell counts, and their changes during and after treatment were retrieved from the patients' laboratory records. The locoregional failure rate was 54.7% with a median follow-up and time-to-failure of 11 months and 6 months, respectively. Actuarial models based on machine learning algorithms were developed for predicting LRC. These models were based on variable shrinkage analysis with Cox proportional hazard (Lasso-Cox) and ensemble methods with Random survival forests (RSF). Relative variable importance was assessed in Lasso-Cox by hazard ratios (HR) and in RSF by the Gini impurity index. To avoid overfitting pitfalls, the bootstrap 0.632+ resampling method was used to validate prediction, and performance was measured using the concordance statistic (c-index).
When modeling with only clinical variables for LRC, the Lasso-Cox selected in descending HR significance order: previous recurrences, age, Child Pugh, ALBI,
as relevant factors, achieving an average bootstrap c-index = 0.538 (95% CI: 0.496-0.564). The RSF achieved an average bootstrap c-index = 0.687 (95% CI: 0.666-0.704) with previous recurrences, Child Pugh, age, ALBI, gender, tumor mean dose, tumor volume
as the top important factors according to their Gini impurity index. The addition of ICPs, available in about half the population, yielded an average bootstrap c-index = 0.569 (95% CI: 0.496-0.67) using Lasso-Cox with pre-treatment hematocrits, ALBI,
and changes in lymphocyte
and neutrophil counts
as statistically significant variables according to their HRs (p<0.05). Whereas the RSF, with immune cells included, achieved the best performance overall with an average bootstrap validated c-index = 0.696 (95% CI: 0.638-0.751). Changes in lymphocytes
and platelets counts, pre-treatment hematocrits, neutrophil counts, ALBI,
were the most important factors according to the Gini impurity index ranking.
Machine learning methods based on RSF can provide a robust framework for estimating locoregional failure risk in HCC patients post-SBRT. The predictive power improves by including immune cells profile and their changes during treatment. These new LRC models can be used to personalize and guide new regimens for combining local (SBRT) and systemic (chemo- and/or immuno-) therapy in HCC patients.
Author Disclosure: I. El Naqa: None. D. Owen: None. K.C. Cuneo: Service Chief; Ann Arbor Veterans Hospital. C. Mayo: Research Grant; Varian Medical Systems. Chair TG-263 Radiation Oncology Nomenclature; AAPM. T.S. Lawrence: royalties; Lippincott, Williams and Wilkins. Honoraria; Massachusetts General Hospital, Pfizer Oncology Innovation Summit, Sidney Kimmel Foundation for Cancer Research. Consultant; Pfizer Oncology Innovation Summit. Advisory Board; ASTRO Radiation Oncology Institute, Dana Farber Cancer Institute, Massachusetts General Hospital, Sidney Kimmel Compreh Cancer Ctr at Johns Hopkins, Sidney Kimmel Foundation for Cancer Research, St. Jude Children's Research Hospital, University of Wisconsin Comprehensive Cancer Ctr. Travel Expenses; AACR, ASTRO Radiation Oncology Institute, Dana Farber Cancer Institute, Lippincott, Williams and Wilkins, Massachusetts General Hospital, Pfizer Oncology Innovation Summit, RSNA, Sidney Kimmel Compreh Cancer Ctr at Johns Hopkins, Sidney Kimmel Foundation for Cancer Research, St. Jude Children's Research Hospital, University of Wisconsin Comprehensive Cancer Ctr. Patent/License Fees/Copyright; Pi Squared Therapeutics. Editor, Cancer Discovery; AACR. Member, Editorial Advisory Board, Cancer Today; AACR. Senior Editor, Cancer Research; AACR. Member, External Advisory Board for Lung SPORE; Dana Farber Cancer Institute. Co-Editor of Principles and Practices of Oncology; Lippincott, Williams and Wilkins. Member, NCI Board of Scientific Advisors; NCI - BSA. President; ROI. Member, External Advisory Board for the Cancer Ctr; Sidney Kimmel CCC at Johns Hopkins University. Member of the Medical Advisory Board; Sidney Kimmel Foundation for Cancer Research. Vice-Chair, St. Jude Scientific Advisory Board; St. Jude Children's Research Hospital. Member, V Foundation Scientific Advisory Board; V Foundation for Cancer Research. R.K. Ten Haken: Research Grant; NIH-NCI. Honoraria; University of Copenhagen. Travel Expenses; Varian Medical Systems Inc, University of Copenhagen.