Oops, you're using an old version of your browser so some of the features on this page may not be displaying properly.

MINIMAL Requirements: Google Chrome 24+Mozilla Firefox 20+Internet Explorer 11Opera 15–18Apple Safari 7SeaMonkey 2.15-2.23

E-Poster Display

1381P - Spatial-statistics-based modeling for predicting treatment response in non-small cell lung cancer (NSCLC) patients using H&E pathology images

Date

17 Sep 2020

Session

E-Poster Display

Topics

Tumour Site

Non-Small Cell Lung Cancer

Presenters

Xiao Li

Citation

Annals of Oncology (2020) 31 (suppl_4): S754-S840. 10.1016/annonc/annonc283

Authors

X. Li1, F. Gaire2, G. Jansen2, R. Copping1, T. Bengtsson1, J. Dai1

Author affiliations

  • 1 Personalized Healthcare, Genentech, Inc., 94080 - South San Francisco/US
  • 2 Personalized Healthcare, F. Hoffmann-La Roche AG, Basel/CH

Resources

Login to get immediate access to this content.

If you do not have an ESMO account, please create one for free.

Abstract 1381P

Background

IMpower150 study in NSCLC (NCT02366143) found no significant overall survival (OS) benefit between the Atezolizumab+Carboplatin+Paclitaxel (ACP, treatment) arm and Carboplatin+Paclitaxel+Bevacizumab (BCP, control) arm, with hazard ratio (HR): 0.85, 95% confidence interval (CI): 0.71-1.03. We applied spatial statistics algorithms to characterize the spatial interaction between tumor cells and lymphocytes in the tumor microenvironment and improved the prediction of response of ACP therapy using the generated spatial features.

Methods

A proprietary image analysis algorithm was applied on H&E pathology images of baseline tissue samples from IMpower150 patients to detect the coordinates of tumor cells and lymphocytes. To systematically extract features that capture the spatial heterogeneity of the tumor microenvironment from these cell coordinates as input, we implemented spatial statistics algorithms based on spatial point, spatial lattice and geostatistical process methods. To investigate the association between the derived spatial features and OS, Cox proportional hazard model with L2 regularization was fitted for the Atezolizumab-treated patients. The high and low response group were further identified using nested Monte Carlo Cross Validation to prevent over-fitting.

Results

284 ACP patients and 271 BCP patients with H&E pathology images and OS in the IMpower150 study were used in this analysis. 41 spatial features including Ripley’s K-function, Morista-Horn index, etc. were derived to capture the cell-cell interaction. In the identified high responder group, the HR between ACP patients and BCP patients is 0·64 (95% CI 0·45–0.91), and the p-value of the log-rank test is 0.012.

Conclusions

We developed the spatial statistics algorithms to identify biologically relevant features in the tumor microenvironment such as immune-cancer cell interactions from the H&E pathology images. Our results indicate the method can better stratify patients who benefit from the Atezolizumab treatment in comparison with standard of care therapy.

Clinical trial identification

NCT02366143.

Editorial acknowledgement

Legal entity responsible for the study

F. Hoffmann–La Roche/Genentech, a member of the Roche Group.

Funding

F. Hoffmann–La Roche/Genentech, a member of the Roche Group.

Disclosure

X. Li, R. Copping, T. Bengtsson, J. Dai: Full/Part-time employment: Genentech, Inc. F. Gaire G. Jansen: Full/Part-time employment: F. Hoffmann-La Roche AG.

This site uses cookies. Some of these cookies are essential, while others help us improve your experience by providing insights into how the site is being used.

For more detailed information on the cookies we use, please check our Privacy Policy.

Customise settings
  • Necessary cookies enable core functionality. The website cannot function properly without these cookies, and you can only disable them by changing your browser preferences.