Article Text

Download PDFPDF

1296 ­Radiomics-based multi-modal prediction of treatment response to PD-1/PD-L1 immune checkpoint inhibitor (ICI) therapy in stage IV non-small cell lung carcinoma (mNSCLC)
Free
  1. Ravi Parikh1,
  2. Petr Jordan2,
  3. Rita Ciaravino2,
  4. Ryan Beasley2,
  5. Arpan Patel3,
  6. Dwight Owen4,
  7. Arya Amini5,
  8. Brendan Curti6,
  9. Ray Page7,
  10. Aurelie Swalduz8,
  11. Jean-Paul Beregi9,
  12. Jan Chrusciel10,
  13. Eric Snyder3,
  14. Pritam Mukherjee11,
  15. Heather Selby11,
  16. Soohee Lee12,
  17. Roshanthi Weerasinghe13,
  18. Shwetha Pindikuri12,
  19. Jakob Weiss14,
  20. Andrew Wentland15,
  21. Anish Kirpalani16,
  22. An Liu5,
  23. Olivier Gevaert11,
  24. George Simon17 and
  25. Hugo Aerts2,14
  1. 1University of Pennsylvania, Philadelphia, PA, USA
  2. 2Onc.AI, San Carlos, CA, USA
  3. 3University of Rochester Medical Center, Rochester, NY, USA
  4. 4The Ohio State University, Columbus, OH, USA
  5. 5City of Hope, Duarte, CA, USA
  6. 6Providence Cancer Institute, Portland, OR, USA
  7. 7The Center for Cancer and Blood Disorders, Fort Worth, TX, USA
  8. 8Centre Léon Bérard, Lyon, France
  9. 9CHU Nimes, Nimes, France
  10. 10Centre Hospitalier de Troyes, Troyes, France
  11. 11Stanford University, Palo Alto, CA, USA
  12. 12Providence Health and Services, Renton, WA, USA
  13. 13Providence St. Joseph Health, Portland, OR, USA
  14. 14Massachusetts General Brigham, Boston, MA, USA
  15. 15University of Wisconsin, Madison, WI, USA
  16. 16University of Toronto, Toronto, Canada
  17. 17Moffitt Cancer Center, Tampa, FL, USA

Abstract

Background Currently approved biomarkers that predict response to ICIs in mNSCLC are limited to PD-L1 expression levels by immunohistochemistry (IHC) and tumor mutation burden (TMB). However, the predictive performance of PD-L1 IHC and TMB are limited, and rates of testing are suboptimal. Radiomic biomarkers may offer an automated and scalable method to predict ICI response.1,2 We developed and validated multi-modal models predicting responses to ICIs in mNSCLC. In contrast to previously published models, our work focuses on generalizable models using a large multi-institutional “real-world” dataset and combines radiomics features with demographic, molecular, and laboratory values routinely available in patients’ electronic medical records [EMR].

Methods We analyzed radiomic characteristics of 6,028 primary and metastatic lesions from 1,169 mNSCLC patients treated with anti-PD-1/anti-PD-L1 ICIs from 8 institutions across the US and Europe. Data were randomly split into training (N=707 patients, n=3,625 lesions) and validation (N=462 patients, n=2,403 lesions) sets. Baseline and follow-up CT scans were manually annotated by board-certified radiologists using RECIST 1.1 criteria and all lesion volumes were manually segmented. We developed two predictive models using gradient-boosted decision tree algorithms, using 1) only manually curated baseline radiomic features quantifying textural heterogeneity and spicularity; and 2) a multi-modal model with radiomic features combined with known demographic, molecular (e.g. PD-L1 IHC), and laboratory (e.g. neutrophil-to-lymphocyte ratio) predictors of ICI response. Primary endpoints were 3- and 6-month radiological progression, defined by a 20% increase in lesion diameter. The primary evaluation metric was the area under the receiver operating characteristic curve (AUC). Models predicting response of lung lesions and lymph nodes were validated on two cohorts: ICI monotherapy and ICI plus concurrent chemotherapy. Patients with unavailable PD-L1 IHC, imaging follow-up, or oncogenic driver mutations were excluded from analysis.

Results The radiomics model showed predictive accuracy comparable to tissue-based PD-L1 IHC for both endpoints and patient cohorts (tables 1, 2). However, the multi-modal model predicted lung and lymph node radiological progression with significantly higher AUC than PD-L1 IHC in all cohorts and endpoints, with 3- and 6-month progression AUCs of 0.86 (P=.00007) and 0.79 (P= .00001) in lung lesions and 0.78 (P=.003) and 0.80 (P=.002) in lymph nodes.

Conclusions Radiomics-based multi-modal prediction of ICI response is feasible and accurate and may provide an opportunity for more personalized management, such as risk-based escalation/de-escalation of concurrent chemotherapy in mNSCLC patients. We will evaluate this methodology in prospective studies.

References

  1. Trebeschi S, Drago S, Birkbak N, Kurilova I, Cǎlin A, Delli Pizzi A, Lalezari F, Lambregts D, Rohaan M, Parmar C, Rozeman E, Hartemink K, Swanton C, Haanen J, Blank C, Smit E, Beets-Tan R, Aerts H. Predicting response to cancer immunotherapy using noninvasive radiomic biomarkers. Ann. Oncol. 2019; 30(6): 998–1004.

  2. Sun R, Limkin E, Vakalopoulou M, Dercle L, Champiat S, Han SR, Verlingue L, Brandao D, Lancia A, Ammari S, Hollebecque A, Scoazec J, Marabelle A, Massard C, Soria J, Robert C, Paragios N, Deutsch E, Ferté C. A radiomics approach to assess tumour-infiltrating CD8 cells and response to anti-PD-1 or anti-PD-L1 immunotherapy: an imaging biomarker, retrospective multicohort study. Lancet Oncol. 2018; 19(9): 1180–1191.

Ethics Approval Ethics approval for US data:

The study was conducted under IRB-approved procedures using de-identified data for patients diagnosed with Stage-IV NSCLC and treated between Jan. 1, 2017 and December 31, 2021. All records were de-identified per HIPAA guidelines at the institution level. Upon transfer, the data was quarantined and then re-inspected by authorized personnel prior to ingestion to ensure compliance and that no PHI was present in the records.

Ethics approval for EU data:

The study was conducted under IRB-approved procedures using de-identified data for patients diagnosed with Stage-IV NSCLC and treated between Jan. 1, 2017 and December 31, 2021. All records were de-identified per GDPR requirements at the institution level. The patients were also notified that their de-identified data would be part of a study and were given the required time and opportunity to respond if they had any objection. Upon transfer, the data was quarantined and then re-inspected by authorized personnel prior to processing to ensure compliance and that no PHI was present in the records.

Consent N/A

Abstract 1296 Table 1

Lung lesion assessment.

Abstract 1296 Table 2

Lymph node assessment.

Statistics from Altmetric.com

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.