Lead Investigator: Diego Chowell, Icahn School of Medicine at Mount Sinai
Title of Proposal Research: Machine learning model to predict cancer immunotherapy response
Vivli Data Request: 7810
Funding Source: None
Potential Conflicts of Interest: None
Summary of the Proposed Research:
Recently, as a treatment of cancer, immune system based therapy, also known as immunotherapy, is widely used. Immune checkpoints are part of the immune system and they exist to prohibit an immune response from being too strong that it kills normal cells in the body.
Immune checkpoints start working when proteins on the surface of T-cells, a type of immune cells that are developed from stem cells in the bone marrow and have an essential role in the adaptive immune response, recognize and bind to partner proteins on other cells such as tumor cells. When the binding is solid, an “off” signal is sent to the T-cells which can prevent the immune system from attacking the tumor cells.
Immunotherapy treatments called immune checkpoint blockade “blocks” the checkpoint proteins from binding with their partners, hence, prevents the “off” signal from being sent. This allows the T-cells to kill tumor cells.
Immune checkpoint blockade (ICB) has saved the lives of many patients with metastatic cancers. Metastatic cancer is a cancer that has spread from the part of the body where it started (the primary site) to other parts of the body. When cancer cells break away from a tumor, they can travel to other parts of the body through the bloodstream or the lymph system. In the United States, 6% of women have metastatic breast cancer when they are first diagnosed.
Melanoma is a type of cancer that ICB therapy is widely used for the first line treatment these days. The 3-year overall survival for advanced melanoma has increased from 12% before 2010, when standard of care was chemotherapy, to approximately 60% using ICB therapy such as PD-1/PD-L inhibitors. Although the effect of this treatment is significantly evident, the cost of treatment is still quite high and only less than 20% of cancer patients show response to this type of immunotherapy. Therefore, it is crucial that we build a guideline that can aid a patient to have the most beneficial treatment with positive outcomes. An accurate method for pre-therapy identification of patients whose tumors will respond to ICB would be of significant benefit.
Statistical Analysis Plan:
To calculate the probability of response to immunotherapy of each patient, we will apply our trained learning random forest classifier with several input features of the tumor and clinical information of the patient. The aim of this research proposal is to validate our machine learning algorithm to generate an accurate prediction of a patient’s probability of immunotherapy response by comprehensively integrating multiple biological features associated with immunotherapy efficacy, and to assess their individual contribution to response when combined in a single machine learning predictive framework.
We are going to impute missing values using MissForest (https://doi.org/10.1093/bioinformatics/btr597) which implements RandomForestRegressor. For this, external dataset will be merged into our training set and MissForest will be applied.
Ten clinical trial cohorts (IMbrave150, IMspire150, IMpower133, OAK, IMpower132, IMpower131, IMpower130, IMpower150, IMvigor211, and IMmotion151) will be used as the external datasets for the validation of our machine learning model. The ten clinical trials will be analyzed individually to demonstrate the predictive power of our trained model. The reason why we aim to analyze each cohort separately is that they have different combinations of treatments: 1) Atezolizumab + bevacizumab for IMbrave150, 2) Atezolizumab+ cobimetinib + vemurafenib for IMspire150, 3) Atezolizumab + Carboplatin + Etoposide for IMpower133, 4) Atezolizumab alone for OAK, 5) Atezolizumab + Carboplatin or Cisplatin + Pemetrexed for IMpower132, 6) Atezolizumab + Paclitaxel + Carboplatin or Atezolizumab + Nab-Paclitaxel + Carboplatin for IMpower131, 7) Atezolizumab+Nab-Paclitaxel+Carboplatin for IMpower130, 8) Atezolizumab+Paclitaxel+Carboplatin or Atezolizumab+Bevacizumab+Paclitaxel + Carboplatin for IMpower150, 9) Atezolizumab only for IMvigor211, 10) Atezolizumab + Bevacizumab for IMmotion151. A meta-analysis of four clinical trial cohorts won’t be performed.
Machine learning model will be built using python’s sklearn or sksurv packages and model performance regarding prediction of clinical benefit will be performed with “pROC” R package.
The search criteria was to find a cohort with patients treated with immunotherapy and the availability of laboratory test results.
R packages needed:
R: caret, survcomp
Python: sklearn , sksurv, ngboost
Requested Studies:
A Phase III, Open-Label, Multicenter, Randomized Study to Investigate the Efficacy and Safety of Atezolizumab (Anti-PD-L1 Antibody) Compared With Docetaxel in Patients With Non-Small Cell Lung Cancer After Failure With Platinum Containing Chemotherapy
Data Contributor: Roche
Study ID: NCT02008227
Sponsor ID: GO28915
A Phase III, Open-Label, Randomized Study of Atezolizumab in Combination With Bevacizumab Compared With Sorafenib in Patients With Untreated Locally Advanced or Metastatic Hepatocellular Carcinoma
Data Contributor: Roche
Study ID: NCT03434379
Sponsor ID: YO40245
A Phase I/III, Randomized, Double-Blind, Placebo-Controlled Study of Carboplatin Plus Etoposide With or Without Atezolizumab (Anti-PD-L1 Antibody) in Patients With Untreated Extensive-Stage Small Cell Lung Cancer
Data Contributor: Roche
Study ID: NCT02763579
Sponsor ID: GO30081
A Phase III, Double-Blinded, Randomized, Placebo-Controlled Study of Atezolizumab Plus Cobimetinib and Vemurafenib Versus Placebo Plus Cobimetinib and Vemurafenib in Previously Untreated BRAFV600 Mutation-Positive Patients With Unresectable Locally Advanced or Metastatic Melanoma
Data Contributor: Roche
Study ID: NCT02908672
Sponsor ID: CO39262
A Phase III, Open-Label, Randomized Study of Atezolizumab (MPDL3280A, Anti-Pd-L1 Antibody) in Combination With Carboplatin or Cisplatin + Pemetrexed Compared With Carboplatin or Cisplatin + Pemetrexed in Patients Who Are Chemotherapy-Naive and Have Stage IV Non-Squamous Non-Small Cell Lung Cancer
Data Contributor: Roche
Study ID: NCT02657434
Sponsor ID: GO29438
A Phase III, Open-Label, Multicenter, Randomized Study Evaluating the Efficacy and Safety of Atezolizumab (MPDL3280A, Anti-PD-L1 Antibody) in Combination With Carboplatin+Paclitaxel or Atezolizumab in Combination With Carboplatin+Nab-Paclitaxel Versus Carboplatin+Nab-Paclitaxel in Chemotherapy-Naive Patients With Stage IV Squamous Non-Small Cell Lung Cancer
Data Contributor: Roche
Study ID: NCT02367794
Sponsor ID: GO29437
A Phase III Multicenter, Randomized, Open-Label Study Evaluating the Efficacy and Safety of Atezolizumab (MPDL3280A, Anti-PD-L1 Antibody) in Combination With Carboplatin+Nab-Paclitaxel for Chemotherapy-Naive Patients With Stage IV Non-Squamous Non-Small Cell Lung Cancer
Data Contributor: Roche
Study ID: NCT02367781
Sponsor ID: GO29537
A Phase III, Open-Label, Randomized Study of Atezolizumab (MPDL3280A, Anti-PD-L1 Antibody) in Combination With Carboplatin+Paclitaxel With or Without Bevacizumab Compared With Carboplatin + Paclitaxel + Bevacizumab in Chemotherapy-Naïve Patients With Stage IV Non-Squamous Non-Small Cell Lung Cancer
Data Contributor: Roche
Study ID: NCT02366143
Sponsor ID: GO29436
A Phase III, Open-Label, Multicenter, Randomized Study to Investigate the Efficacy and Safety of Atezolizumab (Anti-PD-L1 Antibody) Compared With Chemotherapy in Patients With Locally Advanced or Metastatic Urothelial Bladder Cancer After Failure With Platinum-Containing Chemotherapy
Data Contributor: Roche
Study ID: NCT02302807
Sponsor ID: GO29294
A Phase III, Open-Label, Randomized Study of Atezolizumab (Anti-PD-L1 Antibody) in Combination With Bevacizumab Versus Sunitinib in Patients With Untreated Advanced Renal Cell Carcinoma
Data Contributor: Roche
Study ID: NCT02420821
Sponsor ID: WO29637
Public Disclosures:
Yoo, S.K., Fitzgerald, C.W., Cho, B.A., Fitzgerald, B.G., Han, C., Koh, E.S., Pandey, A., Sfreddo, H., Crowley, F., Korostin, M.R. and Debnath, N., 2025. Prediction of checkpoint inhibitor immunotherapy efficacy for cancer using routine blood tests and clinical data. Nature Medicine, pp.1-12. Doi : 10.1038/s41591-024-03398-5