Machine Learning to select patients benefiting from prostate cancer screening
Principal Investigator
Name
Jean-Emmanuel Bibault
Degrees
M.D., Ph.D.
Institution
Laboratory of Artificial Intelligence in Medicine and Biomedical Physics, Stanford University
Position Title
Postdoctoral Research Fellow
Email
About this CDAS Project
Study
PLCO
(Learn more about this study)
Project ID
PLCO-541
Initial CDAS Request Approval
Oct 24, 2019
Title
Machine Learning to select patients benefiting from prostate cancer screening
Summary
The effect of screening with prostate-specific–antigen (PSA) testing and digital rectal examination on the rate of death from prostate cancer has been explored in several randomized trials, including the PLCO trial. After 7 years of follow-up, the incidence of prostate cancer per 10,000 person-years was 116 (2820 cancers) in the screening group and 95 (2322 cancers) in the control group (rate ratio, 1.22; 95% confidence interval [CI], 1.16 to 1.29). The incidence of death per 10,000 person-years was 2.0 (50 deaths) in the screening group and 1.7 (44 deaths) in the control group (rate ratio, 1.13; 95% CI, 0.75 to 1.70).
If prostate cancer screening did not show a benefit for an unselected cohort of patients, it could still be very relevant to screen patients who have high-risk clinical profiles. In this project, we intend to use machine learning methods to tailor a phenotype profile of the patients who are most likely to benefit from prostate cancer screening. Machine Learning techniques can be leveraged to unravel clinical entities and relationship that have not been explored. In the first step, we will use the data from the PLCO project to create unsupervised and supervised patient profiles and establish phenotypes. We will then correlate these profiles to several endpoints:
- prostate cancer diagnosis,
- prostate cancer stage at diagnosis,
- prostate cancer specific survival,
- overall survival.
To do so, we will perform two different analysis: one on all patients included and one with the experimental arm vs a subset of patients from the control arm without baseline screening contamination.
Once these profiles will be done, we will assess the relevance of several supervised machine learning methods to classify patients between benefit and no-benefit from PSA screening: these methods will include Random Forest, Gradient Boosting and Deep Learning. Class imbalance correction techniques will be used before training the models.
The main outcome of the project is to be able to individualize ten clinical features strongly linked with a benefit of prostate cancer screening in order to guide the physicians and patients before deciding to perform a PSA testing.
If prostate cancer screening did not show a benefit for an unselected cohort of patients, it could still be very relevant to screen patients who have high-risk clinical profiles. In this project, we intend to use machine learning methods to tailor a phenotype profile of the patients who are most likely to benefit from prostate cancer screening. Machine Learning techniques can be leveraged to unravel clinical entities and relationship that have not been explored. In the first step, we will use the data from the PLCO project to create unsupervised and supervised patient profiles and establish phenotypes. We will then correlate these profiles to several endpoints:
- prostate cancer diagnosis,
- prostate cancer stage at diagnosis,
- prostate cancer specific survival,
- overall survival.
To do so, we will perform two different analysis: one on all patients included and one with the experimental arm vs a subset of patients from the control arm without baseline screening contamination.
Once these profiles will be done, we will assess the relevance of several supervised machine learning methods to classify patients between benefit and no-benefit from PSA screening: these methods will include Random Forest, Gradient Boosting and Deep Learning. Class imbalance correction techniques will be used before training the models.
The main outcome of the project is to be able to individualize ten clinical features strongly linked with a benefit of prostate cancer screening in order to guide the physicians and patients before deciding to perform a PSA testing.
Aims
- Create unsupervised and supervised patient phenotype profiles
- Correlate these profiles to several endpoints:
- prostate cancer diagnosis,
- prostate cancer stage at diagnosis,
- prostate cancer specific survival,
- overall survival.
- Link them to benefit vs no benefit of prostate cancer screening
Collaborators
Lei Xing, PhD, Jacob Haimson Professor of Medical Physics and Director of Medical Physics Division of Radiation Oncology Department, Stanford University
Maxime Bassenne, PhD, Postdoctoral Research Fellow, Laboratory of Artificial Intelligence in Medicine and Biomedical Physics, Stanford University
Related Publications
-
Development and Validation of an Interpretable Artificial Intelligence Model to Predict 10-Year Prostate Cancer Mortality.
Bibault JE, Hancock S, Buyyounouski MK, Bagshaw H, Leppert JT, Liao JC, Xing L
Cancers (Basel). 2021 Jun 19; Volume 13 (Issue 12) PUBMED