A Machine Learning Approach to Bladder Cancer Screening and Cancer Grade/Stage Differentiation
Principal Investigator
Name
Herbert Chase
Degrees
M.D., M.A.
Institution
The Trustees of Columbia University in the City of New York
Position Title
Professor of Clinical Medicine (Biomedical Informatics)
Email
About this CDAS Project
Study
PLCO
(Learn more about this study)
Project ID
PLCO-597
Initial CDAS Request Approval
Mar 16, 2020
Title
A Machine Learning Approach to Bladder Cancer Screening and Cancer Grade/Stage Differentiation
Summary
Bladder cancer is the 6th most common cancer in the United States, with a 2016 prevalence of 699,450 and a predicted 2019 incidence of 80,470. The 5-year survival rate for all stages of bladder cancer is 76.8%, with localized cancers having a 69.4% survival and distant cancers having a 4.8% survival. The management and prognosis of bladder cancer is highly dependent on the tumor grade and stage. Nearly a quarter of new bladder cancer cases are muscle-invasive, with 50% of non-muscle invasive cancers progressing to muscle invasion during the disease course. Cancer recurrence is a common issue in bladder cancer, with recurrence rates ranging from 20-30% for stage pT2 to more than 50% for pT4.
Despite a wide discrepancy in survival rates based on cancer stage, early screening tests for bladder cancer are not currently recommended by the US Preventive Services Task Force because of insufficient evidence for diagnostic accuracy in asymptomatic individuals. A recent meta analysis from Fernández et al. found that screening is helpful for high risk populations, but there are no current screening modalities recommended for the general population. The primary symptom for bladder cancer is the appearance of hematuria, so studies have looked at the effectiveness of home screening kits in asymptomatic individuals, however the results showed low diagnostic yield.
Given the current lack of dependable screening modalities for bladder cancer and the divergent courses of management according to tumor stage and grade, we believe that a machine learning classifier could help to identify patients at risk of bladder cancer and high grade/stage bladder cancer. Current risk factors for bladder cancer include smoking, occupational exposures, pelvic radiation, age, and gender, but the data available from the PLCO database, in combination with machine learning techniques, could help to elucidate new risk factors or novel combinations of risk factors that are indicative of advanced disease.
Despite a wide discrepancy in survival rates based on cancer stage, early screening tests for bladder cancer are not currently recommended by the US Preventive Services Task Force because of insufficient evidence for diagnostic accuracy in asymptomatic individuals. A recent meta analysis from Fernández et al. found that screening is helpful for high risk populations, but there are no current screening modalities recommended for the general population. The primary symptom for bladder cancer is the appearance of hematuria, so studies have looked at the effectiveness of home screening kits in asymptomatic individuals, however the results showed low diagnostic yield.
Given the current lack of dependable screening modalities for bladder cancer and the divergent courses of management according to tumor stage and grade, we believe that a machine learning classifier could help to identify patients at risk of bladder cancer and high grade/stage bladder cancer. Current risk factors for bladder cancer include smoking, occupational exposures, pelvic radiation, age, and gender, but the data available from the PLCO database, in combination with machine learning techniques, could help to elucidate new risk factors or novel combinations of risk factors that are indicative of advanced disease.
Aims
1. Identify new risk factors and novel combinations of risk factors for bladder cancer in general
2. Identify new risk factors and novel combinations of risk factors for high grade and advanced stage bladder cancer
Collaborators
Herbert S. Chase, MD (Professor of Clinical Medicine in Biomedical Informatics, Columbia University Irving Medical Center)