In This Story
Currently, recommendations for cancer screening are primarily based on the age of the patient. Therefore, practitioners may not encourage younger at-risk individuals to be screened for cancer. They may unnecessarily encourage older low-risk individuals to screen for cancer. Artificial intelligence (AI) can change this. Farrokh Alemi at George Mason University has edited a collection of five articles by colleagues and students on how data science can be used to predict the risk of cancer and enable risk-based AI systems to recommend cancer screening. Their research shows that risk-based models have predicted between 60-90% based on the type of cancers:
-
Carcinoma of the skin ~90%
-
Malignant brain tumors ~80%
-
Kidney cancer ~80%
-
Breast cancer remission ~70%
-
Liver cancer ~60%
Despite being up to 90% effective, risk models are not in the U.S. Preventive Services Task Force’s (USPSTF) recommendations. Alemi, colleagues, and students want to integrate AI models into clinical practices, bypassing USPSTF’s recommendations, and increasing patients’ access to risk-based cancer screening.
“Risk models and AI systems are well-suited to reach patients at home through online services and provide crucial information to patients on whether they should get screened for cancer. Such a system will encourage patients at elevated risk to discuss their situation with their primary care clinicians and, when necessary, go ahead with cancer screening. It will also empower patients at low risk to avoid unnecessary cancer screening,” said Alemi. “Risk-based models are the realization of predictive medicine, much dreamed about but seldom used in clinical practice. AI can lead to wider adoption of these risk models in the care of patients.”
“Predictive risk-based AI models are non-invasive, more accurate than age-based recommendations, more cost-effective, universally applicable, and a pragmatic method of informing patients,” said Alemi.
In a special issue of Quality Management in Health Care, Alemi and colleagues gathered a body of evidence in support of the routine use of AI predictive modeling to better inform those at high risk of cancer.
Grounded in evidence
The issue, edited by Alemi, highlights findings from five peer-reviewed papers written primarily by George Mason University College of Public Health students and faculty (see below for titles and authors). These studies predict the risk of cancer from a comprehensive review of the patient’s medical and social background.
“Integrating these predictive models into clinical practice represents a promising strategy for improving the management and care of patients,” said Yili Lin, a PhD student at George Mason University and a contributing author.
Alemi is a professor in George Mason’s Department of Health Administration and Policy. He was trained as an operations researcher and industrial engineer and has worked in both academia and the health industry. His research focuses on causal analysis of massive data available in electronic health records. Alemi’s work has contributed to the fields of predictive medicine, precision medicine, comparative effectiveness of medications, natural language processing, and artificial intelligence.
Papers authored primarily by faculty, alumni, and students of the Department of Health Administration and Policy at George Mason University include:
-
Early Detection of Basal Cell Carcinoma of Skin From Medical History by Yili Lin, MS
-
EHR-Based Risk Prediction for Kidney Cancer by Kyung Hee Lee, PhD; Farrokh Alemi, PhD; Xia Wang, MD, PhD
-
Predicting Risk of Malignant CNS Tumors From Medical History Events by Aaron J. Hill, MHA, MBA, FACHE
-
Predicting Liver Cancer Risk Using Comprehensive Medical History by Tumen Sosorburam
-
Prediction of Breast Cancer Remission by Vladimir Cardenas, MBA; Yalin Li, BSN, RN; Samika Shrestha, BSN; Hong Xue, PhD
Alemi’s editorial commentary is also available: USPSTF Dismisses Predictive Medicine and Data Science
##
MEDIA INQUIRIES: For reporters who wish to speak to Farrokh Alemi about predictive artificial intelligence, please email media contact Michelle Thompson at mthomp7@gmu.edu.