Improving Lung Cancer Screening and Prediction Through LDCT and Causal Modeling

Lung cancer is the second most prevalent form of cancer in the United States, with an estimated 228,820 new cases diagnosed each year and more than 135,000 mortalities.¹ Prevention and modifiable risk factors, such as smoking, remain the keys to decreasing lung cancer rates and accompanying deaths. Screening for individuals at high-risk for the disease also is paramount, and much work has been done to increase these rates in recent years, in large part due to the rise in the use of low-dose computed tomography (LDCT) and newly devised screening protocols. 

However, LDCT is not a perfect screening tool, and the current parameters and modeling used to classify those at high-risk for the disease are suboptimal, potentially excluding many individuals for whom screening may benefit. While LDCT reliably detects lung nodules, it has been shown to produce a high rate of false positive results – detecting nodules that are not lung cancer. The ability to discern or predict the presence of screen detected cancerous nodules from those that are not cancer is paramount to avoiding unnecessary and costly follow-up care in benign cases. So, too, is accurately predicting who is most at risk for lung cancer based on validated variables and modeling techniques.

Research from David O. Wilson, MD, MPH, associate director of the Lung Cancer Center at UPMC Hillman Cancer Center, seeks to improve significantly upon lung cancer screening for high-risk individuals through the use of predictive and causal modeling techniques coupled with findings from LDCT that can more accurately detect the presence or likelihood of malignant lung nodules from noncancerous ones and to more precisely define the parameters used to gauge risk status of individual patients. Dr. Wilson and colleagues published their data on a new type of lung cancer detection model in 2019 in the journal Thorax.¹

Using Probabilistic Graphical Modeling to Develop and Validate a Lung Cancer Causal Model

To develop their Lung Cancer Causal Model (LCCM), Dr. Wilson and colleagues used data from a cohort of lung cancer patients from the Pittsburgh Lung Screening Study. Using probabilistic graphical models derived from patient data, including findings from LDCT scans, the patient’s clinical history, and their demographic makeup, they developed a model (the LCCM) that included variables determined to be causally tied to malignant lung nodules. These variables included the number of years since the individual quit smoking, the number of nodules detected on LDCT, and the number of blood vessels surrounding the nodule(s) as seen on LDCT. 

The latter variable is a novel imaging biomarker developed at UPMC by Jiantao Pu, PhD, and his radiology informatics team. It has previously been shown to differentiate benign from malignant lung nodules in that malignant nodules were more often surrounded by increased numbers of vessels versus benign nodules.2

The LCCM model was then validated against another similar but separate patient cohort from the Pittsburgh Lung Cancer Screening Study. 

Overall, the LCCM showed improved lung cancer detection rates compared to existing modeling techniques (specifically the Brock parsimonious model). Comparing the model’s assessment against the actual diagnoses of these patients, the researchers found that they would have been able to save 30% of the people with benign nodules from undergoing additional testing without missing a single case of cancer.

Dr. Wilson and colleague’s research constitutes the first use of artificial intelligence methodologies to better identify malignant versus benign lung cancer nodules in a screening protocol. 

More About Dr. Wilson

David O. Wilson, MD, MPH, is an associate professor in the Division of Pulmonary, Allergy, and Critical Care Medicine at the University of Pittsburgh School of Medicine. He holds secondary appointments in cardiothoracic surgery and clinical & translational science.

Dr. Wilson is the director of the Georgia Cooper Lung Nodule/Cancer Proteomics/Genomics Research Registry and the co-director of the Lung Cancer Center at UPMC Hillman Cancer Center. He is the Director of the Pittsburgh Lung Cancer Screening Study (PLuSS).

Dr. Wilson’s research interests include lung cancer screening and chemoprevention, diagnosis, staging, and treatment; COPD, especially related to lung cancer; occupational lung diseases; general pulmonary medicine; and nutrition support. Dr. Wilson's current research focus is on developing predictive tools, beyond emphysema, for risk stratification in lung cancer screening. This work includes risk prediction formulas and artificial intelligence, surrogate tissue biomarkers, and imaging biomarkers.

As a clinician, Dr. Wilson was the first to introduce endobronchial ultrasound (EBUS) to the Pittsburgh region, and he has a longstanding interest in lung cancer staging and treatment.

References and Related Research

1. American Cancer Society: Cancer Facts and Figures 2020. Atlanta, Ga: American Cancer Society, 2020.

2. Raghu VK, Zhao W, Pu J, Leader JK, Wang R, Herman J, Yuan J-M, Benos PV, Wilson DO. Feasibility of Lung Cancer Prediction From Low-Dose CT Scan and Smoking Factors Using Causal Models. Thorax. 2019 Jul; 74(7): 643-649.

3. Wang X, Leader JK, Wang R, Wilson D, Herman J, Yuan J-M, Pu J. Vasculature Surrounding a Nodule: A Novel Lung Cancer Biomarker. Lung Cancer. 2017; 114: 38-43.

4. Wilson DO, de Torres JP. Lung Cancer Screening: How Do We Make It Better? Quant Imaging Med Surg. 2020 Feb; 10(2): 533-536.

5. Warsavage Jr. T, Xing F, Baron AE, Feser WJ, Hirsch E, Miller YE, Wolf HJ, Wilson DO, Ghosh D. Quantifying the Incremental Value of Deep Learning: Application to Lung Nodule Detection. PLoS One. 2020 Apr 14; 15(4): e0231468.