Blood Test-Based Machine Learning Model for Predicting Lung Cancer Risk

Lihi Schwartz¹Naor Matania²Matanel Levi³Teddy Lazebnik^4*Shiri Kushnir¹Noga Yossef¹Assaf Hoogi³Dekel Shlomi³

• ¹Clalit Health Services, Tel Aviv, Israel
• ²Bar-Ilan University, Ramat Gan, Tel Aviv District, Israel
• ³Ariel University, Ariel, Israel
• ⁴University College London, London, United Kingdom

Background: Individual cancer prediction is the goal of early detection. In lung cancer (LC), age and smoking background are the strongest criteria for annual low-dose CT for screening, leaving other populations at risk. Machine learning (ML) is a promising method to find complex patterns in the data which can reveal personalized disease predictors. Methods: An ML-based model using blood tests before the diagnosis of LC and socio-demographic factors such as age and gender among LC patients versus controls that predict the risk for future LC diagnosis.Results: Apart from age and gender, we found 22 blood tests that contribute to the model. For the entire study population, the ML model had an accuracy of 71.2% to predict LC with a sensitivity of 63% and a positive predictive value of 67.2%. Better accuracy was found for females than males (71.8 vs. 70.8) and for never-smokers than smokers (73.6% vs. 70.1%). Age was the most significant contributor (13.6%) followed by the red blood cell distribution (5.1%), creatinine (5%) gender (3.6%), and mean corpuscular hemoglobin (3.3%). Most of the blood tests had a chaotic contribution to the complex ML model however, some tests, such as red cell distribution width, mean corpuscular hemoglobin, prothrombin time, hematocrit, urea, and calcium had a slightly better dichotomic contribution. Conclusion: Blood tests can be used for the proposed ML model to modestly predict LC. More studies as are needed such as in the basic science fields to find possible explanations between the specific blood results and LC prediction.