Across nearly 187,000 lung cancer cases spanning more than a decade, researchers have trained machine learning models to read the hidden logic of how American oncologists treat non-small cell lung cancer — and to ask whether patients who received the expected treatment fared better than those who did not. The findings confirm that patterns exist and can be mapped with impressive accuracy, yet they also surface a deeper caution: knowing what happened is not the same as knowing what should happen. The study stands as both a demonstration of machine learning's descriptive power in medicine and a