Lung cancer patients face a hard choice before treatment. Some respond well to immunotherapy, while others do not. Doctors need a way to know who will benefit before starting these powerful drugs. A recent look at 44 different computer models tried to solve this problem. These programs use machine learning to predict how a patient will react to neoadjuvant immunotherapy given before surgery.
The models looked at scans and other data to guess the outcome. On average, the programs correctly identified the right patients about 79 percent of the time. They were very good at spotting those who would not benefit, correctly identifying them 91 percent of the time. Different types of models performed differently, with some doing slightly better than others.
However, the study has serious problems. The quality of the research was low, and there was a high risk of bias. The sample size was too small, and data handling was improper. Validation procedures were defective, and reports on radiomics had multiple issues. Because of these flaws, we cannot yet trust these tools for real patients. More work is needed before doctors can rely on these computer predictions.