From survival prediction to treatment decision in lung cancer

Abstract

More than 14000 people are diagnosed with lung cancer in the Netherlands every year, and for most this means that they will die from lung cancer within a few years. Nevertheless, there are important differences between patients with lung cancer, both in treatment options and in survival. This thesis focuses on two questions that are crucial after the diagnosis of lung cancer is made: what is the prognosis of this patient and what is the best treatment?

The first part of the thesis presents two studies on predicting survival for an individual lung cancer patient. First, we examine characteristics of the lung tumor on medical imaging. Second, we investigate characteristics of the patient that are visible on medical imaging, such as the amount of muscle tissue present, and how this relates to survival.

The second part of this thesis is devoted to estimating the individual treatment effect. Depending on the disease stage, large groups of lung cancer patients are prescribed the same treatments. However, because no patient and no tumor are alike, a particular treatment will work well for one patient and bring about long-term healing, while for another patient it will only have side effects. By analyzing experiences of historical patients with different treatments with advanced statistical methods and artificial intelligence, it is possible to provide more individualized treatment advice for future patients. Of crucial importance here is to disentangle the causal effect of the treatment from other differences between patients who have had different treatments.