This study is the first to examine the prognostic role of physician empathy in interaction with the type of consultation (TC) (TC, bad news versus follow-up consultations) in cancer patient survival. Between January 2015 and March 2016, 179 outpatients with thoracic cancer and a Karnofsky performance status ≥60 assessed their oncologist’s empathy using the CARE questionnaire, which provides a general score and two sub-dimensions: listening/compassion and active/positive empathy. Survival was recorded until April 2018. Usual medical, social and psychological confounders were included in the Cox regression. The median follow-up time was 3.1 years. There was a statistical interaction between listening/compassion empathy and TC (p = 0.016) such that in bad news consultations, higher listening/compassion predicted a higher risk of death (hazard ratio (HR) = 1.13; 95% confidence interval (CI): 1.03–1.23; p = 0.008). In follow-up consultations, listening/compassion did not predict survival (HR = 0.94; 95% CI: 0.85–1.05; p = 0.30). The same results were found with the general score of empathy, but not with active/positive empathy. In bad news consultations, high patient-perceived physician compassion could worry patients by conveying the idea that there is no longer any hope, which could hasten death. Further studies are warranted to confirm these results and find out the determinants of patient perception of physician empathy.