Objective In oncology, research remains unclear as to whether physician empathy is associated with patient outcomes. Our goal was to answer this question and explore potential moderators of the association. Methods In this meta-analysis on adult cancer care, we excluded randomised controlled trials, and studies of survivors without active disease or involving analogue patients. Eight databases were searched, in addition to reference lists of relevant articles and grey literature. Two reviewers independently screened citations, extracted data, assessed risk of bias and graded quality of evidence by using the AXIS tool. Effect size correlations (ESr) were chosen and pooled by using a random effect model. Subgroup analyses were performed, and statistically significant variables were introduced in a meta-regression. Several methods were used to explore heterogeneity and publication biases. Results We included 55 articles, yielding 55 ESr (n = 12976 patients). Physician empathy was associated with favourable patient outcomes: ESr 0.23, 95% confidence interval (CI) [0.18 to 0.27], z = 9.58, p < 0.001. However, heterogeneity was high, as reflected by a large prediction interval, 95% CI [-0.07 to 0.49] and I2 = 94.5%. The meta-regression explained 53% of variance. Prospective designs and physician empathy assessed by researchers, compared with patient-reported empathy, decreased ESr. Bad-news consultations, compared with all other types of clinical encounters, tended to increase ESr. Conclusion Patient-reported physician empathy is significantly associated with cancer patient outcomes. However, the high heterogeneity warrants further longitudinal studies to disentangle the conditions under which physician empathy can help patients. Recommendations are proposed for future research.