Patient classification has widespread biomedical and clinical applications, including diagnosis, prognosis, disease subtyping and treatment response prediction. A general purpose and clinically relevant prediction algorithm should be accurate, generalizable, be able to integrate diverse data types (e.g. clinical, genomic, metabolomic, imaging), handle sparse data and be intuitive to interpret. We describe netDx, a supervised patient classification framework based on patient similarity networks that meets the above criteria. netDx models input data as patient networks and uses the GeneMANIA machine learning algorithm for network integration and feature selection. We demonstrate the utility of netDx by integrating gene expression and copy number variants to classify breast cancer tumours as being of the Luminal A class (N=348 tumours). In a simplified comparison using gene expression, netDx performed as well as or better than established state of the art machine learning methods, achieving a mean accuracy of 89% (2% s.d.) in classifying Luminal A. netDx uses pathway features to aid biological interpretability and results can be visualized as an integrated patient similarity network to aid clinical interpretation. Upon publication, netDx will be made publicly available via github.