An unsupervised learning approach to optimising tumour therapy through clinical data mining

Abstract

Clinical data mining and healthcare analytics enable systematic evaluation of treatment strategies in precision oncology. This study analysed a harmonised multi-centre dataset of 150 patients treated with linac-based stereotactic radiosurgery for vestibular schwannoma across three institutions in Ireland and the UK. A data-driven framework combining descriptive analytics, unsupervised clustering (K-means and Gaussian Mixture Models), and Random Forest modelling was used to assess treatment plan consistency, explore dose–response patterns, and estimate clinical outcomes. Clustering identified four treatment plan groups with distinct profiles of tumour reduction and organ-at-risk exposure. Random Forest models linked these clusters and dosimetric factors with tumour control and functional preservation. While internal performance was high, results are interpreted cautiously due to the limited sample size and absence of external validation. By integrating unsupervised learning with interpretable predictive modelling, this study provides a reproducible approach to characterising dose–response heterogeneity across centres. The findings support the future development of decision-support tools, while recognising that prescriptive optimisation requires further causal or optimisation-based modelling beyond the present work.