Improving Providers’ Survival Estimates and Selection of Prognosis- and Guidelines-Appropriate Radiotherapy Regimens for Patients with Symptomatic Bone Metastases: Development and Evaluation of the BMETS Model and Decision Support Platform

In the management of symptomatic bone metastases, selection of appropriate palliative radiotherapy (RT) regimens should be based on patient-specific characteristics including estimated survival time. Yet, provider predictions of patient survival are notoriously inaccurate. Moreover, available evidence- and consensus-based guidelines do not provide clear criteria for selecting between the range of palliative RT regimens available. In an effort to improve selection of prognosis- and guidelines-appropriate palliative bone treatments, we developed the Bone Metastases Ensemble Trees for Survival (BMETS) model. Built using an institutional database of 397 patients seen in consultation for symptomatic bone metastases, this machine-learning model estimates survival time following RT consultation using 27 prognostic covariates. Cross validations procedures revealed excellent discrimination for survival, and the BMETS outperformed validated, simpler statistical models, justifying its use in this population. To better characterize a component of decisional uncertainty faced by providers, we next sought to identify the prevalence of “complicated” symptomatic bone metastases across a breadth of possible operational definitions. Our efforts identified up to 96 possible definitions of “complicated” bone metastases, present in up to 67.1% of patients in our database. Given that such “complicated” lesions may have been excluded from clinical trials in this setting, these data highlight the difficulty faced by providers when attempting to select appropriate RT regimens using inadequately defined selection criteria. Informed by these insights, we developed the BMETS Decision Support Platform (BMETS-DSP). This provider-facing, web-based tool was created to (1) collect relevant patient-specific data, (2) display an individualized predicted survival curve as per the BMETS model, and (3) provide case-specific, evidence-based recommendations for treatment of symptomatic bone metastases. We then conducted a pilot assessment of the clinical utility of the BMETS-DSP. In this preliminary assessment, the BMETS-DSP significantly improved physician accuracy in estimating survival and increased prognostic confidence, likelihood of sharing prognosis, and use of prognosis-appropriate RT regimens in the care of case patients. Collectively, this research provides early justification for the use of a machine-learning survival model and resultant decisions support platform to guide individualized selection of palliative RT regimens for symptomatic bone metastases. These data support a multi-institutional, randomized trial of the BMETS-DSP