Machine Learning Applied to EHRs Data of Patients With Sarcoma

Overview

Application of computational statistics and machine learning methods to data derived from electronic health records of patients diagnosed with sarcoma.

This observational, retrospective, multicenter study will be conducted on a group of patients treated at the Rizzoli Orthopedic Institute in Bologna and followed throughout their treatment. The study population includes patients of both sexes and all ages, affected by the two types of bone sarcoma typical of young people, with histologically confirmed diagnoses. The musculoskeletal tumors referred to in the study are osteosarcoma (OS) and Ewing's sarcoma (ES). Both are rare and very aggressive tumors, with a prognosis that remains unsatisfactory. These characteristics limit the possibility of conducting ad hoc studies on large case series that would allow the characterization of patients affected by these conditions in order to identify prognostic predictors. The clinical registries of specialized centers such as the Rizzoli Orthopedic Institute (IOR), which has always been a reference point for the diagnosis and treatment of sarcomas, are a source of very relevant data in this regard, allowing the collection of observational data gathered prospectively over time. The aim of this retrospective observational study is to characterize clusters of patients with different prognostic profiles and, secondarily, to identify the most predictive characteristics with respect to the prognosis of patients, applying computational intelligence algorithms using the open-source programming language R to already available data. At the Simple Departmental Structure (SSD) of Anatomy and Pathological Histology of the Rizzoli Orthopaedic Institute (IOR), two datasets containing these variables are available and ready for use: * patients diagnosed with osteosarcoma at the IOR between January 1, 2003, and December 31, 2012. * patients diagnosed with Ewing's sarcoma at the IOR from 01/01/2003 to 31/12/2012. Following ethical approval, access to these data will be requested, to be subsequently analyzed with computational intelligence algorithms (e.g., Random Forests) to determine the characteristics most predictive of prognosis (using a technique called "recursive feature elimination").

Conditions

Interventions

Eligibility

Outcomes