P.091 Synthetic data reliably reproduces brain tumor primary research data

Abstract

Background: Synthetic data has garnered heightened attention in contemporary research due to confidentiality barriers and its capacity to simulate variables challenging to obtain. This study aimed to evaluate the reliability and validity of synthetic data in the context of neuro-oncology research, comparing findings from two published studies with results from synthetic datasets. Methods: Two published neuro-oncology studies focusing on prognostic factors such as serum albumin and systemic inflammation scores were selected, and their methodologies were replicated using MDClone Platform to generate five synthetic datasets for each. We used Chi-Square test to assess inter-variability between synthetic datasets. Survival outcomes were evaluated using Kaplan-Meier and t-test was used to determine statistical significance. Results: Findings from synthetic data consistently matched outcomes from both original articles, with serum albumin and systemc inflammation scores correlating with survival prognosis in glioblastoma and metastasis patients (p<0.05) Reported findings, demographic trends and survival outcomes showed significant similarity (P > 0.05) with synthetic datasets. Conclusions: Synthetic data consistently reproduced the statistical attributes of real patient data. Integrating synthetic data into clinical research offers excellent potential for providing accurate predictive insights without compromising patient privacy. In neuro-oncology, where patient follow-up pose challenges, the adoption of synthetic datasets can be transformative.