Models could predict meaningful individual-specific clinical outcome variables for survival, malignancy
THURSDAY, Feb. 13, 2020 (HealthDay News) — An open-source smartphone and web application has been developed to help predict clinical malignancy and survival outcomes for meningioma, according to a study published online Jan. 30 in npj Digital Medicine.
Jeremy T. Moreau, from McGill University in Montreal, and colleagues developed methods and a practical app to assist with the diagnosis and prognosis of meningiomas. Models were trained and validated on 62,844 patients from the Surveillance, Epidemiology, and End Results (SEER) database. Patterns of association between malignancy, survival, and a series of basic clinical variables were examined.
The researchers found that the models could predict meaningful individual-specific clinical outcome variables; across 16 SEER registries, they showed good generalizability. For the survival model, age at diagnosis affected survival and increased tumor size was associated with worse survival; malignant tumors predicted worse survival than borderline malignant tumors, which predicted worse survival than benign tumors. Specific surgeries predicted the greatest improved survival versus no surgery. The two most important features in the malignancy model were tumor size and age at diagnosis, and these were the only features retained in the final model. To demonstrate true clinical utility, the researchers noted future model improvements and prospective replication are necessary.
“We provide an original open-source smartphone and web application to illustrate the translation of complex nonlinear predictive models to real-world practice,” the authors write. “The ability of our statistical learning models and app to provide individual-specific predicted survival curves could be valuable for patient counseling.”
Copyright © 2020 HealthDay. All rights reserved.
