In recent study, VAE-Surv achieved a median C-index of 0.78, outperforming classical approaches
By Elana Gotkine HealthDay Reporter
WEDNESDAY, Feb. 26, 2025 (HealthDay News) — A multimodal computational framework for patient stratification and prognosis prediction enables automatic identification of patient clusters, outperforming classical approaches for patients with myelodysplastic syndromes (MDS), according to a study published online in the April issue of Computer Methods and Programs in Biomedicine.
Cesare Rollo, from the University of Torino in Italy, and colleagues introduced a multimodal computational framework for patients’ stratification and prognosis prediction (VAE-Surv). VAE-Surv integrates a variational autoencoder (VAE), which reduces the high-dimensional space that characterizes molecular data, with a deep survival model, combining embedded information with clinical features. The clinical robustness of the algorithm was tested by applying VAE-Surv to the Genomed4All cohort of MDS.
The researchers found that VAE-Surv achieved a median C-index of 0.78 when tested on 2,043 patients in the GenomMed4All cohort, outperforming classical approaches. Compared with a traditional approach that applies the clustering directly to the input data, the latent space enhanced the clustering performance. The analysis of the identified clusters showed that the proposed framework can capture existing clinical categorizations compared with the World Health Organization 2016 MDS subtypes while also suggesting novel, data-driven patient groups. VAE-Surv achieved a good prediction performance when tested in an external MDS cohort of 2,384 patients (median C-index, 0.74), preserving the interpretability of the main clinical and genetic features.
“VAE-Surv framework demonstrates the power of deep learning in handling the intricate genetic landscape of MDS, offering a novel, robust methodology for patient stratification and survival prediction,” the authors write. “It stands to contribute significantly to personalized medicine in the context of hematologic disorders.”
Copyright © 2025 HealthDay. All rights reserved.
