Unsupervised machine learning models reveal predictive clinical markers of glioblastoma patient survival using white blood cell counts prior to initiating chemoradiation

Wesley Wang ¹, Zeynep Temerit Kumm ¹, Cindy Ho ¹

¹Department of Pathology, Ohio State University Wexner Medical Center, Columbus, Ohio, USA.
²Molecular Oncology Research Center, Barretos Cancer Hospital, Barretos, Brazil.
³Department of Pathology, Barretos Cancer Hospital, Barretos, Brazil.
⁴Life and Health Sciences Research Institute (ICVS)/School of Medicine, University of Minho, Braga, Portugal.
⁵ICVS/3B's-PT Government Associate Laboratory, Braga-Guimarães, Portugal.
⁶Departamento de Neuropatología y Biología Molecular, Instituto de Investigaciones Neurológicas Dr Raúl Carrea (FLENI), Buenos Aires, Argentina.
⁷Department of Pathology, University of Mississippi Medical Center, Jackson, Mississippi, USA.
⁸Department of Neuro-Oncology, University of Mississippi Medical Center, Jackson, Mississippi, USA.
⁹Department of Neurosurgery, University of Mississippi Medical Center, Jackson, Mississippi, USA.
¹⁰Department of Radiation Oncology, Ohio State University Wexner Medical Center, Columbus, Ohio, USA.
¹¹Department of Neurosurgery, Ohio State University Wexner Medical Center, Columbus, Ohio, USA.
¹²Department of Neurology, Ohio State University Wexner Medical Center, Columbus, Ohio, USA.

⁰Department of Pathology, Ohio State University Wexner Medical Center, Columbus, Ohio, USA.

Neuro-Oncology Advances +

|

February 26, 2024

View abstract on PubMed

Summary

Serum white blood cell (WBC) count and PD-L1 expression are simple, low-cost biomarkers that predict glioblastoma survival. Machine learning identified these novel predictors, aiding equitable patient care and clinical decision-making.

Area Of Science

Neuro-oncology
Biomarker Discovery
Machine Learning in Medicine

Background

Glioblastoma (GBM) management requires ongoing monitoring.
Personalized medicine relies on molecular biomarkers for prognosis and treatment decisions.
Limited accessibility of molecular testing necessitates low-cost predictive biomarkers for equitable GBM care.

Purpose Of The Study

To identify accessible, low-cost predictive biomarkers for glioblastoma patient survival.
To explore the utility of machine learning in uncovering novel clinical relationships in GBM data.

Main Methods

Retrospective analysis of 581 glioblastoma patient records from multiple institutions.
Application of unsupervised machine learning, including dimensionality reduction and eigenvector analysis.
Objective quantification of PD-L1 expression using immunohistochemistry algorithms.

Main Results

Baseline serum white blood cell (WBC) count significantly predicted overall survival in glioblastoma patients.
Patients in the upper quartile of WBC count had over a 6-month median survival advantage compared to the lower quartile.
Elevated PD-L1 expression was observed in glioblastoma patients with high serum WBC counts.

Conclusions

Serum WBC count and PD-L1 expression serve as simple, effective biomarkers for predicting glioblastoma survival.
These biomarkers can aid in clinical decision-making and improve equitable patient care.
Machine learning effectively distills complex clinical data to reveal meaningful prognostic factors.

Keywords:

clinical decision-making glioblastoma machine learning survival outcomes