Disease Free Months, Karnofsky Performance Score, and Sex Help Predict Glioblastoma Patient Survival

Katie Harbaugh

Overview

Glioblastoma is one of the most lethal cancers, associated with a bleak prognosis and no cure. Survival estimates given to patients are often inaccurate, as actual survival can range from months to years, but estimates given at diagnosis are often presented as a generalization. We hypothesized that by determining the connection between specific patient data and individual survival time, we could begin developing a better prediction process. The goal of this project was to determine the most influential attributes in patient survival and use them to create and test a process to output a more specific and accurate survival curve. The data which produced the final concordance score and attribute analysis came from the cBioPortal for Cancer Genomics 2018 sample titled Glioblastoma Multiforme (TCGA, PanCancer Atlas). The original sample size was 592, which decreased to a total of 288 after the data was cleaned and split into a train and test set. For the patients who had such data, overall survival time had a large range, so Python was used to draw correlations that could not be determined visually. For example, patient data corresponding to “Disease Free (Months)” (p<0.005) was concluded to be significant in predicting accurate survival curves, while “Mutation Count” (p=0.96) had little effect on curve accuracy or patient survival. Through utilization of the Cox Model, survival curves were produced based on additional attributes such as sex, “Diagnosis Age,” and “Karnofsky Performance Score.” To numerically represent curve accuracy, a concordance score of 0.76 was drawn, meaning the curves were accurate to actual patient survival 76% of the time. Here, a process that predicts survival at different intervals, considering multiple patient attributes was developed. Additional data and time could increase the accuracy of predicted results and help lead to usage in professional diagnosis settings, or to produce specific curves for other cancers.

My Story

I have always been passionate about researching Glioblastoma Multiforme (GBM), and this is something I was able to do through the Summer Stem Institute research program of 2020. This is a type of cancer that has affected my family directly, so I turned to research to hopefully give myself more insight about GMB and how it affects patients on an individual level. This inspiration provided the foundations for my research project. When I did this project, most of the world was closed, so the project itself was completed 100% virtually through the use of Python programming. My mentor and I found available datasets to experiment with and ultimately found some significant correlations between patient attributes and survival time. Then, we used existing programs to display individualized survival curves based on those patient attributes.

From the student

In this study, Python programming was used to determine the correlations between specific attributes of Glioblastoma patients, and overall survival time. Attributes such as Disease Free Months and Karnofsky Performance Score were highly correlated variables in patient survival. We then used these findings to produce individual patient survival curves based on the attribute values for each patient.