Project Abstract

Neurological recovery following cardiac arrest remains highly unpredictable, and early prognosis is crucial for guiding treatment decisions and managing family expectations. This research is motivated by the urgent clinical need to provide timely and accurate predictions on whether patients will regain consciousness, especially given the growing availability of large-scale ICU datasets and the advancement of machine learning methods. This study aims to develop a predictive model that estimates the likelihood of neurological recovery in comatose patients after cardiac arrest, using time-series physiological data. The research uniquely explores the application of ensemble learning, particularly XGBoost, to forecast patient outcomes at different time intervals post-admission. The study utilizes a dataset derived from the PhysioNet Challenge, consisting of 60 anonymised patient cases with multivariate time-series data. The approach involves preprocessing the data using existing scripts from the original challenge (e.g., helper_code, train_model) and training classification models on reduced hourly snapshots (e.g., 12, 24, 48, 72 hours) to evaluate predictive accuracy over time. Preliminary results suggest that model performance improves with longer time windows, with the 24th-hour model producing the most balanced validation and test scores in comparison to earlier or later windows. While findings are limited due to smaller dataset size, trends indicate that early prediction (within 24 hours) may still hold valuable prognostic potential. This work contributes to the growing field of predictive medicine by demonstrating that neurological outcomes can be forecasted with reasonable confidence using only early ICU data. It also highlights the feasibility of replicating and adapting large-scale research methods on smaller, more manageable datasets without significant loss in insight.

Keywords: Data Analysis, Neurological Recovery Prediction, Post-Cardiac Arrest

 Conference Details

Session: A

Location: Sir Stanley Clarke Auditorium at 11:00 13:00

Markers: Xianghua Xie, Anton Setzer

Course: BSc Computer Science 3yr FT

Future Plans: I’m looking for work