Скачать или смотреть Intracranial Pressure Monitor Placement Prediction in Children with Traumatic Brain Injury

Description
Traumatic brain injury causes approximately 2,200 deaths and 35,000 hospitalizations in U.S. children annually. Clinicians currently make decisions about placing an intracranial pressure (ICP) monitor in children with traumatic brain injury without the benefit of an accurate clinical decision support tool. In a prospective observational cohort study, we developed and validated models that predict placement of an ICP monitor.

Patient data was gathered from multiple sources and discretized into 5-minute intervals. We divided data into four combinations of nurse documented and chart extracted input data, all including patient level and vital sign variables, and with inclusion or exclusion of data from brain computed tomography imaging reports and invasive blood pressure readings. Using R, we built machine learning models using logistic regression, support vector machines, generalized estimating equations, generalized additive models, and LSTMs. We trained each model with each combination of data. Optimal parameters were identified based on the highest F1.

The best performing model, an LSTM deep learning model, achieved an F1 of 0.71 within 720 minutes of hospital arrival. The best non-neural network model, standard logistic regression, achieved an F1 of 0.36 within 720 minutes of hospital arrival. While non-RNN models did not achieve the best F1, their coefficient size and direction provide insight into factors predicting ICP monitor placement. Additionally, the generalized additive models allow for visualization and interpretation of the marginal impact (after integrating out the impact of the other variables) of a variable over time.

Slides:
https://github.com/magic-lantern/2025...