Back

Machine Learning Methods to Predict Survival in Patients Following Traumatic Aortic Injury

Shiban, N.; Gaul, J.; Zhan, H.; Elhabr, A.; Kokabi, N.; Johnson, J.-O.; Hanna, T.; Schrager, J.; Gichoya, J.; Banerjee, I.; Trivedi, H.

2021-07-31 health informatics
10.1101/2021.07.28.21261166 medRxiv
Show abstract

The National Trauma Data Bank (NTDB) is a resource of diagnostic, treatment, and outcomes information in trauma patients. We leverage the NTDB and machine learning techniques to predict survival following traumatic aortic injury. We create two predictive models using the NTDB - 1) using all data and, 2) using only data available in the first hour after arrival (prospective data). Seven discriminative model types were tested before and after feature engineering to reduce dimensionality. The top performing model was XGBoost, achieving an overall accuracy of 0.893 using all data and 0.855 using prospective data. Feature engineering improved performance of all models. Glasgow Coma Scale score was the most important factor for survival, and thoracic endovascular aortic repair was more common in patients that survived. Smoking, pneumonia, and urinary tract infection predicted poor survival. We also note concerning disparities in outcomes for black and uninsured patients that may reflect differences in care.

Matching journals

The top 4 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.