Predicting Postoperative Delirium Risk for Intracranial Surgery: A Statistical Machine Learning Approach

Juliet Aygun, Alaina Bartfeld, Sahana Rayan
2020 Journal of Purdue Undergraduate Research  
Delirium has a high morbidity rate and is common; around 10% of older, hospitalized patients have delirium, and fi fteen to fi fty percent of patients experience delirium during hospitalization. This puts delirium at the forefront of problems for which doctors and nurses look. Most journal articles about delirium focus on postoperative delirium (POD) in the Intensive Care Unit (ICU); however, none are specifi c to post-intracranial surgery. In this way, this research is distinct from others in
more » ... nct from others in this area. The purposes of this research project are to employ machine learning methods, which accurately predict whether a post-intracranial surgery patient will be diagnosed with POD in the ICU and identify the key predictors of POD. If POD could be predicted, many patients would experience a shorter hospital stay, less marginal complications, and a greater life expectancy. With our model, the onset of POD could ultimately be stopped. We fi rst conducted dimensional reduction on our dataset by employing factor analysis and elastic net classifi cation to prevent overfi tting of the model. After, we trained a neural networking model to predict POD. This model was 85% accurate, and we found the key predictors of POD are whether the patient had delirium when they were admitted into the ICU, the type of lesion they had (if any), and if their blood was carrying enough oxygen. This was supported by chi-square analysis, which proved that the predictors are statistically signifi cant. This image illustrates how our neural networking model functions to predict POD.
doi:10.7771/2158-4052.1466 fatcat:vcabe3eai5do3ewagplk5tpn74