Predicting Delirium in the ICU

**Data Compilation** The first part of this experiment consisted of the compilation of data using SQL queries of a publicly available database. The data was then merged using R software and Excel. This large data table was the basis for the majority of the experiment. **Data Balancing** Due to the overwhelming majority of negative labels for delirium (lack of delirium diagnosis) much of the data had to be balanced by deleting thousands of negative label entries. **Initial Stages** The experiment began with a single branch, selecting several columns from the 65 available ones. Multiple algorithms were tested and the two-class decision jungle appeared to be the most effective and accurate. **Branching Out** Using this algorithm as a basis, multiple branches were created within the experiment in order to test which factors were most significant. We used prior research studies as a baseline from which we determined which factors would be best to investigate. Some of the factors that we are investigating include: 1. Oxygen Saturation 2. Blood Glucose Levels 3. Psychoses 4. Depression 5. Alcohol Abuse 6. Drug abuse **Data Analysis** The results are still being compiled and analyzed. We hope to use the results from this retrospective analysis in a future study of hospital data in order to predict the development of delirium earlier in the ICU setting.