AAI_2025_Capstone_Chronicles_Combined

4

with older ICU patients generally exhibiting poorer outcomes. Dynamic measurements, including

Glasgow Coma Scale (GCS) scores, capture neurological status, where persistently low scores are

associated with high mortality in critically ill patients (Teasdale et al., 2014). Laboratory measures such

as serum lactate are strong indicators of tissue hypoperfusion and have been linked to higher mortality

rates in sepsis and shock patients (Jansen et al., 2009). Conversely, variables with high missingness but

limited clinical interpretability may contribute little to predictive performance.

Correlation analysis, using Pearson’s correlation coefficient (r; Rodgers & Nicewander, 1988),

revealed strong associations among hemodynamic measures. For example, mean arterial pressure (MAP)

correlated strongly with both systolic and diastolic arterial blood pressure (see figure 2), consistent with

its derivation from these measures. These correlations are physiologically consistent but indicate potential

multicollinearity, which could influence certain model types and may require dimensionality reduction or

feature selection. Moderate correlations were found between some metabolic and respiratory parameters,

while many clinically important features such as GCS, temperature, and urine output exhibited weak

correlations with other variables. This indicates that they provide unique, complementary information, an

important characteristic for improving predictive coverage by capturing aspects of patient condition not

explained by other measurements (Knaus et al., 1985).

Background Information

The prediction of in-hospital mortality in the Intensive Care Unit (ICU) is a critical task that can

aid in clinical decision-making, resource allocation, and patient risk stratification (Yeh et al., 2024). Our

project addresses this challenge using the PhysioNet 2012 Challenge (Silva et al., 2012), which contains a

rich collection of demographic data, clinical scores, and time-series measurements for 4,000 ICU patient

150