Rationale:
Triphasic waves (TW) are known markers of encephalopathy on EEG, and are associated with a variety of metabolic, infectious, structural and drug related etiologies. We tested these associations using a combined dataset of EEG reports, laboratory values, radiology reports and medication records.
Methods:
Our dataset included the text of EEGs reports performed at the University of Chicago between September 2012 and April 2023 (n = 17,718) and associated demographic, laboratory values, medication order data, and the text of radiology reports. From these data we generated multiple predictors: age and biological sex of patients; CBC and BMP values (if obtained with 2 days of EEG); cefepime exposure, which was considered present if patients were exposed to at least 3 days of cefepime prior to EEG; and one binary variable indicating the presence of microvascular ischemic disease (MID) on brain imaging reports, and another scalar variable reflecting the severity of MID where 1 is mild, 2 is moderate and 3 is severe. These predictors were used in a multivariate logistic regression with the presence of TWs on EEG reports as the dependent variable. Significant predictors from the regression were then used in a LASSO regression with 10-fold cross validation to generate a formula to predict odds of TW on EEG reports.
Results:
10 variables - age, sex, white blood cell count, hemoglobin, estimated GFR, sodium, phosphorus, creatinine, cefepime exposure, and the score reflecting severity of MID - were significantly predictive (p < 0.05) of the presence of TWs in the multivariate logistic regression analysis. These variables were then included in a LASSO regression, and the resulting coefficients from the lambda value with the lowest deviance were used to generate a formula for the risk of TWs on EEG. The AUC for the ROC curve was 0.74, indicating fair predictive ability.