Predicting mortality among patients with liver cirrhosis in electronic health records with machine learning

• DOI: 10.1371/journal.pone.0256428
• Publication Date: 2021-08-31T17:59:33Z
• Authors: Aixia Guo; Nikhilesh R. Mazu...; Daniela P. Ladner; Randi E. Foraker
• Keywords: Categorical variable; Liver disease; Predictive modelling; Medical record

Liver cirrhosis is a leading cause of death and effects millions people in the United States. Early mortality prediction among patients with might give healthcare providers more opportunity to effectively treat condition. We hypothesized that laboratory test results other related diagnoses would be associated this population. Our another assumption was deep learning model could outperform current Model for End Stage disease (MELD) score predicting mortality.We utilized electronic health record data from 34,575 diagnosis large medical center study associations mortality. Three time-windows (365 days, 180 days 90 days) two cases different number variables (all 41 available 4 MELD-NA) were studied. Missing values imputed using multiple imputation continuous mode categorical variables. Deep machine algorithms, i.e., neural networks (DNN), random forest (RF) logistic regression (LR) employed between baseline features such as measurements each time window by 5-fold cross validation method. Metrics area under receiver operating curve (AUC), overall accuracy, sensitivity, specificity used evaluate models.Performance models comprising all outperformed those MELD-NA DNN LR RF models. For example, achieved an AUC 0.88, 0.86, 0.85 90, 180, 365-day respectively compared MELD score, which resulted corresponding AUCs 0.81, 0.79, 0.76 same instances. The had significantly better f1 at points examined.Other alkaline phosphatase, alanine aminotransferase, hemoglobin also top informative besides MELD-Na Machine standard risk cirrhosis. Advanced informatics techniques showed promise