The following is a summary of “Novel pneumonia score based on a machine learning model for predicting mortality in pneumonia patients on admission to the intensive care unit,” published in the July 2023 issue of Pulmonology by Wang, et al.
There was a lack of scores for predicting long-term mortality in severe pneumonia cases. For a study, researchers sought to develop new pneumonia scores using machine learning methods to predict both 1-year mortality and hospital mortality of pneumonia patients upon admission to the intensive care unit (ICU).
The study population was selected from the MIMIC-IV and eICU databases. The primary outcomes evaluated were 1-year mortality and hospital mortality in the MIMIC-IV and eICU databases. Patients diagnosed with community-acquired pneumonia (CAP) and ventilator-associated pneumonia (VAP) were analyzed as separate subgroups. Common shallow machine learning algorithms, including logistic regression, decision tree, random forest, multilayer perceptron, and XGBoost, were utilized.
The MIMIC-IV database comprised 4,697 patients, while the eICU database contained 13,760 patients. A new pneumonia score, “Integrated CCI-APS,” was defined using a multivariate logistic regression model incorporating six variables: metastatic solid tumor, Charlson Comorbidity Index, readmission, congestive heart failure, age, and Acute Physiology Score III. The area under the curve (AUC) and accuracy of the integrated CCI-APS were assessed in three datasets (full, CAP, and VAP) using test sets from the MIMIC-IV database and an external validation set from the eICU database. The AUC values for predicting 1-year and hospital mortality ranged from 0.784 to 0.797 and 0.691 to 0.780, respectively, with corresponding accuracy ranges of 0.723 to 0.725 and 0.641 to 0.718.
The study set a benchmark for using machine learning models to construct pneumonia scores. The proposed integrated CCI-APS score for severe pneumonia outperformed existing scores in predicting 1-year mortality and hospital mortality.
Source: resmedjournal.com/article/S0954-6111(23)00251-2/fulltext