Analysis of prognostic factors affecting adult patients with anemia combined with sepsis and construction of a prediction model

Background The patients with anemia combined with sepsis is strongly associated with unfavorable prognoses. This study aimed to develop a prognostic model for the early prediction of patients with anemia combined with sepsis. Methods Data were obtained from the MIMIC-IV database. Prognostic influences were analyzed on the training set using the least absolute shrinkage and selection operator (LASSO) and multivariate logistic regression. The data were randomly divided into training and test groups, and the variables in the training set were screened using the LASSO method. Six machine learning methods were used to develop the models: XGBoost, logistic regression (LR), random forest (RF), support vector machine (SVM), k-nearest neighbor (KNN), and decision tree (DT). The best model was selected based on the area under the curve (AUC). Shapley's additive interpretation (SHAP) values were used to interpret the best model. Results Data from 1314 patients with anemia combined with sepsis were included. Age, history of malignancy, heart rate, systolic blood pressure (SBP), diastolic blood pressure (DBP), arterial blood gas partial pressure of oxygen (PO2), blood arterial blood gas lactate, arterial blood gas partial pressure of carbon dioxide (PCO2), respiratory rate, arterial blood gas oxygen saturation (SPO2), reticulocyte production index (RPI), urine output, and Glasgow Coma Score (GCS) were prognostic predictors; the best model XGBoost model was screened to predict the results ( AUC of 0.735 and sensitivity of 0.643). Urine output had the strongest predictive value, followed by age, SpO2, blood lactate, SBP, history of malignancy, and RPI.RPI had a positive effect, i.e., it drove the prediction of an acute patient death. In contrast, an increase in urine output had a negative effect, i.e., drove the prediction of survival. A clinically characteristic-based model was developed and validated, demonstrating good performance. Conclusion A machine learning model based on clinical features was developed and validated effectively for predicting the prognosis of patients with anemia combined with sepsis. The SHAP approach enhances the interpretability of the machine learning model, allowing clinicians to better understand the reasons behind the results.