A Clinical Prediction Model to Predict albumin Treatment Outcomes and Provide Dosage Recommendations: Development and Validation Study

Objective: Intravenous Albumin is widely used in the intensive care unit in patients with Systemic lupus erythematosus. However, The dose of albumin is still controversial.In this study, we intended to develop and validate a machine learning–based model to predict intravenous albumin treatment outcomes and to provide dosage recommendations to clinicians. Methods: : Using Mimic-III V1.4 and Mimic-IV V0.4, we collected data from of patients with SLE admitted to the ICU. Data were divided into two groups based on death or survival within 30 days.Several machine learning methods were used to build prediction models, evaluating model performance by the model area under the receiver operating characteristic curve(AUC).The best performing prediction model was selected. Applying the model to make predictions. The value of intravenous albumin infusion in the input variable was changed using the actual albumin used by the clinician as the baseline. Find the dose that minimizes the risk of death. Results: : Mimic-iv and Mimic-III was used to establish and validation the model. A total of 427 SLE patients, 109 of whom died. They were divided into two groups according to whether they died within 30 days. Random subspace method(RSM) integrated linear discriminant analysis model(LDA) were established. In the external validation AUC results of the models were 0.79[95% CI 0.78 0.82],the model performance was the best. Predictors were: Age at admission in years, Minimum WBC, Maximum WBC, Mechanical ventilation, Mean heart rate, Minimum systolic BP, Vasopressor, dialysis, Accompanied hypertension, nterstitial pneumonia associated with systemic lupus erythematosus, sofa, sirs, sapsii, apsiii, Intravenous administration of albumin, Immunoglobulin M, Minimum Albumin, Urine volume on the first day after admission to ICU, Infection, Complement C4.The model was used to calculate the recommended dose of albumin for intravenous infusion. In the MIMICIV database:366 patients in the survival group, 73 cases in the death group. The survival group improved 26 cases after applying our recommended dose, In the death group, 30 cases were improved. In the MIMICIII database:61 patients in the survival group,36 cases in the death group. The survival group improved 1 cases after applying our recommended dose, In the death group, 6 cases were improved. Conclusions: : RSM-LDA model could be used in the clinical scenario of albumin was infused intravenously treatment with a dosage pattern. The results provide support for recommending optimal albumin was infused intravenously treatment dosing to clinicians.