Journal of Clinical Hepatology
Albumin is a protein produced by the liver that helps maintain fluid balance and transport substances in the blood. Low albumin is common after partial liver removal and can contribute to fluid accumulation, infection and delayed recovery.
Researchers retrospectively analysed 700 patients who underwent partial hepatectomy. Of these, 283, or 40.42%, developed postoperative hypoalbuminaemia. The data were divided into training and testing groups.
Seven machine-learning methods were evaluated. Eight predictors were selected: age, hepatitis B, fatty liver disease, vascular clamping time, preoperative albumin, operation duration, blood loss and preoperative aspartate aminotransferase.
A K-nearest neighbours model produced the best overall performance. In the testing data, it achieved an area under the receiver operating characteristic curve of 0.835, with 84.0% sensitivity and 65.5% specificity.
The model performed better than two conventional liver-assessment tools. The albumin-bilirubin score had an area under the curve of 0.652, while the Model for End-Stage Liver Disease score achieved 0.524.
Preoperative albumin was the most influential predictor, followed by hepatitis B status, operation duration and age. The analysis also suggested interactions between hepatitis B, age and operation duration.
Identifying high-risk patients before surgery could support earlier nutritional assessment, closer albumin monitoring and more individualised perioperative management.
The model uses clinical information already available before or during surgery and captures interactions that simpler scoring systems may miss. Its high sensitivity may be useful when the priority is avoiding missed high-risk cases.
However, the study was retrospective and conducted at one hospital. The model was only tested on an internal data split and had moderate specificity, meaning some patients would be incorrectly classified as high risk. External validation and prospective assessment are needed before routine use.
Objective: To investigate the application value of a machine learning model based on preoperative clinical indicators in predicting the…
