Comprehensive Summary
In this study, researchers developed a novel early anastomotic leakage (AL) risk predictive model using machine learning (ML) to integrate extensive clinical data. In particular, the researchers obtained data regarding radical gastrectomy and AL diagnoses from a population of 1588 patients. Additionally, demographic, tumor pathology, characteristics, laboratory indicators, comorbidities, and surgical parameters were assessed in this population. Using LASSO regression, researchers found 11 core predictors, including postoperative CRP, and 5 ML models were optimized using this external validation. The model demonstrated a high external validation performance (AUC = 0.871, NPV=96.9%). After optimizing sensitivity, the model had an even higher negative predictive value (NPV = 98.9%). Overall, the LASSO-Logistic model demonstrates accurate early risk indicators for AL.
Outcomes and Implications
Following radical gastrectomy for gastric cancer, anastomotic leakage (AL) occurrence remains high with mortality rate up to 50%. To address AL risk, this study created the LASSO-Logistic model, a machine-learning (ML) model that was found to be precise in determining AL risk early on given patient characteristics and indicators. This has several clinical implications. In the future, health-care professionals can use this model to assess the likelihood of a particular patient getting AL following surgery and use this value to determine if surgery is safe for them. Moreover, if health-care providers know a particular patient is high risk for AL, this can allow them to design a more comprehensive recovery plan. It can also help health-care workers stay vigilant in AL identification during post-operative appointments if they know a patient is more likely to get AL. Overall, this model can help clinicians provide personalized patient care, improving patient outcomes and health.