Application of Machine Learning Models in Forecasting Prolonged Length of Stay After Cancer Surgery Using Electronic Health Records

Postoperative length of stay serves as a crucial metric for managing medical resources and offers an indi-rect measure of surgical complication rates following cancer surgery. In recent years, machine learning techniques have been increasingly utilized for complex medical outcome prediction by leveraging large volumes of clinical data. In this study, we employed machine learning models to forecast prolonged length of stay after cancer surgery using de-identified clinical data from The Cancer Genome Atlas (TCGA), a publicly accessible database. We analyzed records from approximately ۳,۵۰۰ patients who underwent primary surgeries for five types of cancer-stomach, breast, colon, thyroid, and lung. The available clinical variables included surgical details, cancer characteristics, comorbidities, and laboratory assessments. Support vector machine, random forest, and logistic regression models were developed to predict pro-longed postoperative length of stay. The results show that machine learning models can predict prolonged hospital stay for breast and stomach cancers (AUC >۰.۸۵) but perform poorly for colon and lung cancers (AUC <۰.۸), highlighting limitations in generalizability. These findings highlight both the potential and the limitations of using public genomic and clinical datasets for perioperative risk stratification and resource allocation in cancer care.

Keywords:

Machine Learning , Length of stay , Electronic Health Records , cancer surgery , Predictive Modeling