Prediction of Readmissions in Pituitary Adenomas Using Machine Learning Algorithm

Introduction: Pituitary adenomas are common benign intracranial tumors with higher chances of unplanned readmissions. There is a scarcity in the literature on identifying patients who may get readmitted after surgery. The aim of this study is to identify the predictors of readmission within the index admission using machine learning.

Methods: We queried the national readmission data (2018–2019) for the diagnosis of pituitary adenoma with the inclusion criteria: age >18 years, discharged alive between January and September, and underwent surgery in the index admission. Using chi-squaared feature selection algorithm, we selected 20 highest ranked features for classification ([Fig. 1]). We used K-nearest neighbor (KNN) and fine tree classifiers to test the prediction capabilities of the selected features.

Results: A total of 11,965 patients who underwent surgically excised pituitary adenoma surgery were included. Of these, 507 (4.1%) patients were readmitted within 90 days. All the features we described were of the index admissions. Readmitted patients have higher frailty and comorbidity scores. Readmitted patients have higher rates of discharge to long-term rehab (11.1 vs. 3%) centers and home hospice (11.4 vs. 7.2%). The mean length of stay was higher for the patients who were readmitted later (7.23 vs. 4.42 days). The complication rates were higher in patients who were readmitted later (31.3% vs. 18.2%). The coarse KNN classifier showed 96.1% validation accuracy with a total cost of 281 ([Fig. 2] confusion matrix). The area under the receiver operating characteristic (ROC) curve ([Fig. 3]), AUC is 0.63. Further feature extraction methods and additional classifiers (e.g., Support vector machines and neural network) will be used to improve the classification outcomes. All the factors were listed in the [Fig. 1]

Conclusion: Evaluation of predictors can help understand the probable causes of readmissions and prognostication in pituitary adenomas.

Publikationsverlauf

Artikel online veröffentlicht:
01. Februar 2023