Development and validation of three risk prediction models for clinical deterioration of patients after craniotomy: a retrospective cohort study in China.

Abstract

This study aimed to identify a deterioration prediction tool for patients after craniotomy. A retrospective cohort study. Three large tertiary hospitals in Hunan Province, China. Between January 2018 and March 2020, 1576 patients who underwent craniotomy at three tertiary hospitals in Hunan Province were selected and randomly allocated to either the training or validation sets at a 7:3 ratio. Comprehensive demographic and disease-specific data were collected. Logistic regression, Bayes classification and back propagation neural network were used to construct the models. The Technique for Order Preference by Similarity to an Ideal Solution method was used to evaluate the efficiency of the models. The performance of all three models was commendable. In both the training and validation datasets, the back propagation neural network model demonstrated the highest efficiency, achieving a sensitivity of 77.1%, specificity of 91.7%, correctness of 86.8%, positive predictive value of 82.3% and negative predictive value of 88.9%. Conversely, the Bayes classifier exhibited the lowest predictive efficiency among the models evaluated. We developed three models to predict clinical deterioration in patients following craniotomy. Among these, the back propagation neural network model demonstrated superior predictive performance. This model serves as a valuable reference for clinical nurses, aiding them in identifying high-risk patients who may deteriorate post-craniotomy.