Comprehensive systematic review and meta-analysis: Evaluating artificial intelligence (AI) effectiveness and integration obstacles within anesthesiology

Abstract

Background: Artificial intelligence (AI) is a multidisciplinary field that focuses on developing intelligent computer algorithms to carry out simple to complex tasks traditionally performed using human intelligence. In anesthesia, AI is rapidly becoming a transformative technology. However, its efficacy remains unknown. Therefore, this study aims to analyze the efficacy of AI in anesthesia by studying two main applications of AI: predicting anesthesia-related events and assisting with anesthesia-related procedures. Methods: A systematic search was performed for English-language records published from inception until June 2024 in PubMed, Google Scholar, IEEE Xplore, and Web of Science databases. Studies were included in this meta-analysis if they examined the role of any AI model in predicting hypotension, hypoxemia, and post-operative nausea and vomiting (PONV). Moreover, studies investigating the role of AI in guiding anesthesia-related procedures, such as tracheal intubation and ultrasound-guided nerve blocks, were included. The CMA software and STATA 16.0 were used for statistical analyses, while the Newcastle–Ottawa Scale was used for quality evaluation. Results: Twenty studies meeting the eligibility criteria were included in the analysis. The pooled results indicated that AI demonstrated good discrimination ability in predicting hypotension (area under the receiver operating characteristic curve (AUROC): 0.81), with the subgroup analysis showing that models incorporating machine-learning algorithms outperformed other models (AUROC: 0.93). Similarly, the pooled analysis showed that AI models had a good discriminatory capacity for predicting hypoxemia (AUROC: 0.81). However, AI demonstrated poor discriminatory capability in predicting PONV (AUROC: 0.68). Our analysis also showed that robotically assisted intubations were successful in both mannikins and humans (success rate: 98% and 92%). Similarly, robotically assisted ultrasound-guided blocks were successful in mannikins and humans (success rate: 96% for humans and mannikins). Conclusion: This study suggests that AI is useful for predicting anesthesia-related events and automating procedures such as intubation and ultrasound-guided nerve blocks. However, multiple barriers hindering the integration of AI into anesthesia, such as cost, privacy and security concerns, data quality, “black box”, and ethical issues need to be addressed.

Ähnliche Arbeiten