The operating room of the future may look very different from today's. Artificial intelligence could flag a patient’s risk of bleeding before the first incision is made. Big data platforms might continuously integrate vital signs, lab results, and imaging studies in real time, feeding actionable insights to the anesthesia team. And large language models could answer a patient's last‑minute questions, generate tailored educational materials, and even help surgeons weigh evidence‑based options during a complex procedure.
This is not science fiction. According to a new perspective article published in Perioperative Precision Medicine (DOI: 10.61189/319254ilybgm), these technologies are rapidly maturing and beginning to show real clinical promise. But the authors, led by first author Di Wang and corresponding author Shafinah Ahmad Suhaimi of Universiti Sains Malaysia, caution that the path from research to routine practice is strewn with challenges that must be addressed head‑on.
A Unified Framework, Not Isolated Tools
Rather than treating AI, big data, and large language models as separate innovations, the paper presents them as three pillars of an integrated, closed‑loop decision‑support ecosystem that spans the entire perioperative journey – from preoperative assessment to intraoperative management and postoperative monitoring.
AI‑driven models can support precise perioperative risk stratification and complication prediction. Big data technologies facilitate multimodal data integration and real‑time monitoring, while large language models can leverage their natural language processing capabilities to improve clinical communication, support documentation, and assist with knowledge‑based decision‑making.
Taken together, they argue, these technologies could enhance surgical safety, boost clinical efficiency, and deliver truly personalized care.
What AI Already Brings to the Table
Machine learning and deep learning models are already outperforming traditional statistical methods in several key areas. In preoperative risk assessment, they digest diverse data – medical history, imaging, laboratory results, and vital signs – to more accurately predict intraoperative complications, postoperative infections, and organ dysfunction. Natural language processing can further automate the extraction of electronic health record information, reducing human error and freeing up clinician time.
During surgery, intelligent anesthesia systems use machine learning algorithms to analyze physiological data, optimizing the depth of anesthesia, analgesic intensity, and hemodynamic stability. Studies have shown that AI‑assisted systems can shorten stabilization time, reduce fluctuations in the bispectral index, and lower propofol consumption in complex procedures.
Postoperatively, AI models that integrate electronic health records with continuous physiological monitoring have demonstrated the ability to detect complications early – including infection, bleeding, and respiratory failure. When combined with explainability techniques such as Shapley additive explanations, these models not only predict adverse events but also help clinicians understand why a particular risk flag was raised.
Big Data as the Backbone
Underpinning these applications is a robust data infrastructure. The authors advocate for standardized, interoperable platforms that can aggregate heterogeneous clinical information from multiple sources – electronic health records, physiological monitors, imaging systems, laboratory databases, and surgical logs.
Such platforms enable real‑time early warning systems. For example, ankle pump motion monitoring devices provide real-time feedback to support venous thromboembolism prevention, while wearable devices have been shown to improve postoperative activity levels and reduce dyspnea in patients with lung cancer.
Big data also enables personalized medicine by integrating genomic, imaging, and clinical information. Platforms like the Intelligent Perioperative System use high‑throughput data processing to predict complications and generate decision support, while advanced cerebral oximetry tools offer real‑time guidance on brain perfusion and oxygenation during high‑risk procedures.
Large Language Models: The Human Interface
Perhaps the most eye‑catching innovation is the role of large language models. These tools can generate personalized educational materials based on patient characteristics, helping individuals better understand their surgical procedures, risks, and recovery plans. Their natural language capabilities support multilingual communication and enhance patient‑clinician engagement.
Studies have shown, for instance, that ChatGPT‑assisted informed consent can reduce preoperative anxiety and increase satisfaction among total knee arthroplasty patients. But the authors are careful not to overstate the case: model accuracy and language complexity vary, and patients with low health literacy may struggle to understand the output.
In clinical workflows, large language models can also assist with documentation and evidence retrieval. More broadly, natural language processing-based deep learning models such as BioBERT have shown strong performance in abstract screening, with the potential to reduce the burden on human reviewers.
The Obstacles That Can't Be Ignored
For all the promise, the authors identify critical barriers that could slow or derail clinical adoption.
Data heterogeneity tops the list. Perioperative data come from diverse sources – medical histories, imaging, laboratory data, and real‑time physiological monitoring – and different institutions use varying formats, collection standards, and quality controls. This heterogeneity limits model generalizability across settings.
Equally concerning is the "black box" problem. Many AI systems offer little explanation for their predictions, undermining clinician trust – especially in high‑stakes decisions where lives hang in the balance.
The integration of multimodal data also raises serious privacy and ethical concerns, including the risk of data misuse, algorithmic bias, and unclear liability when things go wrong. Meanwhile, large language models continue to struggle with hallucinations, factual errors, and inadequate adaptation to medical contexts.
Ensuring safety, transparency, and compliance therefore remains a core challenge for the clinical adoption of these technologies.
The Road Ahead
To move from theory to practice, the paper outlines a clear agenda. Large‑scale, multicenter, prospective studies are needed to validate these tools across different surgical types and patient populations. Developers must prioritize explainable models that provide transparent and traceable explanations for their recommendations.
Privacy‑preserving technologies such as federated learning, differential privacy, and secure multi‑party computation should be integrated into perioperative data platforms. And ultimately, the authors envision a closed‑loop intelligent system that continuously learns from preoperative, intraoperative, and postoperative outcomes, adapting to individual patients over time.
But perhaps the most important message is this: the next generation of perioperative decision support cannot be built by engineers alone. It requires close collaboration among clinicians, data scientists, engineers, and ethicists—working together to ensure that powerful tools are safe, transparent, and genuinely beneficial for every surgical patient.
Publication Details
Journal: Perioperative Precision Medicine
Article Title: A New Era of Intelligent Decision Support in the Perioperative Period: Innovative Applications of Artificial Intelligence, Big Data, and Large Language Models
Article Type: Perspective
Website Link: View Article
DOI: 10.61189/319254ilybgm
Publication Date: March 2026
First Author: Di Wang
Corresponding Author: Shafinah Ahmad Suhaimi (shafinahas@usm.my)
Affiliation: Department of Biomedical Sciences, Pusat Kanser Tun Abdullah Ahmad Badawi, Universiti Sains Malaysia
