Machine learning (ML) is a subfield of artificial intelligence (AI) specializing in the development of algorithms and models that enable computers to learn and make predictions or decisions without being explicitly programmed. It involves using statistical methods and computational models for extracting meaningful insights by identifying patterns during the analyses of large datasets, such as in image analysis for object/features detection, voice analysis through natural language processing (NLP) capabilities to interpret human language, and text analytics for detecting words and numbers, including translation mechanisms (Commun of the ACM 2012;55:78-87). For example, ML with the use of large datasets can perform pattern recognition between an individual’s clinical data (e.g., medical images and blood test indices) and clinical data that represents a certain diagnosed health condition stored in a library. The ML algorithm can identify the matching level between the two datasets (individual’s and library’s) and provide an opinion. Every time a successful matching occurs, the ML algorithm automatically learns from this experience (training dataset) and eventually improves its diagnostic accuracy, thus reducing the requirement for explicit programming. Computer and health scientists can join forces to build and embed ML algorithms in computers and mobile apps to evolve health care decision-making (IEEE Transactions on Engineering Management 2023;70:2809-26). Google Health, Merative, the Laboratory of Computer Science at Massachusetts General Hospital, the Stanford Machine Learning Group, and Aidence are some of the notable organizations/bodies that research and develop ML software to support health care professionals to deliver faster, more efficient, and safer care.

A notable example of a commercial ML predictive tool for physicians is Epic’s Deterioration Index, which is seamlessly integrated with the electronic health record (EHR) system (asamonitor.pub/3t1DnTb). The Deterioration Index uses ML algorithms to analyze patient data in real time available in the EHR (such as vital signs, lab results, nursing assessments, and medication data) and predict a patient’s deteriorating status. The benefits to physicians are multiple: a) it’s an early warning system promoting proactive interventions to prevent adverse outcomes; b) it saves time by circumventing the need to manually review patient charts and by dynamically prioritizing issues and patients who require immediate attention; and c) it uses the anonymized data from the index for research and enhanced health care quality by identifying trends, evaluating the effectiveness of interventions, and ultimately improving care protocols.

Similar tools can be of help to anesthesiologists. An example might be a perioperative predictive analytics tool that could also be seamlessly integrated with the EHR system (Anesth Analg 2018;127:90-4). The ML tool could continuously monitor and analyze real-time data, including patient vitals, medical history, surgical procedure details, and anesthesia-related factors, and predict potential perioperative complications or adverse events. The tool could provide real-time notifications to anesthesiologists for patients at high risk for complications during or after surgery and prompt them for preventive action. The perioperative predictive analytics tool would assist anesthesiologists in developing an anesthesia plan, taking into consideration specific risk factors from the medical history, that includes information about possible adverse drug interactions to minimize complications and improve patient safety.

ML can also help and accelerate drug discovery by virtually screening millions of chemical compounds to identify potential drug candidates. The ML algorithms assess the interactions between these compounds and the target proteins, prioritizing the most promising compounds for further investigation, narrowing down one or more compounds that are suitable for clinical application (Nat Rev Drug Discov 2019;18:463-77). For example, the Atomwise platform uses advanced ML techniques to predict the interactions between small molecules and biological targets (asamonitor.pub/46fIIVg).

In the future, the working environment of physicians in a hospital is expected to be transformed by the integration of ML applications and mobile health (mHealth) apps. Several ML applications are likely to be developed that could ease the work of physicians and anesthesiologists.

  1. Remote real-time monitoring and telemedicine: Physicians will be able to remotely monitor patients’ vital signs via mHealth apps and wearable technology, allowing them to be cared for in distant locations. The streaming data will be examined by ML algorithms to identify irregularities and, if necessary, trigger alerts. With the help of these remote monitoring capabilities, physicians will be able to intervene in a timely and efficient fashion while reducing the number of hospital visits (JAMA Surg 2023;158:699-700). The COVID-19 pandemic significantly advanced both remote monitoring technology and teleconsultation (Pediatr Diabetes 2021;22:1115-9). Same-day surgery and ambulatory patients could be monitored remotely by anesthesiologists in real time. Through the integration of ML algorithms into clinical software and mobile apps, the new intelligent clinical decision support systems will allow anesthesiologists to be swiftly informed and intervene in a timely manner.
  2. Predictive analytics and risk stratification: ML algorithms can identify patients at high risk of contracting specific diseases or encountering complications. They can assist physicians in categorizing patients into various risk groups by analyzing their medical history and enabling evidence-based recommendations to support decision-making processes with focused interventions, preventive measures, and early detection of impending health concerns. Such data-driven methods will enable physicians to diagnose patients with increasing accuracy. Google Health is heavily investing in the development of such tools (asamonitor.pub/3Ps5mD7). A notable example is the ARDA (Automated Retinal Disease Assessment) ML tool that helps physicians detect diabetic retinopathy, a leading cause of blindness (AMA 2016;316:2402-10). For anesthesiologists, ML algorithms can be trained to predict patient-specific responses to anesthesia. These algorithms will assist anesthesiologists in planning the anesthesia care for patients, including their expected recovery trajectory. They can help identify potential complications and aid in planning follow-up care that will help anesthesiologists monitor patients postoperatively and predict their recovery outcomes (JAMA Surg 2023;158:699-700). These ML algorithms will be seamlessly integrated into clinical decision support systems.
  3. Personalized medicine and therapy optimization: ML applications can enable personalized medicine by analyzing patients’ genetic profiles, their family and medical histories, and treatment outcomes (Intern Med J 2021;51:1388-1400). By maximizing therapeutic effectiveness and reducing unfavorable side effects, this personalized strategy is expected to enhance patient outcomes and satisfaction. For anesthesiologists, in particular, anesthesia drug dose optimization because of possible genetic and geographic differences will help during patient care. ML algorithms can learn from large datasets obtained from a wide variety of patient groups. This information should help in decision-making, considering patient-specific factors such as age, weight, genetic and ethnic background, and comorbidities. Along with real-time monitoring data, the algorithms can help in minimizing the risk of adverse events and maximizing patient safety and comfort.
  4. Automated documentation and reporting: ML can automate the documentation and reporting tasks for physicians and anesthesiologists. Natural language processing techniques can extract specific information from unstructured text data such as medical records, clinical and operation notes, monitoring systems, and medical literature (BMC Med Inform Decis Mak 2017;17:155). Through this process, ML algorithms can automatically generate report summaries with specific information, decreasing the administrative burden on physicians and anesthesiologists.
  5. Simulation and training: ML algorithms can enhance training simulation environments by creating realistic patient avatars in virtual reality (VR), augmented reality (AR), and metaverse platforms. These avatars can mimic various medical conditions, clinical reactions, and responses to drugs and anesthesia, including emergencies, thus providing a realistic experience for trainees. Moreover, ML can analyze learner performance metrics and recommend simulation scenarios based on trainee progress, offering more personalized training. A notable example is OSSO VR, a platform developed to teach new surgical techniques through virtual reality (asamonitor.pub/3PnLb9k). These capabilities allow anesthesiologists to hone their expertise and practice critical decision-making in a safe environment (Nat Mach Intell 2022;4:922-9).

It is important to note that while ML applications hold great promise in supporting physicians’ clinical expertise and judgment, these applications still present several challenges that need to be addressed prior to a wider use by physicians. The challenges refer to several issues (Technovation 121;2023:10259):

  • Data quality/security: ML relies heavily on large and accurate datasets for training its algorithms. Much of the health data still exists in silos, in specific settings, and in software with limited interoperability, making it difficult to create unified and diverse datasets for robust algorithms while ensuring patient data privacy and security.
  • Reliability/regulatory compliance: ML algorithms are complex and difficult to explain, but physicians need transparency and interpretation of the process leading to a decision/recommendation, rigorous clinical validation to ensure applications’ efficacy, and compliance with existing health care regulations and standards for approval and deployment. Otherwise, the ML applications are unlikely to gain trust and confidence and to be used for patient care.
  • Accountability/ethical considerations: ML applications present many benefits for augmenting physicians’ capabilities, allowing them to provide more personalized, efficient, and timely care. However, physicians will need to play an active role in ultimately making the final decisions. Physicians need to understand the level of their responsibility when using these technologies and the issues that might affect the performance of the applications, such as data accuracy, algorithmic accuracy, data updates, and system maintenance. Moreover, they need to respond to ethical considerations, including potential overreliance on technology and automation in place of human judgment and patient consent processes for data usage. A notable example is the Samaritans Radar app, which analyzed Twitter posts with the use of ML to detect phrases associated with suicidal behavior. The app then released warning messages to the followers of the posting person so they could obtain help on their behalf. The app was discontinued due to severe criticism for placing vulnerable people at greater risk from cyberbullies (JAMA 2016;315:551-2).

Addressing these challenges requires collaboration between data and health scientists, medical practitioners, and policymakers to create responsible and effective ML applications that augment physicians’ capabilities, allowing them to provide more personalized, efficient, and quality care to patients (JAMA 2016;315:551-2). Finally, physicians must receive training in ML applications, as part of their medical education, in order to foster their use and limit the challenges (IEEE Transactions on Engineering Management 2023;70:2809-26).