Artificial Intelligence in Anesthesia

Educational Objectives

The goal of this program is to improve anesthesiology practices using artificial intelligence (AI). After hearing and assimilating this program, the clinician will be better able to:

1. Identify the scope of AI in the field of anesthesiology.
2. Explain the implications of AI for practicing clinicians.
3. List the techniques used in AI.

Summary

Artificial intelligence (AI): involves the ability of machines to reason and perform functions such as problem-solving, object and word recognition, and decision-making; has been used in radiology, pathology, cardiology, and surgery; AI can facilitate several aspects of clinical care (eg, radiology, pathology); evidence-based data is new; maintenance cost is often high; in anesthesiology, can facilitate improvements in perioperative care and intensive care; can also facilitate assessment of surgical time and human resource requirements, preparation of equipment, preoperative workup of patient (eg, can assist in scheduling if patient has multiple facilities to visit before a procedure; dialysis can be set up, laboratory studies can be obtained, assessments of problematic issues in postoperative care can be made)

Techniques in AI

Standard logic: eg, classical machine learning; emphasis on clarity; uses Boolean logic

Fuzzy logic: used in, eg, higher concept neural networks and deep learning; truth value of variables can be between 1 and 0; nonbinary system

Machine learning: major subdivision of AI; tree-like models (eg, algorithms); easy computing; value is computing power; classified as supervised learning because programmers writing codes and modify them as needed; uses classification system (ie, binary system); eg, in open-loop target-controlled infusion, parameters are set; not very complex and often can be wrong; has a broad ability to interpret numerical data, images, text analysis, and speech recognition; eg, can set patient-controlled analgesia (PCA) dosing based on heart rate, respiratory rate, body mass index (BMI), previous dosing of narcotics, and other factors

Neural networks: unsupervised learning; includes deep learning; machine learns by trial and error; eg, AlphaGo technology; machine learns games by playing itself repeatedly; inspired by biological nervous system; can perform image recognition and data classification

Deep learning: variation of neural networks; consists of neural networks within neural networks; allows for increasing complexity; evolutionary; self-learns and uses sequential data (eg, speech); eg, Face2Gene can analyze child’s face and determine genetic coding

Bayesian methods: coding written based on prior belief system and data is gathered as the method is used; evidence is incorporated into prior belief system

Applications of AI in anesthesia:monitoring the depth of anesthesia — accuracy ofbispectral index (BIS) monitoring in electroencephalography has been refuted in recent publications; AI can use other parameters in addition to BIS, including, eg, patient’s temperature, past medical history, age, heart rate, level of stimulation from start of surgery, surgery type; control of anesthesia delivery similar

Event prediction: new technology developed that can make number of assumptions about patients intraoperatively has been shown to have 95% accuracy; if, eg, the patient has set criteria on hypotension, AI can determine hypotension 15 min prior to the clinical episode, based on, eg, drugs given, propensity score matches

Ultrasonographic guidance: helpful in image assessment; ability to aid in vertebral level placement of epidural anesthesia will likely soon be available

Pain management: to determine opioid dosing and preoperative consultation with acute pain service

Assistance with operating room logistics: can help with, eg, scheduling; reduces overhead

Anesthesia implications of AI: allows for facilitation; AI helps to collect and mine data, and use large amounts of challenging data; may help, eg, in patients in China, Italy, and Iran with COVID-19 who have been treated, to assess whether or not angiotensin-converting enzyme (ACE) inhibitors or angiotensin II receptor blockers play a role (because of the upregulation of ACE 2 by the virus on the lungs originally presumed to be diabetes); similar application possible assessing nonsteroidal anti-inflammatory drugs; could be helpful regarding recommendations

Clinician concerns regarding AI: common fear of being replaced; question of whether goal is augmenting perioperative care rather than replacing the clinician; most literature supports the likelihood that AI will primarily automate cognitive work such as advanced, comprehensive clinical decision support tools, accounting for patients' backgrounds and medication dosages, eg, if patient started on a medication recently vs long time ago; this could be a factor in different responses to intraoperative medicines; perioperative platform called Touch iQ helpful system

Touch iQ: an anesthesia information system, medication management system, and perioperative assessment tool; it suggests perioperative therapies and associated patient profile and risk profiles that may otherwise get missed; could identify, eg, patients with high morbid obesity and high BMI who did not have a sleep apnea study; possible to predict chances of developing pulmonary hypertension based on their age and length of morbidity; this might not be on patient chart because no sleep apnea study was done

Limitations of AI in anesthesia: many consider anesthesia might never be fully automated because of required dexterity-based labor (in, eg, axillary blocks, venous access, pulmonary artery catheterization, intubation); clean decision-making is required; expectations set on medicine high; current robots models do not possess that dexterity

Limitations and ethical considerations: a lot of research and development to still be done; cost is prohibitive regarding risk-benefit ratio; there is duplicate data, missing data, or inaccurate data; the data quality will likely become better with time; AI is a tool that cannot replace clinicians because of the rate at which doctors work, and the decisions they make; but AI can answer and solve appropriate questions and facilitate care for clinicians

Readings

Abdel-Kader et al. Overview and limitations of database research in anesthesiology: A narrative review. Anesth Analg. 2021;132:1012-1022; doi: 10.1213/ANE.0000000000005346; Bellini V et al. Artificial intelligence: A new tool in operating room management. Role of machine learning models in operating room optimization. J Med Syst. 2019;44:20; doi: 10.1007/s10916-019-1512-1; Collaço E et al. Immersion and haptic feedback impacts on dental anesthesia technical skills virtual reality training. J Dent Educ. 2021;85:589-598; https://doi.org/10.1002/jdd.12503; Hashimoto DA et al. Artificial intelligence in surgery: Promises and perils. Ann Surg. 2018;268:70-76; Hashimoto DA et al. Artificial intelligence in anesthesiology: current techniques, clinical applications, and limitations. Anesthesiology. 2020;132:379-394; doi: 10.1097/ALN.0000000000002960.

Disclosures

In adherence to ACCME Standards for Commercial Support, Audio Digest requires all faculty and members of the planning committee to disclose relevant financial relationships within the past 12 months that might create any personal conflicts of interest. Any identified conflicts were resolved to ensure that this educational activity promotes quality in health care and not a proprietary business or commercial interest. For this program, members of the faculty and planning committee reported nothing to disclose.

Acknowledgements

Dr. Mody was recorded at the 47th Annual American Osteopathic College of Anesthesiologists’ Mid-Year Seminar, held March 13-15, 2020, and presented virtually by the American Osteopathic College of Anesthesiologists, Chicago, IL. For information on future CME activities from the American Osteopathic college of Anesthesiologists, please visit aocaonline.org. Audio Digest thanks the speakers and the American Osteopathic College of Anesthesiologists for their cooperation in the production of this program.

CME/CE INFO

Accreditation:

Lecture ID:

AN632802

Qualifies for:

ABA MOCA, Risk Mgmt/Patient Safety/Medical Errors

Expiration:

This CME course qualifies for AMA PRA Category 1 Credits™ for 3 years from the date of publication.

Instructions:

To earn CME/CE credit for this course, you must complete all the following components in the order recommended: (1) Review introductory course content, including Educational Objectives and Faculty/Planner Disclosures; (2) Listen to the audio program and review accompanying learning materials; (3) Complete posttest (only after completing Step 2) and earn a passing score of at least 80%. Taking the course Pretest and completing the Evaluation Survey are strongly recommended (but not mandatory) components of completing this CME/CE course.

Estimated time to complete this CME/CE course:

Approximately 2x the length of the recorded lecture to account for time spent studying accompanying learning materials and completing tests.