Artificial intelligence in the intensive care unit

doi:10.31744/einstein_journal/2024EDS2

Artificial intelligence (AI) has great potential to improve the care of critically ill patients and enhance clinical outcomes.(¹) The Intensive Care Unit (ICU) represents a complex environment in which patients frequently exhibit clinical instability and a high risk of death.(¹) Timely and precise medical decisions are vital to delivering the best available treatment to patients.(¹) AI has emerged as a transformative tool in several fields of medicine, and its application in critical care medicine is particularly promising.(²)

Theoretically, AI may be applied to all routines and process involved in the care of critically ill patients, including diagnosis, prediction, and support for clinical decisions.(¹) Additionally, AI may optimize ICU management by improving resource utilization, supplies, and personnel, thereby enhancing ICU efficiency. (¹) Therefore, AI has the potential to transform and improve patient care safety in the ICU.(¹)

AI algorithms can process and analyze large volumes of patient data, identifying patterns that humans are likely to overlook by, and providing crucial insights for clinicians at the bedside. (¹,²) These algorithms could support clinical decision-making by offering evidence-based recommendations.(¹,²) For instance, AI has shown promising results predicting mortality in patients with traumatic brain injury,(³) clinical deterioration in step-down units,(⁴) myocardial infarctions,(⁵) sepsis,(⁶) and quantifying pulmonary edema.(⁷) However, robust prospective studies are necessary to determine the clinical applicability of these models.

One study developed a logistic regression model based on intracranial pressure, mean arterial blood pressure, cerebral perfusion pressure, and Glasgow Coma Scale to predict 30-day mortality in patients with traumatic brain injury.(³) The model discriminated between survivors and non-survivors with an accuracy up to 84%.(³) Moreover, a machine learning model (random forest classification) was development to analyzed admissions in a step-down unit and predict clinical deterioration, such as hypotension, tachycardia, or desaturation.(⁴) Remarkably, this model could detect clinical deterioration 90 minutes before its occurrence.(⁴)

Another significant innovation introduced is a deep learning model designed to predict myocardial infarctions from electrocardiograms.(⁵) By utilizing six-lead electrocardiography, this model can detect myocardial infarctions, potentially reducing the time to treatment initiation.(⁵) Furthermore, an algorithm has been created for early prediction of sepsis, enabling potential detection of sepsis in ICU patients up to 4 hours prior to clinical recognition.(⁶) Additionally, a machine learning model has been developed to quantify pulmonary edema and differentiate congestive heart failure from other lung diseases using chest radiographs.(⁷) These advancements in AI models showcase their potential to revolutionize medical diagnostics and improve patient outcomes.(³–⁷)

Beyond the promising applications already explored, AI holds immense potential for broader transformations across the healthcare perspective. Two key applications stand out: (i) personalized medicine and precision healthcare, and (ii) drug discovery and development.⁸ In the first scenario, AI can revolutionize patient care by enabling a shift towards personalized medicine.(⁸) By analyzing a patient's unique medical history, genetic makeup, and real-time health data, AI algorithms can generate personalized treatment plans and predict individual responses to medications.(⁸) This approach can optimize treatment efficacy while minimizing side effects. Additionally, AI can assist in risk stratification, allowing for targeted preventive measures for individuals with a higher susceptibility to certain diseases.(⁸)

The traditional approach, to drug discovery is time-consuming and expensive.(⁸) AI can significantly accelerate this process by analyzing vast datasets of molecular structures, clinical trials, and patient data. AI algorithms can identify promising drug targets, predict potential drug interactions, and optimize drug development pipelines.(⁸) This has the potential to expedite the development of life-saving medications and therapies.(⁸)

However, incorporating AI in the ICU presents several challenges and considerations. Firstly, it raises important ethical and legal questions regarding patient privacy, data sharing, transparency of algorithms and responsibilities.(¹) It is essential that patient data is handled ethically, with utmost respect for confidentiality. Healthcare professionals must consider these ethical implications while ensuring that AI-driven decisions align with patient values and preferences.¹ Additionally, strict measures must be implemented to protect patient data privacy and ensure compliance with data protection regulations.(⁹)

Secondly, to ensure the responsible and effective use of AI in critical care, it is crucial to verify that the data is accurate, reliable, and representative of the patient population.(²) Maintaining methodological rigor is indispensable in developing and deploying AI models to ensure their precision, reliability, and reproducibility in clinical practice.(²) The accuracy of AI systems is heavily reliant on the quality of the input data.(²) However, accuracy of medical AI should not be mistaken for efficiency.(²) While accuracy is necessary, it alone does not guarantee efficiency gains. Rigorous validation processes and continuous monitoring are essential to ensure that the data used is accurate, reliable, and representative of the patient population. These measures are crucial to maintaining the effectiveness and safety of AI algorithms in clinical practice.(²,¹⁰)

Thirdly, rigorous validation of AI systems is crucial to ensure their effectiveness and safety.(¹⁰) Similar to new drugs and medical devices, AI algorithms must undergo rigorous clinical evaluations to validate their performance and identify potential risks.(¹⁰) Regulatory bodies such as the FDA have developed specific guidelines for the validation of AI and machine learning software as medical devices.(¹⁰) Finally, integrating these systems into existing clinical practices requires careful consideration of human factors.(¹⁰) It is essential that AI systems are designed to be intuitive and user-friendly, thereby minimizing the cognitive load of healthcare professionals and fostering acceptance and confidence in these systems.(¹⁰)

In conclusion, with the ongoing advancement of AI technologies and the increasing availability of big data in healthcare, AI tools have the potential to revolutionize critical care medicine by improving diagnostic accuracy, customizing treatments and optimizing resource management. To achieve this, healthcare professionals, engineers, and data scientists must collaborate closely to ensure that AI solutions are developed and deployed ethically, safely, and effectively.

REFERENCES

¹ Yoon JH, Pinsky MR, Clermont G. Artificial Intelligence in Critical Care Medicine. Crit Care. 2022;26(1):75. Review.
² Cui X, Chang Y, Yang C, Cong Z, Wang B, Leng Y. Development and Trends in Artificial Intelligence in Critical Care Medicine: a Bibliometric Analysis of Related Research over the Period of 2010-2021. J Pers Med. 2022;13(1).
³ Raj R, Luostarinen T, Pursiainen E, Posti JP, Takala RS, Bendel S, et al. Machine learning-based dynamic mortality prediction after traumatic brain injury. Sci Rep. 2019;9(1):17672.
⁴ Chen L, Ogundele O, Clermont G, Hravnak M, Pinsky MR, Dubrawski AW. Dynamic and Personalized Risk Forecast in Step-Down Units. Implications for Monitoring Paradigms. Ann Am Thorac Soc. 2017;14(3):384–91.
⁵ Cho Y, Kwon JM, Kim KH, Medina-Inojosa JR, Jeon KH, Cho S, et al. Artificial intelligence algorithm for detecting myocardial infarction using six-lead electrocardiography. Sci Rep. 2020;10(1):20495.
⁶ Nemati S, Holder A, Razmi F, Stanley MD, Clifford GD, Buchman TG. An Interpretable Machine Learning Model for Accurate Prediction of Sepsis in the ICU. Crit Care Med. 2018;46(4):547–53.
⁷ Horng S, Liao R, Wang X, Dalal S, Golland P, Berkowitz SJ. Deep Learning to Quantify Pulmonary Edema in Chest Radiographs. Radiol Artif Intell. 2021;3(2):e190228.
⁸ Johnson KB, Wei WQ, Weeraratne D, Frisse ME, Misulis K, Rhee K, et al. Precision Medicine, AI, and the Future of Personalized Health Care. Clin Transl Sci. 2021;14(1):86-93.
⁹ Mamdani M, Slutsky AS. Artificial intelligence in intensive care medicine [Editorialk]. Intensive Care Med. 2021;47(2):147-9.
¹⁰ Pinsky MR, Bedoya A, Bihorac A, Celi L, Churpek M, Economou-Zavlanos NJ, et al. Use of artificial intelligence in critical care: opportunities and obstacles. Crit Care. 2024;28(1):113.

Publication Dates

Publication in this collection
15 Nov 2024
Date of issue
2024

This content is licensed under a Creative Commons Attribution 4.0 International License.

[1] ¹ Yoon JH, Pinsky MR, Clermont G. Artificial Intelligence in Critical Care Medicine. Crit Care. 2022;26(1):75. Review.

[2] ² Cui X, Chang Y, Yang C, Cong Z, Wang B, Leng Y. Development and Trends in Artificial Intelligence in Critical Care Medicine: a Bibliometric Analysis of Related Research over the Period of 2010-2021. J Pers Med. 2022;13(1).

[3] ³ Raj R, Luostarinen T, Pursiainen E, Posti JP, Takala RS, Bendel S, et al. Machine learning-based dynamic mortality prediction after traumatic brain injury. Sci Rep. 2019;9(1):17672.

[4] ⁴ Chen L, Ogundele O, Clermont G, Hravnak M, Pinsky MR, Dubrawski AW. Dynamic and Personalized Risk Forecast in Step-Down Units. Implications for Monitoring Paradigms. Ann Am Thorac Soc. 2017;14(3):384–91.

[5] ⁵ Cho Y, Kwon JM, Kim KH, Medina-Inojosa JR, Jeon KH, Cho S, et al. Artificial intelligence algorithm for detecting myocardial infarction using six-lead electrocardiography. Sci Rep. 2020;10(1):20495.

[6] ⁶ Nemati S, Holder A, Razmi F, Stanley MD, Clifford GD, Buchman TG. An Interpretable Machine Learning Model for Accurate Prediction of Sepsis in the ICU. Crit Care Med. 2018;46(4):547–53.

[7] ⁷ Horng S, Liao R, Wang X, Dalal S, Golland P, Berkowitz SJ. Deep Learning to Quantify Pulmonary Edema in Chest Radiographs. Radiol Artif Intell. 2021;3(2):e190228.

[8] ⁸ Johnson KB, Wei WQ, Weeraratne D, Frisse ME, Misulis K, Rhee K, et al. Precision Medicine, AI, and the Future of Personalized Health Care. Clin Transl Sci. 2021;14(1):86-93.

[9] ⁹ Mamdani M, Slutsky AS. Artificial intelligence in intensive care medicine [Editorialk]. Intensive Care Med. 2021;47(2):147-9.

[10] ¹⁰ Pinsky MR, Bedoya A, Bihorac A, Celi L, Churpek M, Economou-Zavlanos NJ, et al. Use of artificial intelligence in critical care: opportunities and obstacles. Crit Care. 2024;28(1):113.