Critical transformation in critical care

The coming of age for big data and artificial intelligence in the ICU

Are we ready to let of the old and embrace the new? As artificial intelligence permeates almost all aspects of our lives, learn how AI is transforming our practice of critical care by offering us access to a “digital second opinion” with predictive analytics enabling us to be proactive and improve outcomes.

Challenging the age-old methodology

The practice of medicine has, in essence, operated the same way for decades with no fundamental changes in how patients are treated.  The concepts of rounding and awaiting nurse calls for urgent issues is taught in medical school and remain fully ingrained in clinical practice. Unfortunately, this practice model means that medical interventions are often implemented in a reactive rather than proactive process 1; that is, deteriorations in patients are only treated once they have already occurred. Individual patients may deviate from the path of what is expected from treatment and physicians may not realize this until a patient is subject to morbidity or mortality.

Heralding in the new age of AI in critical care

Today, we are seeing the start of an exciting and revolutionary change in the way we practice medicine. In our digitized age, we have accrued massive amounts of data or ‘big data’ which has been exploited by the combination of statistics and computer science. This statistical modeling has become ‘machine learning’, and with ever more sophisticated and unprecedented performance, we have now reached the ability to utilize this ‘artificial intelligence’ or AI 2. These powerful technologies of big data, machine learning, and AI have penetrated into and transformed many facets of our lives 3. Now they are being used in health care due to the availability of open databases containing vast amounts of reliable and usable patient data 2. Such data has been systematically captured and analyzed from both individuals and populations. A particularly robust source of this big data is in the realm of critical care medicine where patients are continually monitored every second as a key component of their care infrastructure 4. Therefore, harnessing this big data with advanced processing technology and analytics can be hugely beneficial for clinical data research purposes as well as for creating data-driven applications for clinical decision support 2

The new age of personalized and precision care

Data-driven AI applications are particularly beneficial in the ICU where the practice of individualized, personalized precision medicine is vital 2,4. Such AI technology has the potential to save lives by recognizing clinical patterns and using predictive models of clinical outcomes. Physicians are able to draw on their own knowledge and experience, current research findings, and clinical illness scores such as APACHE and SOFA, but they are not technically supported or monitored, leading to a high rate of diagnostic errors 5.   AI technology can assist physicians in clinical decision making by drawing on the evidence of relevant big data to predict patient outcomes that may have otherwise been unforeseen. ‘Assist’ is a key word here since truly thinking machines remain beyond our reach; such technology does not negate the vital importance of physicians’ experience and expertise 3. Every patient is important to us, yet we are often blinded by the complexity of the multiple medical problems, that we may sometimes overlook the seemingly trivial factor, that may in fact have great significance for the individual patient. 

AI technology allows more attention to be focused on the individual patient, ensuring each patient stays on the desired and expected path of recovery. As such, if a patient is predicted by an appropriate AI application to have a clinically pertinent event such as physiological deterioration, a clinical decision for treatment can be given speedily to prevent a future deviation from recovery. Employing such AI technology therefore promotes proactive rather than reactive treatment; significantly improves the system of alerts and alarms in the ICU; and is a quantitative on-site second opinion to validate a physician’s sense or opinion. Moreover, this technology may be used to predict whether complications are not going to occur in individual patients. Such positive predictions are also extremely useful because patients consequently need less frequent monitoring and can avoid unnecessary interventions. The confidence building of an encouraging prognosis reduces patient anxiety and catastrophizing, and most importantly, improves the quality and duration of their sleep. All of these can significantly contribute to an accelerated recovery and discharge from the ICU 6,7.  Accordingly, AI technology can improve patient outcomes with both negative and positive clinical predictions.

Critical steps towards a predictive and proactive approach

One example of using a machine-learning technique was carried out at an academic hospital in the United States. They wanted to derive and validate an ICU readmission model and compare it to previously published algorithms 8 Using data available in real time in the electronic health record, a  validated machine learning model was developed to predict the likelihood of ICU readmission. . They found that 11% of 24,885 ICU transfers to the wards were later readmitted to the ICU. The machine-learning–derived model had significantly better performance than either the Stability and Workload Index for Transfer score or Modified Early Warning Score (using area under the receiver operating curve analyses). They therefore concluded that a machine learning approach to predicting ICU readmission was significantly more accurate than previously published algorithms. This is important because implementing such an approach can identify patients who may benefit from additional time in the ICU or more frequent monitoring after transfer to the hospital ward 8

Challenges to transformation

Substantial challenges continue to exist in the development of AI models in medicine.  Clinical data needs to be easily obtainable but there continue to be outdated privacy laws, a certain wariness or reluctance to the rapid uptake and adoption of innovation and collaboration, and disparate data representations between different international open databases 2,4. The availability of affordable and accessible high-performance computer hardware is also vital 2. It is also important to evaluate the process of validation and testing 9 Lui

Overcoming Physician-Data Science Disconnect

Another current challenge is the disconnect between physicians and data scientists. There is a persistent gap between physicians required to understand the clinical relevance of the data and data scientists who are critical to extracting and analyzing usable information from the exponentially growing amount of generated health care data 10. MIT Critical Data has made some steps towards bridging this health-data divide by holding ‘datathons’ where clinicians and data scientists are paired up and challenged to work together to solve a clinical problem 2,4. Indeed, many research papers have resulted from such collaborations which have robust design structures and support in place. However, the adoption of such technologies has consistently lagged the discovery.  Utilization of such models in medical practice has been limited, likely because of the way physicians are educated in decision making. 10. The next generation of clinicians need to be educated beyond traditional biostatistics and able to understand and work with complex statistical methods, and grasp the computational approaches that will undoubtedly be incorporated into their practices 2,11

Transforming our mindset is critical ..big data , big impact, better care

Despite the challenges, seen with all new clinical advances, the benefit clearly outweighs the risk. The prospects of AI in the critical care ICU setting, including precision medicine, proactive and decision making, and thus improved patient outcomes, are transforming the way we practice medicine. These continually evolving technologies may achieve medical feats that we can only dream of today. It is up to us, as clinicians to step up to the challenge, upskill ourselves and keep up with the accelerating pace of this exciting transformation in the new-age practice of medicine. We owe it to the patients under our care.


  1. Cosgriff CV, Celi LA, Stone DJ. Critical Care, Critical Data. Biomed Eng Comput Biol. 2019; 10:1179597219856564. doi: 10.1177/1179597219856564. eCollection 2019. Review.
  2. Beam AL, Kohane IS. Big data and machine learning in health care. JAMA.2018;319:1317–1318.
  3. Organizing Committee of the Madrid 2017 Critical Care Datathon, Núñez Reiz A, Martínez Sagasti F, et al. Big data and machine learning in critical care: Opportunities for collaborative research. Med Intensiva. 2019;43(1):52-57. 
  4. Graber ML. The incidence of diagnostic error in medicine. BMJ Qual Saf. 2013;22 Suppl 2: 21–27.
  5. Pisani MA, Friese RS, Gehlbach BK, Schwab RJ, Weinhouse GL, Jones SF. Sleep in the intensive care unit. Am J Respir Crit Care Med. 2015;191(7):731-738.
  6. Ding Q, Redeker NS, Pisani MA, Yaggi HK, Knauert MP. Factors Influencing Patients’ Sleep in the Intensive Care Unit: Perceptions of Patients and Clinical Staff. Am J Crit Care. 2017;26(4):278-286.
  7. Rojas JC, Carey KA, Edelson DP, Venable LR, Howell MD, Churpek MM. Predicting Intensive Care Unit Readmission with Machine Learning Using Electronic Health Record Data. Ann Am Thorac Soc. 2018;15(7):846-853. 
  8. Lui Y, Chen P-H C, Krause J, Peng L. How to Read Articles That Use Machine Learning. Users’ Guides to the Medical Literature. JAMA. 2019;322(18):1806-1816.
  9. Celi LA, Davidzon G, Johnson AE, Komorowski M, Marshall DC, Nair SS, Phillips CT, Pollard TJ, Raffa JD, Salciccioli JD, Salgueiro FM, Stone DJ. Bridging the Health Data Divide. J Med Internet Res. 2016;18(12): e325. doi:10.2196/jmir.6400.
  10. Waldman SA, Terzic A. Health Care Evolves from Reactive to Proactive. Clin Pharmacol Ther. 2019;105(1):10-13. doi: 10.1002/cpt.1295.
  11. Moskowitz A, McSparron J, Stone DJ, Celi LA. Preparing a new generation of clinicians for the era of Big Data. Harv Med Stud Rev.2015;2: 24–27.


Like what you see? Drop us a line.

Schedule a time with us here:

* We want to see you. Register before we’re filled.

(*) indicates mandatory field

One moment while we direct
you to the CLEW website

Patient View
Rapid, clear, consistent patient data summary and visualization

  • Patient risk stratification (high, moderate and low risk patients)
  • Intubated patients
  • Vasopressors/inotrope support
  • Empty beds
  • Key patient and unit information
  • Occupancy, ventilator status and vasoactive medication
  • User selectable vital sign display
  • Summary of all relevant lab results
    Dynamic and collaborative caregiver task list
  • Aggregate patient demographicsand PMH (past medical history) andpredicted risk level

Patient View
Rapid, clear, consistent patient data summary and visualization
High Risk Patients

  • Tasks
  • New Admissions (pulled from a left sidebar)
  • User selectable vital sign display
  • Summary of all relevant lab results
  • Dynamic and collaborative caregiver task list
  • Aggregate patient demographics, PMH and risk level

Unit View
Full TeleICU situational awareness, displaying patient data and predicted state from all units and highlighting all notifications

  • Patient risk stratification (high, moderate and low risk patients)
    Intubated patients
  • Patients on vasoactive medication
    Unit occupancy
  • Patients on vasoactive medication
  • High-level estimation of the overall unit acuity

    Key patient and unit information


Work List
Integration with the TeleICU workflow by gathering all notifications, tasks, new admissions and high-risk patients

  • Notifications
  • High Risk Patients
  • Tasks
  • New Admissions (pulled from a left sidebar)

Unit View by Layout

Full ICU situational awareness, displaying patient data and predicted risk level, for one unit or multiple units. The display highlights all notifications, low risk patients and other key clinical information, to provide multi-dimensional situational awareness.

  • Patient risk stratification (high, moderate and low risk patients))
  • Intubated patients
  • Vasopressors/inotrope support
  • Empty beds
  • Key patient and unit information
  • Occupancy, ventilator status and vasoactive medication
Thank you for requesting our demo.

You’ll be hearing from our specialist shortly.