“There is no doubt that Formula 1 has the best risk management of any sport and any industry in the world.”

Jackie Stewart, three time Formula 1 World Drivers’ champion


This is a timely short article on the role of artificial intelligence in the intensive care setting. The COVID-19 pandemic has shown that while artificial intelligence performed reasonably well in the domain of drug repurposing and vaccine design, it did not demonstrate its full potential in the area of decision support due to data and IT deficiencies.

Talha Burki, a regular contributor to the Lancet, argues that current ICU clinicians have human cognitive limitations due to the high number of variables for patients on mechanical ventilation and the large number of patients in the ICU on mechanical ventilation, especially during the pandemic.

He then elucidated how artificial intelligence is ideally suited for the ICU setting as it has vast amounts of data available for machine and deep learning techniques, and ”there are so many possibilities at every stage of translation, from understanding the trajectory of disease to understanding when we are not doing things we should be doing (quote from Brijesh Patel, an intensivist).

In addition, the pulmonologist Gary Weissman advocated use of predictive models rather than conventional large randomized controlled trials to help advance frontline knowledge of favorable interventions in the ICU. This point was painfully obvious during the very early months of the pandemic when patients were dying on mechanical ventilation and yet very few were being proned, a strategy that turned out to be of some benefit for some patients.

While machine learning techniques have been deployed to predict length of stay, risk of mortality, and likelihood of readmission, more real-time models can be of great assistance for clinicians. Patel terms a project using a device that is in the ventilator circuit as “deep physiology”, a smarter way to understand the lungs and track disease perhaps better than human clinicians can in a busy ICU setting.

An AI professor Aldo Faisal further explains that “it has become increasingly evident that you cannot just look at the admission data and that you need to look at the trajectory of the patient on a day by day, hour by hour , and minute by minute basis.”

The racing cars in Formula 1 with their 300-plus sensors generate over 1.1 million data points per second. The teams use historical race data and combine it with live data streamed from the cars for insights. We desperately need to bring some of this analytics technology and philosophy to our patients in biomedicine.

The AI models in the ICUs can mimic this Formula 1 strategy to minimize morbidity and mortality while maintaining a human-machine synergy. But one of the first obstacles to overcome would be to set up a system that will enable ventilators, or at least insights from each ventilator, to communicate with each other. Additional challenges include: variations in machines and data standards, biases in practice, and issues in regulatory practices.

The full article can be read here