Q&A: Preventing further harm in pediatric patients on ventilation with machine learning

Key takeaways:

• Ventilators may unintentionally harm patients; research is lacking in children.
• A new study will test how well machine learning algorithms can recognize breathing asynchronies in pediatric patients.

For children in critical condition who struggle to breathe, ventilators are essential machines; however, it is important to recognize that they may unintentionally hurt the patient.

According to a press release from Children’s Hospital Los Angeles, if air from a ventilator is not in sync with a patient’s breathing, they are at risk for infection, lung injury and brain damage. While this has been described in adult patients, it is unknown whether the same is true in pediatric patients, and this has led to the development of a study — funded by the NIH — that will assess the prevalence of breathing asynchronies and identify factors that may place children at higher risk for this event.

Researchers also plan to validate machine learning algorithms that are capable of recognizing various types of breathing asynchronies in this patient population.

Healio spoke with Robinder Khemani, MD, MSCI, leader of the research team and attending physician in pediatric intensive care at Children’s Hospital Los Angeles, to learn more about ventilator support in children, the use of machine learning in critical care and his hopes for the study.

Healio: What are possible harms of ventilator support among children?

Khemani: Using a ventilator is absolutely necessary to save the lives of many children, but there can be unintended harm from the ventilator such as injury to the lungs or respiratory muscles. If the ventilator support is not matched completely with the needs of the patient and the disease state, this harm may occur. For example, when children have trauma or diseases such as pneumonia, the ventilator settings need to be prescribed very carefully and individually for the patient to make sure they get just the right amount of pressure or air to go into the lungs. Too much or too little may lead to worse injury in the lung or inadequate delivery of oxygen to vital organs of the body. Similarly, there are times when the ventilator needs to do all the breathing for the patient, but if this happens for too long, the muscles of breathing (such as the diaphragm) can get weak, which may make it harder for the child to come off the ventilator.

Healio: What is patient-ventilator asynchrony (PVA)? How does it differ between adults vs. children?

Khemani: Patient-ventilator asynchrony generally means that the ventilator does not provide its help in the precise way the patient wants it. This can occur broadly in three areas. The first is if the patient is trying to take a breath, but the ventilator does not recognize this and does not deliver assistance. The second is if the duration of the breath delivered by the ventilator is either too long or too short for what the patient wants. The third is if the degree of help the ventilator is providing for the patient (amount of pressure or airflow) is either too much or too little compared with what the patient wants.

All of these factors can make the patient uncomfortable when they are breathing and may make injury to the lungs or respiratory muscles more common. These same basic types of asynchronies occur in both adults and children, but there are likely differences in how common each type of asynchrony is and how harmful it may be between adults and children. But much of this is still not fully known, which is why it is important to do research in this area.

Healio: What inspired you to research PVA in critically ill children and use machine learning algorithms in this research?

Khemani: We have been continuing to improve the way we ventilate critically ill children and adults over the last several decades, which has included shifting toward the patient being less sedated when on the ventilator and participating more in the breathing process. We have done this to try to prevent them from getting weak while on the ventilator and preventing harms from high doses of sedation medications which may prolong recovery or harm the brain. The net result is a patient on the ventilator who is controlling their own drive to breathe. When this happens, it is very important for the ventilator to match the way the patient wants to breathe. It is increasingly clear that many times the ventilator is not aligned in this way with the patient. The challenge is that detecting this misalignment is very complex and is easy to miss at the bedside unless providers are highly trained and are looking for it. This is why machine learning algorithms can be particularly helpful. These algorithms can be trained to detect these sometime subtle differences in ventilator waveforms, and they can consistently identify them, especially when a trained expert is not at the bedside.

Healio: Has machine learning been used in other aspects of critical care?

Khemani: Machine learning is increasingly being used in critical care although much of it is still in the research space. A number of algorithms and predictive tools have been developed using machine learning to identify patients who may be at high risk for complications, or to potentially identify patients who may benefit from certain management strategies. It is an exciting time in critical care because the critical care environment really lends itself to machine learning because we generate a huge amount of data every minute in the ICU. Machine learning provides an opportunity to harness this information and transform it into knowledge.

Healio: What do you hope to achieve in this study?

Khemani: With this study, we hope to improve our knowledge of PVA in children, understand what types of PVA are most common and which ones may be most harmful. We also hope to develop tools that can one day be at every patient’s bedside, identify when PVA is occurring and eventually help practitioners decide how and when to make adjustments to improve it.

This is a multi-center effort with two other expert hospitals from the Netherlands and Montreal, Canada. These hospitals and collaborators also have a lot of expertise in PVA, and this collaboration will truly accelerate the research and increase the speed at which we can get this to the bedside.

Healio: How will this research impact clinicians caring for children requiring ventilator support? How will the care they provide potentially change based on the results?

Khemani: This research will help us prioritize which types of PVA are most important or harmful to patients and help clinicians identify when it is happening. Based on these findings, it will help clinicians adjust the ventilator to better meet the needs of the patient, improving patient’s comfort on the ventilator and hopefully preventing unintentional injury which may be occurring from the PVA.

Reference:

• Using machine learning to make ventilator support safer for children. https://www.chla.org/newsroom/press-release/using-machine-learning-make-ventilator-support-safer-children. Published Aug. 7, 2023. Accessed Aug. 8, 2023.

For more information:

Robinder Khemani, MD, MSCI, can be reached at rkhemani@chla.usc.edu.