Using AI, researchers at New York University have predicted which patients that were diagnosed with COVID-19 would eventually develop a serious respiratory disease, according to a clinical study released in Computers, Materials & Continua. (Healthitanalytics.com)
The spread of the coronavirus around the globe means there is a vital need to pinpoint which cases will make people seriously ill, according to the research team. (Techscience.com). Approximately 80% of the virus cases seem to be mild, but those who get the worst symptoms, usually need oxygen and ventilation for days.
AI May Be Able To Predict Who Gets ARDS
Acute respiratory distress syndrome or ARDS is a fluid buildup in the lungs that can be lethal in the elderly. It is a common feature in virus patients who decline after they are diagnosed. The research team at NYU wanted to see if artificial intelligence could be used to predict you who would get ARDS after being infected with COVID-19.
The NYU researchers gathered demographic, laboratory, and radiological findings from 55 patients who tested positive for the coronavirus at two hospitals in China. Their average age was 43. The research team then used the information to train artificial information models to get smarter as they collected more information.
The goal of the study was to design and deploy an AI tool using predictive analytics that could be used to support healthcare decision making. If future COVID-19 severity can be predicted, doctors may be better able to assess which patients are sick enough who need beds and which ones can be sent home safely.
Traditional Illness Characteristics Not Useful to Predict Critical COVID-19 Patients
One of the most vexing problems with the coronavirus is that common illness characteristics of COVID-19, such as specific patterns in lung images, fever, and a strong immune response were unhelpful in deciding which patients with mild symptoms would become critically ill.
Age and gender also were not useful to predict who would get really sick; however, previous studies have shown that males over 60 have a higher risk of severe respiratory illness.
AI Tool Found 3 Critical Parameters That Indicate Higher Risk of ARDS
The team’s AI tool noted changes in three important features – the level of liver enzyme called alanine aminotransferase, reported myalgia, and levels of hemoglobin were predictive of severe respiratory disease. With these critical factors in mind, the team was able to predict who was at risk of ARDS with 80% accuracy.
The clinical researchers also noted that ALT levels increase quickly as diseases such as hepatitis ravage the liver. ALT levels were only a little higher in COVID-19 patients. But they still were important in predicting the severity of the disease.
Myalgia is a deep ache of the muscles and was more common in virus patients. Past clinical research also suggests myalgia is connected to higher inflammation levels in the body.
Elevated levels of hemoglobin, a protein that contains iron that allows blood cells to carry oxygen, were also strongly connected to respiratory distress. The team noted that this may be explained by other factors, such as the patient not reporting tobacco use; smoking has been connected to higher hemoglobin levels.
The study had some limitations, such as the small dataset and the limited severity of the disease in the population that was studied. More refining of the model using more information from other settings will help to enhance its power to predict events.
Model Shows How AI Can Support Physicians
Scientists on the project noted that while there needs to be more work to validate the model, it does hold promise as another helpful tool to predict who will be most vulnerable to the virus. But the AI tool only can be used to support doctors’ clinical experience in the treatment of viral infections.
The study shows how AI and predictive analytics can be used to support doctors and other clinicians, especially during a global health crisis. Researchers say predictive analytics can play a critical role in enhancing clinical skills to distinguish being people who are sick and not sick.
They stress that just as predictive text is intended to augment and not replace writers, the goal with the AI tool is not to replace clinical reasoning. Instead, the intent is to devise models that can provide helpful insight to doctors. Clinical acumen is based on personal and collective professional learning. Machine learning can offer even more insight to aid in positive patient outcomes.
The introduction of AI in healthcare is opening up new possibilities to track and monitor the COVID-19 pandemic, which may help improve clinical care responses and outcomes.
The use of this AI tool will encourage doctors to pay attention to the data during clinical practice and to watch patients more to see if they, for example, complain of severe myalgia. Being able to share data with the field in real-time is extremely useful in a pandemic such as we face now.