• Thu. Aug 11th, 2022

Microsoft’s AI model identifies where medical treatments hurt more than they help – IoT World Today

Dead-end Discovery uses reinforcement learning ML framework as it is well suited for healthcare

Microsoft researchers have built an AI model that can identify when medical treatments intended to help patients ended up harming them.

The model, Dead-end Discovery, uses the reinforcement learning ML framework – where an agent learns through trial and error – as it is well suited for healthcare. Health care is characterized by sequential decision-making: after reviewing a patient’s condition, providers apply treatment and observe the results. If the patient improves, the process repeats.

Medical care today is characterized by an emphasis on what needs to be done to help the patient recover. But the researchers argued that this may be an “unachievable” goal. Instead, they propose a reverse approach: identify treatments to avoid and prevent a point of no return for the patient.

The researchers applied their model to publicly available real-world medical data, focusing on critically ill patients with sepsis in intensive care.

Their goal was to help providers “identify which subset of treatments could statistically lead to further deterioration in health so that they can eliminate them when deciding on next steps,” the researchers said in a blog post.

They used a dataset of 53,400 hospital admissions between 2001 and 2012 and extracted a group of nearly 20,000 intensive care patients with sepsis. (They applied reinforcement learning offline because they were using a fixed data set.) The researchers studied 72 hours of ICU patient stay, 44 observational variables, and 25 treatments. They also used their AI model to identify patients showing signs of slipping into death.

The results: More than 12% of treatments given to patients who died later could have been harmful – 24 hours before their death. Additionally, the AI ​​model identified up to 10% of patients slipping to a point of no return up to 48 hours before death.

“While these percentages may seem low, more than 200,000 patients die of sepsis each year in US hospitals alone, and any reduction in this rate would potentially result in tens of thousands of people who would otherwise survive,” they said. .

The researchers said other uses for their AI model could include other areas of healthcare. Another possible application could be in finance, to alert investors when certain buy or sell decisions are likely to lose money – a financial dead end.

To access the code: https://github.com/microsoft/med-deadend

This article first appeared in IoT World Today’s sister publication, AI Business.