November 29, 2022 — Artificial intelligence can help improve diversity, fairness, and inclusion in clinical trials and drug development by overcoming some traditional human bias in these areas, but we’re not there yet, experts say. The technology can also help clinicians gain insights into the data to make diagnosis and treatment more accurate.
It starts with quality. Artificial intelligence (AI) relies on large amounts of data to create algorithms – or computer instructions – to develop best practices and predictions. But the instructions are only as good as the data used to create them. And people make the data.
“The development of AI technologies is driven by people, and these people have their own biases,” says Nahid Korji, chair of the AI in Healthcare Alliance. “As a result, algorithms will have their own biases.”
Technology that uses speech to diagnose disease is an example.
“There are many cases and examples where companies fail to recognize differences in speech across different cultures,” says Corgi. When a technology is based on the speech patterns of a limited population, “then when this model is applied in the real world to a different population with a different accent, that model fails.”
“As a result, it’s not charade.”
Another example is genetic and genomic data.
“Give or take, more than 90 percent of genetic and genomic data originated from people of European descent. He’s not from a continent,” says Corgi, who is also president and CEO of Cyclica Inc., a data-driven drug discovery company based in Toronto. Africa, Southeast Asia, Asia or South America.
Therefore, “a lot of the research that’s been done at this level of data is inherently biased,” he says.
To be fair
Creating data that takes diversity, equity, and inclusion of people and cultures around the world into account is not a hopeless challenge. But experts say it will take time. Once this is achieved, AI should be closer to being freed from human and systemic biases.
Raising awareness is essential.
“The solution to the problem comes from people’s inherent understanding that there is bias,” Corgi says, and then only fair and balanced data that passed the test of diversity are included.
A wiser choice?
Another promising avenue for AI is to streamline the drug development process, narrow the pool of potential drug candidates, and make clinical trials more cost-effective.
“If source data has challenges and limitations, AI will continue to propagate those limitations,” agrees Sastri Chilokuri, co-CEO of data-driven clinical trials company Medidata and founder and president of Acorn AI. “The source data needs to be more representative and it needs to be fairer to the AI in order to reflect what’s going on.”
When it comes to human or systemic bias in drug development, “it would be an understatement to say that AI or machine learning can fix it,” says Angeli Mueller, Ph. D., head of data and insights-generating integrations at Roche in Berlin. “But the responsible use of AI and machine learning can help us identify biases and find ways to mitigate any negative effects they may cause.”
At the same time that AI aims to streamline the drug development process, the technology can also help improve all doctors’ performance at their jobs, experts say. AI will help, for example, by widely disseminating knowledge and experience, sharing best practices from doctors who have a lot of experience with more complex patients. This should help guide those who treat only a few of these patients each year.
Chilokuri says the volume of surgeries in New York City or Delhi can run into hundreds of patients annually. “But if you go into the interior of the United States like Nebraska, the surgeon doesn’t see that much.”
AI can help clinicians “by providing the kind of tools that allow them to be able to deliver the same first-class care to their entire population a lot faster,” he says.
AI can help target treatment by using data to identify patients most at risk. The technology could also improve some areas of bottlenecks in medicine, such as the time it takes to interpret radiographs, Corgi says.
There is an AI company whose “business model is not entirely about replacing your radiologist but optimizing your radiologist,” he notes. One of the company’s goals is to “prevent death or severe illness from radiological examinations that are missed or that are piled on a heap and not dealt with quickly enough for that patient.”
Chilokuri says radiologists are so busy, they may only have 30 seconds or less to interpret each scan. The AI can flag a potential lesion of concern, but it can also compare an image to previous scans of the same patient. This AI-enabled insight applies not just to radiology but across data-driven fields of medicine.
Development of personalized medicine
AI can also guide a personalized approach to surgery, “because it’s not like humans come in small, medium, and large sizes,” says Chilokuri. This technique can help surgeons determine exactly where they should operate on a patient.
Mueller agrees that AI holds potential to advance personalized medicine.
“AI can help diagnose and predict risk, which can mean early interventions,” says Mueller, who is also vice chair of the board of directors of the Alliance for AI in Healthcare. If you look, for example, at a diabetic patient, what is the likelihood that they will have eye problems from diabetic macular edema? “
Technology can also help you look at the big picture.
“Machine learning can look for patterns in a population that may not be in your medical book,” Mueller says.
In addition to diagnosis and treatment, AI can also aid recovery by tailoring rehabilitation to each patient, Chilukuri predicts.
“It’s not like everyone is going to rehab the exact same way. So, you have very individualized AI plans that allow you to stay on track and predict where you’re going.”