What’s the most creative way researchers have used big data or AI to strengthen evidence-based medicine in recent years?
For example, are there cases where AI discovered hidden patterns in old clinical trial data, predicted drug interactions before human testing, or identified new risk factors from wearable health devices?
3 Answers
Haitham Jowah
I believe that the most creative and important use of AI has not been to find new patterns, but to truly challenge how we think about diseases. We are switching from asking AI to answer our questions to letting AI ask its own. This will show us that conditions we have thought of as one disease for decades are actually a group of different diseases.
The renaming of diabetes by AI is a great example of this.
We have used a general "Type 1 vs. Type 2" model for years. One important study from Lund University in Sweden, on the other hand, used an unsupervised machine learning model to learn from thousands of patient data points, such as genomics, metabolic markers, age of start, BMI, and more. They did not tell the AI to look for Type 2 diabetes; they just told it to find the natural groups in the patient data.
The end product was beautiful. The AI did not care about the usual labels and found five separate groups, which are:
1- A group of people who are severely insulin-resistant and have a very high chance of kidney disease.
2- A group that is severely insulin-deficient and has a high risk of blindness.
3- Mild Age-Related group with a lot of people and a much better outlook.
Two things make this a big deal for evidence-based medicine:
- It explains why some research trials do not work. A lot of drugs may not have worked in large-scale tests because they were very good at treating one group but not the others, which would have cancelled out the positive signal.
- It makes precision treatment possible. We can now plan studies that are just for these new, biologically distinct subtypes. We do not have to give a drug that protects the kidneys to all "Type 2" people. Instead, we can focus on the Severe Insulin-Resistant group, which is where it will work best.
People are now using this method, called "digital phenotyping," to study other difficult illnesses, like Parkinson's and heart failure. It is creative because it uses AI to change the present as well as predict the future. This gives us a much better way to plan our study and clinical work.
#Link for the Sweden study " ( https://pubmed.ncbi.nlm.nih.gov/29503172/) "
The renaming of diabetes by AI is a great example of this.
We have used a general "Type 1 vs. Type 2" model for years. One important study from Lund University in Sweden, on the other hand, used an unsupervised machine learning model to learn from thousands of patient data points, such as genomics, metabolic markers, age of start, BMI, and more. They did not tell the AI to look for Type 2 diabetes; they just told it to find the natural groups in the patient data.
The end product was beautiful. The AI did not care about the usual labels and found five separate groups, which are:
1- A group of people who are severely insulin-resistant and have a very high chance of kidney disease.
2- A group that is severely insulin-deficient and has a high risk of blindness.
3- Mild Age-Related group with a lot of people and a much better outlook.
Two things make this a big deal for evidence-based medicine:
- It explains why some research trials do not work. A lot of drugs may not have worked in large-scale tests because they were very good at treating one group but not the others, which would have cancelled out the positive signal.
- It makes precision treatment possible. We can now plan studies that are just for these new, biologically distinct subtypes. We do not have to give a drug that protects the kidneys to all "Type 2" people. Instead, we can focus on the Severe Insulin-Resistant group, which is where it will work best.
People are now using this method, called "digital phenotyping," to study other difficult illnesses, like Parkinson's and heart failure. It is creative because it uses AI to change the present as well as predict the future. This gives us a much better way to plan our study and clinical work.
#Link for the Sweden study " ( https://pubmed.ncbi.nlm.nih.gov/29503172/) "
ZJT
One of the most groundbreaking use cases of AI in evidence-based medicine is to extract latent patterns from historical clinical trial and EHR data—e.g., finding subsets of patients who were responsive to once-useless drugs. AI has also predicted human drug interactions on the basis of knowledge graphs before testing and discovered preclinical indicators of disease (e.g., atrial fibrillation or sepsis) from wearable sensor data. These have led to faster drug repurposing, precision therapies, and earlier diagnosis.
Dr Chisom
Now, to some people, that might sound quite simple. But for those of us working in low- and middle-income countries, it’s a big deal. Many facilities here can’t even run the basic blood tests, so having a low-cost, portable way to get that information could change everything.
If we can use AI to predict and detect illnesses early, we can cut costs, reach more patients, and improve survival rates. In places where every naira or dollar counts, that could mean the difference between a child getting treatment on time or not at all.