Could Artificial Intelligence Catch What Doctors Miss? Precision Medicine and Its Role in Brain Health

By | November 15th, 2018

What if a machine could determine the source of your brain fog, offer a targeted treatment plan to reduce your risk of developing Alzheimer’s or slow the disease’s progression? 

According to Dr. John Walker, who founded uMETHOD Health, which uses artificial intelligence (AI) and precision medicine to help treat Alzheimer’s, a machine can now do just that. After providing basic information about your diet, lifestyle, genetic history and a blood sample, you’ll receive a report with detailed information about prescriptions you should stop taking, supplements you should start taking, recommended diets, exercise programs and other lifestyle suggestions. 

  • After plugging your information into a machine, you’ll get two personalized reports in easy-to-understand language
  • The machine will detect if you have conditions that may be inducing cognitive decline, like sleep apnea or depression, then provide a targeted treatment plan. According to Walker, some depression medications could induce brain fog
  • Based on your blood test results, the machine will also determine how high your inflammation levels are, a factor associated with Alzheimer’s
  • While Walker said AI could slow the progression of Alzheimer’s, the ultimate goal is to encourage early prevention and treatment methods

Being Patient spoke to Walker about how AI could diagnose or treat Alzheimer’s, what this process looks like and what AI could identify that doctors may miss.

Being Patient: How is artificial intelligence being used to diagnose or treat a disease like Alzheimer’s?

Dr. Walker: Artificial intelligence is a big field. If you are an expert cook, you may have many areas of expertise. For example, you may know everything about baking bread, like yeast or how long things need to set. If you make sauces, you may know how to properly mix the flour, oil, eggs and so on. Saying that someone’s an expert cook actually implies a lot of expertise in different areas. Similarly, artificial intelligence has a lot of techniques in its arsenal. It’s a long list and they’re all pretty technical, but I’ll discuss three of them.

There’s a whole family of things in artificial intelligence called rules-based systems. For example, if your vitamin D levels are low, there is probably a rule that says you need to take a vitamin D supplement to raise it. That would be the first rule. But there’s another rule that says that if you take too much vitamin D, it’s toxic. There’s a rule that overrides the first rule, and then there’s another set of rules that say if you’re taking another drug or medication, it might heighten the effect of that medication, so now you need to change the dosage. You end up with these stacks of rules and you have to weigh or prioritize them to see which ones are most important in certain situations.We’ve had to do that with artificial intelligence for the field of medicine, including genetics, medications, different diseases, allergies and so on. That’s step one for AI.

Now, I’ll discuss a different field in AI called machine learning. You may have seen a lot of examples of that in the press, but I’ll share one, too. If your HDL blood level, or “good” cholesterol, is 25, you’d like to have it raised to 50. If I come back and measure you in three months, your HDL has gone up five points and the expectation in a normal population is that someone who works hard to raise their HDL should be able to raise it by five points in that amount of time, then that’s a success. But if the expectation is that someone needs to get from 25–50, three months have passed and you’ve only gone up by five points, then you’ll feel like a failure. There’s a coaching aspect to machine learning, including what the expectations are and what can people expect as we look at data. There’s no good answer in the medical books that says, “Here’s what someone’s HDL should look like,” because it depends on their age, gender and more.  

Being Patient: How much do we have to tell a machine before it can start drawing conclusions?

Dr. Walker: My company designed software to start with nothing, so if you only told me your age and gender and we’re talking about Alzheimer’s, we actually know a lot of information. We know that if you’re at this point, what the next test you should ask for is, or the next thing you should get measured. Yet, everything added on top of that fills out the picture more completely. If you give me blood numbers, then I know a lot about your blood chemistry and can say, “OK, interesting, your cholesterol is in good shape, but we’re seeing a problem with kidney disease.” If I get genetic information, then I know a lot about propensities, what’s pushing you, what the right drugs to be matched to are and so on. Every one of these adds to the picture.

What’s the bottom line for how much information we should have? If you just tell me your age and gender, you won’t get very specific information. If you give me your age, gender and say you’re concerned about Alzheimer’s disease, I’ll tell you the basics that say you need to sleep better, exercise well and improve your diet, but I don’t know anything about your diet. So if you tell me more about your diet, then I can improve the recommendations from there.

Being Patient: What can artificial intelligence tell us today? Can it make a diagnosis or give us appropriate treatments to delay the onset of Alzheimer’s and improve brain health?

Dr. Walker: It can predict and give you the recommendations for what to do now, next month and over the coming year to slow the progression of Alzheimer’s, if that’s the case, prevent it and actually improve your cognitive situation under a wide range of conditions. It looks like a report to your doctor.

We’re not just looking at the two rows for ApoE. We’re looking at your propensity for diet and exercise, other diseases you may have, how you react to medications and so on. Then we ask about your medications, family history, whether you had meningitis when you were six years old or not, what allergies you’ve had, and the list continues to unfold.

Being Patient: So the approach involves applying precision medicine, but a machine is doing a lot of the work for us. Is that correct?

Dr. Walker: In fact, the machine’s doing all the work. A third aspect of artificial intelligence is something called natural-language generation. It’s another thing in that toolkit. We use that technique to have the machine write reports where the doctor knows what every word means. Also, with that same information, we can write a report that goes to the patient or their caregiver. It’s the same information, but in a totally different language. It’s all written so that someone without a medical background can read the same thing.

Being Patient: How many factors do you enter into the equation to determine a precision medicine approach?

Dr. Walker: There’s no simple answer, but we’ll take an entire 23andMe genome file; that’s a spreadsheet with 700,000 rows. We’re not just looking at the two rows for ApoE. We’re looking at your propensity for diet and exercise, other diseases you may have, how you react to medications and so on. The first 700,000 are the easy ones. Then we ask about your medications, family history, whether you had meningitis when you were six years old or not, what allergies you’ve had, and the list continues to unfold. Often, it’s too many questions to start with at one time so you get the edge and go, “What’s the follow up on these?” Let me pick an example. Sleep apnea is incredibly important to consider and if you have sleep apnea, it trumps everything else. Anything I can tell you about diet, exercise or lifestyle, you have to fix the sleep apnea because it’s literally starving your brain.

Being Patient: How do we know that about sleep apnea?

Dr. Walker: It’s based on extensive scientific research. It’s one of several things on my list that we’ll get to that trumps everything else. It turns out there’s a really simple way to understand if someone has sleep apnea or not. The formula’s called STOP-Bang. Every one of those eight things refers to a question you ask to determine what someone’s likelihood of having sleep apnea is. The S stands for ‘snoring.’ Do you snore? The N stands for ‘neck circumference.’ It’s a simple world, but like a lot of things, it’s a formula, and it’s hard for your doctor to run that formula in a ten-minute annual visit.

Being Patient: Can a machine determine how severe someone’s sleep apnea is? 

Dr. Walker: Yes, it could determine that, but we believe a better path is to look at the formula. A formula says someone can have a high, medium or low likelihood of having sleep apnea. The simple recommendation to their doctor is that they should get involved in a sleep study. Then that’s where everything gets quantified and the sleep apnea gets addressed.

Being Patient: How does AI help in this example?

Dr. Walker: In this example, it’s simply the rules-based system that says, here are the rules to determine someone’s likelihood of having sleep apnea. It comes as answers to some of these millions of data points that we’ve collected along the way.

Being Patient: When you say millions of data points, what does that mean?

Dr. Walker: We’ve got genetic information, we collect blood chemistry samples and the results of urine tests that you may have done. If you’ve done scans, we take that information and information about your allergies. What’s really important is the medications that you’re on. What we’ve seen is that people are on an average of 15 medications with a standard deviation of 12. Taking a large number of medications has extensive effects on cognitive health. There’s an entire unwinding that has to occur there. Part of the analysis that goes into the rules related to medications.

The number one thing that affects cognitive health, other than genetics, is inflammation.

Being Patient: Plaques and tangles may appear in the brain, but it’s not until inflammation sets in that we see Alzheimer’s symptoms. Will AI help us to determine what role inflammation plays in all of this?

Dr. Walker: The number one thing that affects cognitive health, other than genetics, is inflammation. In addition to inflammation, we look at infection, insulin resistance and diabetes, stroke, heart disease, thyroid, kidney and liver disease, HSV, HIV, depression, smoking history, midlife obesity, brain trauma, hormonal balance, homocysteine, metal balance, metal toxicity and mycotoxin exposure. The answer is how to bring them all together under one umbrella and say, “All right, out of everything we analyze, let’s start by addressing inflammation.”

Being Patient: What can a machine do better than humans when addressing inflammation’s effect on cognitive decline? 

Dr. Walker: It can identify and prioritize inflammation as the next thing to work on by looking at your blood chemistry, hs-CRP, IL-6, etc. Those are the names of blood tests. It would look at different blood markers in your blood test panel that your doctor may not regularly ask for. We have a longer list of things to request and the first one of those is this thing called high sensitivity, or a c-reactive protein test. If that’s high, it makes sense to follow up with other inflammation tests, and then the recommendation at that point is to work with a specialist in inflammation to identify exactly what’s causing this: allergies, internal disease and more.

A second thing that AI has done is to go through your genetics and look at your propensity for a number of autosomal recessive diseases. It looks and says, do you have the genetics for Crohn’s disease, for example? If you do, then we’ve arrived at something we can address that shows up as inflammation.

Being Patient: At what point can AI help someone who isn’t having memory problems yet, but has a genetic link to Alzheimer’s? Or is AI better suited for someone who is experiencing memory loss and needs to understand how they can treat related conditions that could speed up the progression of Alzheimer’s disease?

Dr. Walker: The related conditions affect everyone. It affects you without being ApoE4 and me as well. A simple answer is people should start this process around age 55. Extensive charts show that if you have ApoE4, then some percentage of the population will start to see symptoms at age 60–65, and if you can get ahead of the train at age 55, you can push that curve further to the right.

There’s a class of drugs that increase brain fog. Brain fog is real; it can be scored and counted. There are hundreds of medications, and these are pretty common drugs; you can readily look and see how much brain fog they’re causing and how much they’re lowering your score.

Being Patient: How could artificial intelligence help people who have already been diagnosed with MCI or the early stages of Alzheimer’s?

Dr. Walker: One of the first steps in this is unwrapping what their current status is. You may be diagnosed or have taken tests that show you have MCI, but let’s unwrap a lot of things before we get there. It may be driven by things that can be undone. For example, there’s a class of drugs that increase brain fog. Brain fog is real; it can be scored and counted. There are hundreds of medications, and these are pretty common drugs; you can readily look and see how much brain fog they’re causing and how much they’re lowering your score, all of which is predictable on a standardized test. For each of those drugs, there are alternatives that are a good thing for your doctor to recommend and to say, instead of taking this, take this because it has a lower score. All of that is built into the system that says knowing your medications, we can start from that point. It’s surprising how common this is. It’s all well-documented; every drug is scored from 3, 2, 1 or 0.

Being Patient: Does AI take comorbidities like diabetes into account?

Dr. Walker: Yes, it does. First, don’t get diabetes. Akin to that is, don’t have a stroke. But that’s a lifelong struggle and both of those are pushing your brain chemistry in the wrong direction in terms of Alzheimer’s. The recommendation is, if you’re young, not diagnosed with diabetes yet and your A1C [a form of hemoglobin] is at 6.3, let’s do low-touch versions to get your A1C down to 5.6 and build a lifestyle that keeps it there for the rest of your life.

Being Patient: Where is the proof that AI and precision medicine are effective?

Dr. Walker: This is a relatively new field—the extensiveness of it. There’s a name for this field and when I started several years ago, there was no name, so it’s nice to be in a field that finally has a name. It’s called multidomain interventions. The leading group that’s doing this worked on the FINGER study. Now, there are FINGER studies around the world going on and they talk about giving your ApoE4, diet, lifestyle and so on, as well as expectations you can have. Kristine Yaffe at UCSF understands this process well and has her own track at the Alzheimer’s Association International Conference (AAIC). There’s a group we work with in New York; Dr. Richard Isaacson has an Alzheimer’s prevention center. All of these are well-known and well-documented techniques and in each of the cases, the amount of improvement and prevention you can affect are well-measured. 

Being Patient: How do people access artificial intelligence and how expensive is it?

Dr. Walker: I am a founder of a company called uMETHOD Health, so I do have a commercial interest here. Everything we do, our years of work, is covered by Medicare codes. So every aspect of screening, blood testing, your doctor getting your report and reviewing it with you and getting coaching on a regular basis—all of these steps have Medicare reimbursement associated with them. So this has been a delightful take on this new field called multidomain interventions. The U.S. government says, yes, this is real. That’s all new in 2018. You access this through your doctor. You can sign up through our website and we’ll point you toward an appropriate doctor, but all the interaction is done directly with your physician or neurologist.

Being Patient: Would a doctor have to prescribe this technique or be using the uMETHOD technology? 

Dr. Walker: That’s right, because of reimbursement and FDA regulations. We’re not practicing medicine on our own. We’re doing it with respect to your doctor.

Being Patient: What if you’re 65 or younger and don’t have Medicare?

Dr. Walker: Then you can pay for this individually. It’s on the order of hundreds of dollars, not thousands. The biggest challenge is if you end up with a set of expensive supplements and those can be more expensive than the report generation itself.

Being Patient: How much information do you need from us for the report, and what type of report do you get?

Dr. Walker: We ask for a specific blood test panel that includes a comprehensive metabolic panel and complete blood cell count. Those only cost $5 to the labs. Cholesterol is simple. But we ask for a longer list. We start with homocysteine. Homocysteine is a brain toxin readily treated with vitamin B12 and yet it might be the first thing to address in blood chemistry. We do an extensive hormone panel for males and females. We’ll look at thyroid extensively, vitamin deficiencies, zinc and selenium and things like that. It’s a bit longer test than what’s normally done on a regular annual visit, but every one of those has its contributors, drivers or follow-up questions. It’s about 100 items on the blood test.

We’ll ask for genetics and that can come in a number of ways: from 23and Me or AncestryDNA or Genoese, which will give you an entire genome for under $500. Another way to do this is with your doctor, who can do a cheek swab. The cheek swab looks at ApoE and knowing ApoE for this disease is a good place to start as well. A medication list is the next most important thing. What are you taking, why and what dosages? We’ll ask for your vital signs, and we have something called ‘The 35 Questions,’ which are follow-up questions, like do you snore a lot? We work down through the list of these things.

There are recommendations related to each of the things that we’ve seen and they’re prioritized. There’s recommendations on prescription medications you should be taking, supplements and extensive information on lifestyle and diet. I saw you had an interview on the MIND diet. So when the physician gets their report about diet, it’s about a page and a half. When the patient gets a report about diet, it’s about six or seven pages. Because it says, let’s start with the MIND diet, but then here’s all the changes we’re going to make to it because we’ve looked at your genetics and understand how you metabolize saturated fats. We know about your gluten sensitivity or how you respond to histamines. Don’t eat Swiss cheese, for example. Here’s how you respond to coffee or exercise. All of those things end up in the diet report. With lifestyle, we talk extensively about sleep, fasting and exercise, given where you are and what you’re capable of at that time.

Being Patient: Should people be worried about getting cut off by insurance when they get their results, if it becomes public information and shows they have a high risk for Alzheimer’s?

Dr. Walker: I hate to say this is beyond my field of expertise, but I’ve never heard those concerns from people. We end up with an extensive amount of data and my terms are that the data has to be protected better than the NSA. This is aggregated and de-identified, but it’s used to look at trending and machine learning in the future.

Being Patient: If people use uMETHOD Health, how long does it take to get results?

Dr. Walker: Once their blood has been taken, it’s normally just three or four days back to their physician. If they’re starting from getting their genetics test from 23andMe for example, that can take six to nine weeks. The way I describe this is that you can get your blood test, vitals and first report back, and then nine weeks later, we can add in the genetics and you can see how it changed or responded to what your genetics are. So it can be done in stages. But view it as essentially what is the same timeline as getting your blood drawn.

Being Patient: Does the AI become more accurate as you get more information from more people?

Dr. Walker: I wouldn’t have said accurate. I’d say a better understanding. We see more outliers and then spend time with the medical team to understand what this really means. You and I talked earlier about one of the things in a blood draw called hs-CRP; it’s related to inflammation. Whoever you are and at what age, get that done as part of your next blood panel. You should have a number under 1. A number under 1 says you have very little inflammation—great, keep it that way. A number above 3 says you have inflammation that may be leading you towards cardiovascular risk. We saw someone this week with an hs-CRP of 99; our target’s under 1. This result suggested we needed to stop and look at the entire vista of what this means. There’s a population out there that has unbelievable levels of inflammation; we sat and worked through the entire process of, we have to guide someone in this. I know the answer is to say, go back to your physician because you have high levels of inflammation, or your rheumatologist. But we have to look at, is this [chronic obstructive pulmonary disease], a blood disease or a reaction to an operation. There’s a fairly long list that we built into the system because we had never seen that situation before. Maybe it makes it more accurate for certain situations, but it improves the understanding at every level.

We also found that 60 percent of the females we see are depressed. That’s a jaw-dropping number.

Being Patient: How long has uMETHOD been taking in data?

Dr. Walker: We’ve been doing it commercially since April of 2015—about three and a half years.

Being Patient: What do you know today that you didn’t know when you first started?

Dr. John Walker: Number one and two are kidney disease and medications. The kidney disease was unexpected. When you get your blood drawn and get it back from the lab, there are a couple of numbers in there called creatinine and GFR. They’re talking about your kidney health, but they’re wrong. They’re wrong because your kidney health is related to your age and gender, and this creatinine level, but also your body: your height and weight. Since no blood lab ever asks you for your height and weight, they just make a best guess. They’ve missed part of a fairly substantial equation that needs to take your body mass into account. Your doctor could readily do that, but it’s a big, hairy equation. A high percentage of people we see have kidney disease and the kidney disease needs to be addressed before they take their next pill. I don’t want to recommend that they take a new prescription or supplements because each of those is now a bullet to the kidneys. So you have to unwrap kidney disease before you proceed with the next step. That was one discovery. Going back to what the report looks like, I said, well, it starts with prescriptions, supplements, lifestyle and so on, but it doesn’t do that if you have kidney disease. I want you to address the kidney disease because I don’t want you to take another drug until that situation is addressed.

We also found that 60 percent of the females we see are depressed. That’s a jaw-dropping number. How do we know they’re depressed? They tell us; we see they’re on an antidepressant; they take a standardized test. There’s a number of ways to see they’re depressed, but nonetheless, they’re depressed. Now I have to unwrap depression before we proceed with the rest of the things we need to address that look like other imbalances, since depression has been linked to cognitive decline. There are drugs that induce depression. An article in JAMA found 200 common drugs that induce depression, and showed how. Not only did 100 of those drugs induce depression; the other 100 induced suicide. If I see you’re depressed, I need to look at what medications you’re on that could be inducing depression. All of those examples are rules. All those rules have to be prioritized with what you do first, second or so on.

Now, wrapping genetics and depression together. Someone may tell us they’re depressed and we’ve looked at their medications. But just looking at genetics, I can see the genetics that may be predisposing someone to depression, which has a sense of “not my fault.” The reverse of that exists too that says, oh, you’re depressed, but it’s not caused by genetics. So both sides of that argument are worth understanding. We look at whether someone’s on the wrong antidepressant medication. You can look at the genetics for depression-inducing drugs or exercise, and see if people are doing the right exercise in terms of depression. All of those things are wrapped together and become part of the analysis.

Being Patient: Can AI and precision medicine help someone who is in the later stages of Alzheimer’s disease?

Dr. Walker: If some of this is related to medications and other comorbidities, those can be unwrapped. Otherwise, the true intention here is to start early, prevent cognitive decline as much as you can; the moment you think you are stepping into MCI or past that, that’s when to address these things aggressively. Your ApoE is your guide in terms of urgency. If you have ApoE4, it’s urgent, a slap in the face, that you need to address these things now.

This interview has been edited for length and clarity.

If you find our articles and interviews helpful, please consider becoming a supporting member of our community. Frustrated by the lack of an editorially independent source of information on brain health and Alzheimer’s disease, we decided to create Being Patient. We are a team of dedicated journalists covering the latest research on Alzheimer’s, bringing you access to the experts and elevating the patient perspective on what it’s like to live with dementia.

Please help support our mission.

2 thoughts on “Could Artificial Intelligence Catch What Doctors Miss? Precision Medicine and Its Role in Brain Health

  1. I have read so many articles or reviews regarding the
    blogger lovers but this post is really a pleasant piece of writing, keep it up.

Leave a Reply

We are glad you have chosen to leave a comment. Please keep in mind that comments are moderated according to our comment policy.