Part 2: Innovations in AI Amid COVID-19

Drug Topics®: Tim, thank you so much for joining me today.

Mok: My pleasure.

Drug Topics®: So what are some of the evidence based prevention strategies amid COVID-19 that have the ability to incorporate AI?

Mok: I think probably in terms of evidence basis of pre-loaded term, I think that, more or less, in the world of treatment because once the diagnosis is made, we can better understand and talk about science.

I think what we're seeing a lot with AI and thinking about how it can help with treatment: going back to some of the early identification, some of these mobile devices—your smartwatch or your fitness tracker—many of them actually have an oxygen sensor in there that can measure your oxygen saturation. Oftentimes the tech changes in oxygen saturation earlier before the onset of symptoms, or, even though you may have a diagnosis, these sensors can detect that you may be going into trouble into developing a more complicated or severe case before you feel it symptomatically. That is one area where we can see a lot of potential promise, in terms of notifying the individual patient, we can say, “Hey, we notice there's been a change in your oxygenation level, maybe you should get this checked out.”

Similarly, we certainly are seeing a lot of research interest for the patients who do have the diagnosis and get admitted to the hospital. You all may have heard that a significant portion of the mortality associated with COVID-19 are those who end up getting put on the ventilator in the ICU. What if, using AI, we're reading the records in the electronic medical records and reading those data, and then identifying patients who are about to experience these clinical demise and pick them earlier, instead of waiting until they have developed into a full blown respite failure?

If we can't identify those patients sooner, we can do other things that may have an impact on changing the trajectory of the outcome. We're talking here in the third week of June, I don't know when people are going to encounter this information. But early this week, we've seen some very positive results from the UK talking about the use of dexamethasone in changing outcomes or improving outcomes in patients with COVID-19.

Wouldn’t it be nice if we can actually utilize AI to identify those patients that may benefit the most from this treatment, so that we can administer them much earlier, and make sure that those patients have the best chances of surviving this?

Drug Topics®: So AI can also help decide how to distribute the different potential treatments.

Mok: I think that is a fair way to summarize that because right now, I think we are still in the early stages of understanding this disease.

We're even trying to just find out what drug may work. Perhaps some of these drugs may work better in one subset of patients versus the other set of patients. So those are things that we're still trying to learn right now. Certainly, everyone who has received some of these drugs will not have a good outcome. What is the difference between those patients who didn't have a good outcome versus those who do have a good outcome? We don't indiscriminately give these medications that have potential side effects.

Some of your audience of pharmacists are well aware that steroids can change your glucose metabolism. It can have a significant side effects as well. So, if the use of this drug is not likely to lead to a good outcome, why expose our patients or put them at risk for having some of these side effects. That's very important work that needs to be done in the next few weeks, as we understand this disease more and more, to identify who are likely to benefit and who are not.

Drug Topics®: You mentioned that pharmacists can use AI to identify side effects. What role do you see pharmacists having in implementing AI into their practice, during the pandemic and then going forward as well?

Mok: Great question. Some of the issues related to COVID-19 is that some of the patients will inevitably develop what we call secondary bacterial infections, because their body is being attacked by the virus. So as a result, the immune system is weakened. So a certain subset of those patients will develop other more common [ailments], like bacterial pneumonia, for example, or things that we see in hospitals all the time, maybe bacteria in the blood, stuff like that.

I think leveraging AI and then thinking about some of the early detection systems and natural language processing. Electronic Health Record contains a lot of data, including a lot of the laboratory results, vital signs, and a lot the patient status. A lot of human input, for example, when the nurse sees something or a doctor sees something on their physical exam, all that data is still captured in a format that is not discrete. In other words, it's not a yes, no, it is not a number. So a lot of these are just natural sentences that we will say this particular patient has a noted rash, or this patient is non-responsive, those are things that are quite tricky for what we would call a rule-based system to pick up. Using AI, using that natural language processing, where you can leverage AI to use some of that non-structured data, can drive very meaningful outcomes.

So, you mentioned earlier, how can pharmacists use this type of technology in improving the patient care? Well, if you have a toolbox with natural language processing, that can help them understand which of their patients experienced side effects are some of the treatments. I mentioned earlier, that dexamethasone sometimes can have significant side effects. So if they go in some sort of central nervous system reactions or steroids, these are not things that you can capture in the laboratory results.

So really, it will be dependent upon the ability of having the program algorithm to read the notes that the nurses are charting, or the doctors are recording so that the system can pick it up and notify the pharmacist: “Hey, this is a patient who may be experiencing side effects with that medication and maybe we need another pair of eyes on it, to see if there's an alternative treatment for that.”

Similarly, looking beyond the pandemic, how would that change the way that pharmacists would utilize technology to help their patients? I think, again, going back to the way that we think about that larger framework when we first talked about this: how can we identify patients who may potentially have some of these medication-related problems? Because of genetics, pharmacogenomics is another area where we hear a lot of interest and activity. So how can we leverage some of this best technology to read very complex information in the human genome, and unlock those sequences that have been correlated with additional side effects from the medications so that we can change the utilization of those drugs? Looking at the identification of the side effects once they have occurred? Are there any early signals that we can leverage to identify a patient who's about to have a kidney failure during the antibiotics before you can see the manifestation? Even if you didn't catch it 24 hours earlier, you can stop the drug earlier and stop the continual assault on the kidneys.

Those are areas that are very promising for pharmacists to be an advocate for the patient in managing their medication therapies.

Drug Topics®: Are there currently any impediments or challenges in being able to expand AI capabilities?

Mok: With regard to the obstacles and challenges, I think a lot of the health records especially, when we're going to be translating it from the hospital level to the individual clinics or outpatient areas where most of our patients are seeking the care, I think the interoperability is still a problem. We are seeing that if your primary care physician and your hospitals are not on the same electronic health records, they’re unable able to see a data element that is incomplete. That remains a challenge for the healthcare industry that we need to resolve to make sure that it doesn't matter if the information is on one brand of the electronic health record that needs to be easily translated into a second electronic health records. It's not dependent upon the patient's provider to be in the same ecosystem to get all the information.

The other thing that we need to have a more frank conversation about is with ourselves as healthcare professionals. I think there's a lot of hesitancies and fears about using a database or data science-driven approach to some of these innovations. So how can we reduce the friction where we are not positioning the technology as a replacement or as a competition to the prescriber or pharmacist autonomy, but rather, these are tools that will help them in the clinical decision-making so that they can make the best decision based on the best data available. I think that change management will be a big hurdle for us to overcome, both on the provider side as well as on the healthcare consumer side.

Drug Topics®: How do you envision AI functioning in healthcare in the near future?

Mok: We had talked a little bit already about the treatment and the management of these different diseases. I think precision medicine is gonna become more and more prevalent. I think people will demand that the treatment plan be tailored and be specific to you as an individual instead of, if you have this particular heart disease go through our drug 1, 2, and 3, and everyone gets put on it. I think people would expect and demand that that the treatment will be based on the best data available for your demographic, your age, your organ dysfunctions, the other diseases that you may have, other drugs that you're taking, so that the treatment will be much more personalized, and your treatment plan will be much more robust in terms of monitoring because as there are more and more events, medications become available in the marketplace. I think the management of these drug interactions and side effects are becoming more and more complex, so having that robust technology to help your pharmacists and physicians to identify any potential problems before they start, or if there is a problem that has started that they would expect that the identification will be much earlier in the course of the disease and this treatment plan.

There's a lot of promise, and certainly, a lot of applications that we have already been able to even deliver in the marketplace. One of the things that we're just wrapping up is a pilot project in being able to use an AI algorithm to identify patients who may be at risk for developing an infection that is common in a hospital called C. diff. We are so excited to see the precision and how much earlier our model can detect someone who may be at risk for having that. We can even identify those patients, even before they have symptoms. So a project that has really rapidly moved from a proof of concept to being piloted to being made available to customers as early as next month.

We're not really talking in the abstract here, we're talking about things that are coming to market and people will have access to these tools. I think that's why it excites me to have the opportunity to talk with you about things that we are putting in the hands of our consumers and use it to improve patient care.

Drug Topics®: Steve, thanks so much for joining me today and your continued work as well.

Mok: Thanks very much.