COVID-Flu Vaccine; AI for Diagnosing Diabetic Kidney Disease

TTHealthWatch is a weekly podcast from Texas Tech. In it, Elizabeth Tracey, director of electronic media for Johns Hopkins Medicine in Baltimore, and Rick Lange, MD, president of the Texas Tech University Health Sciences Center in El Paso, look at the top medical stories of the week.

This week’s topics include large language models and discharge summaries, treating female urinary incontinence, using artificial intelligence (AI) to diagnose diabetic kidney disease, and a COVID/flu vaccine.

Program notes:

src:4src COVID and flu vaccine

1:4src More immunogenic response

2:4src Female urinary incontinence treatment

3:4src Midurethral sling or botulinum toxin (Botox)

4:4src Doesn’t require surgery

5:4src Combined therapy group

6:1src Large language models and discharge summaries

7:1src 1srcsrc hospital encounters

8:1src Maybe provides a draft

9:srcsrc AI and diagnosing diabetic kidney disease

1src:src1 Large number of retinal images to train

11:src1 Can actually see blood vessels

12:29 End

Transcript:

Elizabeth: Can AI help us make the diagnosis of diabetic kidney disease?

Rick: Using large language models to provide a hospital discharge summary.

Elizabeth: What’s the best way to treat urinary incontinence in women?

Rick: And a multicomponent vaccine that takes care of the flu and COVID.

Elizabeth: That’s what we’re talking about this week on TTHealthWatch, your weekly look at the medical headlines from Texas Tech University Health Sciences Center in El Paso. I’m Elizabeth Tracey, a Baltimore-based medical journalist.

Rick: I’m Rick Lange, president of Texas Tech University Health Sciences Center in El Paso, where I’m also dean of the Paul L. Foster School of Medicine.

Elizabeth: Rick, with vaccines, so much in the news, why don’t we turn to JAMA and take a look at this study looking at this combination vaccine?

Rick: Despite the fact we’re not doing a bunch of talking about either the flu or COVID, probably important that our listeners understand there are approximately 1 billion influenza cases and 2.3 billion COVID cases annually worldwide.

Now the current recommendation is to provide vaccines for both. These investigators have actually developed a vaccine. It’s investigational. It’s an mRNA-based vaccine that is called multicomponent. It combines certain antigens from the flu, the hemagglutinin antigen, and certain antigens from the COVID virus. These are the spike proteins.

Is it as effective, or perhaps even more effective, than using the standard vaccines? These investigators took over 8,srcsrcsrc participants and they administered either routine vaccines or this investigational mRNA vaccine. What they discovered is that the multicomponent vaccine actually produced a more immunogenic response. For those individuals age 5src to 64, when they looked at four different flu strains, this multi-component was more immunogenic, and for individuals over 65, it was more immunogenic in three of the four different blue strains. So this is very promising.

Elizabeth: This is a promising development and, of course, we know that people would prefer to have a single visit versus multiple visits in order to get up to date on their vaccines.

Rick: It is. When you get something that’s more immunogenic, you have more side effects. So it had slightly more side effects — injection pain, a little fever, a little aches, but nothing serious at all. The value of this particular vaccine is because it’s mRNA, we can do different antigens as the flu and COVID evolve, and we can do it fairly quickly.

Elizabeth: That’s right, and that’s way better than raising vaccines in eggs. That time-honored, but antiquated, I would point out, method for raising flu vaccine.

Rick: Well said.

Elizabeth: Thank you. So since we’re in JAMA, let’s talk about this issue of female urinary incontinence, which is just a ginormous issue.

I didn’t realize that up to a third of women, as they age, actually have an issue with urinary incontinence. There are two different types. There’s urgency and then there’s stress incontinence, where urine can leak with effort sneezing or coughing, and a large number of women have a mixed urinary incontinence issue. So that’s what this study is addressing.

We’ve talked about this issue before. There has been just a dearth of really effective treatment for this. I would also note that the authors say that this is an industry — this attempt to try to deal with this with different products on the part of women — that’s worth literally billions of dollars a year.

What they were looking at here was Botox injection into the urinary bladder in the detrusor muscle or what is called a midurethral sling procedure. How did they compare to each other? Was Botox noninferior to the midurethral sling procedure?

They enrolled women who had moderate to severe mixed urinary incontinence, who had had previous unsuccessful conservative treatments and oral medications. They were able to get data on 137 of the patients who were treated.

What they showed was that the Botox injection and the midurethral sling were each of them fairly successful in helping to ameliorate some of the symptoms, but not entirely. And that a lot of these women crossed over after the study and received the other treatment, and still neither one of them is 1srcsrc%, and in combination they’re not 1srcsrc% either.

Rick: The value of Botox is it doesn’t require surgery. It does require a cystoscope so that you can give the injection into the muscle. The other surgery is usually it’s an outpatient surgery. It’s usually three small incisions. We thought that the Botox would only be useful for the urgency incontinence, but it seems like it works for the stress as well. Somewhere between 2src% and 3src% of patients who have either one of these procedures are going to need to have another procedure. The patient can decide which procedure they want, giving them all the data, because these are equivalent in many ways. By 8src years of age, 2src% of women usually have surgery for stress or mixed urinary incontinence. This is, again, a big issue not frequently talked about.

Elizabeth: Exactly, nobody talks about it. Let’s just note that in this study over 3src% of the folks in the sling group ended up getting Botox and almost 16% in the Botox group also elected to have a sling. The editorialist notes that the pathophysiology of stress and urge incontinence are likely intertwined, which is what accounts for the fact that they were both impactful to some degree and a significant degree in either of these types.

I would say one thing that emerged that was concerning was the combined therapy group had more patients who required self-catheterization and more UTIs [urinary tract infections] ultimately. Those, of course, can be very problematic in women as they age.

Rick: Right. I think this is the kind of thing where the physician sits down with the patient, lays it all out there, and lets the patient be involved in the decision-making process, knowing that if they choose one particular method and it’s not successful, the other method is still an option.

Elizabeth: Right. That’s the positive news for sure. Let’s move on to your next one that’s in JAMA Internal Medicine.

Rick: When a patient is in the hospital, caring for them in the hospital and getting them out of the hospital is only the beginning of their treatment. Because after they leave the hospital, usually they have continued care as a result of whatever brought them in and they have other physicians they’re going to be seeing. And so the question is how do you get that information from the hospital into the hands of the physicians that will be caring for them afterwards?

Usually, what happens is the physician dictates a discharge summary. They go through all the records. They try to summarize what brought the patient in, what tests they did, what kind of follow-up needs to be done, medications. Quite frankly, it’s tedious. It’s necessary and it’s important, but it takes time. And if you have multiple physicians caring for a patient — let’s say you start caring for the patient, then you leave and I’m caring for the patient, then I got to go back and figure out what was done. And so these discharge summaries, it’d be nice if there was some other way of doing it.

Well, large language models provide an opportunity to support physicians by doing that. The question is, are they as good? What these physicians did was they looked at 1srcsrc different hospital encounters. They had the large language model provide a discharge summary and they had the physicians provide a discharge summary. And they asked some simple questions. What was the overall quality? How comprehensive was it? How concise was it? Then how accurate was it? They had an independent person or two different people grade them.

When they compared the two, overall quality, reviewer preference, comprehensiveness was the same between the two. And the large language model was actually a little bit more concise. Unfortunately, it had a few more errors. They were usually an error of omission.

What were the caveats of this? Well, these were patients that were in the hospital for a relatively short period of time, from 2 to 6 days, and it wasn’t patients in the ICU [intensive care unit] where things can be a little bit more complex. But overall, I think it does show that this holds some promise.

Elizabeth: The editorialist, of course, opines that this offers us the opportunity for further study and for further development of these models.

Rick: Right, and I tell you what. As a physician, I’d love to have this study refined. Now, maybe what it does is it provides a draft that allows me to review it.

Elizabeth: As you know, I’m currently involved with a large language model for the development of an app that’s going to be doing some editing. What’s interesting to me are sort of two things I would put on the wish list. One would be that it retains a sense of humor. And the other thing I would ask for is, does it ask the question, does it know what it doesn’t know? Is there a way to implement that, particularly in scanning a discharge summary?

Rick: The large language model only has access to patient information it can see and oftentimes the physician knows things that are not in the medical records, so they may have access to things and provide a view and a perspective — and a little bit of humor, by the way — that the large language model can’t.

Elizabeth: Right. Finally, let’s turn to The Lancet, speaking of large language models and machine learning.

This is a look at whether an AI-driven model is able to take a look at people’s retinas and determine whether they have kidney disease that’s related to their diabetes or whether it’s non-diabetic kidney disease. The rate of developing diabetic-related kidney disease is quite high and a lot of these folks also have this non-diabetic kidney disease, and they’re treated differently.

The other problem with trying to assess this difference is the need for a biopsy. This issue of the retina… I’m starting to believe, you know, we have that adage about the eyes are the window of the soul, and it seems clear that the retina is really the window of health. We’re looking at the retina for so many different things now and this, sure enough, is a retinal image-based AI deep learning system. They pre-trained it using over 7srcsrc,srcsrcsrc retinal fundus images, and then for the diabetic kidney disease detection, they used almost half a million retinal images from 121,srcsrcsrc+ participants to train this thing. They also used data sets from China, Singapore, Malaysia, Australia, and the U.K. for external validation.

So they tried to account for the things that we would call inherent bias, I think, and then they used 1,srcsrcsrc+ retinal images from 267 participants for both their development and internal validation. What’s this look like if we try to differentiate isolated diabetic nephropathy from non-diabetic kidney disease with 4 to 6 years of follow-up?

This model was able to detect about 84% of that area under the curve, which is pretty good, and then they achieved almost 9src% on their internal validation data set. They could also detect diabetic kidney disease with higher sensitivity, almost 9src% versus about 66% with that model.

Rick: For our listeners, why the retina ends up being important is that’s one of the few places you can actually see blood vessels. Diabetes affects the blood vessels. What they’re saying is, if you’re likely to have eye disease from blood vessels involved from diabetes, you’re likely to have kidney disease, and those go hand in hand. Your ability to see things clearly depends upon what kind of camera you’re using to take pictures of the eyes.

Elizabeth: These authors speculate this further development of potentially cheap mobile phone-based fundus cameras that would enable this technology to be used all over the place.

Rick: They even mentioned that the accuracy of the determination will depend a lot on the pictures that are taken.

Elizabeth: I would just note for you that we’ve talked about studies where examination of the retina for diagnosis of Alzheimer’s disease, for progression of MS [multiple sclerosis], there have been multiple applications of this. I suspect it is going to get to the cell phone.

Rick: It’s really the only place you can look very closely at blood vessels, take a picture, and store it and compare it over time, and that’s the value of the retinal scans. If you have a demyelinating disease or a disease that can affect the neural apparatus, oftentimes it can be seen, as you mentioned, in the eye.

Elizabeth: More of this co