PodMed Double T is a weekly podcast from Texas Tech. In it, Elizabeth Tracey, director of electronic media for Johns Hopkins Medicine, 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. A transcript of the podcast is below the summary.
This week’s topics include bias in algorithms used to manage population health, new agents for cystic fibrosis, CBT versus antidepressants, and pain and weather.
Program notes:
0:50 Bias in algorithms
1:50 Impacting on black patients disproportionately
2:50 Variations on how the algorithm was trained
3:49 We can lower cost of medical care
4:49 Need to look at health
5:02 CBT and medications for treating major depression disorder
6:02 At five years CBT costs were lower
7:02 What do you prefer?
8:06 New medications for cystic fibrosis
9:12 About 70,000 persons worldwide with CF
10:10 90% of those with CF can benefit
10:27 Pain and weather and citizen scientists
11:20 Correlate symptoms with weather
12:06 I can tell the weather is going to change
13:17 End
Transcript:
Elizabeth Tracey: Can we detect racial bias in algorithms that are used to manage the health of populations?
Rick Lange, MD: Comparing the cost effectiveness of CBT or medications in major depressive illness.
Elizabeth: Really incredible news for many, many people with cystic fibrosis.
Rick: Using a smartphone app to determine, “Does weather affect chronic pain?”
Elizabeth: That’s what we’re talking about this week on PodMed TT, your weekly look at the medical headlines from Texas Tech University Health Sciences Center in El Paso. I’m Elizabeth Tracey, a medical journalist at Johns Hopkins, and this will be posted on November 1st, 2019.
Rick: I’m Rick Lange, President of the Texas Tech University Health Sciences Center in El Paso, where I’m also Dean of the Paul L. Foster School of Medicine.
Elizabeth: Rick, how about if we turn right to the journal Science. This was published late last week, but I thought it was so important I really wanted to talk about it on this week’s podcast. This was looking at the idea of bias that’s inherent in algorithms that are used to manage health of populations. The authors note in this extremely dense and academic article that these algorithms are already widely used not just in health care, but in lots of other places in our society in order to predict, “Hey, what’s going on and how can we manage things?” This issue of bias is particularly concerning to me right now because at Hopkins there are precision medicine models where algorithms are being employed, and I’ve been hearing an awful lot about what many use the phrase “garbage in, garbage out” to describe, where the data, how do we select the proper thing so we can create an algorithm that doesn’t have an inherent bias that’s already in it?
In this case, they were taking a look at, “Are there biases that are impacting on black patients disproportionately when compared to white patients?” They took a very, very large dataset working with a large academic hospital, and they identified all primary care patients enrolled in a risk-based contract from 2013 to 2015. They distinguished between black and white patients. They had just over 6,000 patients who self-identified as black and 43,000+ who self-identified as white. They had 71%+ who were enrolled in commercial insurance and about 29% who were in Medicare. On average, these people were about 51 years old and 63% of the sample were female. They basically said, “It’s reasonable to assume we can take these models and we can say, ‘Based on all of these different factors, you’re going to be a cost-intensive person, and therefore, we can create an intervention that could help keep you out of the hospital.'”
How well do these things really predict this and what happens with regard to race? To make a long story short, they basically found out that depending on how they trained the algorithm, there were really rather substantial variations in blacks or whites with regard to what kind of treatment and what percentage of people would actually be included in this program to reduce their risk. They come to a conclusion that I think you and I would both ratify, which is that what we really need is some holistic way of measuring health, not necessarily something that just pinpoints specific conditions. They say they think they created a few models that could actually work with regard to overcoming this racial bias.
Rick: As you mentioned, they picked the 2.5% of patients that were the most expensive. Their hypothesis was they must be the ones that are most ill, and if we can address them, if we can make them better. We can lower the cost of the medical care to the practice or to the insurer. Actually, for the same risk prediction, the African Americans or the blacks, had more significant burden of illness and yet their costs were lower. You’d say, “That’s counterintuitive.” Their costs may have been lower because maybe they didn’t have access to care because they didn’t have transportation or they had daily jobs or they weren’t as educated. These things inherently provide a bias. It’s not intentional, but it’s systematic and it needs to be addressed.
Elizabeth: I think it’s especially important that we address this as we are employing these artificial intelligence and machine learning models more and more in health care to help with these interventions.
Rick: The reason why these authors were able to point this out is they did this information from a database that has about 200 million people involved in it. They were able to look at the input, the output, and the outcomes as well. As you mentioned, if we just look at costs, we’ve missed the picture. We need to look at health. Interesting article, difficult to read, but we need to recognize that unintentionally there may be biases that we need to address, especially if we’re going to address the sickest individuals that most likely needed health care. Where do you want to go next, Elizabeth?
Elizabeth: That’s your choice. Where would you like to go?
Rick: Let’s talk about CBT, which is cognitive behavioral therapy, and pharmacology, specifically the medications for treating major depressive disorder in the United States. We’ve talked before about this. We’ve covered this over the almost two decades we’ve been reporting. I think most of our listeners would recognize in comparison studies cognitive behavioral therapy and pharmacology for the initial treatment of major depressive disorders or illnesses are about equal. But we really haven’t done an economic analysis to say which of these is the most cost effective, and that’s what these authors did.
They looked at adults with newly diagnosed major depressive disorder in the United States. They actually looked at relative effectiveness data from a meta-analysis of randomized, controlled trials. These data not only had clinical data, but also the economic data. They followed these individuals over a one- to five-year period. Ultimately, here’s what they found. They were similar in terms of efficacy. Over the course of one year, the CBT was a little bit higher. But at five years, the costs were actually lower, about $1,800 lower. This is counterintuitive because oftentimes insurance companies don’t offer cognitive behavioral therapy because it’s thought to be more expensive. It is in the short term for one year. But over five years, not. Part of that is because with cognitive behavioral therapy, you’re teaching individuals how to deal behaviorally with their major depressive disorder. The medications you don’t do that. Over the course of a longer term, CBT is more cost effective.
Elizabeth: Did we mention this is in Annals of Internal Medicine? I find this to be supportive of things we’ve talked about in many, many other arenas where we’ve seen the efficacy of CBT and helping people really cope with lots of different things.
Rick: There are places that may not have access to people that are trained in cognitive behavioral therapy. From a doctor standpoint, it’s easier just to give a medication, but I think what this article points out is we need to make sure that CBT is available to individuals. In terms of the shared decision-making, we allow the individual to decide, “Do you want to be treated with medications or do you prefer to have CBT?” Making CBT available needs to be a priority and the payers need to realize that over the long term it is more cost effective than pharmacotherapy.
Elizabeth: I would ask you to speculate on a couple things. One is does the pill work faster or are they about the same? The other one is what do you think about a model that would deliver CBT via telemedicine?
Rick: Two good questions. First of all, yes. The medications do work quicker. In terms of administering CBT, there are a lot of different ways to do it. There’s individual. There’s group therapy as well. There is some evidence, although it’s not conclusive, that it can be delivered via telemedicine. We’re very interested in that in rural west Texas where we have many counties where there’s no health provider at all, let alone someone that gives CBT.
Elizabeth: Ultimately, I would wonder about a model that might administer medicine concomitantly with CBT with the idea that the medicine would be tapered off.
Rick: There are many hybrid ways of approaching this.
Elizabeth: Let’s turn to the New England Journal of Medicine. This one is great news, a celebration for nearly 90% of patients who have at least one copy of a common mutation in cystic fibrosis. This was a study that employed three different agents — elexacaftor, tezacaftor, and ivacaftor — for cystic fibrosis with people who, as I said, have at least one copy, but also people who had two copies of this mutation that’s common in cystic fibrosis. There were a total of 403 patients who underwent randomization and received either this treatment or placebo. They basically, I have to say, found this was really effective at dealing with the impact of cystic fibrosis on their lives, which is really huge.
As we know, many people with cystic fibrosis will die when they’re in their 30s. In order to manage much of the symptomatology of CF, I’ll abbreviate it, they have to have a number of rather uncomfortable and unpleasant treatments. They’re susceptible to lots of infections, and there’s about 70,000 persons worldwide today who have cystic fibrosis. This combination treatment is a phase 3, multicenter clinical trial that documents that it actually is effective in this. It’s just an amazing piece of news for those people.
Rick: It was literally 30 years ago when the genetic defect in what’s called the CFTR, that is the cystic fibrosis transmembrane conductance regulator protein, was identified. This protein affects how water and salt are transported across various membranes. What these small molecules do is they allow the protein to be refolded correctly, to be transported, and then, by the way, they potentiate the effects as well. The improvement was rapidly seen. These were administered for 24 weeks, but at the first four weeks, people felt better and the amount of chloride transported across these membranes significantly improved as well.
Elizabeth: I would also note that there was another paper that was published in the Lancet that also documents this same really stellar impact of this combination. This is really great news for people. For those 90% of people with cystic fibrosis, I guess we’ll keep working on the 10% who still need help.
Rick: Kudos to the NIH investigators who have been working on this for three decades and the companies that have been partnering with them.
Elizabeth: Let’s turn now, then, to yours from a journal that we have never talked about before. That’s Digital Medicine and citizen scientists, a concept that I absolutely love, using a smartphone app and participating in this study.
Rick: The thing I really enjoyed about this was not so much the results, which are kind of interesting, but how the study was done. They’re citizen scientists. Here’s the question: patients with chronic pain commonly believe their pain is related to the weather. Is that true or not? Let’s create an app. Let’s call it Cloudy with a Chance of Pain. What this app does is it tracks symptoms. People can talk about their pain severity and rate their fatigue, morning stiffness, impact of pain, sleep quality, their physical activity, and all these things. By the way, since it’s on a smartphone app and we can GEO locate where individuals are, we know what the weather is during that exact same time, so that allows us to correlate symptoms with weather.
When the individuals did that, they had about 10,000 people initially sign up for the app and give all the information. A little under 2,700 collected enough information they could actually analyze. What they discovered was that there was a modest correlation between pain and weather, specifically humidity, wind speed, and atmospheric pressure. They were all associated with increased pain severity. If you had to put it into a percentage, it looks like the weather increased the odds of a pain event by just over 20% compared to an average day.
Elizabeth: 20%. That’s not bad. Is that worth an acetaminophen or an ibuprofen?
Rick: It’s worth kind of answering the question because our older individuals say, “I can tell when the weather is going to change because my pain gets worse.” It’s not a perfect study, but I do think what we’re going to find is we’re going to be using applications to tell us more about disease, disease processes.
Elizabeth: I guess the one piece that I’d like to add to this one would be an age piece because, as you already mentioned, it does seem to be people as they age that they start to complain about pain and the weather. I think that would be a really revelatory piece of data that I suspect is not in here. I would have to say I’d be happy to participate in stuff like this as long as it’s anonymized or I don’t feel like Big Brother is watching me more than Big Brother already is.
Rick: It’s interesting. The mean age of the individuals in here was about 51 years old. You can imagine there probably weren’t very many 80- and 90-year-olds downloading the app and using it.
Elizabeth: But watch out for bias. On that note, that’s a look at this week’s medical headlines from Texas Tech. I’m Elizabeth Tracey.
Rick: I’m Rick Lange. Y’all listen up and make healthy choices.