skip to Main Content

Dr. Faith Fitzgerald once quipped that prognostic modeling is the “punctilious quantification of the amorphous.”  She has a point.  Prognosis is inherently uncertain.  As Alex Lee says on our podcast today, all prognostic models will be wrong (in some circumstances and for some patients); our job is to make prognostic models that are clinically useful.  As Sei Lee notes, the argument for developing prognostic models has won the day, and we increasingly use prognostic scores in clinical decision making.  What makes prognostic models for mortality different from models used for anticoagulation or risk of renal injury?  James Deardorff replies that there is something inherently different about predicting mortality.  Death is different.  For some reason clinicians who might be perfectly comfortable using an anticoagulation risk calculator might be skeptical of a mortality risk calculator (see this recent terrific JAMA IM study from Nancy Shoenborn on this issue).  And yet, the only thing that may be worse than a prognostic calculator is a clinician relying solely on their clinical intuition.

Today our guests Alex Lee, James Deardorff, and Sei Lee, talk to us about the uses, limitations, and clinical use cases for prognostic models.  As a springboard for this conversation we discuss new prognostic models developed to predict (simultaneously) mortality, disability, and mobility impairment (Alex Lee first author, JAGS) and mortality for people with dementia residing in the community (James Deardorff first author, JAMA IM).  

Both new models are now available and free to use on ePrognosis.  

And Sei and Eric reminisce about slow dancing to “Forever Young” by Alphaville in their teenage years.



Eric: Welcome to the GeriPal podcast. This is Eric Widera.

Alex Smith: This is Alex Smith.

Eric: And Alex, we’ve got a full house today.

Alex Smith: We have a full house today and we are delighted to welcome Alex Lee to the podcast. She’s an epidemiologist and assistant professor of medicine in the UCSF Division of Geriatrics. Welcome to the GeriPal podcast, Alex.

Alex Lee: Thank you. Happy to be here.

Alex Smith: And we’re delighted to welcome back James Deardorff, who is a geriatrician and research fellow in the UCSF Division of Geriatrics. Welcome back to GeriPal, James.

James: Thanks for having me.

Alex Smith: And we’re delighted to welcome back Sei Lee, who is a geriatrician and palliative care doc and is professor of medicine in the UCSF Division of Geriatrics. Welcome back, Sei.

Sei: Howdy.

Eric: So we’re going to be talking about prognostication and prognostic indices. But before we jump into that topic, James, I think you have a song request.

James: Yes, I do. I have a request for Forever Young by Alphaville.

Eric: I love… I actually just went to New Order and Pet Shop Boys. So the fact that you’re bringing up synth-pop warms my heart.

Alex Smith: Were you alive when this song came out? [laughter]

James: Definitely not-

Eric: 1982.

James: … alive in the 1980s, but yes, you sang Forever Young by Bob Dylan around three years ago, and then Eric jokingly started singing this song, and so now it’s about time to fulfill Eric’s request.

Alex Smith: Sei, do you have any comment?

Sei: I was thinking that this shows James’s true older soul at heart, even though chronologically, he may be a lot younger than me. Because this song is near and dear to my heart because this was my first slow dance when I was in… I think it was ninth grade. It was kind of the junior high party. So this is a song that is kind of burned into my brain.

Eric: Yeah, this is definitely… If you grew up in the ’80s, this is the slow dancing song. Alex, give us a little bit.

Alex Smith: Yeah, all right. Here we go.

Eric: Alphaville.

Alex Smith: There is no guitar in the Alphaville version.

Sei: That’s what I was thinking.

Alex Smith: No guitar at all. So this is my rendition, with generous support from people who’ve covered it on YouTube.

Sei: It is also very high, so I am very curious to see how you’re going to hit the notes.

Alex Smith: (Singing).

Eric: That was awesome, Alex.

James: Well done.

Eric: James, why’d you pick this song?

James: Well, I think today we’ll be talking about how it might not be possible to live forever. And I mean, also, it was German Day in Golden Gate Park yesterday, so-

Eric: Oh, it was?!

James: … 1980s German synth-pop sounds great.

Eric: Well, I think going back, I doubt that me and Sei, while we were slow dance… Me and Sei weren’t slow dancing together [laughter] but the fascinating thing about German synth-pop in the ’80s, we dance it, but it’s really about nuclear annihilation. Like Forever Young is about that prospect of dying in a nuclear Armageddon, which is like 99 Luftballoons.

Alex Smith: That’s what it’s about?

Eric: Yeah.

Sei: Absolutely.

Eric: Read the lyrics.

Alex Smith: Well, here’s some from the end, we’re going to sing the second verses and, “It’s so hard to get old without a cause. I don’t want to perish like a fading horse.” One wonders if there is some translation issues from the original German.

Eric: Hoping for the best, but expecting the worst. Are you going to drop the bomb or not?

Alex Smith: Okay.

Eric: Yeah.

Sei: Yeah, I think especially in Germany where they had such a heavy U.S. military presence, being so close to the Warsaw Pact.

Eric: Yeah. 99 Red Balloons also.

Sei: Absolutely.

Eric: Have we done… No, I think we have done 99 Red Balloons.

The German version I want to hear Alex do.


Alex Smith: That’s not going to happen. [laughter]

Eric: But that’s not our topic today, is it? I could talk about ’80s synth-pop all day. By the way, Pet Shop Boys in concert, amazing. Yeah. So we’re not going to talk about Pet Shop Boys. Anyways, we’re going to talk about prognostication and prognostic indices. James, I’m going to start with you since you picked Alphaville, and thank you for that. Why is prognosis important when we’re thinking about caring for older adults?

James: Yeah, so I guess just to start… I mean, prognosis in general is kind of a term for predicted course of disease, and I think people commonly use it to refer to an individual’s life expectancy or how long the person has to live. I think this is a really difficult topic to talk about with patients. And oftentimes, I mean, patients and clinicians to be appropriately skeptical about our ability to predict the future, and an individual’s life expectancy can change dramatically. But we know from previous studies that a lot of older adults actually want this information, but might not feel comfortable bringing it up.

And really providing these estimates of how long a person has to live affects a lot of the decisions we have to make clinically. So if you can think about certain decisions like continuing to screen for certain cancers like colon cancer or breast cancer, or how strict you should control diabetes or blood pressures.

We know that a lot of times with cancer screening, you don’t actually get benefits right away and it can cause potential harms right away. So we want to make sure that we’re targeting these interventions to those most likely to benefit.

So I mean, say Leah’s done a lot of work on a topic of time to benefit. So for colon cancer, it takes about 10 years before you start to see a mortality benefit for cancer screening. And if you have an older adult who might have a life expectancy in the range of two to three years, it might not make sense to continue screening for cancer. That’s a difficult decision to make, and providing these estimates can help clinicians and patients with that type of shared decision-making.

Eric: So maybe it would be helpful as we just think about this, because the topic of time to benefit and thinking about how does prognosis fit into that, maybe we can just go around kind of quickly and just mention what decisions that we think about involve time to benefit. I know there was a huge colonoscopy study that just came out in the New England Journal of Medicine. We won’t talk about that, but James mentioned colon cancer screening, which includes flex sig and other forms. Generally we’re looking at a time to benefit of… Let’s say you did this study about 10 years?

James: Yeah.

Eric: What other decisions? I’m going to turn to Alex. Alex, what other decisions? Alex Lee.

Alex Lee: Sure. For diabetes example, a lot of the diabetes prevention that the controlling blood sugar works for is for things like preventing chronic kidney disease or dialysis. Those have a very, very long time horizon. Someone who’s healthy in their 60s and just got diagnosed with diabetes, yeah, they might benefit, if they have a very long life expectancy, to prevent that decline in kidney function. But for someone who already has significant cardiovascular disease or other comorbidities, they might have a shorter life expectancy, and so the benefit that they would get from lowering their blood sugar from, say, an A1C of eight to an A1C of seven, that might improve their kidney function if they lived another 20 years. But if they’re not going to live another 20 years, then there’s no point in giving them the extra medication for lowering their A1C. So I think that’s a great example of how time to benefit is really important in terms of comparing that with the expected life expectancy that a patient has.

Eric: Sei?

Sei: Yeah, I think one example would be more intensive blood pressure treatment. And this time to benefit issue I think really becomes really important when the treatment leads to immediate risks, and then the benefits happen later.

So for example, for blood pressure treatment, we know that the risks for the blood pressure becoming too low and leading to lightheadedness and sometimes even falls and fractures, we know that that is much more likely when you just start a new blood pressure medicine or intensify a blood pressure medicine in the first 30, 60, 90 days. So there’s an immediate risk, but we also know there are delayed benefits, and the delayed benefits of decreased heart failure exacerbations, decreased strokes. They’re so really important benefits, but the benefits are a little bit later.

So that time to benefit, as best as we can tell, is somewhere in the one to two year range, and we’re still kind of working out exactly what that is. So if somebody’s life expectancy is substantially shorter than that, then the chances that they’re going to be exposed to all the risks with a low likelihood of getting the benefits, that’s less likely that we should do that for them.

And I would just like to point out one kind of tangent is Alex Smith is one of the people who’s done work on how there’s a lot of heterogeneity among older adults on how much they want to hear about their own prognosis. And certainly from my experience, there have been patients who when I have kind of cautiously broached this topic, it’s kind of like the floodgates open and they’re like, “Thank God somebody is talking to me about this. My best friend just died, and this is something that they’ve been thinking about and nobody is wanting to talk to them, because everybody is so…” And then there are other patients that I’ve talked to and they’re like, “I don’t really think you know, or anybody can know,” and they’re not interested in talking about prognosis. But there is this large heterogeneity, and I think it behooves us as clinicians to actively engage with patients and really get a sense of is this something that they want to talk about, and try to make the best decisions for them, given their values and preferences.

Eric: Alex Smith, you’ve got another example?

Alex Smith: I’ll build on what Sei just said and go with a non-clinical example. That is when we interviewed a diverse group of 65 older adults who were living with disability in English and Spanish and Cantonese, we found that two-thirds of them wanted to know if their doctor thought that they had less than five years left to live, and it was for a whole host of reasons. Most of which, almost all of which were non-clinical, right? They wanted time to prepare, to get their spiritual house in order, to get right with God, to think about moving near to the grandkids, to get their finances in order, to prepare their wills. It was all about preparation, preparation, preparation.

I think that’s something that is probably underappreciated as we’re thinking of, “Do I disclose prognosis to patients in the context of a clinical encounter?” And we are so focused on these clinical decisions and how prognosis is important on clinical decisions, and yet older adults, patients, our patients may want to know for a very different set of reasons. Eric, you’re up.

Eric: Well I’m going to actually switch topics and ask Sei, because Sei, there is a website that I heard about called ePrognosis that has a list of time to benefit not from what Alex was talking about, from a clinical decision standpoint. You want to describe what that website is?

Sei: Yeah, so this is something that Alex and I have been kind of, I guess, working on on the side for the past 10, 12 years. ePrognosis started out as just a list of different prognostic indices, so things to predict mortality in hospitalized adults, things to predict mortality in community dwelling adults, things to predict mortality in nursing home residents. And we’ve been slowly building that out.

Really the two things that we felt like really needed to be built out was number one, focusing on patients with dementia, because it’s such a common population and really, we have good studies to show that the trajectory of decline in dementia is pretty different than almost any other disease. So we thought kind of focusing specifically on dementia patients was important. And then the other thing that we really wanted to do was to come up with an index that looked at things other than mortality, because older adults, as Alex mentioned, care about a lot of things, and the clinical decision-making is probably less than half of what they care about. So thinking about other things like the mobility impairment, the ability to take care of themselves, live independently, those sorts of things, outcomes other than mortality we thought were really important.

Eric: And we’ll get to both of those, but can you describe the time to benefit page?

Sei: Yeah. So the time to benefit page is trying to make the mortality indexes link more closely to the time to benefit. So we have a list of various interventions like hospice referral, like more intensive treatment for blood pressure, more intensive treatment for diabetes, for dysglycemia, hyperglycemia, statins, bisphosphonates. All of these different preventive medications which we highlight the time to benefit, so that if somebody’s estimated life expectancy is greater than the time to benefit, then that intervention would be recommended. Whereas if somebody’s life expectancy is less than the time to benefit, then that intervention would generally not be recommended.

Eric: Alex Lee, Sei was bringing up prognostic indices and all this stuff. Why use prognostic indices? You’re on mute.

Alex Lee: Sorry. One reason to use prognostic indices is really to kind of standardize prognosis across clinicians. I think there can be a lot of variability across different clinicians. Maybe their experiences, what sort of comorbidities they see. Maybe an endocrinologist has a different opinion about what life expectancy might be compared to a geriatrician compared to an oncologist. And maybe they’re all even seeing the same patient, but they could have different ideas of what that prognosis looks like. So that’s one reason.

I think also there’s been studies that show that clinicians often have a biased view of what prognosis, how long people have to live. I think the more that they feel connected to the patient, it’s often that they’ll kind of overestimate, optimistically, the person’s life expectancy. So I think prognostic models are very important when we’re doing evidence-based medicine to really kind of standardize the playing field and make it so that these are objective projections of what someone has left in their life.

Eric: Okay, I’ve got a question for Alex Smith. Alex, way back when you published a paper in JAMA on prognostic indices with Lindsey Yourman, and the editorial was rather harsh on prognostic indices. Do you remember what they said?

Alex Smith: It wasn’t the editorial. Tom Gill wrote the editorial.

Eric: Okay, it wasn’t Tom Gill.

Alex Smith: He was very kind and supportive. But we did hear from former UCSF faculty member and I think she was the dean at UC Davis at the time, Faith Fitzgerald, who said these prognostic indices, they’re the punctilious quantification of the inherently amorphous, right? How can you… It’s this attempt to tidy up, systematize, quantify what is inherently unknowable.

We appreciate that criticism, because prognosis, like the weather, is inherently uncertain. There is tremendous amount of uncertainty not only in these prognostic models, as Alex Lee and James will talk about, as well as uncertainty of whether it applies to that patient who’s sitting right in front of you. And yet our rejoinder would be, “Well, it’s probably better than the clinician’s estimate,” which studies have consistently shown are off, right? By as much as a factor of five, right? Nicholas Christakis’s landmark series of studies in this area.

So yes, it’s imperfect and we recognize that, and we’re always clinically making decisions that are based on imperfect information. So the idea is that we’re reducing as much as we can the uncertainty, acknowledging that it still exists and acknowledging that with our patients.

Eric: Well, let me push you on that. We have a fair amount of certainty. I know, based on our lyrics that we just did, we are not all going to be forever young. I also know that everybody who is listening to this podcast or who’s on this podcast will be dead in 120 years. I feel like what we’re often talking about is where our confidence intervals are as far as the likelihood that somebody will survive a particular area. I’m pretty sure everybody’s going to make it through this podcast today, but it’s around those confidence intervals. Sei, what do you think about that?

Sei: Yeah, I mean, I think there’s the Hemingway quote about all stories continued long enough end in death, and he is no true storyteller who will keep that from you. Kind of similarly, I’ve made the remark in various forums to trainees about how, “Yeah, a survival model, if you’re going to go out 150 years, the survival model is going to be pretty boring because everyone is going to die and the only factor that really is going to matter at that stage is going to be age.”

I think fundamentally, I will say that I think Dr. Fitzgerald’s comment probably has not won the day, because increasingly what we are seeing is more and more clinical decisions that are tied to prediction. So the two examples that I would talk about are treatment for osteoporosis that hinges on the 10 year risk of fracture, and then treatment for hyperlipidemia that hinges on predicted 10 year cardiovascular events.

So the idea here is that generally the risks of bad things, like side effects, are pretty constant throughout the population. And then so if you try to treat people at very low risk of disease, then their risk of the side effects may actually be higher than their risk of avoiding that disease. Whereas if you treat people at high risk of the disease, their chances of benefit are going to be much higher than their chances of side effects. So this is all about trying to make sure that we are identifying people at high enough risk of the bad thing that we’re trying to prevent, so that they’re more likely to benefit than be harmed by what we’re doing to them.

Eric: I’ve got a question for you, James. I mean, Sei just mentioned a couple like FRAX. We use predictive indices all the time. Whether or not to start anticoagulation for atrial fibrillation, CHA2DS2-VAS. What’s the risk for bleeding, HAS-BLED. There’s no huge editorial saying, “We can’t predict the future when it comes to AFib.” Why is it that when we talk about mortality, all of a sudden we hear these arguments that this is… What was the line again, Alex?

Alex Smith: The punctilious quantification of the amorphous.

Eric: I don’t think anybody ever said that about CHA2DS2-VASc.

James: Yeah, and I mean-

Sei: People have said that I am punctilious.

James: Yeah, I mean, I think you run into these issues with all types of models. CHA2DS2-VASc and kind of 10 year cardiovascular disease risk calculator’s all can be misleading or miscalibrated in certain subgroups and populations, and they’re all prone to these same errors that deal with mortality.

I think probably it might just be the gut feeling of you’re predicting someone’s life expectancy rather than where there’s so much… I mean, it’s difficult to talk about someone’s mortality, and so there might just be that kind of knee jerk reaction to using some of these models. Whereas, I mean, even those strokes and heart attacks also can be equally problematic. I think mortality is the final end path. It’s a more sensitive subject.

Alex Smith: Before Eric gets to the articles, tremendous title from The Onion from ’97, “World Death Rate Holding Steady at a Hundred Percent. World Health Organization officials expressed disappointment Monday, as the group’s finding that despite the enormous efforts of doctors, rescue workers and other medical health professionals, the global death rate remains constant at 100%.” I love The Onion. Okay, that’s my interjection.

Eric: Okay, I’m going to jump to the articles. Let’s start off with your article, James. What, published in JAMA IM?

James: Yeah.

Eric: You did a prediction model for community dwelling older adults with dementia. Why did you do this?

James: Yeah, I mean, I think in older adults with dementia, I mean, we know that the clinical course is highly variable. And if you look at previous kind of population-based studies, the median survival time after diagnosis is really broad and can range based on the study from 3 to 12 years. That’s been reported previously.

And there have been a few mortality prediction models that have been done in the past. Most of them are dealing with older adults with dementia who are living in the nursing home or have severe dementia. But we know that actually most adults with dementia are living in the community, and this represents a really heterogeneous population that kind of span the disease severity, and a lot of these are older adults with mild to moderate disease. So just given that there weren’t any well-known or well-used models, we decided to create one for specifically this population of those living in the community at home with… Most of them had more of in the mild to moderate stages.

Eric: And just for our listeners, how do you create a prediction model? Can you simplify it into… We don’t want to get too wonky on this podcast, but what does that process look like?

James: Yeah, so there is no exact like there’s this one model that’s going to say, “This is the one that predicts mortality.” What you really do is you look at a whole bunch of different factors that we kind of know are associated with increased risk of mortality. So age is an obvious one from previous studies. We know that functional difficulties, so needing help with dressing or bathing, is an important risk factor for mortality.

What you do is essentially just look at a lot of these kind of predictors, come up with one through some statistical techniques that seems reasonable based on something that can be used in clinic that’s not too large that has so many variables it’s just really difficult to calculate, and make sure that it performs well. So once you come up with something that seems pretty reasonable, that is ultimately the model. But we’re not saying that this is the only model that would predict mortality. We’re just saying that these are some of the factors that are likely to be associated with increased risk of mortality and may be helpful for basing these predictions.

Alex Lee: I’ll just interject to say it’s often said that all models are wrong, but some are useful. I think that’s helpful to keep in mind, that nothing is perfect. There is so much randomness in all of these outcomes that no model’s going to be perfect. But it’s a matter of creating something that’s going to be useful for whatever decision it’s going to be used for and convenient also for the people who are going to be using it, both in terms of how they can do it practically, but also what needs to be assessed to go into the model, whether that’s age or disabilities or comorbidities.

Eric: There’s a lot of association around socioeconomic status and social determinants of health. Should that be part of prognostic indices?

Alex Lee: Well, I think it’s a very important issue that I think today really needs to be discussed and thought about very seriously. I think we’ve seen a clear backlash among medicine in terms of using race specifically in these prognostic models, for good reason. That you would actually change some clinical decisions based on race alone, if that’s being pushing the model one way or the other that would make a clinical decision different.

But I think socioeconomic status, a lot of those indicators are very closely tied to race. So it could be that introducing some socioeconomic status predictors, say ZIP code, if you have a highly segregated city, you’re basically going to have that ZIP code could actually just basically be a proxy or something that’s essentially capturing a black race, for example.

So I think it needs to be done sensitively. I think obviously, we know that people who live in poverty or have had difficult circumstances during their lives, they’re often more likely to have a shorter lifespan. And even with a similar comorbidity level, they might still have a shorter lifespan. So that’s certainly something to be considered.

But we don’t want to have kind of a feedback loop, whereby people who are living in difficult circumstances are then kind of, in a sense, discriminated against in medicine and given less intensive treatments that might actually benefit them. So I think it’s something that has to be done very, very carefully and thoughtfully in terms of both what you think the inputs you’re using, like ZIP code are actually capturing, and then also what kind of decisions are going to be made from that model that we may or may not want socioeconomic status to play into those.

Eric: So you’ve thought about this a lot. I mean, I can imagine that you create a prognostic indices that people may use, let’s say for colon cancer screening. You don’t want that then to worsen disparities of care. What are your thoughts?

Sei: Oh boy, this is such a mess. Meaning I don’t have any good answers. I think this is incredibly hard. Kind of the thing that I keep coming back to was an article I read about how when… This was in the legal space when people were talking about parole and using recidivism rate in the prediction model for parole. And then there was a lawsuit because they were saying, “You’re using gender, and gender is a protected characteristic. So you can’t use gender to make this parole decisions.”

What they found was that if you don’t use gender, the black and white truth of the matter is women are much less likely to recommit crimes than men. And isn’t that an important and valid reason to include a factor like gender in decisions about parole? So they said when they decided not to include gender, what it ended up doing was keeping women, who are very unlikely to recommit crimes, in prison longer, and then letting men, who were more likely to recommit crimes, out earlier, because they were not allowed to use gender.

So I think about this in terms of medicine in that if I do believe that somebody who has been historically marginalized and has had all of these additional poor food, no physical environment for exercise, next to pollution. If that means that they’re more likely to die sooner, does that mean that they should not get screening? On some level, if I feel like all of those things are truly unchangeable, then I think the pragmatic or empiric answer would probably be yes. But the problem is I have no idea if those things are truly unchangeable. So all of this is to say I don’t know what the right answer is, but I do think that whenever we develop these models, we try to think through these questions and we try to stay away from things that would exacerbate or perpetuate disparities, which I think everyone can agree are things that we should try to minimize.

So I feel like I’ve talked myself into circles. I don’t know what to do. I do think that we try to think about this and when we decide which factors to incorporate and which factors to not incorporate, we try to balance all of these different considerations. But I think it’s just really, really these are thorny topics and I don’t have any good answers.

Eric: Yeah, I just want to throw out to… We did a Updates in ID and Nephrology GeriPal podcast, and I think Rasheeda Hall was talking about how to think about eGFR and where we are with race-based algorithms for eGFR versus thinking about newer calculators. So I encourage all of our listeners to listen to that podcast as well.

Okay, I’m going to switch back to topics. I know we could go on for this. James, how good was your calculator or indices?

James: Yeah, I mean, I think there’s a few different ways that you can measure model performance. I don’t need to get into much of the details. A lot of it’s related to… The two main ones are discrimination and calibration, and essentially just how closely the model’s predicted event probability matches the observed probability, and whether or not you can kind of correctly identify those who do and do not die during follow-up.

So the model was pretty good compared to previous ones. We calculated something called the integrated area under the receiver operating characteristic curve, and it was 0.75, which is a pretty reasonable metric. And then on calibration, the observed probability seemed to align fairly well with the models predicted probability. So we felt pretty good that our model did a pretty good job at predicting mortality from 1 to 10 years for these older adults. And we also used another cohort to validate the model, just to make sure it also applied to other individuals. That seemed to perform pretty well.

Alex Smith: Yeah, just to put this in context, what does an iAUC of 0.75 mean, for our listeners, Ken Covinsky put it succinctly for us the other day when he said Louise Walter, who’s been on this podcast, she has a prognostic model for hospitalized older adults, and she had same 0.75, right, was her AUC. So Ken Covinsky liked to explain it as, “Well, if 0.5 is a coin flip, which is useless clinically, right? And a hundred percent would be perfection, 0.75 means it’s halfway between useless and perfection.”

Eric: Does anybody know how does that compare to other prognostic indices that we use, like CHA2DS2-VASc or FRAX or any of that?

Sei: I would say it’s fairly similar to many indices that we use. I think most indices, pragmatically, that are used in clinical medicine are in the around 0.75, 0.7 to 0.8 range. Alex Lee, were you about to say something?

Alex Lee: No, I think, yeah, that’s correct. I think it’s good. I think it’s hard to get, in some sense, models that do too well. Like above 0.9, you kind of question what exactly are they predicting. Is it even a useful model if you can really tell that well? In a sense.

Eric: And Alex, you did another study published in JAGS, but you didn’t look at mortality, right? You looked at disability risk. Why did you do your study?

Alex Lee: Well, so our model is what we’re calling a comprehensive prognostic index. So it predicts actually three outcomes. It predicts mortality, it predicts time to walking disability, so time until someone can’t walk across a room, and time to ADL disability. This is really important because a lot of indices predict mortality in different populations, but there’s very few that predict disability. There’s some that predict disability for patients with stroke or other very serious diseases, but not a model that predicts in the general population of older adults.

This is important because a lot of older adults actually care more about their functional abilities than their time to death. I think that’s largely because it influences so much about their quality of life and how independent they feel in their life, whether they’re still able to go out and take their grandchildren on a walk or if they need help cooking dinner every night, things like that that really change kind of their life circumstances.

So we wanted to create a model that’s going to reflect that sort of risk and what it also means for their families and financial decisions. Because I think a lot of times when someone develops significant disabilities, they need to either move into an assisted living home or get in-home supportive services. So it’s important for a wide range of both health and non-health related decisions that older adults need information on. And again, so this is a gap in the literature that we thought we could help out with, so that’s why we did the paper.

Eric: And since we are getting close to the hour, how good was this indices?

Alex Lee: This index, it did pretty well for mortality. Similarly about 0.7, 0.72 for mortality. It wasn’t quite as good for disability and walking disability. I think it’s a harder outcome to predict, in some sense. It’s based on self-report. So it could be that it’s something that maybe varies over time a little bit or we don’t have perfect reporting, because we do have… This is based on a survey that’s done every two years, so maybe it happened at some point, but it… I don’t know. People can go back and forth on disability, so it’s something that it kind of wavers more and it’s not quite as yes or no. Mortality is very binary. You’re either dead or you’re alive. Whereas being able to walk across the room, well, maybe you can do it some days, but not all. Maybe you need your walker sometimes, but not always. Things like that. So I think it’s just harder outcomes to predict. So the discrimination statistics was more in the 0.6 to 0.7 range, which is lower, but still useful, we believe.

Eric: Both of these indices are on ePrognosis now, right?

Alex Lee: Correct.

Eric: How do you imagine clinicians actually using this? Either one of these indices. Alex?

Alex Lee: Yeah, so I think this could be used in a variety of circumstances. Again, I think there is some clinical treatment decisions that also rest on how good we believe someone’s functioning to be. For example, with diabetes treatment, we’re really concerned whether they have some IADLs or ADLs that might make them at higher risk of hypoglycemia and have less likelihood to benefit from intensive glycemic treatment.

So I think it could be used to inform some of those treatment decisions, but I think it could also be used just generally if a patient’s kind of wondering, again, “What’s my prognosis in a bigger sense? Not just mortality, but what do you think my life is going to look like in five years? Am I going to be living independently or am I not? Am I going to need help with getting dressed or am I not?” Things like that where I think patients might want to know this information. So I think this will be very helpful for clinicians for those sorts of circumstances.

Alex Smith: I really like in the discussion of your paper, how you talk about these two different patients. You name them hypothetical patients, Mark and Nancy, and how they make very different decisions about where they choose to live because of their future risk of mobility disability and mortality. One chooses senior living, and I think the other chooses… I don’t remember, but this is how it could impact the life choices that older adults are making.

Alex Lee: Right. Yeah, I mean, maybe someone shouldn’t be continuing to live in a two-story house by themselves, for example, if they think they’re going to be at high risk of walking disability. Whereas someone who’s fairly low risk of walking disability, but is quite old and maybe a fairly high risk of death, yeah, they’re fine where they are and they don’t really need to change things. So again, it can make big differences in life decisions.

Eric: And James, how do you think people should be using these indices?

James: Yeah, I think I would echo the same kind of comments. Yeah, I think in the clinic, really as clinicians, we should be empowered to ask some of those questions, like how much do you want to know about the likely course of this illness. Because as Sei was saying earlier, there’s a really wide range in terms of what people want to know and how comfortable they are asking. Some people really want this information, and it can help frame a lot of the screening and treatment decisions that we’ve talked about over the last hour. So just using in that context of a trusted kind of clinician-patient relationship and seeing how we can better improve shared decision-making.

Sei: I was going to say that I think one of the ways that these indices are really helpful is it almost allows clinicians to not be the bearer of bad news, but have the news be more kind of one step removed. If the clinician has a good relationship with this patient, but really doesn’t want to be the person saying, “Your kind of diseases and you’ve fallen a couple of times. This makes me worried about what’s coming up.” I think talking about, “There’s this study that shows that for people like you, there are some bad things that can happen in the future. Maybe it’s about time that we start talking about some of these things.” It allows that relationship. It allows the clinician to not be the bad guy or the bear of bad news. And I think that may be another way that this work could be helpful.

Eric: Yeah. Almost like in the song, “hoping for the best but expecting the worst. Are you going to drop the bomb or not?”

Alex Smith: Maybe that’s where our palliative care language came from. Here we thought that whoever wrote that seminal article, Bob Arnold, whoever it was, they just made it up. No, they were listening to Alphaville. ’80s synth-pop-

Eric: It all comes back to Alphaville. It all comes down to synth-pop. That is the underlying communication skills training program. You’ve just got to listen to a lot of synth-pop.

Alex Smith: That’s right. Here’s a little bit more.(Singing).

Eric: Alex, I loved how you brought in the melancholy into that song.

Alex Smith: I can imagine a young Sei and a young Eric rocking back and forth, eighth grade. [laughter]

Sei: Just trying not to step on feet is what I was thinking, just don’t do that.

Eric: Yeah. Alphaville. Love it.

Sei: Supposedly, don’t step on feet.

Eric: James, Alex, thanks for joining us. Sei, it’s always great to have you on.

James: Thanks for having us.

Sei: Thanks for having us.

Alex Lee: Thank you.

Eric: Excellent song too, James. Thank you to all of our listeners for your support. If you have a second, follow us on YouTube. We’re trying to get those numbers up. And have a great night, everybody.

Back To Top