Podcast – What is normal? Using language and big data to inform diagnosis of dementia

Voice Over:

Welcome to the NIHR dementia researcher podcast brought to you by dementiaresearcher.nihr.ac.uk in association with Alzheimer’s research UK and Alzheimer’s Society. Supporting early career dementia researchers across the world.

Dr Anna Volkmer:

Hello, I’m Anna Volkmer. Thank you for listening to the Dementia Researchers podcast. In a change to our usual host and panellists’ format, this podcast will feature myself and my good friend and colleague in the department of language and cognition at UCL, Dr Vitor Zimmerer. And today we’re going to have a conversation about a topic which we’ve been discussing quite a bit, and it’s working around the title, what is normal, using language and big data to inform the diagnosis of dementia. But first we should really introduce ourselves. So I’ll go first. I’m a senior teaching fellow in the department of language and cognition at UCL, and I’m also a senior speech and language therapist at the national hospital for neurology and your sciences at University College London Hospitals specialising in dementia. And I’ve recently finished my PhD on developing an intervention for people with language lead dementia and their families. So Vitor, tell us a little bit about yourself.

Dr Vitor Zimmerer:

Happy to do that. Hi, I’m Vitor Zimmerer. I’m a teaching fellow in the department of language and cognition at UCL. So same department, and I carry out research on language and different clinical populations. And I am particularly interested in language profiles associated with specific disorders, such as dementia, but also stroke, schizophrenia, depression. And I’m interested in the relationship between language and other aspects of behaviour and condition, which of course is also important for work on dementia. I have developed tools that allow us to look at language profiles in a slightly more efficient way. And perhaps that’s the vision detect early change that can be identified as a symptom of dementia.

Dr Anna Volkmer:

Yeah, really exciting work. And it kind of does link in with the topic we’ve discussing, doesn’t it, which is this importance of understanding the range of normal behaviours in our population to better inform the diagnosis and the treatment of dementia. And I wonder if Vitor, maybe you could start us off by explaining this a little bit more. So why is it important to understand normal language in terms of diagnosis, would you say?

Dr Vitor Zimmerer:

Well, perhaps we should start by just quickly explaining why understanding language in general is important because obviously one of the important questions in our efforts in coping with dementia, dealing with dementia is early detection. So, we want to detect behaviour that is not typical. And language is a quite rich aspect of human behaviour, right? So, we produce it all the time, so language is natural behaviour, and we’re not just measuring your response to some kind of abstract or very artificial test. And, it’s behaviour that is related to quality of life. So we’re not just measuring, we’re measuring something that directly relates to the person’s ability to socialise, to follow the news, to be a part of society or nowadays, to follow government guidelines, for example. And language is cheap. So all it takes is a microphone and perhaps a computer, if you want to run more sophisticated analysis, but computers by now also very cheap relative to stuff like MRI or some types of DNA analysis of blood sampling.

Dr Vitor Zimmerer:

Language, I call it well connected. So it involves large networks in the brain on both hemispheres. Cortical, sub-cortical. So changes in the brain are likely to manifest at some level of language processing, language use. And also it’s complex enough to allow description of different profiles. So we have the way people produce speech. So for phonetics, that’s an important part of language use, but then also the grammar, the vocabulary what’s called pragmatics. So whether what we’re seeing is appropriate or not, or the way we kind of package information, our message into sentences. All of that, each one of those sub domains is itself very, very complex. So we have a lot to work with that allows us to define very specific profiles that we can associate, not even with dementia, as opposed to other disorders such as depression, but even look at sub-types of dementia. For example, distinguish between Alzheimer’s type dementia, perhaps like a lot of panic prominent variants of progressive aphasia. So that’s great. So language is useful and there’s a lot of potential and a lot of it has not been explored yet, right?

Dr Anna Volkmer:

Yeah.

Dr Vitor Zimmerer:

Super long answer to a very short question, I apologise. So why is understanding what normal is, so important? Well, that complexity of language, as I said, it’s a blessing because it allows us, it has these many dimensions along which we can describe individuals, but it’s also a curse. So because complex systems in psychology are very sensitive to a range of variables, but biographical variables, age, gender, education, socioeconomic status, of course dialect, but also other clinical variables that may not be that important for dementia diagnosis.

Dr Vitor Zimmerer:

So we have a normal range. So it’s not just your typical bell curve along one dimension, but it’s very, very complex. And so far it’s not been explored enough. So, if we want to identify atypical behaviour, well, how certain can we be if it’s atypical, especially if we want to detect early change, right? Of course, at some point [inaudible 00:06:48] that which we say, okay, this is very kind of typical for Alzheimer’s disease, but by that time we don’t need the diagnosis tool that much anymore. But we can still use it to track changes, for example, related to an intervention, but we want to detect early change and that will be very subtle. And when it comes to subtle change and the variety of that we find within the behaviour of within the normal population, there’s still so much work to do.

Dr Vitor Zimmerer:

And I’m going to stop talking now because I think I’ve said so much, but I think that’s why understanding what normal is in languages is important for the diagnosis of dementia. But I understand that it’s also relevant for speech and language therapists and an intervention, right?

Dr Anna Volkmer:

Absolutely.

Dr Vitor Zimmerer:

[inaudible 00:07:40] talk a bit about that.

Dr Anna Volkmer:

Yeah, no, as we are talking about this today, I’ve actually thought of some new things. So we’ve discussed this a lot, Vitor and I, and one of the things I’ve actually been reflecting on is so firstly, in my clinical setting where I work with people with dementia and their families, people might often come in and they say, gosh, I think I’ve got dementia. I’m experiencing occasional word finding difficulties, for example, or they’ll feel that their speech or their family members speeches is not normal, so to speak. And sometimes that is indeed a sign of something else going on. But sometimes what people don’t understand is actually the breadth of normal speech and language that we use in conversation. Word finding difficulties is a great example because word finding difficulties, is something we can all relate to.

Dr Anna Volkmer:

Everybody experiences a word finding difficulty once in a blue moon. And it might be because the dog’s barking while you’re trying to have a conversation and you need to rush out the door in 10 minutes. It could be simply that you’ve forgotten somebody’s name and the word finding difficulty really interrupts the conversation and causes conversation to break down. We have these specific rules in conversation as to how we repair conversational breakdowns. So for example, you might say, if you’ve got somebody’s name wrong, you might say, oh Adam, I mean Vitor, and you might self-correct yourself. You might simply repeat yourself, you might apologise, but we have different rules and strategies to do that. But what can happen is that when we are more fatigued or if we’re going through a stressful period, or if we have something else going on, other life changes, it may also be because we have a diagnosis of dementia that we may experiencing be experiencing more word finding difficulties.

Dr Anna Volkmer:

And I think that when that happens, often family members will automatically identify some of these behaviours as not normal and thus indicative of some other kind of underlying pathology. And the other thing that then impacts upon is the way family members cope with it. So we often find that in therapy, family members arrive asking for guidance on what they should do. And what they feel is useful often to deal with this kind of word finding difficulty is often rather than allowing the person, I just gave an example of a person kind of self-repairing. Well, often family members will do is they’ll correct a person, or they will tell the person exactly how to say it. And they’ll become very focused on supporting the person to produce a perfectly normal version of a word or a sentence. And that can actually cause conversation to break down further. But I guess… Vitor and I are very good at talking. So I’ve now given a very long answer.

Dr Vitor Zimmerer:

But I have a question-

Dr Anna Volkmer:

Yeah, go on.

Dr Vitor Zimmerer:

And it’s related to that because I think one important, well, I don’t know if it’s an important question or not actually that’s my question. Important whether someone has word finding difficulties because of dementia or because of something else. If we think about what it does to a conversation, do you treat word finding difficulties and people who are experiencing dementia, do you think you treat this differently than when you want to repair that conversation, than you do if someone that’s not experiencing dementia?

Dr Anna Volkmer:

That’s a really good question. So do you mean as a clinician or would we provide different strategies if a person has got dementia than if they hadn’t?

Dr Vitor Zimmerer:

Mm-hmm (affirmative).

Dr Anna Volkmer:

I guess if a person hadn’t got dementia, so if the person didn’t have any underlying diagnosis to speak of at all, then generally one would expect that they would have strategies to repair their own conversations. And actually we might focus less on the conversation per se, but more on how they are feeling about that conversation. So you and I have spoken a little bit about this idea of the worried well, and sometimes that means that we are working with the individual on what their areas of concern are. What’s triggered those concerns? How can they kind of cope more broadly with conversation in an interaction? Whereas if somebody had an underlying diagnosis that is caught, so a dementia that is causing the word finding difficulty, we would be much more likely to actually specifically examine their strengths, areas of strengths and difficulty and base any strategy so that there would be a broader range of therapy options available.

Dr Anna Volkmer:

So obviously we could tackle the word finding difficulties potentially at the root with some maintenance work. But I started talking about conversation analysis. That’s what I’ve been describing when I’ve been describing the way we interact and the way conversation breaks down. And what we often use is an applied conversation analysis approach to an intervention. So we would examine the conversation of the couple together and we would be much more likely to actually tackle strategies or identify repair sequences that allow the conversation to flow, to maintain some kind of normality for the person to help them identify what they feel is not normal so to speak.

Dr Anna Volkmer:

In some ways using the phrase normal in this case then becomes a bit more stigmatised and difficult because everybody is so different. I’ve met people who when we video recorded their conversations, it always springs to mind, a couple of I worked with and we video recorded their conversations. And I asked them about a certain behaviour, which we call a testing question. So essentially that means that the partner, so it was the unimpaired partner, who didn’t have any difficulties. She would continuously text her husband in a [inaudible 00:14:28] so their entire conversation was based around kind of asking questions that they both knew the answer to almost like a teacher student relationship. That isn’t typical of the majority of our population. And we know that because of the CA, the conversation analysis literature that has examined the breadth of the population. But for that couple, it was typical.

Dr Anna Volkmer:

Which brings us right round to this idea that of the breadth of what’s normal. And there’s a huge breadth of normal in our society in terms of how we communicate and how communication influences our relationships. And you kind of talked about that right at the beginning of this conversation. And actually, because I think conversation forms a key part of our relationships, it’s also very personal, and I know you and I have thought about how we could potentially find out what is normal and what methods we could use to gather data like this. And some of that data that maybe you could speak to this? We’ve debated what is the best, what’s the safest way that people might feel most comfortable gathering this type of the data?

Dr Vitor Zimmerer:

Well, I think it starts by recognising again, a strength of collecting language data in that it’s naturalistic behaviour. So, what we’re moving towards too is not, it’s not using a naming test or sentence picture matching, or picture description or anything like that, but just recording people as they speak as they’re having a normal conversation. Because not only is this behaviour, people feel comfortable with, right, at least in many cases, but also, it is rich behaviour. It gives us great depth of data. Having you talk about your job gives us much more information even if we just focus on the language, than having you name a whistle, right. So I think it starts there. And the second point, I guess, something that links to what you just said about whether someone is having a good day or bad day and how that influences language, right?

Dr Vitor Zimmerer:

So we, right now, it seems that for a variety of reasons, diagnosis can be very snapshot based. So in many studies you do perhaps one test, unless you do an intervention and you test and retest, and you may be using several baselines. Often, you have that person on that one day and you’re supposed to make, to infer from that. So the next step, I guess, would be to use more longitudinal data. And again, language, data is really good for that, because if we can just record conversations regularly, and I’m not just talking about perhaps a bunch of weeks, so that we have a more stable baseline, and we can understand where that person is in this phase of their life. I’m talking about perhaps recording over years and understanding how progression looks like and typical development, including ageing. So, I turned 40 last year, no, not last year, the year before, Jesus. At least I think so. [crosstalk 00:18:09]. Once you turn 40, the GP ask you to come in for them to get a blood sample from you. Did this happened to you, because of cholesterol?

Dr Anna Volkmer:

Not yet.

Dr Vitor Zimmerer:

Not yet?

Dr Anna Volkmer:

Two and a half weeks. In two and a half weeks they’ll start asking. I’ll know what to expect.

Dr Vitor Zimmerer:

Yeah, it’s like a rite of passage, oh, okay, so I am getting regular check-ups now. [inaudible 00:18:33]. But what if we don’t just do that, we don’t just look at a blood sample, but we also have participants give a language sample every few years from a certain age on, and that would help us really understand the combination with all the data that we have about a person’s background. Not just the range of normality in the population, but also the range of normality and progression in ageing. And with that, we would have a wonderful model with which we can use to detect deviation from normal, right? Because I guess we will find out that the variance in the population is so wide, that it’s almost impossible to detect early change. Everything that is slightly off will probably be found in a substantial part of a population if you just measure one.

Dr Vitor Zimmerer:

But what we want to find is whether the trajectory is different. And I think that will be easier to, that will be the more reliable finding. This is someone who’s slope if we map the variables over time, looks different from how it should look like regardless of where that slope is on the absolute axis, whether that number is high or low, it’s decreasing too quickly. It’s rising too quickly. And I’m my intuition is that this will be key. But for that, obviously, first of all, we’d need to have the infrastructure to collect all this. We still don’t know which variables to pay attention to. So we still don’t have the proper models, but also it’s an intrusion obviously.

Dr Anna Volkmer:

It is.

Dr Vitor Zimmerer:

And we don’t know whether people would be ready for this.

Dr Anna Volkmer:

Yeah. Yeah. And it’s interesting, isn’t it? Because the link with this idea of recording somebody and it is an intrusion and also the implications of it. If we’re suggesting there is a point when our trajectories indicate that our linguistic competence, so to speak, is somehow linked with a dementia diagnosis, that has such an enormous implication. And I certainly, as a clinician, I have walked out of clinical rooms following somebody or a family where a person’s just been diagnosed with dementia. And I can hear relatives saying right now, dad’s got dementia now. So we’ll pay all the bills, he won’t be able to go shopping anymore. We won’t give him any money. He won’t have to do anything. So I think that it’s not only the intrusion of data collection of collecting that information, but also I can imagine the implication, the kind of implication in terms of competence and mental capacity in one’s life would be for some people mammoth. Because there’s a huge proportion of people we already know don’t want to know if they’ve got dementia.

Dr Vitor Zimmerer:

Yeah. And this is going to be a risk. And we’re now entering this kind of sci-fi territory, a scenario in which we already have a model. Right. And we already have that knowledge. But we already having the discussions, right. So right now, if we look over the Atlantic, right, we may be facing maybe, it’s election year in the US and both candidates are from different sides, almost accused of having some neuro pathology, right? Both.

Dr Anna Volkmer:

[inaudible 00:22:35]. Yeah.

Dr Vitor Zimmerer:

And one way, and one subject that is really important during these discussions is of course, language, right? Look at Trump’s language in the eighties. Look at how he once produced these very complex sentences and look at the rambling now, or, sleepy Joe and his gaffs, has that increased? And we’re going to enter territory in which we may have a model to actually just use the public data and start making conclusions, right? To different degrees of, obviously of confidence, but that may be happening at some retroactively, these studies have already been carried out. So Ronald Reagan who died of Alzheimer’s.

Dr Anna Volkmer:

Yep. Thank you.

Dr Vitor Zimmerer:

Yeah. Was compared to George H. W. Bush. So daddy Bush and their languages were compared and there was a finding that Ronald Reagan’s language changed over his presidency while Bush’s language remained more stable. And whether this is already a sign of something. And the more we know, the more we need to be aware of the consequences to society, and that involves having debates about this, what should be possible and whatnot. And also of course, just having the kind of the legislative foundation to do things responsibly so that yeah, with every technology it’s used for good.

Dr Anna Volkmer:

Yeah. And that’s the research, isn’t it, the research component is the ethical responsibility, but certainly it’s more and more common now if I meet somebody clinically, actually they will be able to present me with a video recording of them having a conversation, pre injury, pre diagnosis, pre whatever. And we are more and more able then to measure difference in a clinical sense. And that’s just people volunteering examples. But if people actually, I guess, as a society, do more of this, for example, I can imagine, you and I Vitor, we’ve done a couple of podcasts together now they’re really conversational aren’t they? And actually there’ll be more, there’ll be more sources of our incidental interaction conversation that could be mined for this.

Dr Vitor Zimmerer:

But it’s never gone so far that you asked someone to provide you with videos or audio from a few years ago or so doing diagnosis, have you ever done this actively or are you only talking about people already coming in with…

Dr Anna Volkmer:

No, we’ve done it actively, but not, I’ve never used it as a diagnostic tool I’ve only ever used it, well, I say diagnostic tool, I’d never used it as a diagnostic tool for a medical diagnosis of like diagnosing a dementia, but I’ve certainly used it as a tool to identify the symptoms of the dementia. So by that, I mean, did this person really use complex sentences? Did these people, this couple, really have test questions routinely? Did this family really, were they real jokers? So there was a couple, I can think of a couple of examples where somebody has presented me with a family video to demonstrate what a joker, inverted brackets, that their family member was to demonstrate that they were the life of the soul of the party, that they could hold court and tell stories.

Dr Anna Volkmer:

And that after their diagnosis, they become a very quiet person who didn’t participate in that and that person would become their entire conversation was then meaning that they were being identified as a quiet, serious person. So, that’s just a great example of how conversation actually reflects your personality and your relationship. And his goal was to become, he identified as somebody who told anecdotes, who was a funny person. So then we were able to identify what the family were doing to perhaps, so they were often talking over the top of him, or they were talking really quickly and not giving him opportunities to interject.

Dr Anna Volkmer:

Whereas previously, the conversation was often dominated by him so actually he was given loads of opportunities to contribute. So in a way, we were then drawing the family’s attention to what strategies, what were you doing in conversation? Not only what was your partner doing, the person with dementia, but what was the family members strategies to enable this style of conversation? And that’s always really helpful. So, yes. In short, we would ask people to bring samples in absolutely. And more recently we have in the current climate, also explored conversations via Zoom, which has been also very interesting.

Dr Vitor Zimmerer:

But then you may not have enough data to compare.

Dr Anna Volkmer:

No, no. In those cases, we haven’t really had any data to compare back at all. And really it’s been very novel because how often do you use Zoom with your spouse?

Dr Vitor Zimmerer:

Well, but then there’s Skype. I think about that. I mean, obviously there is a history too, by now you have so many grandparents talking to the children over Skype, et cetera. So there is experience already. But it’s true, when we ask what is normal, there’s also a new normal that we will have to cope with.

Dr Anna Volkmer:

Absolutely.

Dr Vitor Zimmerer:

But it’s just interesting to see that there are these two notions of normal that we are talking about. There’s the population normal, which may be relevance to detect abnormality. And we need to kind of build solid models of language change of typical behaviour, but then there is the individual normal. So is this normal for that person. And do you want to go back to that, to that normal? Right. And in both cases just seems that having a history of recordings, maybe very, very beneficial, especially since, I mean, I don’t know what you would say about this, but how much do you trust if you don’t have a recording, just people’s recollection of how someone used to be [inaudible 00:29:47] their descriptions. How useful is that? How accurate is that? What do you think?

Dr Anna Volkmer:

Gosh, well, clinically with my clinical hat on, we always, we have to trust that their description of what they perceive their, I guess, conversation to have been like. And actually what we tend to use then is more of a much broader brush stroke. So there’s a number of rating scales that you can use with both the person and their partner. And they’re certainly also used in research around interventions. So they are often kind of more descriptive rating scales. Did you consider yourself somebody who would regularly interrupt people? Did you consider yourself somebody who was a chatty person? What did you consider yourself somebody who was a listener? And then partners then, we try and validate that with partner’s opinions as well. I guess that’s more qualitative which is also really helpful because that’s the impression that’s been given, it’s less objective.

Dr Anna Volkmer:

But in research in terms of like, I would often use qualitative research methods in the work I’ve done. And actually it’s really important because you’re asking people about their experiences of conversation in qualitative research often. And then actually their experiences of conversation are often what influence their quality of life. And we’ve talked about having data from lots of people, but actually often the data, what matters to the individual is quite different than what matters to the population at large. Yeah. Interesting.

Dr Vitor Zimmerer:

I mean, obviously we’ve been talking about language a lot, but I think we should also add that this is not just an issue language sciences work with. I mean, there are of course, many behaviours, many aspects of behaviours that change as the result of dementia. And in each of these domains, you can actually make a similar case.

Dr Vitor Zimmerer:

So I had some contact with researchers who are looking at gait in Alzheimer’s disease. So the way people walk and how that may change, whether your steps are symmetrical, how many steps you take for a certain distance, even how much you walk and their solution looks very similar to what we’re doing. Just, their optimal solution would, I guess, be just attach a device to the person that just gets you a lot of data about when people walk, how much they walk, how they walk and have models of typical walking and then identify atypical walking, perhaps walking that’s related to Alzheimer’s disease. And it doesn’t have to be as fine grained as that, it could be, when we think about tracing apps, for example, how many social contacts do you have? How long are your social contacts?

Dr Vitor Zimmerer:

We have to keep in mind that a lot of these data are already being collected, right? And people actually with sometimes without knowing just voluntarily donate these data often for advertisement purposes. So they just want you to think about Burger King more often or whatever. But what a lot of researchers are proposing is just to use that, to look at the wide range, not just language, but a wide range of behaviour that may help us identify dementia earlier and make a real impact on people’s lives. And yeah, we just spoke about that with, technical challenges, but also challenges to society when it comes to carrying out that work.

Dr Anna Volkmer:

And that reminds me of one of the points from our previous discussion, which is actually the importance about the breadth, having a vast number of samples in this, because often we use formalised standardised psychometric tests, which have been developed and they’ve been tested on a group of normal people. They’ve been tested on a group of people with various diagnoses, but often that samples of about 30 or 40 people. And then I use them clinically, or we use them in research studies and a whole group of people who may have a diagnosis of dementia that hit ceiling, or maybe they experiencing in their domains, their lives. So I’ve had a number of people I’ve spoken to recently for whom they are scoring ceiling on all these tests. And yet there’s something that’s changed. And so actually this group of normal 30 or 40 people that they’ve tested, just isn’t adequate to reflect the range of normal.

Dr Vitor Zimmerer:

No. These let’s say naming tests, for example, I mean the other classic example for a useful, but flawed test, good performance in the naming test, we both know it, it’s dependent on so many things that are not directly related to the health of the person. Even just being familiar with the test situation can make a huge influence. And that gives you a lot of education effects. So you have a double education effect. You have the education effect of perhaps someone just knowing more words and that being normal for the person. I think if you look at the Boston Naming Test, many educated people in their late sixties, early seventies will do really, really, really, really well. Right. And you can probably give them a harder test and you will still, so it’s, they reached ceiling too earlier, right? So just someone reaching ceiling doesn’t mean that there’s nothing going on.

Dr Anna Volkmer:

And somebody failing similarly [inaudible 00:35:52] failing or getting very low scores also there’s plenty of people. There’s an item on the Boston Naming Test, which is a pretzel. And there are many people I know who’ve never seen a pretzel, let alone a black and white drawing of a pretzel.

Dr Vitor Zimmerer:

Well, it’s an American test, right? It’s the Boston Naming Test.

Dr Anna Volkmer:

It is. That’s it, exactly. Or even a camel, there’s a picture of a camel on it. And there’s lots of people who don’t necessarily know what a camel is or a yoke, but there’s all of these items. When I’m talking to people I ended up apologizing sometimes. Seems a bit ridiculous.

Dr Vitor Zimmerer:

Sometimes I keep explaining that it’s an American test because it is the pretzel, it’s the Beaver, just keep spoiling the test. Some items are really difficult. So what we want is individual baselines and some knowledge about typical progression and also their role in a conversation, which is, I guess, even harder to measure? It would be more your domain.

Dr Anna Volkmer:

Yeah.

Dr Vitor Zimmerer:

Yeah.

Dr Anna Volkmer:

Well, I feel like we’re drawing to a close in terms of our allotted time. It’s been great to chat to you today Vitor. Thank you for joining me.

Dr Vitor Zimmerer:

Well, it’s good to see you, especially during lockdown.

Dr Anna Volkmer:

Absolutely.

Dr Vitor Zimmerer:

Or partial lockdown or whatever we’re having right now.

Dr Anna Volkmer:

[inaudible 00:37:16] up and down isn’t it. So you can find profiles on myself and Vitor on the NIHR Dementia Researcher website, including details of how to find us on our Twitter accounts. I also have a few blogs there, so please take a look. It’s dementiaresearcher.nihr.ac.uk. And finally, please remember to like, subscribe, and leave a review of this podcast through our website, iTunes, Spotify, SoundCloud, and all the other places you find podcasts. Thank you for listening

Voice Over:

Brought to you by dementiaresearcher.nihr.ac.uk in association with Alzheimer’s Research UK and Alzheimer’s Society. Supporting early career dementia researchers across the world.

END