MQDataSCience - Suicidal Ideation in Dementia

So yeah, so I'm going to talk about suicidality in dementia, and this is something I've been working on with some people at UCL and KCL who are listed here, and also wanted to thank the CRIS team, the service user groups that have helped me with the project as well. So yeah, and the study is not yet published as well, but it's been accepted for publication, and there's a preprint link which appears later on. So basically the starting point for this was there was a Danish registry study that was published a couple of years ago, and it looked at the risk of suicide in neurodegenerative disorders, and I was particularly interested in dementia as one of those disorders, and there was a pattern of generally sort of possibly lower overall risk of suicide, but an increase in the few months after diagnosis, and then coming down again. So that was a Danish registry study. We don't know if that's the same in the UK or not, but we're interested in that really, and looking at suicidal ideation as a sort of linked topic to that. So yeah, and generally a lot of the previous studies have been on smaller samples, and Alzheimer's dementia has been better researched. Other types of dementia that are less common, so things like vascular dementia and Lewy body dementia are generally under research in this area. And yeah, we opted to use a CRIS database for this. I mean, I think just to give a bit of the context, when we were thinking about why the risk might potentially be higher around the time of diagnosis, you know, it could be that people have more, we know that depression is common in dementia, so it could be related to that. It could be something to do with the impact of receiving sort of news about the new diagnosis that can be concerning to people. Although sort of interestingly that some people have varying insight into whether they do have a dementia or recognize their memory problems or not. And also specific to the different types of dementia, it may be that there's, you know, some types may be associated with more symptoms such as hallucinations or depression or fluctuations in things like Lewy body dementia. So that's really what we're investigating. We chose to use CRIS for it. So yeah, so it's now been rolled out to a number of trusts around London and other places as well, but the area we looked at was four boroughs in London. It has access to, it's a population of about 1.3 million people and there's 400,000 anonymized records as part of the CRIS database. And yeah, so there's a de-identification process and there's then both structured data and free text data available. And in terms of what we used it for, we used structured data for some of the dementia diagnosis and for, there's also natural language processing algorithms available and they cover a wide range of things such as difficulties with sleep, with agitation, with delusions. We used it specifically for suicidal ideation and also for picking up some dementia diagnoses where they're not coded in a structured way. So for some things like Alzheimer's there's well-recognized codes from the ICD-10 that are used and reliably kind of entered by clinicians but for things like Lewy body dementia there's not. And we also use the HONOS, which is this Health as a Nation Outcome Scale. There's one for older adults and we use that to look at depression, agitation and hallucinations. And just say, I think some of the advantages of using CRIS apart from the fact that it's a large data set, it's also that it's routinely collected data and in terms of dementia specifically, the vast majority of people are diagnosed through mental health services. So we kind of capture that. Yeah, so as I mentioned, some diagnoses are coded in a structured way but for Lewy body dementia, we instead used natural language processing and that we're able to get, yeah, this is a technique developed previously for other studies that are listed below, but they were able to get sort of only 4.3% of records having false positives, so generally pretty accurate identification of dementia with Lewy bodies. And for suicidal ideation, these are some of the examples of these positive and negative annotations, which are coded. So basically we found the results that 15.1% of people had recorded suicidal ideation and this was in the one year window around the time of dementia diagnosis. And if I just come onto the table now, because we use logistic regression to look at different potential confounders, oops, hold on. So yeah, so basically the main findings I think for this were that depressed mood was strongly associated, Lewy body dementia was also strongly associated, there were, yeah, it was less common in people with impaired activities of daily living, that may be related to potentially how the dementia being a more progressed or advanced stage, it could reflect that. The mini mental state examination, which is a short test of cognition including memory and orientation, that wasn't significantly associated in the final model. We did think that people with more preserved cognition, so better memory when they're diagnosed, might be associated with higher risk, but I think the MMSE is a relatively short test, so it may be that if a more detailed memory test and cognitive test was done, that there would be different findings in terms of that element of cognition and the link to suicidal ideation. So, and in terms of the, I mean, I think depression is the sort of expected finding, in terms of the dementia with Lewy bodies, we did hypothesise that it might be increased, and I think the reasons for that are that people can often get quite distressing neuropsychological symptoms, such as hallucinations, and often it's diagnosed at a later stage, that can be distressing for families and their carers as well. So, yeah, so I think the advantages, as I mentioned, with being able to use a large data set and look at subtypes, and I think some of the limitations or things that I was thinking about in particular with the, some of the PPI groups helped think about this as well, was that suicidal ideation may be expressed differently in people with dementia compared to, say, younger adults. So, for instance, some behaviours like not taking regular medications for physical health conditions, that's, the patient may not be trying to end their life by doing that, but it's a form of self-harm, and that's something that is commonly reported in this group of patients. There may also be, in terms of the dementia with Lewy bodies, with it being an underdiagnosed condition, in many places, it may be that it's the people with more severe symptoms who come to medical attention and are the ones who get a diagnosis, and so it may reflect that as well. Yeah, and so I've discussed this in a number of PPI groups, and I've mentioned some of the themes that came up. I think other points that they helped me think about were some issues that may not be captured in the data, so things like the impact of sensory impairment, which can be more common in older adults, the impact of different ethnicities, so often dementia is underdiagnosed in certain ethnicities, and the impact of the things that, so, I mean, religion might be something that's coded in data, but in terms of getting a more kind of holistic understanding of that, that's was something that the PPI groups were interested in, but is difficult to capture in this sort of data. So, yeah, so I think that those are my main findings. So, I think the CRIS database was a very helpful tool for looking at this, and particularly using the natural language processing to identify the suicide ideation and DLB. As I mentioned, the strongest associations were for depression and dementia with lewd bodies, so I think, you know, it's important for clinicians to pay attention to those issues, and when they're seeing patients, and I think, yeah, kind of future directions for it would be looking at some more qualitative work to sort of help contextualize and, yeah, as a sort of synthesis with the data science work, and to do more longitudinal work as well. So, yeah, thanks very much, and happy to take any questions. Hey, Mish, thank you very much indeed. Thank you for taking the time. Excellent presentation, and I must say, a really nice example of how PPIE will bring that research forward, and really some nice suggestions coming through that group.