Using artificial intelligence (CognoSpeak™) in memory assessments: a GP interview study

System-wide problems

GPs spoke about issues within the MAP reflecting wider problems across the NHS, namely lack of funding and lack of resources. GPs spoke about secondary care work being shifted onto primary care without the corresponding funding. Many GPs expressed frustration about the ‘referral hoops you've got to jump through’ (GP11) for patients to be seen by secondary care describing them as overly rigid and in some cases impractical, for example, rejecting referrals because of co-existing depression or excess alcohol consumption: ‘I think if someone is very rigid about their use of guidelines then potentially they might not use a service when a patient could still benefit from it’ (GP10).

Some GPs spoke about their experience of having to work around the guidelines to get their patient to secondary care: ‘it’s a bit of game-playing’ (GP16). They also said that the interface between primary care and secondary care did not allow good shared care and communication especially while waiting for a diagnosis:

‘... once I’ve kind of referred them I feel like after that, it’s completely out of my hands. There’s no collaboration between primary and secondary care after that point.’ (GP13)

Despite growing referral numbers, GPs thought that claims of over-referral were wrong:

‘I’d still be very hesitant about secondary care taking less and less patients … I think secondary care need to be seeing as many if not more.’ (GP05)

‘I have a patient that told me about their memory concerns, it sounds like they're relatively mild but significant, still significantly impacting their day-to-day life. It might be quite early memory impairment but it’s definitely different from before, I haven't found an alternative cause. I don't really want anything else in the way of me making a referral.’ (GP01)

Overwhelmingly, GPs felt that the biggest issue in the MAP was the wait time patients face following a referral to secondary care: ‘people are dying before they get [there]’ (GP06). This was seen as not only difficult for patients and families but also rendering the memory service ultimately useless:

‘I think it’s the waiting list, the waiting list cannot; like if it just carries on going up it basically isn’t a service, because by the time they get there they can’t have the medicines or it’s too late or they’ve missed the boat.’ (GP14)

Many were pessimistic about the possibility of improvement without drastic changes to the whole NHS system, and many seemed resigned to just working through the system already in place: ‘Um [sighs] it’s, it’s, it’s OK, it’s OK, it’s what we’ve got’ (GP10).

GPs are overworked yet underutilised

Despite being experienced doctors, often with lifelong relationships with their patients, GPs felt that their ability to make a pragmatic preliminary diagnosis, or in some cases a definitive diagnosis, was underutilised and they disliked the lack of trust in their clinical judgement compared with the ‘tick-box’ (GP12) referral process:

‘I think it’s my, my history and conversations with the, the patient that really make my decision ...’ (GP07)

GPs understood and to some extent agreed with the NICE dementia guideline,³ which states that diagnosis should only be made in a specialist memory clinic:

‘... as GPs we’ve always been told you can’t diagnose dementia.’ (GP05)

‘I wouldn’t feel confident to do it based solely on sort of level of assessment that I do in primary care.’ (GP08)

But GPs spoke about the need for pragmatic diagnoses to be made more readily in primary care when their patient’s neurodegeneration was obvious and had progressed to the extent that waiting to be seen in a specialist clinic came with very few benefits:

‘... us in primary care are becoming more confident and developing skills to be able to diagnose dementia.’ (GP15)

GPs reported having very little time available for complex cases of suspected dementia and wasting the available time on unnecessarily complex referral forms, with the to-ing and fro-ing of referral rejections and clarifications. Managing expectations during very long waiting periods for memory clinics is particularly difficult for GPs:

‘So the initial assessment itself is that, it’s predictable, but kind of then coming back in the interim is, is the additional kind of time burden.’ (GP14)

‘ ... it just seems like this endless task of them coming back and back and back and back, when you can literally do nothing. Like so it, you just feel so, I don’t know, useless and helpless, and like they get, and the family’ll get more and more frustrated and angry, and it’s all directed at you [laughs] often when you’re like, there’s nothing that you can do.’ (GP14)

GPs’ assessment and referral dilemmas

Many GPs spoke about the difficulties they encountered when performing the initial assessment, especially in patients with significant comorbidities:

‘... it’s very rare to get someone who’s just got a memory impairment, nearly always they’ve got a previous history, and again you’re trying to figure out is this a vascular condition or what is it.’ (GP11)

Specifically, a few GPs spoke about the difficulty of disentangling the impact of co-occurring mood disorders from a neurodegenerative memory loss:

‘... is this just anxiety or depression, or just dementia? Cos quite often, patients will have both. And it’s really complex to unpick how much impact each is having.’ (GP07)

GPs spoke about the anxiety they felt about a ‘wasted referral’ (GP13) for people who ultimately did not have dementia, and on the other hand the huge impact of a confirmed dementia diagnosis for individuals and their families: ‘I think it is just cos it’s such a life-changing diagnosis, more than most other diagnoses’ (GP14). Some GPs also spoke about the difficulty of managing patients unwilling to proceed with a primary care assessment or a referral for a diagnosis:

‘I would say it’s often quite a sensitive subject, people are often not completely willing to come in … cos they’re afraid of getting a diagnosis.’ (GP14)

And that family members often push for an assessment or referral despite their relative’s reluctance, ‘often the, the agenda is being pushed by the family member’ (GP12).

Patients’ and families’ perspectives

Many GPs talked about the difficulties from the perspective of patients and their families:

‘They know that there are medicines that help prevent, er like slow down the deterioration, and they recognise that they’re missing out on that, which is really challenging.’ (GP14)

GPs also spoke about the difficulty of families witnessing their loved one deteriorate while waiting to be assessed in a memory clinic, ‘it’s just so sad, um, and it’s just so hard for the families to watch them deteriorate for two years. It’s just a massive ask that, isn’t it, without any help at all’ (GP14), because access to dementia services is often restricted until there is a formal diagnosis made in a memory clinic.

GPs also spoke about the ‘stigma and a lot of fear’ (GP12) that some patients and families face when seeking a dementia diagnosis. GPs spoke about the lack of consideration for the additional barriers patients from ethnic minorities can face:

‘I think it’s that embarrassment and shame in um families in ethnic minorities tends to focus on, um, and it prevents them getting the support or going to their GP.’ (GP16)

‘... in my community, dementia isn’t a thing, White people get dementia, Brown people apparently don’t get dementia. So it was almost like, really, our dad’s got dementia? No, they’ve got that wrong, kind of thing.’ (GP16)

Optimal implementation site(s)

Many GPs spoke about the potential for using CognoSpeak™ in primary care with GPs ‘gatekeeping’ (GP12) access. Some suggested that a CognoSpeak™ assessment could be offered ‘before even seeing the GP’ (GP12), or during a GP consultation. Other GPs spoke about the potential for incorporating CognoSpeak™ into the current GP assessment process with the help of a healthcare assistant (or other healthcare professional):

‘... how I could see it working is you do the history and examination. Then you take a — go to reception now and reception will show you how to do CognoSpeak™. You do that. Go for your blood test and then come back and we’ll talk about the results and then in that result appointment you would — either reassure them or refer them on.’ (GP16)

Some GPs felt that CognoSpeak™ would be more appropriately placed in secondary care, ‘when I look at that and think about that, I sort of see far more benefit in secondary care than primary care. Erm, perhaps because I think primary care’s already quite full of these sorts of pre-assessments and things like that’ (GP03). Or as a ‘pre-screening for the memory clinic’ (GP03) following a GP referral.

A few GPs spoke about the potential for CognoSpeak™ to be rolled out as an openly available memory screening tool:

‘I think make it directly accessible for patients so that they can use it without needing to wait for a GP appointment or have a GP refer them.’ (GP05)

Some GPs saw no place for CognoSpeak™ at all:

‘Unless … Cognospeak negates the whole need to see a GP, I can't see, I can't see where it’s gonna get implemented.’ (GP09)

Barriers and difficulties

While many GPs were open to using a tool such as CognoSpeak™, almost all expressed concerns especially about the possibility of it adding to GPs workload, increasing patient waiting times and other difficulties that may cause patients to ‘drop out of the process and then never end up getting a referral to the service they need’ (GP08).

The potential for technological barriers excluding patients from completing CognoSpeak™ was a common concern across GPs. Given the lower levels of technological confidence seen in the age group most at risk of developing dementia, there was concern that many patients with memory concerns would be unable to complete CognoSpeak™ without assistance:

‘I think there are a lot of patients that don’t, especially elder, older patients that don’t, don’t use technology.’ (GP17)

‘... you’re gonna create inequalities with people who don’t have access to computers, internet or, you know, maybe even language barriers.’ (GP06)

GPs spoke about patients’ preference to speak to a person when they have a health concern, and resistance to engaging with online alternatives to traditional clinical appointments, as a reason that many may be unwilling to use CognoSpeak™:

‘... you’re gonna sit on a computer, I think would be a really hard sell.’ (GP14)

‘I just think we just cannot lose completely the human touch on it; like I just think that patients would struggle so much if we say, oh go and speak to this person, and then they get another AI thing and they get another AI thing, I think it’s just a really sad way that it’s going if that’s the case.’ (GP14)

At the time of the interviews, the proposed threshold for referral in CognoSpeak™ was a probability of dementia of>50% which GPs thought was unacceptable:

‘Am I missing something, I don't think I am but am I missing something, I'd be interested to know what other people think about it— where they see it in the referral process. And how do they feel about the 50% yes, 50% — it’s like flipping a coin.’ (GP09)

Opportunities and benefits

Many GPs recognised the potential benefits for both patients and clinicians in using tools such as CognoSpeak™ in the MAP. Some saw CognoSpeak™ as a tool to both save GP time and lower anxiety when deciding whether to refer a patient to memory care:

‘I think, like I mentioned before, it would help with the borderline cases. I already feel quite confident with diagnosis of both ends of that spectrum. But in the — in the unclear cases, I think it would really help.’ (GP13)

GPs also spoke about the benefit of using CognoSpeak™ to monitor patients discharged from memory service with mild cognitive impairment (MCI):

‘...in terms of monitoring it’d be quite helpful, particularly if you’re able to get sort of a more standardised score, you know, “oh well you had this assessment a year ago, we’ll repeat it and we’ll see if that standardised score has dropped”, I think that’s quite a useful thing to be able to do and I can certainly see that that would drop down significantly the numbers of those patients who go on to have a re-assessment by memory clinic perhaps unnecessarily.’ (GP08)

GPs spoke about CognoSpeak™’s potential to help combat the distress that comes with the uncertain and long waits to be seen in secondary care. The potential to provide faster reassurance or confirmation of patients’ concerns and to help families to understand the deterioration of their loved one was described as a large potential benefit for patients and their families to better manage the condition:

‘So it’s a really good, objective tool, isn’t it, to be able to assess how much worse, not only from the clinician point of view but also from the family’s point of view about the rate of deterioration.’ (GP16)

Future development of CognoSpeak™

While GPs were hopeful for a tool such as CognoSpeak™, they also expressed scepticism and highlighted unanswered questions:

‘I think it’ll be, it’ll be exciting to see where it goes um, but, as I say, I think it’s, it’s, it, it’s working out how to make it so it’s actually helping streamline a process rather than adding more information and more steps to a process.’ (GP12)

Most GPs wanted more information on exactly how to implement CognoSpeak™ for their patients, and how to deal with the output of the assessment: ‘the question I'm always thinking is about where is this gonna fit into my referral process, my time pressure’ (GP09), ‘So um the question again would be if somebody says, I’m worried about my memory, and then the, the screening thing, the CognoSpeak™ says, well you don’t meet threshold for referral, where’s that then meant to go?’ (GP12). Many GPs suggested pilots or tests for CognoSpeak™ to help to establish the optimal method of implementation.

There was also a desire to see more research to determine the accuracy and usability of CognoSpeak™ in a diverse patient population:

‘Does it work with different languages was another thought I had; … how’s that gonna pick between someone with learning difficulties and somebody who’s just not as bright, who can't name animals quickly and somebody who isn't used to speaking in this way in assessments, how does, is CognoSpeak any good at picking up on that?’ (GP09)