What do individuals with visual impairment need and want from a dialogue-based digital assistant?

ABSTRACT Clinical Significance Optometrists are well-placed to provide helpful advice and guidance to patients with visual impairment but may not know how best to do this. The availability of a reliable and comprehensive conversational agent to which patients could be directed would be a valuable supplement to clinical intervention. Background The Artificial Intelligence in Visual Impairment (AIVI) Study is a proof-of-concept study to investigate whether ongoing information support for people with visual impairment (VI) can be provided by a dialogue-based digital assistant. The phase of the AIVI Study reported here explored the different dimensions of the information-seeking behaviour of individuals with VI: in particular, their need for information, the methods for obtaining it at present, and their views on the use of a digital assistant. Methods Qualitative data were collected from 120 UK-resident adults who responded to an online survey who were either visually impaired (86.7%), a carer or family member of someone with VI (5.8%), or a professional involved in the support of those with VI (7.5%). In addition, 10 in-depth 1:1 semi-structured interviews explored opinions in more detail. Thematic analysis was used to analyse the findings. Results Analysis of information needs identified 7 major themes: ocular condition; equipment, technology and adaptations; daily activities; registration; finance/employment; emotional support; and support for the carer. Participants used a wide variety of methods to access information from many sources and explained the barriers to access. Participants accepted the merit of a dialogue system aiding in a goal-directed search for specific information, but expressed reservations about its abilities in other areas, such as providing emotional support. Conclusions Participants highlighted potential benefits, limitations, and requirements in using a digital assistant to access information about VI. These findings will inform the design of dialogue systems for populations with VI.


Introduction
Globally, over 596 million individuals have visual impairment (VI) based on their presenting distance visual acuity. 1 In the UK the number of individuals with visual impairment (VI) is estimated to be 1.93 million: 2 or about 3% of the adult population. 3 The combination of the social and emotional, 4 and functional 5 impact of VI, has been shown to have a significant bearing on many aspects of an individual's life. In the UK, help and support for those living with VI is provided by a complex multi-agency system consisting of a wide range of organisations including hospitals, Social Services, charities, commercial companies and volunteer groups. However, there is no standard care pathway 6 and accessing these services, and the advice they provide, can therefore be complex and difficult. 7 For those individuals who are eligible for certification as sight impaired (SI) (VA < 6/60) or severely sight impaired (SSI) (VA < 3/60), the subsequent registration 8 can be a gateway to both financial benefits, and an assessment of needs by a rehabilitation worker. In addition, it may include referral to an Eye Clinic Liaison Officer (ECLO), also sometimes called a sight loss adviser. ECLOs can facilitate understanding of the regulatory processes and ensure that "care continues beyond the medical treatment of the clinic". 9 However, 43% of UK Ophthalmology departments have no accredited ECLO in place. 10 As there are less than 100 ECLOs in the UK 9 with each seeing around 1100 clients per year, 11 it is apparent that not all individuals with VI can access support from this source. It is also important that individuals have ongoing support, as their requirements are likely to change considerably over time even if their eye condition remains stable. 12 Of course, even if individuals are told about different types of support, it does not mean that they will be willing to engage with it, 13 although for some, an initial lack of acceptance of their diagnosis will be overcome in time, and they would benefit from information at a later date.
The requirements of easy access to comprehensive information, at a time of the individual's choosing, seems an ideal match to provision of information online. Henshaws Knowledge Village, 14 Sight Advice FAQs 15 and Sightline Directory 16 are examples of such sources in the UK, although awareness of these resources is far from universal. Access is also an issue with one-third of individuals with VI reporting their use of devices was limited by their disability, 17 since it can be difficult to enter queries, or to access search results. 18 The Artificial Intelligence in Visual Impairment (AIVI) Study is a proof-of-concept study to investigate whether ongoing support for people with VI can be provided by a dialoguebased digital assistant. The potential applications of conversational agents with natural language input capabilities are being investigated for a variety of different purposes, across a wide range of healthcare related fields (for reviews see Laranjo et al. 19 and Schachner et al. 20 ) .
The phase of the AIVI Study reported in this paper aimed to determine the information needs of those with VI; explore how they currently source and receive information, and any barriers to obtaining information; and investigate what they would want from a personal digital assistant. These findings can be used to inform the design of dialogue systems for the VI population.

Methods
Two methods of qualitative data collection were used: a survey was used to collect a wide range of opinions, and interviews were used to explore opinions in greater detail. Participants were UK resident adults and either: visually impaired; a carer or family member of someone with VI; or a professional involved in the support of those with VI. Some participants fell into more than one category.
The survey investigated the information needs of people with VI, covering all the aspects in the model proposed by Moore: 21 that is, what are the topics of information, who provides it, and in what formats, how do needs differ between individuals, and do individuals have enough understanding of the information, and the wider context, to make appropriate decisions. An Expert Advisory Group included eye-care professionals, rehabilitation workers and individuals from a range of charities that specialise in providing support for those with VI (see Acknowledgements). The Group provided guidance on the content, format and accessibility of the survey (including the design for compatibility with screenreaders). The initial survey questions captured participant demographic data and these were followed by 5 free text response questions exploring aspects of the information needs of those with VI ( Figure 1). There was also a final space for (optional) additional comments. Open questions were used to allow for a broad range of opinions to be expressed, without the constraints of a pre-determined options list. The questionnaire (see Supplement) was distributed online through The University of Manchester using the SelectSurvey platform (SelectSurvey.NET™, ClassApps Inc.) in February-April 2021, and publicised through newsletters and/or social media of UK-based charity partners and through professional networks (see Acknowledgements).
In addition, in-depth 1:1 semi-structured interviews were conducted between February-August 2021 to explore opinions in more detail. An interview schedule was developed (see Supplement) to structure the interviews, in consultation with the Expert Advisory Group. Participants for the interview were recruited through the Macular Society and professional network groups. Purposive sampling was used to target a mix of people with differing demographic and visual status to investigate a range of experiences regarding the information needs of those with VI, from different perspectives.
The survey responses and interview transcripts were combined and analysed using thematic analysis. 22 This analysis method involved a sequential multi-stage process that included familiarisation with the survey data set by re- Figure 1. The five free-text questions which were asked in the online survey.
reading the transcripts, initial manual coding of the data, grouping codes into themes (by JT), reviewing themes (in discussion with the research team to ensure the validity of the analysis), and naming and defining themes. In the final stage of report production appropriate quotes were chosen from the dataset which justified the research findings. Illustrative quotes were assigned by participant number, with designations of "QP" and "IP" being used to represent questionnaire and interview participants respectively. An extended illustrative data set of quotes for all the topics discussed in the paper is available in the Supplement.
The study was conducted in accordance with the Declaration of Helsinki. It received a favourable opinion from the University of Manchester Research Ethics Committee, and all participants gave consent.

Results
A total of 120 individuals completed the survey: 119 answered all 5 questions, and 69 (57%) provided additional comments. A total of 10 participants were invited to take part in the interviews: at this point it was judged that data saturation had been reached. Table 1 shows the demographic characteristics of the study participants for both the questionnaire and the interview.
The vast majority of the participants had VI: although of those responding to the questionnaire, 49% (51/104) had difficulties with vision but were not registered, whereas all of the interviewees with VI were registered with 5/8 being SSI. The interviewees were evenly split between male and female. A broad range of ages were represented in both groups, although the majority (58.3%, 70/120) of questionnaire responders were 70 years of age or over. In this group 90.8% of the group had access to the internet and 75.4% had access to a smartphone, with 58.3% and 53.3% reporting use of a laptop and desktop respectively. All interviewees reported access to the internet and either a smartphone or computer, and the majority reported access to, or use of, voice-activated technology. Figure 2 summarises the seven broad themes which emerged as the most important topics of information. Participants felt it was important to have an understandable explanation of what their eye condition was, including details of relevant test results, why it had happened, and the availability and effectiveness of any treatment. Information on the speed and magnitude of possible future deterioration, and advice on managing their condition were also important (e.g. how to alleviate symptoms or prevent deterioration, what to do if their condition changes or deteriorates, how they will be monitored, and if/when to get new spectacles).

What information is important?
Participants reported a need for information, advice and guidance on how to perform a wide range of practical daily living activities (e.g. cooking, shopping, hobbies and looking after children). Interlinked with this was the need to know about assistive equipment and technology, and the training to enable devices to be used optimally. Information on the registration process (i.e. where and how to register) and the benefits (i.e. how registering would help them) were also important. This was also related to what financial support and government benefits were available. This included the support to individuals to remain in work; plus the discounts, concessions and cost saving schemes that are available across a broad range of everyday activities (e.g. travel and entertainment). Emotional support needs were also identified by participants, as they expressed the fears and concerns that VI causes. Formal support (e.g. counselling) was mentioned, but peer support was also very important. Providing information for people that help support and care for those with VI was also highlighted. This included signposting to support organisations, and guidance on how carers could provide support while also allowing those with VI to maintain their independence.

How should information be delivered?
The analysis identified four overarching themes that ran through the seven broad topics of information (Table 2). Participants often described a distinction between generic information (applicable to all or many) and targeted information (applicable to the individual, and their condition, interests and circumstances). Whilst generic information was Table 1. Demographic characteristics of study participants. (In the questionnaire, total participants = 120). *for these characteristics multiple categories were possible. # These data were not available from the anonymous survey. recognised as important, some participants felt that much of the information they received was too generic and not specific enough for their needs. The format of information was another key theme emphasised by many. It was key that information should be in an accessible format, and frustration was experienced when this was not the case. The timing of access to information during their sight loss journey was particularly important. There were some that felt they wanted a lot of information at diagnosis (or as soon as possible afterwards), to avoid any delay in moving forward. However, more commonly, individuals reported needing a period of adaptation after diagnosis before being in a position to receive and absorb further information beyond the initial overview and signposting. Results suggest that people's information needs evolve during the course of their sight loss, based on circumstances at the time, and when they are ready to process that information. The timing of information was also linked to the quantity of information received at  different points (i.e. not receiving enough information, or conversely, being overwhelmed by too much information). It is therefore clear that there is an important balance to be found between the type and quantity of information someone receives, and when they receive it. Signposting to sources of support and information was often mentioned by participants. People wanted to know about relevant sight loss organisations, the type of support they offer and how to access it. The distinction was often made between national and local organisations, with the perception being that the former tended to provide more generic advice (applicable to many), and the latter tending to provide more targeted information (e.g. travel schemes within a particular locality, or access to local support groups).

How is information accessed?
Participants commonly reported accessing information using multiple sources and/or multiple methods, suggesting that no single source provided all their information needs (Table 3). Although use of online information sources broadens access to available information, participants highlighted the challenges associated with navigating the plethora of information sources and identifying the most appropriate. Some sources appeared to be associated with certain types of information (e.g. diagnosis or prognosis of condition, from healthcare professionals). Other sources were viewed as more multi-faceted, providing information across a range of the most important topics. Particularly notable was the wide variety of advice and information sourced through peer support (in organised groups, or as acquaintances or friends). Peers could describe how they managed with daily activities, their use of new and existing technology, employment experiences, the benefits process, and the organisations they had been in contact with. Peers also provided a social connection, and were understanding and empathetic.
The sources of information used, appeared to evolve with time: for example, information from hospitals tended to be earlier in the sight loss journey, with information from other organisations following at later stages. The methods which participants used to access information from the different sources was dependant both on how organisations provided the information, and on the technology people had access to and were able to use. A broad range of barriers to obtaining information were described (Figure 3). Sometimes information was felt to be simply lacking, or it was in an inaccessible format. Difficulties using technology became barriers to accessing information: other factors included the limited availability and prohibitive cost of some technology. Participants commonly reported a lack of awareness of what services were available, and uncertainty as to what questions to ask, particularly in the early stages of VI. There were some participants who reported not accessing any information, and not identifying any good sources to go to. The lack of a joined-up approach to the provision of information, was a source of frustration and stress for some. The global COVID-19 pandemic leading to the temporary suspension of some services, was a further barrier.  How could a digital assistant be helpful? Table 4 shows positive and negative opinions on whether a digital assistant would be helpful to accessing information.
The main positive reasons were: ease of use in using voice recognition (in comparison to difficulties reading or typing) and the timing of access to information. The timing was important both in terms of continual access (i.e. 24hrs/day, 7 days/week), and obtaining information in a timely way (i.e. fast responses to questions). Reservations expressed included the reliability and relevance of the information provided, so provenance was important. There was doubt as to whether the information provided could be targeted to the individual, and whether certain types of information could be provided via a dialogue system interface (e.g. difficult/bad news and emotional support). Potential difficulties with device use, and the impersonal nature of a computerised device, were also mentioned. Some also felt that a further device was unnecessary for them as they could manage access to information with existing assistive low vision aids: some suggested that it may become useful if their condition progressed. Interestingly, the theme of privacy led to potential reasons for and against the use of a digital assistant. While some reported that they would find it easier to talk to a device like this without the fear of being judged, others reported security concerns with respect to personal data and the internet.
When questioned about the format/platform for the digital assistant, the preferred choices covered all possible options. It was therefore more relevant to look at the reasons for those choices (Figure 4), showing that this was largely due to ease of use, accessibility, familiarity or functionality. "Ease of use" was often identified based on visual access to the display and controls, and this may be less relevant for interacting with this system. In general, people wanted the digital assistant to be available on a range of devices or platforms. This was primarily because they often have access to multiple devices and/or use different devices for specific tasks or in particular situations: they therefore wanted something that suited their needs at that point in time.

Discussion
This study explored the information needs and behaviour of individuals with VI in relation to their rehabilitation and adaptation to sight loss. In particular, the information they deemed most important, the current methods they use to access it and how they would view the use of a dialogue system as a supporting device. Many of the sources of information reported by participants were in broad agreement with those reported in previous studies (see e.g. Beverley et al. 23 ). The barriers to accessing information were also similar (for a review see Lam & Leat 24 ). despite the relative familiarity of responders with technology. The wide variety of information sources reported by participants may be reflective of the lack of a formal support pathway, leaving individuals to use their own initiative to source the information they need. 7 Peer support (both in organised groups and informal one-to-one relationships) was repeatedly identified as an extremely important, and very adaptable, information source. The advice is often trusted by those with VI, because it is drawn from the lived and relevant experience of others with VI.
The topics of information that participants identified in this study were also very similar to those identified in previous studies, 23,25-27 falling into the category of "social information" which is that required to function in everyday life. 21 The exact spectrum of information priorities varies between individuals and they want targeted information (in terms of topic, timing and quantity). Despite the reservations of participants, this personalisation is an aspect which digital assistants are wellplaced to deliver, 28 although providing the appropriate level of granularity in responses is challenging. 29 The study confirms other challenges and gaps identified previously in the AI-based dialogue-systems literature. To deliver appropriate information about a wide range of diseases, extensive data collection of conversations between patients and healthcare professionals is needed to build and evaluate healthcare dialogue systems, but such collection is difficult to obtain due to data privacy concerns. The lack of training data in such a domain is a challenge when providing users with relevant information. 30 Healthcare professionals should be involved in designing such systems, providing data closer to the

Themes
Illustrative quotations Positive aspects (1) Ease of use.
QP21: "Most certainly as using limited sight to access computer/phone/tablet can be very stressful".

QP82: "Yes as sometimes using a screen for a long while makes my eyes tired and gives me a headache".
QP119: "Yes that would be easier to access if voice activated". QP47: " It is a relief to be told things rather than always having to mess about with magnification devices".
QP26: "Yes, because it would be easier than typing a question and hopefully would receive a fast response".
QP10: "Yes. Digital assistants are becoming widely available. On demand access to information is extremely powerful and more cost effective than manning a round the clock helpline". patient's real information needs. 31 Users have also previously raised trust as an essential factor in deciding whether to use a system. Opaque systems tend to decrease user confidence since people tend to be uncertain if the output information provided is correct. 32,33 Concerns about the use and storage of personal data are also prevalent. 34 In order to be accepted in sensitive domains such as health, conversational interfaces must extend their assistance capabilities and have human-like assistant behaviour. 35 Users of dialogue systems emphasise the convenience they bring into their daily routine, significantly simplifying arduous tasks. 34 However, visually impaired users will often have difficulties discovering new functionalities in these types of systems. 36 It would be a major strength if the digital assistant was able to facilitate undirected search which was more exploratory in nature and may be open-ended and iterative. 18 This would address the needs of some users who are unaware of what support is available and are unsure what question to ask. In both these aspects, using AI algorithms to provide the digital assistant with the ability to initiate some of the conversations with the user, may provide a major advantage over a conventional "internet search" for information. The digital assistant could become a very useful supplement to existing methods for obtaining information.

Recommendations for the next stage of the AIVI study
Through the study analysis, it was possible to identify the following specific requirements for the next stage of the AIVI study when developing a dialogue system to assist VI individuals: (1) Relevance via context understanding: The information delivered should be contextualised within the unique disability setting of the user. (2) Disability modelling and patient pathway understanding: the system needs to commit to an ontological (conceptual) model of the clinical problem space and possible trajectories of support and progression of the condition and rehabilitative process. (3) To enable highly contextually sensitive and concise answers, achieving completeness with conciseness. (4) Transparency, explainability and trust: End-users should know enough about the system to form a mental model of its operation. Trust will be achieved by understanding how the information provided by the system is curated, how their data is stored, and how the answer was obtained. (5) Accessibility to a diverse audience: The user expects that the system can understand their utterances and return the answers conveniently and matched to their understanding level. It should cope with speech and language variations and communicate back in an accessible discourse.

Limitations
There was no attempt here to recruit all types of individuals with VI, and these would be the subject of follow-up studies.
The online questionnaire was biased to responses from those more familiar with technology, and individuals with communication difficulties were not specifically recruited. The questionnaire was only available in English, so it is likely that ethnic minorities were under-represented. Although carers were recruited to the study, and issues regarding support for carers emerged, the emphasis here was on the individuals with VI themselves (carers were asked about the needs of individuals with VI, rather than their own needs). Other research studies 37,38 have investigated the needs of carers, and these needs can be specifically addressed in digital assistant design.

Conclusion
Conversational agents could meet an identified need in the provision of personalised information to those with VI, and offer important advantages including: ease of use (accessible format), control over the timing, quantity and type of information provided, help with interpreting the large array of information sources available, and selecting information to help support those who do not know what to ask about. The findings will be useful for the development of dialogue systems for the VI population to support disease and rehabilitation management.