Chapter 2: Disability data

What is disability data and why is it important?

Disability data refers to information regarding an individual’s disability (eg type, severity and support requirements); a disabled person’s other personal data (eg demographic details, medical history, behavioural data and individual preferences); and national or international information on disability prevalence within a population.

Disability prevalence data can be used to better estimate support needs for a population, evaluate policies and interventions and understand demand for DigAT products and services (footnote 71). This data is often commissioned by policymakers to understand service provision needs and by innovators to investigate potential markets. Given its relevance for ensuring disabled people’s needs are met, it is important the data collected is trustworthy and reliable. The following six criteria, adapted from a list created by the Data Management Association (footnote 72), provide guidance on the characteristics underpinning high quality data:

1. Accuracy
   Data is accurate when it reflects reality. For disability data, this can refer to details being correct for an individual’s health condition, demographic information, or support needs.
2. Completeness
   Data is complete when all necessary information is included. For disability data, this can refer to disabled people being properly represented in large population-level datasets, as well as the data collected reflecting critical relevant information about their disabilities.
3. Uniqueness
   Data is unique if it appears only once in a dataset. For disability data, this can refer to duplicate records of disabled people arising from merged datasets or double counting in data collection.
4. Consistency
   Data is consistent when there are no conflicts within or across different data sets. For disability data, this can refer to an individual’s disability being recorded inconsistently (or not at all) across datasets held by different public bodies.
5. Timeliness
   Data is timely if it is available when expected and needed. For disability data, this can refer to an individual’s health data being tracked on a real-time basis, rather than through ad hoc health appointments.
6. Validity
   Data is valid if it conforms to the expected format, type and range. For disability data, this can refer to specific conditions being listed in a format recognisable by other systems.

In addition to these guidelines, creating data according to the FAIR principles (footnote 73) (where data is findable, accessible, interoperable and reusable) may aid the development of AI-based DigAT trained on data from a diverse range of sources. Data linking can potentially enable the creation of more comprehensive DigAT products as well as mitigating issues with small datasets, however this needs to be done in ways that avoid potential reidentification of individuals in datasets.

Approaches to disability data collection

There are, broadly, two approaches to disability data collection (footnote 74). The first is perceived measures, where researchers or analysts apply their own definition of disability to individuals within a population dataset. This approach involves identifying certain clinically assessed conditions and labelling individuals with these as disabled. It is based in the medical model of disability where disability is considered a health condition to be avoided, managed, or eliminated. The second approach is self-reported measures, in which individuals self-identify as disabled based on their experience of difficulties in undertaking specific activities. In this approach, individuals describe to researchers (eg through surveys) how disabling a condition is in the context of their daily life. This approach accounts for both the medical model of disability as well as the social model (where an individual is impacted by societal barriers in their environment).

Disability-adjusted life years and quality-adjusted life years

Perceived measures draw upon health records and registries for disabling conditions. Health records can include information held by hospitals and general practitioners. Registries can include social welfare initiatives where individuals are required to be registered as disabled to receive specialised support (eg accessible documentation, welfare payments).

Information derived from these datasets is used for calculating disability-adjusted life years (DALYs) and quality-adjusted life years (QALYs). One DALY represents the loss of the equivalent of one year of ‘full health’ due to disability (footnote 75). One QALY represents the equivalent of one year of life in perfect health (footnote 76).

These units are used to determine the ‘burden’ of disability and inform decisions related to resource allocation. They are also used to evaluate health interventions with a reduction in DALYs or an increase in QALYs considered to be a success. These approaches have been criticised for dehumanising disabled people. By optimising for QALYs and DALYs and considering disability a ‘burden’, they are defining disabled people’s lives as less valuable than non-disabled people’s lives (footnote 77), (footnote 78). In particular, evaluating interventions by how they decrease DALYs can result in disabled people having less claim to resources than non-disabled people by limiting the types of support and interventions available to them(footnote 79). These issues are particularly important to consider, in view of the widespread use of QALY and DALY measures in global public health through the ‘Global Burden of Disease’ study (footnote 80) and cost-effectiveness evaluations (footnote 81).

Perceived measures for data collection have also been used by statistics agencies. As of 2021, 43 countries rely on self-reported health conditions in census data collection to ascertain disability prevalence (footnote 82). The UK government’s research and decision-making is frequently based on a medical model, reflected in the Office of National Statistics’ Census question requiring a respondent to confirm whether they have “…a physical or mental health condition or illness expected to last more than 12 months,” before proceeding to answer about activity limitations. If an individual does not identify as having a condition or impairment, but does have activity limitations, they will not qualify as disabled (footnote 83).

Disability data based on health records and registries can result in inaccuracies and bias. For example, reported estimates may be underestimates due to health diagnostic services being difficult to access due to user costs, lack of resources in the healthcare system, or inaccessible health facilities (footnote 84). Other factors, such as distrust in medical services, privacy concerns and stigma around registry data collection, can affect the accuracy of this model of data collection.

Self-identification

Disability can form part of people’s perception of their identity, akin to gender, ethnicity and age. This identity can be subjective and motivated by numerous factors (footnote 85). It is, therefore, too inconsistent to be used as a method for making decisions on resource allocations. For example, older people who experience changes which introduce functional challenges to their everyday lives may not renegotiate their identity or consider themselves ‘disabled’ (footnote 86).

Wearable sensors

Wearable technologies (eg smart watches, sensors on prosthetics) can be used to collect data on the prevalence and nature of disability (footnote 87). The use of these technologies has been applied to analyse wheelchair users’ quality of mobility (footnote 88), (footnote 89), detect dementia (footnote 90) and to capture levels of pain (footnote 91). In recent years, major technology companies including Google, Apple and Samsung, have developed proprietary health platforms through which to securely collect, visualise and automate reports on health-related activity (footnote 92).

Functional assessments

The diverse, inconsistent and non-standardised approaches to disability data collection present significant challenges to policymakers and the developers of new DigAT. Some methods offer simplicity over utility, while others risk misrepresentation and exclusion. To overcome this, the United Nations’ Washington Group on Disability Statistics developed a functional assessment (known as the Washington Group questions) designed to be integrated into existing national data collection (footnote 93). The questions aim to focus on difficulties people may have undertaking functional activities, applicable across all nations, cultures and societies. There are several question sets, including sets focused on children, education and work. The short set (see Box 1) includes six questions.

As of 2023, 70 countries have used the short set in at least one wave of national data collection (footnote 94). It has also been integrated into the World Health Organization’s Rapid Assistive Technology Assessment survey (footnote 95).

Another global standard is the World Health Organization’s International Classification of Functioning, Disability and Health (ICF) (footnote 96). This framework adopts the biopsychosocial model of disability (encompassing biological, psychological and social factors) (footnote 97) and contains detailed, comprehensive questions which can be used by healthcare professionals, as well as policymakers, to design interventions or rehabilitation plans for patients.

The self-report bias that can occur with the Washington Group questions, the ICF and other functional assessments, has led to criticism of the approach due to the potential of estimating a higher prevalence of needs compared to clinical assessments (footnote 98). Furthermore, the subjective nature of the questions can lead to inconsistencies in how the responses are interpreted (ie ‘some difficulty’ or ‘a lot of difficulty’) (footnote 99). However, they have advantages compared to perceived measures. By avoiding explicit mention of disability identity or diagnosis, functional assessments can reduce misreporting resulting from stigma (footnote 100) or from those who have functional challenges but do not identify as disabled, such as older people (footnote 101).

Functional assessments are also more easily linked to specific supporting interventions, such as DigAT for mobility, rather than assuming the same needs apply across all individuals with the same disability (footnote 102).

Box 1

The Washington Group Short Set on Functioning

Respondents are asked to answer the following questions with one of four options: ‘no difficulty’, ‘some difficulty’, ‘a lot of difficulty’, or ‘cannot do at all’.

Vision

[Do/Does] [you/he/she] have difficulty seeing, even if wearing glasses?

Hearing

[Do/Does] [you/he/she] have difficulty hearing, even if using a hearing aid(s)?

Mobility

[Do/Does] [you/he/she] have difficulty walking or climbing steps?

Cognition

[Do/Does] [you/he/she] have difficulty remembering or concentrating?

Self-care

[Do/Does] [you/he/she] have difficulty with self-care, such as washing all over or dressing?

Communication

Using [your/his/her] usual language, [do/does] [you/he/she] have difficulty communicating, for example understanding or being understood?

Challenges leading to representation gaps

Irrespective of which approach is used to measure disability, major challenges exist with the collection of data from disabled people. Drawing on the Danemayer and Holloway review (footnote 103), the following five core challenges to disability data representation have been identified.

Exclusionary survey designs

Surveys can lead to representation gaps when they do not account for a broad range of disabled respondents from diverse backgrounds. This includes a failure to support respondents who may have intellectual and communication disabilities, which may require sign language interpreters or adapted survey modules. Other factors include an overreliance on institutionalised populations (eg those in education or work) and accessibility barriers at physical or online data collection points. Data collection may prioritise more common disabilities at the risk of excluding those with rarer disabilities, where data is harder to collect or less easy to represent in datasets.

Furthermore, in some cases, disabled people may be ineligible to enrol in population cohort studies altogether. Disabled people may have less time or energy available for tasks and be more selective about how to spend their time, for example, by opting-out of long surveys if these are perceived to not directly benefit them. Co-design of research with disabled people can help address issues around inclusivity and consider creative methods of data collection (footnote 104).

Social stigma

Internal stigma surrounding disabilities can lead to some respondents not wishing to share their own or their children’s disability to researchers. Meanwhile, external stigma from researchers may lead them to exclude disabled respondents from their study or to not consider questions which may be relevant to the lives of disabled people. Due to discrimination, disabled people may also be underrepresented in the organisations conducting research which may contribute to low prioritisation of inclusive research design.

Machine learning trained on biased data

AI systems using machine learning algorithms trained on unrepresentative data can lead to poor quality outputs for disabled people. If disabled people are absent or underrepresented in the data used to train and develop AI applications for research or decision-making, the outputs are likely to also be unrepresentative and exacerbate further exclusion of disabled people in future research projects.

Low resources of disabled people

Disabled people are less likely to be part of the labour market than non-disabled people and when they are, they tend to earn less (footnote 105). Overall, an estimated 80% of disabled people globally live in low-resource settings. Digital exclusion, where disabled people lack access to the internet or internet-connected devices, can make it challenging to develop sufficient image and language datasets for disability. It also presents difficulties for creating inclusive datasets based on wearable technologies. Data collection using wearable technologies or DigAT needs to account for differential access to technology and differential capacity to use it.

Trust and engagement

Without effective methods for community engagement, it can be difficult to identify, reach, or gain consent from disabled people for a research project or public sector data collection. In addition, historical experiences of discrimination from institutions or disillusionment based on participation in previous research activities can contribute to lower levels of trust and engagement from disabled people. For example, disabled people may be cautious about sharing information with public sector bodies or researchers due to concerns about how this data may be shared or used by other government bodies (such as for disability benefit assessments). Overcoming issues related to trust will require clear data governance and consent policies alongside contextual factors related to cultures, regions and nations.

Case study 2: DigAT for gaming

Disabled people are a large but currently underserved part of the gaming community with nearly a third of gamers in the UK and US identifying as disabled, with mental health conditions the most reported disability (footnote 106). Two thirds of disabled UK gamers report experiencing challenges related to gaming (footnote 107). With the average age of gamers rising, a significant proportion of future gamers will need more inclusive gaming options. This case study draws on a roundtable jointly organised with PlayStation conducted for this report in July 2024.

Opportunities

Developers are increasingly using DigAT to create more accessible gaming hardware and software options for disabled people. At the Game Developer’s Conference in 2024, in a significant increase from previous years, nearly half of surveyed attendees reported their current products including accessibility measures, such as closed captioning or colourblind modes (footnote 108). More customisable features allow disabled players to adjust gameplay, visuals and controls. For example, Grounded includes an arachnophobia safe mode, designed to make gameplay safer for people with phobias and Sea of Thieves includes additional audio settings for Blind and partially sighted gamers (footnote 109).

Examples such as PlayStation’s Access Controller, the Xbox Adaptive Controller and the Microsoft Proteus Controller allow players to tailor controllers to their needs. This can be particularly helpful for gamers with limited mobility who can use alternative controls, such as large buttons or foot pedals, to interact with games in more accessible ways. Community feedback, through Discord, is being used by developers to seek ideas for accessibility upgrades to continuously improve gaming products.

Disabled gamers often struggle to determine whether games are accessible and meet their needs. A 2021 survey suggests 40% of disabled gamers have purchased games they are not able to play due to poor pre-purchase accessibility information, with some gamers unable to return inaccessible games (footnote 110). Improving the availability of accessibility information before purchase would allow disabled gamers to make informed decisions. Dedicated platforms, such as the Game Accessibility Nexus and ‘Can I Play That?’, enable disabled gamers to check accessibility data before purchasing and avoid them having to extensively research different sources. Accessibility Tags also allow game developers to provide detailed insight on their games’ accessibility features with over 300 games in the PlayStation Store using accessibility tags to help users make decisions.

Challenges

Although DigAT for gaming has seen significant advances in recent years, there remain challenges with regards to access to DigAT. According to a 2020 survey, the most significant challenge for disabled gamers is the affordability of assistive technology, with 30% reporting it as a barrier (footnote 111). Companies, such as Ubisoft, Sony and Microsoft, often use inconsistent terminology for accessibility, which can lead to difficulty in making appropriate purchases. Organisations, such as Makers Making Change in Canada and the Controller Project, have initiatives to increase access to gaming assistive technologies such as the GAME Checkpoints program for disabled gamers to trial gaming devices with trained professionals before purchasing or using 3D printers to provide free assistive controller add-ons (footnote 112).

Since there is no legislation mandating minimum accessibility standards for games, disabled gamers are reliant on industry-led initiatives which can vary across companies and regions. Progress is often driven by internal accessibility advocates which can lead to burnout when companies are not welcoming to disabled designers.

Making internal business cases for DigAT for gaming can also be challenging, where companies often use data on disability prevalence. This can be misleading as it fails to acknowledge how assistive features are used more widely, such as subtitles often being used by non-disabled gamers. Retroactive accessibility updates are particularly challenging due to obsolescence where the original game developer is no longer in business and there is no ability to update the game.

Interoperability of DigAT is key to allowing users to switch to different platforms and devices without having to reconfigure their accessibility settings. However, interoperability of assistive controllers can enable them to be used in applications beyond gaming, such as e-sports, where there are risks of players cheating by adapting their controllers. Gaming developers can also be reluctant to use tools which may improve accessibility, such as generative AI, due to ethical concerns with how they’re developed.

Example: Eye-tracking technologies for gaming

SpecialEffect, a gaming charity supporting physically disabled gamers, has created a suite of games called Eye Gaze Games, which uses eye-tracking software for gaming (footnote 113). They also offer solutions that can be adapted to other games such as Minecraft by creating an overlay that sits on top of the game.

Conclusion

Gaming showcases the opportunities of an innovative industry recognising that games designed to be accessible are good for all gamers, expanding access to DigAT. However, there are challenges around industry incentives, inconsistent approaches to accessibility and affordability.