Sexual health service users’ perspectives on artificial intelligence applications for identification of lesions associated with sexually transmissible infections: A qualitative study

Abstract

Background: Early identification of sexually transmissible infections (STIs) is central to their control by facilitating timely access to healthcare. A web-based sexual health application using artificial intelligence (AI) to identify the risk of a genital or anal lesion being an STI, from a photographic image, is in development. However, the usefulness, accessibility, and acceptability of this technology to potential users are unknown. Method: This research adopted a developmental evaluation approach to explore potential users’ views of the usefulness of this AI application, as well as factors impacting its acceptability and accessibility. Three focus groups were conducted with attendees of a sexual health clinic. Focus group discussions were recorded, transcribed, and analysed using action coding to identify key concepts. Results: A diverse group of 12 participants provided feedback on the proposed application. Feedback on the application's usefulness highlighted its potential value in supporting access to individualised sexual health information; informing decisions about healthcare seeking; and preventing transmission to others. Factors mediating the accessibility and acceptability of AI-powered applications included potential users’ awareness of the application; concerns about data security; the accessibility of the application to diverse user groups; trust in the technology; and transparency about the application's limitations. Conclusion: AI applications for the identification of visible changes in the ano-genital region have the potential to support access to sexual health information, healthcare seeking and the prevention of onward transmission of STIs. Codesign approaches will increase the likelihood of more widespread uptake and public health impact of AI-powered sexual health applications.

Background

Early identification of sexually transmissible infections (STIs) is central to their control. ¹ The longer an STI goes undiagnosed, the more likely it is to be passed to a sexual partner and result in poorer health outcomes. STIs that present with lesions, such as syphilis, genital warts, and genital herpes, may or may not be recognised by individuals as STIs, resulting in delays in testing and treatment, and an increased risk of onward transmission.

Syphilis and human papillomavirus (HPV) have been named in the top three infections of major global significance by the World Health Organisation ² owing to their high incidence and prevalence, and potential for serious adverse outcomes. It is estimated that 7.1 million people were diagnosed with syphilis globally in 2020, with rising incidences noted in the United States, ³ the United Kingdom ¹ and Australia. ⁴ If left untreated, syphilis can result in disability through neurological damage ⁵ and congenital syphilis is a leading cause of miscarriage, stillbirth, and infant morbidity and mortality.^6,7 HPV, the virus that causes ano-genital warts, is the main contributor to cervical cancer, which globally affected 604,000 people in 2020, and is associated with other cancers. ²

It is estimated 500 million people worldwide have genital herpes simplex virus (HSV) infection. ² Whilst less concerning than other STIs in terms of clinical outcomes, a potential herpes diagnosis has been described as equally distressing as syphilis by members of at-risk populations.^8,9

A review by Tucker et al. ¹⁰ described the potential benefits of digital sexual health interventions in facilitating engagement with in-person sexual health services. Two included reviews focused on facilitating engagement using social media, ¹¹ including the use of short messaging services (SMSs). ¹² In their review of 35 studies on social media, Taggart et al. ¹¹ found high levels of user satisfaction and ease of use, as well as the potential for improvements in knowledge, engagement in STI testing, and identification of infections. Subsequently, Palmer et al.'s ¹² review of 27 studies of communication via SMS found some support for this technology increasing service utilisation. Both reviews highlighted potential risks associated with confidentiality inherent in social media interventions.

Web-based sexual health applications offer another means of providing sexual health information and promoting access to digital or in-person sexual health services. Web-based applications providing assessment of risk based on users’ responses to questions about risk factors are helpful to potential users.^13,14 More recently, similar applications have been improved by the use of a Bayesian-powered mathematical algorithm to identify potential STIs in people experiencing ano-genital symptoms, ¹⁵ and machine learning algorithms to accurately predict individual risk of STIs with acceptable levels of confidence. ¹⁶ Such applications have relied upon users’ interpretation of questions about risk factors^13–16 and symptoms. ¹⁶

The application of artificial intelligence (AI) to the identification of STIs from photographic images represents another leap forward in improving the accuracy of online assessments of STI risk. Researchers at the Melbourne Sexual Health Centre (MSHC) have developed a prototype AI-powered, online, image recognition application, named AiSTi, that enables the identification of possible conditions (including STIs and non-STIs) that might cause a visible abnormality (e.g. lesion, rash, skin changes) in the genital or anal region. However, those working in the digital sexual health space have advocated for human-centred design approaches and transparency in the development of these technologies. ¹⁷ This qualitative research sought to explore sexual health service users’ views of the usefulness of this technology as well as its acceptability and accessibility to potential users.

This sub-study was conducted as part of the STARTOnline (Supporting Timely and Appropriate Review and Treatment Online) study which sought feedback on three web-based sexual health applications, including the AiSTi application that is the topic of this paper. The primary objective of the STARTOnline study was to understand the role of web-based sexual health applications in promoting timely and appropriate STI testing and treatment. A secondary objective of the study was to identify ways to promote web-based sexual health applications to priority populations. The applications evaluated in other sub-studies, MySTIRisk (mystirisk.mshc.org.au) and iSpySTI (ispysti.org), are already available for public use. Thus, further objectives for these applications were to explore participants’ experiences of use and opportunities for improvement. Findings of these sub-studies have been reported elsewhere.^18,19 Application is an umbrella term including web-based, mobile, and desktop applications; however, the term ‘application' in this paper refers to a web-based application, unless otherwise specified.

Methods

This manuscript has been written with reference to the Standards for Reporting Qualitative Research ²⁰ (Supplement 1).

A developmental evaluation approach ²¹ was employed to inform the further development of the AiSTi application for public use, including the user interface, supporting information and promotion. This initial study employed qualitative research design, using semi-structured focus groups and analytic methods described by Saldaña ²² and Bazeley, ²³ to explore the usefulness, accessibility and acceptability of the proposed application. A further study subsequently sought additional feedback via discrete choice experiment ²⁴ and further user testing is planned.

Description of the application development and features

A prototype of the AiSTi application was developed and tested for accuracy ²⁵ with attendees of the MSHC, a publicly funded sexual health clinic in metropolitan Melbourne, Australia, which provides free STI testing and treatment in about 60,000 consultations per year. A database of over 10,000 clinically verified photographic images of visible sexual health conditions taken at MSHC was used to train the application using deep learning algorithms. The application provides the choice of uploading an image they have taken with their own smartphone directly via an anonymous and encrypted weblink, without storing an image in their smartphone, or selecting an image they have taken previously from their smartphone photo library. The deep learning algorithms analyse the morphological features and patterns in the uploaded image and compare it to the training dataset. Following analysis, a short list of possible conditions is provided with corresponding percentage likelihood of being that condition.

An adapted version of the prototype application, hereafter referred to as ‘AiSTi’, is being developed for public use. Another study will assess the value of additional data collected by questions at the start of the AiSTi application, such as age, information on sexual partners, and recency of sexual contact, in improving the application's accuracy. AiSTi will encourage users to seek clinical confirmation of results and provide links to sexual health information, including how to access testing and treatment.

Recruitment and consent

A convenience sample of MSHC attendees who had participated in research about other web-based sexual health applications^18,19 and given permission to be contacted about participation in a focus group were invited by SMS and email. All who expressed an interest in a focus group and provided their availability were invited to a focus group. Participants in previous studies had been purposively sampled to ensure a diverse range of perspectives.^18,19 A Qualtrics survey was used to determine participants’ availability and preference for online versus face-to-face focus group. Participants were contacted by telephone by one of the co-authors (AK) to confirm their availability and preference for same or mixed gender focus group with all indicating they were comfortable with the latter. Participants provided verbal consent via telephone prior to the focus group, or written consent at the time of the focus group, having received participant information via email in advance.

Data collection

Focus groups were chosen as a method to allow participant discussion. AK and NS facilitated two focus groups via Zoom (FG1 and FG2), and one face-to-face focus group (FG3), in the first 2 weeks of November 2023. AK is a female PhD-qualified research fellow with expertise in qualitative health research. NS is a male doctoral candidate with a medical and public health background whose research is focused on AI applications for sexual health. Prior to the focus group both AK and NS had shared their involvement and interest in the research. Neither AK nor NS was involved in participants’ clinical care.

The discussion was guided by a semi-structured focus group guide (see Supplement 2) covering participants’ views on the AiSTi technology and prototype design features (Figures 1–4); their hypothetical use scenarios and responses to results; potential barriers to use; and suggestions for promotion. No pilot testing was conducted of the focus group guide. Given the sensitive nature of sexual health, Zoom participants were advised they could participate with their camera off and change their name. Participants were advised to share only what they felt comfortable about themselves and their sexual practices and asked not to repeat information shared by other participants outside the group.

Figure 1.

Draft landing page shown to focus group participants.

Figure 4.

Draft results page shown to focus group participants.

Figure 2.

Draft questions shown to focus group participants.

Figure 3.

Draft image upload page shown to focus group participants.

Focus group discussions were hypothetical in nature as the AiSTi application has not yet been modified for use outside the clinic. Participants were shown a series of slides showing draft content for pages on the application, including the landing page, instructions, initial questions, image upload and example result pages (see Figures 1–4). Example images of STI (e.g. syphilitic chancre, genital warts) and non-STI (e.g. lichen sclerosis, dermatosis) abnormalities in penile, vulval and anal regions were shown and the relevant probabilities for each. Participants were advised the presented content was open to change and questions about the application's development, capabilities and limitations were answered by NS throughout the focus groups.

Focus groups were audio-recorded and transcribed by a professional transcription service. Transcripts were not returned to participants. Demographic information was also collected outside of the focus group using a questionnaire administered by AK (see Supplement 3).

Data analysis

Focus group transcripts were checked for accuracy and uploaded to NVivo 20 Software for coding and data management purposes. Transcripts were manually coded inductively by AK using action (or process) coding which involves allocating phrases beginning with gerunds (-ing words) to sections of text representing a participant's cognitive, emotional or behavioural responses ²² (see examples in Supplement 4). Similar action codes were grouped together in pattern codes ²² throughout the coding process (see example coding tree in Supplement 4).

Similar pattern codes were spatially arranged into groups representing key concepts in a concept map. ²³ After each focus group was analysed, the concept map was progressively modified to reflect the meaning generated from the data. TP read all three focus group transcripts, and the final concept map was reviewed by TP, EC and CF, resulting in minor changes (e.g. names of concepts). Consistent with Braun and Clarke's position on data saturation, ²⁶ no further focus groups were deemed to be necessary as sufficient meaning had been generated to achieve the study aims.

Once all three focus groups had been coded, pattern codes were further grouped into coding sets representing key concepts. ²³ Data coded to the combined pattern codes within each coding set was re-read to produce a concept description (see Supplement 5 for example) which forms the basis of findings presented below. An example coding tree showing action codes and pattern codes within the coding set for the Informing my healthcare seeking concept is provided in Supplement 4.

Reflexivity

The following measures were undertaken to support the trustworthiness of findings. Reflective field notes were recorded by AK following each focus group, ²⁷ and AK and NS met after each group to discuss initial impressions of the data collected. AK, TP and EC were not involved in the development of the AiSTi tool and thus provided – in addition to participants’ perspectives – new perspectives on the tool. Both positive and negative feedback on the tool's value and impact is included in the reported findings.

Ethical review

Ethics approval was obtained from Alfred Hospital Ethics Committee, Melbourne, Australia (approval no. 39/23) on 31 January 2023. Participants received financial compensation of $50AUD (as an e-gift voucher) for their time.

Findings

Participant characteristics

In total, 14 participants from previous studies^18,19 who had expressed an interest were invited to a focus group but only 12 participated in one of three focus groups of 72 to 81 min duration. Two potential participants withdrew after consent, and prior to participation. Of these, one did not attend without explanation and did not respond to a further invitation. Another was unable to attend due to illness.

Participant characteristics are summarised in Table 1. Each focus group comprised four participants. All groups included a mix of ages, genders, sexual identities, and countries of birth. Gender categories such as woman or man were intended to include trans-women and trans-men; however, the one transgender woman who participated indicated she preferred to be identified as such.

Table 1.

Participant characteristics (n = 12).

Participant group	No. of participants
Age group  -29 years  -39 years  ≥40 years	5 5 2
Gender  Woman or female  Man or male  Non-binary  Transgender woman	4 6 1 1
Assigned sex at birth  Male  Female	8 4
Sexual identity  Straight (heterosexual)  Gay or lesbian  Bisexual  Different term	5 1 2 4
Region of birth  Australia  Maritime South-East Asia  United Kingdom  Southern Europe  South America  Southern Asia	4 3 2 1 1 1
Sex work in past 12 months	1
Living with HIV	1

Key concepts conveyed by participants within focus group discussions are displayed in the graphic in Figure 5 and described in the text below with exemplary quotes. The findings are conceptually divided into potential pathways to better sexual health supported by the AiSTi application (shown in the black shapes in Figure 5) and factors mediating the accessibility and acceptability of AI applications (shown in the white block arrows in Figure 5).

Figure 5.

Potential pathways to better sexual health (black shapes with white text) and mediating factors (white arrows with black text).

Briefly, participants described seeing something unusual (i.e. a lesion) in their ano-genital region and taking the first step to care by looking online for sexual health information. Whether or not they found a web-based sexual health application such as AiSTi would be influenced by how successful developers had been in spreading the word about it online and in clinical and public spaces (i.e. promotion). The choice to use a web-based application would be influenced by their perception of the application protecting [their] online identity (i.e. privacy) and design features making it accessible to them (i.e. accessibility). Use of the application would result in them getting an idea of the cause of their STI lesion (i.e. a list of possible diagnoses). Information and advice provided by the application would impact their own and others' health and wellbeing by informing [their] healthcare seeking, arming [them] with information (i.e. preparing them for seeking care) and decisions about temporarily abstaining from sexual contact with the goal of protecting others. Further description of these concepts and exemplary quotes are provided in the text that follows.

Potential pathways to better sexual health supported by the AiSTi application

Analysis of focus group discussions about the AiSTi application revealed an ideal pathway for users of the application, from noticing a lesion to using the application and, as a result, benefitting their own or others’ sexual health.

Seeing something unusual

Participants imagined finding a use for the AiSTi application when they noticed visible changes in the skin condition of their genital or anal region that were outside their normal experience, causing them concern: ‘Most probably, the user[s] would be the people who may experience that physical sign or change in their skin conditions that will prompt them to do the test’ (FG3-1, indicating the first participant in the third FG).

Taking the first step to care

Using AiSTi was described as a potential first step when concerned about a visible change to seek guidance on what to do next ‘I think the whole thing of using it as a guide and a starting point to know what kind of care to seek or what to do’ (FG2-1). Groups particularly felt to benefit from the AiSTi application included those who may be reluctant, or unfamiliar with, accessing sexual health care (e.g. young people, men): ‘I can also see how a stoic fellow might be like, “I just want to check this out, see what it is”. Because, I guess, some people don’t access healthcare’ (FG3-2). AiSTi was also described as more accessible for those in rural and remote areas: ‘…especially targeting those in remote areas, as well. So, most people living in the city. … They have better access to healthcare. But think about the others in regional Victoria’ (FG3-2).

Participants described a preference for seeking information online due to feelings of embarrassment and desire to avoid judgement from others.

So, when you are getting something [genital symptoms], what you normally do is write the symptoms on Google to try to find out what you have, right? … This is still lots of taboo topic. So, I am not going to go to a friend and say, … “Hey, I have this, can you help me?” (FG2-2)

Without access to the AiSTi application participants were conscious of the potential for inaccurate or biased information found via internet search engines.

There's so much rubbish on the internet. It's enough to terrify you. … All you see are the worst cases. The more benign ones, nobody is interested in, so it always looks like bits are just about to fall off. (FG3-3)

Getting an idea

Participants described the AiSTi application as giving them an ‘idea’ of what might be causing a lesion: ‘It would be very helpful to just get an idea’ (FG1-1). They universally liked that the result page provided multiple possibilities.

It didn't just say, even if it was a hundred percent or 90% on one STI. … It also said all the other percentages for other options. I just like having more detail and knowing that it's likely something, but…it could be other things, as well. (FG2-3)

Participants understood that AI applications were not as definitive as diagnosis by a health professional, confirmed by laboratory results.

… sometimes when a doctor sees something [an STI lesion] he can totally tell that it's this [health condition], but he'll still perform the checks [pathology testing] on that. And this is the AI, we can trust the doctor, at some point, but the AI is totally online. (FG1-2)

As a result, they imagined they would err on the side of caution in seeking care after using these applications: ‘If it came back even as low as 20, 25%, even 15%, I feel that would be enough for me to want to go in to get it checked out and tested’ (FG2-4).

Whilst less definitive, AI applications were seen as more accessible, thus influencing the timeliness of care seeking. Participants felt they would use AiSTi earlier than they would seek professional care and in the intervening period whilst waiting to access care: ‘You take around one, two or three days, at least, to get the appointment, sometimes. If it's a public service, it can take a bit longer, as well’ (FG1-3).

Imagined emotional responses to the results ranged between reassurance and anxiety depending on whether the most likely result was treatable (e.g. non-STI or bacterial STI) or incurable (e.g. herpes, genital warts), respectively.

I guess having that assurance overnight thinking, like, ‘Oh, it's not an infection’, but if you still have doubts, you definitely will go to seek professionals. (FG1-2)

Certain conditions are much - something like HSV-2 [herpes], for example, probably has an even bigger stigma about it, and people think, “Oh my God, that's it, my sex life is over. What do I do?” (FG3-3)

Informing my healthcare seeking

Participants imagined getting an idea of what might be causing a lesion would inform their healthcare seeking, particularly deciding to seek care. They emphasised the importance of the site directing people to seek care, regardless of the result.

This is the moment where you need to be direct and yes, just “Please go to the doctor”. Even if it's nothing, it's better to go and that someone says, “Hey, it's nothing”, than maybe just say like, “Oh, maybe it's not that bad”, and then it turns out it was very bad. (FG2-2)

Whilst most participants recognised the medico-legal limitations of providing more tailored advice, one participant felt more specific guidance on how urgently to seek care would be helpful.

I like the idea of having… treatment recommendations, because…, for some things, it might be you need to go in right now like ASAP to an emergency department. … Or, another thing could be, you know, “Go in, but it can be within the next week”. … or something that's not an STI… like, “Keep an eye on it. Check in with your GP”. … knowing what to do about it. (FG2-3)

Knowing they might have an STI would increase the urgency of seeking care for most participants. However, one participant expressed concern that some people may delay seeking care to avoid confirming they have an incurable STI.

I think that some people are like, “Oh, this looks like genital warts”, and then they still have a fear to go to the real doctor. … So, maybe in the results part, elaborate more how to deal with anxiety or things like that, to put more context, so that people have less fear of going to get a real test. (FG3-1)

Practical information about where to seek care was also highlighted by participants as important in taking the next step towards seeking sexual health care: ‘STI centres or clinicians within the area. … Especially if you're not an Australian citizen or if you're not under Medicare… what kind of providers can you access? Are there any low fee ones?’ (FG2-1).

Participants indicated they might also share their results with a healthcare professional, particularly general practitioners, to support them to get the right care.

I think, that [my doctor] could [get] a second opinion on it. … they could make their judgement and compare it to what the app's saying. And if they're both leaning towards the same thing, then that's a good indication it's correct. (FG2-3)

Lesions unlikely to be an STI might be dealt with less urgently and with less stress. In these instances, participants suggested they might seek care at a pharmacy: ‘Well, if it's a minor thing, I mean, we can usually get the medicines from the chemist, but if it's a major one then, of course, we'll see a health professional’ (FG1-2).

Arming myself with information

Distinct from informing their decision to seek care, participants described accessing information online as a means of preparing themselves for the healthcare encounter. The AiSTi application was described as supporting access to trusted sexual health information, prior to accessing care, by identifying specific conditions that users could read further on.

I’d use it, knowing that I'm going to go in, but just for me to maybe relax myself, have some peace of mind knowing that I could have X, Y, and Z and that's what I could be expecting… I'd prefer to have a direct message as to what I could have, even if it's something quite bad… I'd want to be quite well educated on it, as well. (FG2-4)

Participants suggested the AiSTi application link people to information on STIs (e.g. herpes, genital warts) and non-STI (e.g. lichen sclerosis, balanitis) conditions it identified but also STIs that may present asymptomatically or with other symptoms (e.g. bacterial STIs, HIV), including information on treatments: ‘That chance to get some more trusted information from MSHC on, like, “Hey, this is that condition, this is the treatment pathway for this”. … to, kind of, to pre-arm people’ (FG1-4).

Information regarding the prevalence and implications of incurable but treatable STIs, such as herpes and genital warts, was seen as important in reducing distress prior to accessing care. A participant related this to their experience of finding out a partner had herpes, reducing their own concern associated with this diagnosis.

I can remember my own experience, and I was terrified, the first time. … Got a diagnosis, told a recent partner, who said, “Oh, yeah. I’ve already got that”. So, it was, “Oh, okay” … I think to reduce the barrier to getting people to come in, by putting it into some sort of context, would be good. (FG3-3)

Protecting others

Participants also described the benefits of the AiSTi application beyond their own knowledge and healthcare seeking. One participant reported they would temporarily abstain from sexual activity if the application detected a possible STI, whilst seeking clinical confirmation and treatment.

For me … if something is an STI, I then I'm going to abstain from sexual activity. If it turns out it's contact dermatitis, then I may not abstain from sexual activity, right? So, it's going to really affect my behaviour in terms of other people and putting other people at risk. (FG1-4)

Participants also imagined using an AI application to support others to access sexual health information, in a work context.

It would be a great application for sex workers, as well. I mean, especially when clientele who refuse to think that it was a disease or an infection. … Getting your phone and snap it and just say, “See, I told you so!”. (FG1-1)

The community I work within in my healthcare role is low literacy, no literacy, limited tech skills. I can imagine giving my phone to someone and letting them take a photo of it and send it off. (FG1-4)

In this way, the AiSTi application was seen as potentially supporting others access to healthcare and preventing onward transmission of STIs to others: ‘It really is about capturing those people who aren’t getting tested, who maybe have a chronic condition that they need to deal with or is putting a partner at risk or a future partner at risk’ (FG3-3).

Factors mediating the accessibility and acceptability of artificial intelligence-powered sexual health applications

Factors that participants felt might interfere with users’ progress along the pathway described above were also represented in focus group discussions. These are described below in reference to AI-powered sexual health applications, in general, as they are not exclusive to the AiSTi application.

Spreading the word

Discussion amongst participants indicated that the impact of applications will be limited if potential users are not aware of their existence. Several means of promoting applications were suggested by focus group participants.

Online promotion, via search engine optimisation, dating or ‘hook-up’ apps and social media, was suggested. Finding free and trusted information via a popular search engine was challenging.

I think the single most important thing is that people can find it. … my experience was going through all of these private clinics offering this, offering that, “Come and get tested for 240 dollars”. … Whereas I think if people went, “Right, what STI do I have?” or whatever, and this came up… (FG3-3)

Beyond the online space, participants suggested promoting applications in settings that target populations access, such as clinical and educational settings.

When you put up the QR codes in the Melbourne Sexual Health Centre… that's a really good way for people who are curious, who are waiting, to just go and look it up. So, GP clinics are perfect for that and even waiting spaces in student hubs. … I know in public high schools they have a nurse's office or wellbeing office. That would be a really good place for students to find that information… Even in pharmacies. I think that's a really important place because it's a first point of contact for, I guess, finding treatment for a lot of people. (FG2-1)

Marketing in sex on premises venues (e.g. gay saunas, brothels) and public spaces (e.g. on public transport) was also suggested: ‘You can create a campaign, as well. You can have posters everywhere at the tram stops… and spread the word’ (FG3-1).

Protecting my online identity

Whether or not participants would feel safe in using AI-powered sexual health applications was mediated by their perception that their identity would be protected online. For some, a sense of safety was associated with an explicit association with a trusted institution (e.g. health service, university).

If I have to take a picture of myself to put in a website or an app, I like that idea it's not a dodgy app… it's from Monash University. There's a guarantee that there is a serious someone behind. (FG1-1)

Participants also indicated a preference for parsimony in the information they provided applications. Requiring users to register or provide identifiable information such as names and contact details was seen as a barrier to use, particularly in the context of the potential for data leaks.

No data is secure. … There's a lot of data leaks that happened, and certain people have been targeted, like, they’re rich or something. … They use it as, “We’re going to release this little image of you” trying to scam a bit of money out of you. (FG3-2)

Conversely, providing deidentified data such as images, sexual orientation, sex, age and recency of sexual contact was seen as acceptable, if contributing to a more accurate result: ‘I don’t mind sharing my age or my sexual orientation, as long as you don’t ask for my email or my phone. … Sexual orientation is important, I feel. Gender, age is also important’ (FG3-4).

Transparency around the storage of data, seeking permission if necessary, built trust in applications. However, advanced means of identifying users such as canvas fingerprinting were seen as largely unavoidable for the average user.

There's a trace whenever using a device. We call it fingerprinting canvas. … Unless you are really IT secure, which is you remove all your fingerprinting from the moment that you use the device, or blocking trackers, things like that. That happens, but it's not like generic… a normal user. (FG3-1)

Making it accessible to everyone

The potential impact of AI-powered sexual health applications was described as mediated by their accessibility to a broad range of users. Participants advocated for universal design that would accommodate the needs of specific groups and minimise barriers to use.

Ease of use was supported by minimising the number of steps involved in getting a result such as asking the minimum number of additional questions to achieve an accurate result. Participants queried the inclusion of questions relating to gender, rather than assigned sex at birth or genital anatomy, if they did not contribute to accuracy.

Don’t ask me my gender if it doesn’t matter. I don’t like being asked if it doesn’t matter. … Ask as few questions as possible. … And particularly those questions that may be a barrier, like the binary gender stuff. (FG1-4)

Explaining the relevance of questions was also suggested.

Sometimes you see on surveys, or you’re filling something in, and it will have a little question mark where you can basically go, “What's the relevance of this question?” and it says, “This could help us differentiate between A and B, or give us more information”. (FG3-3)

Other suggestions relating to ease of use included use of menus to allow users to skip to what they needed and providing the option of a saved profile within an application for frequent users: ‘My preference would be towards app, as well. It would be very handy and easier and convenient to upload the pictures’ (FG1-2). Concerns about privacy, however, influenced several participants’ preference for a website (i.e. web-based application) over a smartphone application.

I would prefer to use it on the website. I wouldn’t want to have the application on my phone. … If someone was to go onto my phone… and if they were to see that then, yeah, I think – it's not an embarrassing thing, but I think I’d get quite – yeah, I’m not sure. (FG2-4)

Participants felt charging for use of sexual health applications presented a barrier to use: ‘I think you want zero barriers to using it, which is, again, data security, not having to pay for it. This is a public health initiative, it's not something to raise revenue’ (FG3-3).

To meet the needs of people with limited English language proficiency, participants suggested reviewing applications for Easy English compliance, use of audio description of text, and the option to translate text to other languages.

Concerns about the diversity of skin tones represented within the image data sets of AI-powered applications were raised by participants, acknowledging that this might reduce the accuracy of results for persons of colour.

AI is mostly, it's like a learning algorithm, it keeps on learning with the images. Do you have enough images that it has learned so many? … Let's say for syphilis, or whatever, that can have many different images [appearances] for many different kinds of people, different ethnicities, and one thing can look many different ways. (FG1-2)

Trusting the technology

Trusting the technology behind AI-powered applications was central to participants’ choice to act upon the information they provided. Several participants suggested providing users with information regarding applications’ development and accuracy.

It might be about, “Here are some facts about how we built this app” I think, when people are aware… of where did this come from… it’ll help them understand, yep, this is done through extensive research. There's 10 years of photography and analysis… behind this. (FG2-1)

Knowing health professionals were involved in reviewing the performance of AI-powered sexual health applications also supported trust: ‘I understand that it's getting the images from thousands of images and there's confirming on some point, but is there… a human on the backend who's like, “Yeah, this AI is performing right?”’ (FG1-2). Relatedly, participants felt they would trust an application only once they had tried it for themselves and had the result confirmed by a clinician.

It's an AI so there's always doubts with technologies. Sometimes it's good for us, sometimes it fails. But, yeah, so I guess if we can start testing it out and see how it works, then yeah, I guess then we’ll find the answers. (FG1-1)

Being transparent about limitations

Users’ trust in sexual health applications powered by AI as credible and accurate sources of information was balanced with an understanding of the limitations of AI. As previously described, participants recognised that AI was not comparable to clinical care provided by a trained health care provider and pathological testing. Participants comments supported the need for AI-powered sexual health applications to emphasise that potential conditions identified were possibilities rather than definitive diagnoses.

It could also be a backlash… if the AI said you actually have an infection and then when you go to the professional and they say, “It's actually not.” …, You put people mental health at risk because they're just having a panic attack, having anxiety just to wait to go and meet a GP. … Instead of giving people hope so much, you need to have a bit more transparent and saying “Sometime AI might not have a hundred percent accurate with these answers, so we still advise you to go to seek professional help”. (FG1-1)

This was achieved in the AiSTI application by including more than one possible condition, and the percentage probability it could be a certain condition: ‘Because if you’ve got those extra [possible conditions] there, you think, “This is not that accurate. I should go and see a doctor”’ (FG3-2). Ultimately, understanding that the purpose of AI applications is to provide information rather than replace clinical care was suggested as a goal of information provided to users: ‘It's to channel people in the right direction. It's not per se there to make a diagnosis’ (FG3-3).

Discussion

This qualitative study explored the views of potential users on an AI-powered sexual health application (AiSTi) that provides users with a short list of possible conditions that could be causing a visible change in their ano-genital region, after uploading an image taken with their smartphone. Participants’ feedback on the proposed application suggests it has the potential to support timely and appropriate review and treatment of STIs, improved sexual health literacy, and the prevention of onward transmission to others. However, the realisation of these outcomes was seen by participants as mediated by factors impacting the accessibility and acceptability of the application.

Participants highlighted the usefulness of the application in allowing users to ‘get an idea’ of what might be causing a lesion sooner than they would normally seek care from a healthcare professional, as the application involved less effort than attending a clinic in person. Recent survey research testing the technology acceptance model linked usefulness with behavioural intention to use a healthcare application. ²⁸ Similarly, the perceived benefits of use have been found to overcome privacy concerns in relation to digital health applications. ²⁹ The perceived usefulness of the AiSTi thus suggests it is more likely to be used. In addition to usefulness, ease of use and user satisfaction have also been linked to the intention to use healthcare applications. ²⁸ Thus, usability and user experience testing of the AiSTi application will be essential to ensuring its uptake.

Participants also identified potential barriers to the acceptability of the AiSTi application found in research regarding other health applications, particularly in relation to privacy concerns and trust. Participants’ perspectives echo findings of survey research building upon the privacy calculus model. In addition to confirming the importance of trust in providers, personal attitudes to privacy and perceived control of their data, Von Kalckreuth and Feufel ²⁹ compared the contribution of disease characteristics to intention to use electronic health records. They found that stigmatisation potential did not impact intended use, in the case of acute illness (i.e. gonorrhoea), whereas it did in the case of conditions with a longer course (i.e. depression) did. This supports our qualitative findings about privacy concerns, particularly in the case of treatable but incurable STIs identified by AiSTi (i.e. genital warts, herpes). Eke and Shuibe's recent review ³⁰ further elaborated on the role of transparency in fostering trust in AI healthcare systems. As well as their trust in the health service and university developing the AiSTi application, participants in our research emphasised transparency in relation to how personal data is collected and stored, the dataset on which it was trained, and the application's limitations supported trust.

Beyond acceptability, the accessibility of the AiSTi application to range of users was raised by focus group participants. Whilst offering some unique advantages in terms of convenience of access, digital sexual health services are no different to other health services in requiring people-centred and rights-based approaches to reach those most in need. ¹⁷ Operationalising these principles, alongside known contributors to technology acceptance, ²⁸ requires the engagement of a wide range of potential users from conceptualisation to development and promotion. For example, ensuring people of colour are included in the collection of image datasets to reduce bias in results ³¹ and people with disabilities are included in design processes to ensure user interfaces are accessible. These issues have implications for fairness which have been discussed in relation to other AI technologies such as facial recognition, ³² and the inclusion of people with disabilities in co-design. ³³ Chen et al. ³⁴ suggest incorporation of principles drawn from human-centred design, such as engaging ‘extreme users’ early and often in an iterative design process, may help to improve the reach, adoption and effectiveness of public health interventions.

Should barriers to acceptability and accessibility be addressed, feedback from potential users of AiSTi suggest is potential for impact. Focus group participants reported that finding out they could have an STI would encourage them to seek health sooner and avoid sexual contact with others, in the meantime. A study of the time between symptom onset and healthcare seeking in people experiencing symptoms of 13 common STIs found the median time for seeking care for genital warts was 60 days. ³⁵ Moreover, those living further from a sexual health clinic were more likely to delay care. Being able to identify a potential STI without the inconvenience of attending a clinic may help to reduce delays in seeking in-person care.

Increased access to information afforded by AI-powered health applications is not without its risks. Participants in this and other research have described heightened concern related to a potential diagnosis of an incurable but treatable STI, such as herpes and genital warts.^8,9,18 Despite this risk, participants clearly conveyed a preference to have this information prior to accessing care. Moreover, they preferred a short list of possibilities from a trusted source, over the plethora of irrelevant and extreme possibilities presented by internet search engines. Whilst concerns were raised about users of digital sexual health applications experiencing distress at the prospect of an STI diagnosis, it is clear people seek sexual health information online before accessing care. Digital sexual health applications can support users’ wellbeing through links to online sexual health information that addresses their concerns and corrects misinformation.^36,37

Strengths and limitations

This study sought the views of a small but diverse sample of people already engaged with sexual health services, including members of several priority populations including culturally diverse communities, LGBTIQ + people, young people, women of reproductive age, trans and gender diverse people, sex workers and people living with HIV. ³⁸ Beyond their current sexual health knowledge and practices, participants reflected upon their past knowledge and experiences and those of others they knew. Our recruitment strategy, however, did not directly capture the views of those not already engaged with sexual health services, which may not represent harder-to-reach or digitally excluded groups. Moreover, the use of focus groups likely excluded some potential participants who were uncomfortable attending a focus group and may have resulted in social desirability bias. Future research should seek to engage groups for whom the anonymity of web-based sexual health applications may be particularly appealing.

Conclusion

Digital sexual health applications utilising AI represent a potential leap forward in delivering individualised and evidence-based sexual health information into the hands of those who most need it. Like any health service, the impact of AI-powered sexual health applications will be mitigated by their accessibility and acceptability to potential users. Including users from a range of perspectives at all stages of application conceptualisation, development and promotion holds the key to their success.