Dr. Bot - the new frontier for health consultations?
5 December 2017
Bots have been the subject of much media attention of late, in part due to a growing concern that they are being used to distort online debates and spread misinformation on social media. Beyond the malicious or devious use of bots to influence political agendas, there are many legitimate and well-intentioned applications of so called ‘social bots’ and other types of bots too. From managing schedules, acting as virtual assistants, to supporting e-commerce - bots are widespread. Healthcare hasn’t escaped the attention of bot developers either with some even claiming that bots will be diagnosing us in the future. Is this hyperbole or is it time to prepare for a future of bot-based healthcare?
What are bots and how do they work?
Bots are computer code (software) applications that perform an automated task. The majority of bots automate relatively simple and repetitive tasks, for example ‘feed fetchers’ that ferry web content to provide weather or news updates to mobile devices. Whilst bots are not new, in recent years there has been a significant expansion in applications, driven by the growth of smart phones, social media and messaging services, and by advances in artificial intelligence (AI).
AI approaches such as machine learning that learn directly from data to improve their performance have been key to the improved capabilities of next generation bots. For example, ‘symptom checking’ medical bots can be underpinned by machine learning algorithms trained on existing health datasets of medical symptoms and associated conditions. Bots which simulate conversation, known as ‘chatbots’, tend to be driven by machine learning based ‘natural language processing’ algorithms. These are trained on existing ‘human’ speech to support the bots’ interpretation and response to spoken or written language cues.
How can bots be used in health?
The main developments around medical bots are currently focused symptom checking, patient engagement, and patient triaging. Typically, these bots require individuals to manually input data about themselves through their mobile phone or computers in order to obtain the bot generated response or recommendation. Given that many people now access health information online, bots could offer a more structured approach for querying symptoms and obtaining a suggested course of action. In the case of ‘symptom checking’ bots, their response might be on how to self-manage a minor ailment, or whether to seek medical advice. Similarly, triaging bots might inform individuals or the healthcare worker on whether or not a particular set of symptoms requires medical attention. These types of bots are not diagnostic, although that maybe a longer-term objective for developers. Instead, many companies partner with health systems or practicing physicians to connect people to healthcare services following the bot-based recommendations.
Some medical bots are being integrated into mobile and messaging apps, so users can communicate with the chatbot by text directly through existing messaging services. A range of bots including symptom checkers, mental health trackers, and medication reminders interface with Facebook’s Messenger platform, as have Public Health England’s chatbots for support and advice with breastfeeding and for quitting smoking.
Although the availability of medical bots, especially chatbots, is on the rise, there is some way to go before AI powered diagnosis reaches our fingertips. Even with the existing forms of bots there are critical questions to be addressed before they feature more prominently in healthcare.
Where’s the evidence?
In theory triaging and symptom checking bots could help patients and health services by reducing unnecessary medical consultations and supporting patients to self-manage minor ailments. However there is some debate around the accuracy and effectiveness of these tools. Recently it was reported that a London CCG decided against proceeding with plans to trial Babylon Health’s 'symptom checker' app for triage, after indications that patients may overplay symptoms in order to access a GP appointment. Crucially, building confidence in these types of bots will require evidence to demonstrate that they do not result in unnecessary triage, and at the same time do not overlook those serious cases requiring medical attention
What happens to user data?
Clearly given their vast user-base, existing mobile and messaging services can provide a major outlet for conversational health chatbots to reach and engage individuals around health matters. What is less clear is if and how personal data submitted by users through chatbots will be used by the messaging services and the bot-operating companies.
Face-to-face human interaction is often seen as central to a medical consultation. So, a patient’s and doctor’s perception of bots, including trust in bot-generated guidance, and how comfortable individuals are interfacing with a bot, could also influence their use. A survey by PWC on people’s willingness to talk to an AI device, assuming it was more accessible and efficient than human doctors, revealed that across the surveyed EMEA countries there was more willingness (55%) than unwillingness (38%) to engage with AI and robots. However, results varied markedly between countries - 95% of survey respondents were willing to interact with an AI device in Nigeria, in contrast to 50% in the UK.
A medical-bot based future?
With the growing number of AI medical bots already in use, it is easy to appreciate why AI is seen as one of the most promising and transformational technologies for healthcare. But it is important to not get swept away with the enthusiasm; currently the applications of bots remain limited and questions about the accuracy, effectiveness, privacy, and user perception of these tools remain. and As the technology improves, these these questions are likely to be resolved and bots may yet deliver on their promise to bring us faster and more precise diagnosis, and better support for citizens to live healthier lives.
With the opportunities to combine information from several devices such as wearables, monitors, and fitness trackers using the Internet Of Things, it is very possible to envisage a future where bots automate health alerts with minimal input from individuals. Crucially, access to and availability of health records and information will affect the pace of these developments, since existing datasets are essential for developing and improving the AI algorithms. Still, with the right data, evidence, and refinement, it may well be possible that our future medical consultations will be with a Doctor bot.