Design that spans the digital divide: How could online symptom assessment tools reduce the burden on primary care?

In short, it depends. For a medical emergency like crushing chest pain, you should call an ambulance, but for a stubborn wart, visiting the pharmacy might be better. For everything in between, there’s primary care. A general practitioner (GP) – referred to as a primary care physician (PCP) in the US – is a doctor that can assess, diagnose, treat, and refer to specialist services. They also have a range of other responsibilities including prevention, screening, immunization, and coordination of care with other health and social services.

Long before COVID-19 upended the healthcare system, there was mounting pressure on primary care. The Association of American Medical Colleges (AAMC) projects a shortfall of between 21,100–55,200 primary care doctors in the US by 2032. In the UK, National Health Service (NHS) GPs each look after an average of 1,500 patients, but the UK’s aging population, changes in the workforce, and growth in patient demand means that general practice is ‘in crisis’. In a 2019 survey of NHS GPs, over half revealed they worked beyond what they considered safe levels, seeing one third more patients than they should. Countries such as India and Malaysia face pressures on primary care too, with a quarter of physician posts unfilled and private/public divides in care leading to disparities in health outcomes. The problems are worsened by the exodus of qualified professionals to richer countries.

In the UK, registered NHS patients can book an appointment or visit a ‘walk-in clinic’, with some triage undertaken by receptionists. However, many patients feel uncomfortable disclosing personal medical issues to relatively untrained receptionists and triage based on simple questions like “what’s it for” or “is it urgent?” is ineffective. Within the NHS, GP visits are free at the point of care. As a consequence, there are inefficiencies known as ‘medical overuse’ where some patients visit the doctor unnecessarily for a condition that could have been managed at home. Conversely, others who don’t show up to the GP or are put off by having to negotiate with a receptionist and long waits risk harm by attempting to manage potentially serious conditions at home. Even rich countries like Sweden experience inequalities in healthcare use due to the small payments necessary to access care. These inefficiencies are unevenly distributed in society, with age, sex, and socioeconomic factors influencing whether people seek medical care. The consequences of this include late cancer diagnosis, increased admissions for bedsores, and lower satisfaction with healthcare providers. From a wider health system perspective, the downstream consequences mean busier emergency rooms and the presentation of late-stage cancer in urgent settings.

While training more doctors is one strategy, a recent analysis pointed out that issues like the uneven distribution of doctors, inconvenient hours, inflexible care models, and other barriers to access are far more significant challenges to reform than lack of training. Instead, the solution is likely to include “better targeting of physician care, greater use of non-physician labor where appropriate, and much broader deployment of technology to increase access to primary care.”

Digital symptom assessment tools have the potential to transform the flow of patients and improve their access to appropriate care. These tools ask patients questions about their medical history and new symptoms to help determine what the possible causes might be. Patients can submit health complaints in their own time, day or night. There’s no limit to the number of simultaneous users, and the intelligent software (including artificial intelligence) allows patients to be triaged and receive helpful symptom management information at the same time. In the EU, tools like this must be registered as medical devices with a CE mark and adhere to rigorous controls and standards to remain on the market.

In another recent study (currently in peer review before publication), our research team of doctors and scientists found that our digital symptom assessment tool, Ada, outperformed seven other popular tools on the breadth of issues covered, safety of advice, and accuracy of suggested conditions. In the same study, Ada had the same high level of coverage and safety as seven human GPs (100 percent coverage of medical issues, 97 percent safety). But our previous study used simulated cases, or vignettes in place of real patients. How can we evaluate the effectiveness of these tools in the busy, real-world GP practice?

We expect that different groups of users will vary in their enthusiasm for trying out digital tools to manage their health. For instance, a 2018 survey of over 1,000 Londoners conducted by Healthwatch Enfield found that 63 percent of participating NHS patients would be interested in using a digital symptom checker, but there were significant differences between age groups. Enthusiasm was much higher amongst 18–24 year olds (74 percent were interested) compared to people over 70 (34 percent were interested). But that was for a theoretical symptom checker. How would a similar group of primary care patients in the GP’s waiting room react to having Ada’s digital symptom checker in their own hands? Would it have helped them make a different decision on whether to attend the GP’s surgery? Would older people feel alienated by the technology?

These are the questions that Ada and a team of doctors at a large NHS GP practice sought to answer in a new study published in the Journal of Medical Internet Research: Human Factors. Over a three-month period, we showed a version of Ada to 523 patients in the NHS primary care practice of Paxton Green in South London – a busy practice that looks after 20,000 registered patients in a relatively deprived part of the city. Each participant was asked to enter the symptoms that had brought them to the practice that day and then fill out a questionnaire about their experience with the app.

Overall, most participants (97.9 percent) found the Ada app very or quite easy to use with a much higher degree of acceptance than suggested in the previous Healthwatch Enfield survey (63 percent), albeit in a different sample. Most said they would use it again if given the chance (88 percent) and agreed they would recommend it to a friend or relative (85 percent). Overall, 79 percent of participants agreed that Ada provided helpful advice. But here we started to see more differences across age groups. For instance, 93 percent of users aged 18–24 found Ada provided helpful advice, whereas for those aged 70 and over, the figure was 59 percent – still useful for the majority but a difference worth investigating.

We also wanted to estimate what impact using Ada in the clinic had on the patients’ decisions. As they largely agreed with Ada’s advice to see a doctor the same day, most patients (86 percent) said it would not have changed their decision to see a GP. However, 13 percent of respondents said they would have changed their mind to use a less-urgent form of care. One third of these participants said they’d have stayed at home, a third said they’d have gone to the pharmacy instead, and a third said they’d have delayed their appointment by a few days.

Ada can also help appropriately escalate care when needed, with 1.2 percent of patients saying they would have taken the more urgent care route of visiting the accident and emergency department after taking their digital symptom assessment. Here again we found an age-related trend, with 22 percent of 18–24 year olds saying they would have made a different healthcare decision compared to 0 percent of those aged 70 or over. From this we can conclude that even when a digital symptom assessment tool is easy to use and provides helpful advice across all groups, there might still be more deeply entrenched differences in what they choose to do with the information provided.

As a result of feedback from these users, we made a number of changes to the Ada platform. We’re using machine learning to make it easier for our users to describe the symptom they are experiencing and understand if the user has made spelling mistakes when inputting symptoms, both of which improve the user experience. We also have partnerships underway to test if Ada could help people make more informed decisions before booking an appointment.

An example of this is our partnership with Sutter Health, an integrated, non-profit health network in Northern California. Sutter Health users have access to a customized version of our symptom assessment on Sutter’s website and mobile app. Tens of thousands of people have used the tool. Many visited outside of normal practice hours and were supported to find answers to their health problems when the clinics were closed.

Scale like this offers the most opportunity for digital symptom assessment tools. But challenges remain too – technology alone is not a remedy for changing cultures of practice that have existed for decades and resistance to new methods has been widely recognized in health systems.

Now that COVID-19 is forcing us to re-evaluate how we plan and structure the face-to-face time of our healthcare professionals, digital technology will increasingly play a role in managing both clinical efficiency and risk of infection for patients and medical professionals alike. Those technologies with the most robust evidence base and highest standards are the ones most likely to succeed.

You can read the full study in the published paper in JMIR Human Factors Discussion based on Miller et al. (2020) published article here.

Reference: Miller S, Gilbert S, Virani V, Wicks P. Patients’ Utilization and Perception of an Artificial Intelligence–Based Symptom Assessment and Advice Technology in a British Primary Care Waiting Room: Exploratory Pilot Study. JMIR Hum Factors 2020;7(3):e19713