The Case for Symptom Checkers

June 8th, 2024 | Klemen Vodopivec

The Global Healthcare Workforce Crisis

By now it’s no revelation that there is a worldwide chronic shortage of healthcare workers, with the WHO estimating a deficit of approximately 10 million professionals by 2030. This shortage is unlikely to improve in the near future, as a growing number of workers are leaving the workforce and the population continues to age.

Primary healthcare is the sector most affected by this shortage. For example, the United States alone is predicted to have a shortfall of between 7,300 and 43,100 primary care physicians by 2030. Similarly, in the United Kingdom, a joint report by the Health Foundation, King’s Fund and Nuffield Trust in March 2019 suggested that there were 6% fewer general practitioners (GP) practicing in September 2018 than in September 2015. This report further predicted there would be a shortfall of 11,500 general practitioners by 2028/29. As a result, people are confronted with increasingly limited access to essential medical expertise and knowledge, longer waiting times and increased use of emergency services.

Navigating the Maze of Online Health Information

Due to the aforementioned challenges in accessing medical expertise, the public is increasingly turning to online resources to seek healthcare information and make informed decisions about their well-being. 59% of U.S. adults have searched for health information online within the past year, and nearly 7% of daily Google searches are related to healthcare topics . Furthermore, 35% of U.S. adults can be classified as “online diagnosers,” meaning they have turned to the internet to attempt to identify a potential medical condition they may be experiencing.

However, relying on online information for self-diagnosis raises several concerns, as the vast amount of medical content available on the internet varies greatly in terms of credibility and accuracy. A recent study has revealed that only 26% of Google search results related to health topics are considered high-quality, and these reliable results are not consistently displayed at the top or near the beginning of the search results page. Consequently, among U.S. adults who sought health information online, only 37% reported being able to access the desired information without experiencing frustration.

The Solution

Symptom Checkers: Digital Doors to Healthcare

The issues discussed necessitate the development of scalable technological solutions. One such solution is the implementation of symptom checkers, which are digital health tools designed to help individuals better understand the potential causes of their medical symptoms. By answering a series of questions about their health concerns, users can receive guidance on the most appropriate course of action, such as whether to self-treat, schedule a doctor’s appointment, or seek urgent medical attention. Symptom checkers serve as digital doors to healthcare systems, offering a new and accessible avenue for individuals to connect with medical services. The COVID-19 pandemic has brought new attention to these digital tools as they helped reduce in-person contact and preserve urgent care capacity in the face of surging demand for healthcare services.

Symptom checkers offer the convenience of being available 24/7, ensuring users can quickly assess their symptoms at any time, from any device with internet access. Compared to regular web searches they provide more reliable information and can ask follow-up questions to gather additional details that may not have been included in the user’s initial query. Moreover, symptom checkers are designed to analyze multiple symptoms simultaneously, taking into account the user’s medical history and demographic information to provide a more comprehensive and personalized assessment. Major healthcare providers, including the UK National Health Service (NHS111) and the Australian government’s Healthdirect service, have already integrated symptom checker applications into their digital healthcare offerings.

The Benefits of Symptom Checkers in Healthcare

Symptom checkers empower individuals by providing them with valuable insights into the potential causes of their symptoms and helping them determine whether professional medical attention is necessary. This is particularly beneficial in situations where consulting a physician in person is not feasible due to factors such as time constraints, geographical limitations, or limited access to healthcare services.

Symptom checkers act as navigational aids, helping users navigate the complex landscape of available medical services and guiding them towards the most appropriate course of action. This has a two-fold benefit of redirecting non-urgent cases away from overburdened medical facilities and ensuring that individuals with more severe or acute conditions receive the timely attention they require. Additionally, symptom checkers foster a sense of engagement and trust between patients and the healthcare system which translates into higher levels of satisfaction and loyalty and ultimately contributes to improved patient retention and better overall health outcomes. The de-identified data collected and processed can also be used to discern population level incidence trends and provide value for healthcare provider planning and service improvement.

Next-generation symptom checkers

Limitations of Contemporary Symptom Checkers

Despite their claims of simulating real-world dialogue, contemporary symptom checkers operate within a rigid framework of preset questions and responses. This hinders their ability to truly mimic the complexities and subtleties of human conversation, which in turn imposes significant limitations on their usefulness.

Firstly, while symptom checkers offer improvements over search engines, users still face limitations in fully conveying their health issues since they typically select their symptoms from a predetermined list rather than describing them in natural language. This is particularly challenging for individuals who are not familiar with medical terminology, making it difficult for them to accurately input their symptoms. Since most of their questions are of the “yes” or “no” variety, users often do not view their interactions with symptom checkers as a true dialogue. Furthermore, the evident lack of empathy displayed by these tools makes the assessment feel impersonal and somewhat mechanical.

Most crucially, the accuracy of diagnosis and triage is underwhelming. The rate of correct initial diagnoses (M1 diagnostic accuracy) has a median below 50%. Additionally, the median triage accuracy in 2020 was 55.8%, which is nearly unchanged from 2015’s figure of 59.1%. This issue is especially pronounced for uncommon and complex diseases. Moreover, most symptom checkers do not even cover the full spectrum of diseases.

Additionally, current symptom checkers fall short in their ability to provide explanations. They often fail to clarify, or at best offer only generic explanations for why a particular question is asked. They also do not adequately explain the reasoning behind a given diagnosis or the basis for categorizing a patient under a specific disease classification. This lack of transparency can erode trust in the process.

Transforming Symptom Checkers into Conversational Agents

Imagine a symptom checker that doesn’t just ask you to tick boxes, but actually converses with you like a doctor with infinite time, patience and an encyclopedic knowledge spanning all medical conditions. That’s the exciting potential unlocked by large language models (LLMs). They facilitate the creation of true conversational agents, providing multiple advantages over the traditional controllable interfaces:

They create a natural and familiar interaction for users to communicate their symptoms. This not only makes the system more intuitive but also updates and aligns the user’s mental model with how the system operates.
The design mimics a flexible dialogue, accommodating a wide range of user requests without constraining them to a predetermined path.
The natural language interface facilitates interactions that can offer empathy and understanding. This capability can capture user attention and build trust, as users react favorably to a system that demonstrates an ability to comprehend and empathize with their concerns.

Most significantly LLMs have already demonstrated the ability to rival human experts in extracting relevant information during simulated consultations. Moreover, when provided with the same amount of acquired information, LLMs have proven to be more accurate than human experts in formulating a comprehensive differential diagnosis. Furthermore, a dual evaluation by patient actors and specialist raters revealed that LLMs outperformed human experts on metrics assessing empathy and communication skills.

An additional advantage of LLM-based symptom checkers is their capacity to explain why it poses specific questions and how the symptoms and information provided by users directly contribute to the suggested diagnosis and advice. This way, users tend to perceive a higher diagnostic quality due to the clear post-hoc explanations, and gain a deeper understanding of their symptoms and the clinical decision-making model that drives the diagnosis. The explanations also assist users in organizing medical information, enabling them to make more informed decisions, such as determining the need to consult a medical professional. Finally, medical recommendations that are understandable and transparent further bolster user trust.