Symptom checkers: how AI could transform the field of diagnostics

By Hannah Scheucher

Diagnosis is a patient-specific process to characterize an illness via collection of information, integration of different data and clinical reasoning. In this pursuit, hypotheses of possible conditions explaining a patient’s symptoms are established and eliminated over the course of time.

Even with advances in technology and diagnostic testing, diagnostic error still represents a large obstacle in health care (Goldman et al., 1983). Misdiagnoses are still very common and in other cases, diagnosis comes too late for the condition to be treated (Balogh et al., 2015; Nateqi et al., 2019). A Dutch study found that 6.4% of adverse effects on patients were caused by diagnostic error (Zwaan et al., 2010), and some estimates even suggest that every 7th diagnosis is incorrect (European Commission, 2017). Incorrect medical evaluations are expensive and come at a great cost to hospitals and insurance companies. From a patient’s perspective, it is incredibly challenging to be delivered multiple, perhaps conflicting, diagnoses, and go through many rounds of testing to finally receive the correct diagnosis for conditions impacting one’s life.

In particularly rare diseases, often termed orphan diseases, diagnosis is difficult due to their low incidence and efforts are being made to set up databases allowing facilitated access to information in such conditions (Orphadata, n.d.). Still, due to their infrequency (roughly 1 in 2000 according to European guidelines) and the sheer number of different illnesses (estimates range from 6000 to 8000), patients can often go without receiving the correct diagnosis – a critical step in any treatment plan (Dodge et al., 2011).

Advances in technologies such as artificial intelligence may be able to help address this issue. Clinical decision support systems (CDSS) are computer-based programs developed to support health care practitioners with decision-making through the integration of data, such as patient data and clinical knowledge (Mitchell and Ploem, 2018; Sutton et al., 2020). Similar programs have existed for a considerable amount of time, but had only allowed retrospective analysis of treatment to develop protocols for future diagnoses. Now, CDSSs are being developed that help health care professionals make decisions at the time and point of care (Berner and La Lande, 2007).

Symptom-checkers are specialised CDSSs that can list possible conditions based on the patient’s symptoms which are entered into the system. Their usage is designed for patients as well as doctors, however, limited accuracy restricts their practicality. A 2019 study (Nateqi et al., 2019) analysed different symptom checkers in their ability to diagnose ear, nose, and throat diseases. It was found that “Symptoma”, an artificial intelligence-based digital health assistant developed in Austria that was created throughout one and a half decades of research (About Symptoma, n.d.), showed the best results. This platform would list the correct condition in the top 3 diagnoses 92.9 % of the time and 64.3 % of time as the top hit, making it the most accurate symptom-checker on the market so far with the runner up, Isabel, showing the relevant condition in the top 3 hits only 40.4 % of the time (Nateqi et al., 2019).

Symptoma can provide more accurate results due to three distinguishing factors: Its ability to recognise nearly all symptoms, its extensive medical database containing more than 20,000 conditions, and its ability to identify connections via the use of millions of medical publications (Nateqi et al., 2019). Furthermore, compared to other symptom checkers, this technology utilises a free-text input method, allowing the user to specify as many different symptoms as they wish, and it can also take other factors such as gender, age, and location into account to further improve the accuracy of top suggested diagnoses (Martin et al., 2020).

Due to its accuracy and user-friendly design, Symptoma has been utilised in the COVID-19 pandemic to attempt to reduce infection rates (Martin et al., 2020). One key strategy in a pandemic is to perform as much testing as possible to identify and subsequently isolate cases (Rosenthal, 2020). As limited testing capacities exist, triage of potentially infected citizens must first take place. To do so, governments have set up phone hotlines, however, these are often overrun, and often produce an inaccurate assessment of the patient, especially if the diagnosis system is automated. Analysis of Symptoma’s accuracy to diagnose patients with COVID-19 showed that it could correctly identify cases 96.32% of the time. Further investigation showed that Symptoma had a specificity of 90%, being able to distinguish COVID-19 infection from similarly presenting diseases such as bird flu. These results illustrate that Symptoma performs better than previous implemented methods with pre-defined symptom options for triage (Martin et al., 2020). Importantly, this CDSS is available in 36 languages, allowing wide-scale usage of the technology (About Symptoma, n.d.)

An opportunity for CDSS and other symptom checkers to improve health care has emerged in low-income countries. In such countries, there is a general difficulty to provide sufficient universal health care, especially in rural areas. Experts suggest that the use of symptom checkers may increase the quality of healthcare, by allowing patients direct access to such platforms. Additionally, community members trained to provide essential care could benefit from the implementation of such technologies, and improve the quality of care that they provide (Morita et al., 2017).

As with every new development, a certain amount of scepticism accompanies CDSSs and other symptom checkers. Some question their accuracy and express concern over the potential misuse of such systems, while others have highlighted worries about personal data collection and privacy. In general, it has been noted that such technological developments may impact the level of trust patients have in medical professionals (Mitchell and Ploem, 2018). However, given the potential for CDSSs like Symptoma to dramatically improve millions of lives through the delivery of correct diagnoses, and thereby appropriate treatment, it seems highly likely that this technology will be increasingly adopted in the years ahead. 

References:

About Symptoma | Digital Health Assistant & Symptom Checker (no date). Available at: https://www.symptoma.com/en/about (Accessed: 8 February 2021).

Balogh, E. P., Miller, B. T. and Ball, J. R. (2015) ‘The Diagnostic Process’, in Balogh, E. P., Miller, B. T., and Ball, J. R. (eds) Improving Diagnosis in Health Care. Washington (DC): National Academies Press (US). Available at: https://www.ncbi.nlm.nih.gov/books/NBK338593/ (Accessed: 8 February 2021).

Berner, Eta S. and La Lande, T. J. (2007) ‘Overview of Clinical Decision Support Systems’, in Berner, E.S (ed.) Clinical Decision Support Systems. New York, NY: Springer, pp. 3–22. doi: 10.1007/978-0-387-38319-4_1.

Dodge, J. A. et al. (2011) ‘The importance of rare diseases: From the gene to Society’, Archives of Disease in Childhood. BMJ Publishing Group Ltd, pp. 791–792. doi: 10.1136/adc.2010.193664.

EuropeanCommission (2017) Symptoma. Available at: https://ec.europa.eu/eipp/desktop/en/projects/project-11138.html (Accessed: 8 February 2021).

Goldman, L. et al. (1983) ‘The Value of the Autopsy in Three Medical Eras’, New England Journal of Medicine. Massachusetts Medical Society, 308(17), pp. 1000–1005. doi: 10.1056/nejm198304283081704.

Martin, A. et al. (2020) ‘An artificial intelligence-based first-line defence against COVID-19: digitally screening citizens for risks via a chatbot’, Scientific Reports. Nature Research, 10(1), pp. 1–7. doi: 10.1038/s41598-020-75912-x.

Mitchell, C. and Ploem, C. (2018) ‘Legal challenges for the implementation of advanced clinical digital decision support systems in Europe’, Journal of Clinical and Translational Research. Journal of Clinical and Translational Research, 3(S3), pp. 424–430. doi: 10.18053/jctres.03.2017s3.005.

Morita, T. et al. (2017) ‘The Potential Possibility of Symptom Checker’, Int J Health Policy Manag, 6(10), pp. 615–616. doi: 10.15171/ijhpm.2017.41.

Nateqi, J. et al. (2019) ‘From symptom to diagnosis—symptom checkers re-evaluated: Are symptom checkers finally sufficient and accurate to use? An update from the ENT perspective’, HNO. Springer Verlag, pp. 334–342. doi: 10.1007/s00106-019-0666-y.

Orphadata (no date). Available at: http://www.orphadata.org/cgi-bin/index.php (Accessed: 8 February 2021).

Rosenthal, P. J. (2020) ‘The importance of diagnostic testing during a viral pandemic: Early lessons from novel coronavirus disease (CoVID-19)’, American Journal of Tropical Medicine and Hygiene. American Society of Tropical Medicine and Hygiene, pp. 915–916. doi: 10.4269/AJTMH.20-0216.

Sutton, R. T. et al. (2020) ‘An overview of clinical decision support systems: benefits, risks, and strategies for success’, npj Digital Medicine. Nature Research, pp. 1–10. doi: 10.1038/s41746-020-0221-y.

Zwaan, L. et al. (2010) ‘Patient record review of the incidence, consequences, and causes of diagnostic adverse events’, Archives of Internal Medicine. American Medical Association, 170(12), pp. 1015–1021. doi: 10.1001/archinternmed.2010.146.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s