How symptom checks are made
A simple Google search for ‘coughing and fever’ gives close to 30 million results and searching for ‘headache’ gives you over 104 million websites!
If you might think this is very different in scientific literature, you’d be wrong. Searching on Google scholar (an open search engine for scientific literature) ‘coughing and fever’ prompts 53,000 results and ‘headache’ a whopping 1,000,000+ scientific papers on this symptom everyone has once in a while.
Medical data is regularly combined into guidelines and protocols and then used by most doctors in diagnosing and treating their patients. Our checks are built using the available medical data on symptoms, biological mechanisms and treatment. Just like in a regular doctors consult, our checks ask you questions about your symptoms, phrased in normal language. Through the answers to these relevant questions the system guides you to the relevant information about your personal, specific symptoms, and giving you the same information a doctor would most likely give you.
All of the checks have two stages: triage and diagnostics. In the triage-stage of the check you answer a set of questions relevant to assessing the level of urgency of your symptoms and in the diagnostics-stage you answer questions to indicate which three conditions your doctor would think are most likely causing the symptoms.
For the assessment of urgency we use 4 categories:
- Emergency – Seek medical attention immediately
- Urgent – See a doctor today
- Not urgent – See a doctor on a normal weekday
- Self-care – Wait and see for x number of days
For each urgency level there is a set of (single or combined) criteria. For example, a combined criterium for urgency level 2 (Urgent) in headache is ‘Pregnant for longer than 24 weeks AND Never had a similar headache’ and a single criterium for urgency level 1 (Emergency) in abdominal pain is ‘Throwing up blood’.
These criteria are derived from as many sources as possible. As a basis we use several existing paper based triage systems. These systems are used in either family practice or in emergency rooms. Additionally we use recent studies on triage and urgency assessment. After the first draft of the algorithm is made, we sit down with doctors that have extensive experience with the complaint in question. We try to utilize the doctors’ experience and knowledge to anticipate any situations in the existing systems our algorithm doesn’t account for.
In the diagnostics-stage things are a little bit more complex. As opposed to the triage-stage, here symptoms are individually weighted per condition. This means that each condition has a list of defining symptoms and each symptom (also single or combined) gets assigned a weight to adjust for how specific it is for that disease.
Fatigue for example, is a very common symptom with the flu, but doesn’t distinguish between the flu and lots of other causes of fever, so in the fever check the symptom ‘fatigue’ won’t get a high weight as a symptom for the flu, because it doesn’t help us distinguish between the flu and other diseases.
Sore muscles is a much more helpful symptom in diagnosing the flu, because that is quite common in the flu, but less common in other conditions, so it’s more suitable in distinguishing between the flu and other conditions, so ‘sore muscles’ gets a higher weight than ‘fatigue’.
As a basis for the diagnostic algorithm we use medical guidelines which are used in family practice and recent studies about the diagnostic approach and the presentation and symptoms of all the diseases. To keep the number of questions as low as possible, we focus on finding the symptoms that are most specific for one or two of the possible conditions and because of that are most suitable for distinguishing between all of the conditions in the check. After we’ve made a first draft of the diagnostic algorithm, we sit down with doctors to check whether or not the symptoms we ask about are the same symptoms that are used in actual day-to-day medical practice to distinguish between conditions.
When all the algorithms are written and all the questions are formulated, we start testing the checks. Every check is tested extensively:
- The first step is testing the check on technical functionality: ‘Does everything technically work as it is supposed to work?’
- For the second step we test a wide range of possible cases and presentations, to test if the algorithms display the results we intend it to.
- After we’ve finished the second step, we let the doctors we work with test the check, to account for irregular presentations and cases we might have missed.
- Finally we test the check extensively with end-users. We let people who visit the family practice run through the check, while we watch their behavior and discuss the check with them.
This way, through the use of all this scientific data and using the extensive experience and knowledge of doctors, we can guide everyone with an internet connection and a basic understanding of the english or dutch language, to reliable, relevant information.