We've made the final model available on the web!!!
Click to have a go for yourself. Read more about the final BeNDi model below.
Constructing and designing decision support is of no use if the outputs of the models are not easily accessible and understandable by the people who need to use them. In parallel with the BeNDi elicitation process we have been working on ways to make our models available on the web for people to use.
After logging in you will be presented with three options. Clicking on BeNDi app will allow you to test out the model (see below for details). Here you can create a new patient, add new predictions for a current patient or view/revise previous predictions
Use the tabs at the top to enter the risk factors and signs and symptoms associated with the patient (you can also enter scenarios information). Once satisfied you can click the Calculate button at the bottom. The app will then work out the probabilities associated with the judgement factors and display those to you along with baseline predictions.
We would like to reiterate our thanks to everyone who has taken part in the process thus far. Your help has been invaluable in researching whether MSK models can be constructed using a panel of experts in this way.
For anyone wishing to help us one last time we have two very short questionnaires that test the outputs of the model against expert judgement. These are completely voluntary - you can still use the model but the results will be give us a good indication of how well the model is working.
The first questionnaire gives you three patient case studies (in terms of inputs to the model) and asks your opinion about the risks associated with each possible masquerading condition.
In the second questionnaire, for each condition you will be asked to pick the input (i.e. risk factor or sign and symptom) you believe to give the best indication for the presence of that condition.
The BeNDi process sought to construct a Bayesian Network (BN) for predicting presence of serious underlying pathologies in patients presenting with lower back pain. Split over three stages, domain experts gave their judgements about the most appropriate variables to include, how they are connected and their associated probabilities.
A schematic of the final BN is shown below (you can click on each node to get a better idea about the connections). It is represented causally, meaning the risk factors (in green) cause judgement factors (blue) which in turn cause (signs and symptoms), as shown by the arrows. By using suitable algorithms we can `reverse' the causal direction, i.e. if we have information about the signs and symptoms, we can infer information about the judgement factors.
For those unable to attend the online webinar on 10th May we have recorded the session, which gives a complete overview of the website.