Waiting times have been a focus for all healthcare providers for many years, these were first introduced alongside the NHS constitution and have been implicitly linked to the experience a patient receives. Moreover, research suggests that the longer a patient waits can severely affect the patients perception of the quality of service and overall impression of the healthcare system. Alongside this, a crowded emergency department can lead to a number of other associated factors (see: https://www.sciencedirect.com/science/article/pii/S0716864017300354).

How do we tackle the problem of predicting TTBS?

To tackle this problem the team at Draper & Dash have developed a ML solution to aid in management decision making, planning for long waiting patients and to align capacity to demand.

This predictor is useful in its own right and is also contained within our Command Centre application (see: https://www.draperanddash.com/machinelearning/2019/10/command-centre-amplification-with-predictive-analytics-and-machine-learning/). If you are interested in either, then please enquire at sales-support@draperanddash.com.

What features does the model use to make the prediction?

To make the prediction the machine learning algorithm uses the features bulleted below:

  • The time the patient arrived into ED
  • Patient type
  • Age of patient
  • Referral source
  • Arrival method
  • Among other in department measures

These features, are then assessed, to see which features have the largest impact on the time it takes  a patient to be seen. D&D’s unique Feature Importance Location Engine (FILE) kicks in here and is the driver behind the feature importance identification – to read more (see: https://www.draperanddash.com/machinelearning/2019/08/bucketing-and-highlighting-dominant-predictors-in-your-ml-models/).

What is the algorithm?

We use a combination of algorithms to train our models, because certain models fit certain data distributions better. However, in this case we opted for a Naive Bayes classifier, as this allows for usage across multiple target variables (multinomial) and is an effective way to make estimates of the probability of a group of patients belonging to a certain class.

Why use our algorithm?

Our team of data science experts have not only data science and predictive analytic skills, they also come from a healthcare background, meaning that they can add context to the algorithms D&D are building. In essence, we know what features (data fields) to use for which problem.

The time to deployment, with our algorithms, is swift – meaning that they can be deployed into an existing infrastructure within days. Saving the development time and augmenting the data held within existing systems.


If you are interested in understanding more about our approach to predicting time to be seen, then please contact: analytics-support@draperanddash.com.

Gary Hutson – Head of Solutions and AI | Alfonso Portabales – Data Scientist