Process mining as a framework for AI

I have recently started to learn about process mining and how it can help with analysis of healthcare data.

Process mining needs ‘timestamp’ of activities, for example when a particular ‘activity’ (for example, a surgical procedure or an investigation such as a CT scan) started and was completed.

This allows us to understand the ‘pathways’ – which activities each patient has, when (and how long it took), in which order and which healthcare staff was involved at each stage. This then allows us to understand the common pathways and the ‘variants’ to get insights into what’s working, what’s not and how to improve outcomes.

Process mining is not currently a widely used tool in healthcare, but I think the following will increase its importance:

1. The prevalence of Electronic Health Record (EHR) systems is increasing. These provide the ‘structural’ data that process mining needs
2. The focus on Integrated Care Systems (ICS) and providing joined up care will (hopefully) lead to better sharing of and linkage of datasets across organisations
3. Increase in the number of AI based solutions

The last one is an interesting one. Traditionally, the 3 uses for process mining were:

1. Process discovery – understanding more about the process
2. Conformance checking – comparing the identified process with the expected (or previous knowledge)
3. Enhancement – identifying opportunities to improve throughput time, costs, etc

With the increasing focus and application of AI in healthcare, and building on the above 3 uses, we can look at applying the uses as a ‘Framework for AI’.

This will help in the following ways:

1. Provide a more structured data model to help train AI models
2. Support developing of robust compliance methods to evaluate whether the AI models perform as expected
3. Knowledge of variants to help improve models and develop models that are inclusive / fair

Very happy to hear your comments below or feel free to email me to share ideas – janak@usehealthdata.com