Automating first-patient contact is already on the works – and it is at this point that data enters the conversation: What kind of data should be used for signposting? Does the healthcare industry have that kind of data and how accurate is it? In this article, we explore the backbone of the development of first patient contact automation.
After identifying the needs of our end-users, the patients and the healthcare professionals, it is time to explore which technology choices are reasonable when building automated patient contact solutions. The requirements here are a) they need to serve the end-user needs sufficiently in the first step, b) satisfy the critical success factors – in our case, transparency, medical context awareness, and integration in the digi-physical flow, and c) allow for the long-term evolution of the solution. So, what technologies could fulfil those criteria in primary care?
For many, the cornerstone in addressing this question comes to the patient data sources: what kind of data do you have? The most popular answer and often the first to come to mind is historical data coming from years of patient journal records. The advantage of this type of data is for sure its time-series coverage that we can accumulate on a large scale. In reality though, many overlook the fact that a large portion of such data (not that seldom up to 95% to 99%) is simply not fit for the purpose, let alone the difficulties in obtaining it. Moreover, since its origin is legacy systems, a good pile of money needs to be invested in structuring and aligning the datasets.
The second type worth considering is newly generated data, which is often much more homogenous but takes effort to collect and needs to be put in the practical context of future application. Before any large set of data is collected, which data points we may need to collect need to be imagined. Here the balance lies in being pragmatic enough to “guess” in just the right amount of data points that will be sufficient, yet feasible and viable for the thought application.
While the former two types of data sets are a frequent topic of discussion, there is a third type of data that is frequently overlooked in AI applications: Expert domain knowledge. Primary care is just the field where this valuable resource shouldn’t be ignored, as medicine has centuries of accumulated knowledge describing connections between symptoms and diagnoses in patients. The question for us feels almost rhetorical – Why not reuse that?
Few will argue that despite the years of accumulated data in electronic patient records all over the world, primary care suffers from chronic lack of perfect data sets that would objectively describe all the primary care conditions. At the same time, in literally any application, the type of data, its amount and quality will pre-define the development tactics of the automated solution. Luckily, there are several ways to go while creating intelligent systems. So, based on our data choice, how do we develop our solution? Find out here!