VIDEO | Healthfirst CAO: We Don’t Build Models in Black Box

(US and Canada) Christer Johnson, Healthfirst Chief Analytics Officer, speaks with Jake Dreier, HiLabs Director of Strategy & Growth, about the transformative use cases of analytics, the model-building process, and organizational transformation through advanced analytic tools.

Johnson looks forward to the use cases of natural language processing and text analytics for the transformation in the radiology space. He says the notion of using graph database technology to build robust time-sequenced customer journey databases excites him the most.

He explains that graph database technology brings all the data into one place with a timestamp on each event, forming a dictionary of the interactive journey of the members. It helps organizations proactively identify patterns that produce opportunities to engage with members and improve their health, Johnson suggests.

Johnson sheds light on the importance of attaining health equity. The difficulties members face while accessing healthcare is a challenge, he states. As a result, his focus is on collecting relevant data about people trying to gain healthcare access. Appointments set up through digital channels is one way, Johnson adds. He asserts that, since 70% of the business is value-based care, providers must have better information and insights about access to the care provided.

Johnson maintains that, while building models, it is fundamental to communicate with business stakeholders to understand relevant issues and pain points. Instead, the model-building process is considered a black box, highlights Johnson. He points out that explicit measures are being taken at Healthfirst to ensure that they do not build models in a black box.

Johnson states that the organization focuses on structured collaboration through the process of building, testing, validating, and operationalizing the analytical models. He stresses the importance of formulating structured ways to engage all stakeholders and ensure complete visibility and transparency in the model-building and deployment process.

Speaking on transformation, Johnson pinpoints that models only work if operationalized, and it requires more than uncovering insights. He elaborates that, if these models are to identify patterns at scale in real-time, they must be cohesive with the operating system of the organization.

According to Johnson, Healthfirst has fully adopted and scaled its IT agile methodology. He adds that the growth in the number of models being built and operated poses a challenge for machine learning operations to work in a system. As a result, the need is to re-architect the system to include ML Ops functionality and drive outcomes. In conclusion, Johnson advises that organizations cannot only deploy models and share results; they must also provide the explainability and traceability of the results.