Health Analytics: Buy or Build?

Chris Grenz
Analysts EIM Practice Director

The Analysts team wrapped up HIMSS Big Data and Analytics in Boston recently with conversations on how custom analytics need to be in health. The keynote speakers were split on the topic with at least a couple holding strongly to the "very custom" side and, predictably, the more vendor-aligned skewing towards common analytics models.

Actionable Analytic Models

Analytic models, (predictive and ideally prescriptive), dominate IT conversations as Machine Learning, Deep Learning, and AI trend on Twitter. In practice, these techniques can be applied as Boston Children's has in areas like ICU capacity planning, or, as Atrius Health did, in avoiding preventable hospital admissions via proactive interventions. There's no shortage of applications.

The question for the IT leader though is how to develop and deploy models in support of these initiatives. Many vendors are jumping into the space. Are they, as Leonard D'Avolio of Harvard Medical School asserted, destined to fail, (or at least under-perform), due to their "one size fits all" mentality?

How Unique Are We?

D'Avolio focused on these axes of uniqueness:

- Priorities – unique vision

- Operations – unique ability to execute

- Data – unique data set over a unique population

The first two are linked and indisputable. Each organization has a unique viewpoint on where they want to go and will, if successful, align their ability to execute to this vision. This means we must choose the models that align with these priorities, but it doesn't mean that a shared model isn't a useful one when we share a priority.

The last axis though might, and this is what each presenter in turn explored: how does my unique data environment demand a unique analytics approach?

Interoperability in Analytics

Healthcare interoperability provides lessons that can be applied to health analytics and the portability of analytic models. Repeatedly the unique nature of each system’s health data was explored in terms of:

- Structured Data Semantics – how does data collection philosophy (revenue vs. clinical focus) and the terminologies used in this collection impact the resulting data set and its utility in analytics?

- Unstructured Data – how are images, notes, and genetic information abstracted into meaningful data sets for analysis?

These questions are at the heart of operational interoperability, as well as analytics interoperability and the penetration of standards such as: HL7 C-CDA and FHIR, DICOM, SNOMED-CT, and LOINC into the HIT landscape are critical.

The focus for health analytics practitioners, vendors, and consumers interested in portable, repeatable analytics solutions must be health data interoperability. Semantic interoperability. That's where we're headed.

Moving Forward

Analysts continues to invest in our team and our clients to improve health interoperability. We see analytics as a critical component of this HIT revolution, and we're working with HL7 and its members to define facilities for applying analytics to standardized health data. Health data is complex, health models even more so. Focused, personalized analytics are possible today. Portable analytics will be possible tomorrow.