Full steam ahead. Complete mobilization. The power of people coming together to solve tough problems is described in many ways, but the results are always the same: solutions are found. Proof and point, the US Whitehouse recently included a specific focus on collaboration to solve problems, albeit in the context of security. 23andMe and Ancestry are the largest, direct-to-consumer DNA collectors in the world and what they’re doing is massive for teams researching genomic data – there’s no comparable collection of this type of data in terms of both volume and diversity. The importance of such analysis is the reason public cloud providers are working tirelessly to provide a “safe place” for the mass aggregation and analysis of healthcare data; aka “security clouds”. Today, healthcare companies often operate in silos, with limited collaborative abilities within a single country let alone on a global scale. As we covered in part one of this series, all of the major healthcare M&As over the last 10 years have been in support of initiatives that require data collaboration across lines of business, jurisdictions, or both. To understand the importance and difficulty in such collaborations, it’s important to understand healthcare data ownership: the relationship between data analysis technologies and those who actually own the data.
The buzzwords around data analytics are business intelligence (BI), machine learning (ML), and artificial intelligence (AI). Without going too deep, here’s a basic overview of the technology and some considerations one should take into account when evaluating them:
First, these terms are invoked rather loosely in the buzzword bingo game companies are forced to play, called SEO/SEM. Be sure to ask “how” ML and AI are used rather than accepting the terms at face value, and focus on the quality and speed of outputs rather than the underlying technology used to get there.
Second, you don’t simply summon “AI” out of a magic LAMP (if you know, you know), pack up your bags, and let the genie do the work. Like any software, developing these models requires the time of specialized engineers and data scientists to continuously train and improve models to attain acceptable levels of efficacy (in healthcare, the degree to which an intervention, procedure, or treatment is successful in producing a desired outcome).
Third, achieving high efficacy requires large volumes of diverse data in reducing bias. In this context, a bias is similar to a bug. A problem. For example, a bias in Apple’s facial recognition meant that this feature didn’t work for populations with low representation in the data used to develop and train the model. Now, imagine a bias that misdiagnoses a patient and prescribes an ineffective treatment (or worse yet, one that exacerbates the issue or creates a new one).
For a more technical dive, check out this whitepaper by Osterman explaining how to improve ML model training using secure, collaboration technology. The takeaway is that while we have the analysis tech we need, our data analysis tool sets are only as good as the data with which we can train them. This leads us to our next topic, healthcare data ownership.
Understanding how to apply “teamwork makes the dream work” across an entire healthcare system requires knowing a bit about the “team.” Here’s a high-level summary of the key players in the US healthcare system when it comes to healthcare data ownership:
We have the tools and know how to use them. We know who has useful data. We know who should collaborate to achieve these major health and business objectives. But the reality is that collaboration across these stakeholders is largely still manual and piecemeal, or doesn’t happen at all. In the next post, we’ll cover the regulatory and technical challenges faced when sharing and collaborating in healthcare.