There’s a lot of buzz — rightfully so! — around the potential for Machine Learning (ML) and Artificial Intelligence (AI) in healthcare. I’ve been watching this space for some time now, and I continue to be bullish on the prospects of ML/AI in the healthcare industry. Here, I’m going to write about my views on The Why of ML/AI, some examples of The Who in this space, and, finally, some thoughts on The How these practices are going to disrupt processes in healthcare. Oh, and I’ll also provide some thoughts on the infrastructure that’s needed to make this all happen, because, those who know me will know my thoughts on infrastructure, viz. Infrastructure is Nothing! Infrastructure is Everything!

Let’s start by looking at this clever map of the most well-funded AI startups in each state. Upon tallying the results, you’ll note that 21% of the most well-funded AI startups across the US focus exclusively on healthcare. Wow!


Healthcare is, by far, the leading vertical focus for these AI startups, with the next largest focus being Business Intelligence (BI) @ 12%, followed by Cybersecurity and Marketing, both @ 9%. Interesting. This does, indeed, feed the bullishness. And, it’s not too hard to make the leap that while the Cybersecurity fellows are horizontally focused, they are going to help our industry, too!

Next, let’s look at a very clever application of Machine Learning in healthcare. Ayasdi, whose tagline is Use machine intelligence to become an automated, data and compute driven enterprise, has partnered with Mercy to optimize clinical pathways (read about it here). Using Ayasdi, Mercy was able to identify that the administration of a specific analgesic after arthroscopic knee replacement surgery correlated with lower costs and shorter hospital stays. Woah! Expanding this sort of next-generation, new-stack application opens up a multitude of possibilities. As Seth Barbanell, MD, Vice President of Mercy’s Clinical Performance Acceleration says, “Ayasdi will help us deliver new care pathways that provide improved clinical outcomes, such as higher patient satisfaction, lower readmissions, shorter length-of-stays, and lower costs. Additionally, Ayasdi will enable Mercy to maintain and optimize these care pathways on a frequent basis as industry conditions change.” Yes! Better patient care! Better patient outcomes! Higher patient satisfaction! Lower costs!

Further, considering the continued digitization of imaging and other rich media, AI applications abound in this space, too. For example, Wired Magazine has recently published an article titled “If You Look at X-Rays or Moles for a Living, AI Is Coming for Your Job”  that touches on AI in imaging (read about it here), In fact, Radiomics, which is defined as “the conversion of images to higher-dimensional data and the subsequent mining of these data for improved decision support,1 has exciting applications across all disease states, well beyond the initial application in oncology. Another great example in this space is IBM’s Watson Health medical imaging initiative (read about it here), where the goal is to collaborate across leading healthcare organizations and imaging software vendors to enable “…deep image analysis using image quantification, segmentation, classification, pattern recognition, and characterization can be used to propose statistically significant guidance on what the most-likely diagnoses would be, based on comparisons with a large cohort of other patients’ images and associated reports.2

These are exciting times! Application of AI to help detect and diagnose diseases faster, reducing time-to-treatment, lowering costs, leading eventually to better patient care and outcomes. This is the new-stack equivalent of clinical decision support (CDS)! Yes!

Why is this important? You’re probably thinking, duh, of course we want better care and outcomes with lower costs and higher satisfaction! I understand that, but I’d like to examine another key force at play. Here goes…

The Need for Strategic Data Infrastructure Management

As an industry, we’re getting close to the tail end of the arduous task of digitizing health records. The reality is that most organizations believe that EHRs, while indispensable, haven’t demonstrated a payback commensurate with the massive expenses incurred in implementing said EHRs. In many cases, organizations are struggling to find a reasonable, if at all, payback period. Additionally, advances in scientific computing and analytics — things like next-generation EHR, genomics, precision medicine, predictive analytics, machine learning and trending, managed care, Accountable Care Organizations (ACO), the imaging revolution, radiomics, exploring unstructured data such as notes, and so on — have resulted in an unbounded data explosion in healthcare.

With the big emphasis on cost remediation, waste elimination, and risk reduction, we find ourselves at an inflection point as an industry: we need to leverage all of the digital health data  to generate value from our EHR investments. We need to shorten the payback period by doing things better, cheaper, safer, and faster. We need precision medicine; and genomics; and population health; and reduced admission rates; and so on. The key lies in unlocking meaning from this massive quantity of data, applying rigorous math, statistics, and analytics to understand how to optimize healthcare delivery, make patients and providers happier, and result in better patient care, better patient outcomes. This is an extremely data intensive endeavor, as are these next-generation meaning-unlocking applications, some of which I’ve touched on earlier in this post. We’ll be generating more data on top of a lot of data, resulting in even more data. And the cycle continues. But it’s necessary.

In order to survive, let alone thrive, our data infrastructure management model needs to undergo a fundamental transformation. We need to shift from a tactical model – one which is analogous to the processes of procuring surgical gloves and thermometers – to a strategic framework – one where the data infrastructure melds into the background and allows us to focus on innovating and harnessing value from data. We need to break the never-ending bid-procure-install-configure-tune-run-monitor cycle. We need to move to a set-it-and-forget-it model. This is what I call strategic data infrastructure management.

In a coming post, I’ll dive deeper into strategic data infrastructure management, with a focus on infrastructure and business models that allow us to move to this set-it-and-forget-it paradigm.


RSNA. (2016). Radiomics: Images Are More Than Pictures, They Are Data. Retrieved from:

Physicians Practice. (2016). Radiomics Come of Age at RSNA 2015. Retrieved from:

Join the discussion...