Predictive modeling methods and systems will revolutionize virtually all aspects of how healthcare is organized, delivered, and paid for in the near future. But what will that future look like? What are the skill sets that will be relevant – and less relevant – and what will be the technologies, approaches, algorithms, and systems that will be most common and most successful?
Looking at other domains: Application silos and scarcity of talent.
One way to look at how predictive modeling technology will transform the healthcare sector is to compare it to other industries that were the earliest adopters of these methods; marketing is one such domain. According to Scott Brinker of Chiefmartec.com, the number of companies offering relevant solutions around marketing is truly amazing – 1,876 Vendors across 43 categories, which is roughly twice the numbers of relevant companies only a year earlier (947).
Two things stand out.
The challenges around data lakes and data silos are widely acknowledged today as key hurdles that need to be overcome for most implementations of effective analytics technologies. Likewise, a similar threat to the long-term success of analytics solutions in any environment arises when too many spot-solutions take root in the organization, creating application silos that become unmanageable and immovable when priorities change or new strategic directions emerge.
Scarcity of talent
If indeed we will witness a similar explosion of solutions and providers of analytics in healthcare, then a large number of “smart” and well-trained people will need to devote a lot of cumulative time and effort to build this out. Where are these skilled programmers, project managers, data scientists, and trained data-scientist/physicians going to come from?
According to a recent Forbes article, even though there are now over 100 Master’s and PhD programs in data science world-wide, the shortage for such talent persists and will be getting worse. This is immediately evident when looking at a recent snapshot of the exponentially growing numbers of openings for “big-data” data scientists.
Clearly, these trends cannot continue, and if healthcare organizations implement predictive analytics at rates similar to other industries and domains – such as marketing – then we will run out of resources, and the cost of such technology may even render it impractical for most healthcare providers, except perhaps for the largest and most prestigious organizations with ample public funding.
What if data science and predictive modeling can be automated?
Automation has changed many industries, such as manufacturing, where scarcity of human expertise and skill has been replaced by intelligent machines for many workflows. Can this happen in predictive analytics in healthcare?
Automated analytics as the next stage in analytic maturity
Thomas Davenport recently argued that automated analytics will emerge to address the shortage of data scientists, and perhaps completely change the way in which analytics is implemented. Systems and algorithms are rapidly evolving that enable dynamic learning from streaming content, thus automating the learning from such data sources. Some of my research with Dr. Lewicki on human expertise published years ago seemed to indicate that human learning may not be that different from advanced machine learning implemented in that manner. In other words, why not automate the process of (a) extracting useful information from data, and (b) translating that information into actions?
In fact, if we can be successful in this quest, not only could we ameliorate the shortage of skilled data scientists, but we may also be able to greatly scale up the availability of medical expertise. Perhaps much of the work and assessments that heretofore required skilled physicians can be automated to allow these professionals to focus on unusual cases, new treatments, etc. This possibility was recently explored in an article by Andreas Haimböck-Tichy.
What will automated intelligent analytics in healthcare look like?
Most likely, analytics and prediction models for prioritizing relevant facts, information and treatment options will become fully embedded and “disappear into” existing care delivery workflows, drastically increasing the efficiency of health care delivery. To give just one example, Anesthesia OS has recently integrated Dell Statistica models and decisioning rules into a platform to help anesthesiologists automate the process of identifying and prioritizing risks, and to suggest risk mitigation strategies. While not fully automating the critical data assessments that need to be performed by clinicians, analytics and intelligence embedded into the system will automate much of the information gathering and extraction from the data, to make anesthesiologists more efficient and accurate in risk prediction. Automation also makes the process safer by delivering real-time information to the place and person where it is most relevant.
Governance, fairness, and automated decisioning systems of the future
When computer systems become fully enabled to extract information from data and to prioritize or even act on those data, then how can we ensure that the outcomes are correct, robust, secure (not tampered with), and fair? In the traditional approach to predictive modeling, the data scientist, statistician, or data modeler plays the important role of reviewer and gatekeeper, deciding if a model is good-enough and appropriate to deploy. Who will play this critical role when predictive models are built and applied automatically?
Audit trails, version control, transparency, and validation
This topic has actually been receiving increasing scrutiny recently from regulatory agencies as well as legal scholars and consumer advocates. To satisfy the concerns of these groups, automated algorithmic decision making systems will need to be secure, transparent, auditable (support audit trails, version control), integrated with documented approval processes, etc. In other domains where analytics can potentially affect human welfare — namely in pharmaceutical and medical device manufacturing – similar system features to comply with regulatory oversight requirements are common. In short, the system must be designed to deliver results (predictions, actions, etc.) that are consistent with the anticipated system requirements;, that are fair and non-discriminatory; do good or at least do no harm; are tamper-proof and secure; and have sufficient safeguards and sentinels built in to detect when data inputs change in a way that invalidate models and predictions.
Predictive modeling systems and approaches hold tremendous promise for automating much of the routine diligence, decision making, information gathering and extraction. And they may even automate actions that now require highly skilled experts – be they data scientists, medical experts, or both. Such systems can automate much of the learning from the vast new data that are becoming available in medicine and elsewhere to deliver better, more personalized health care. At the same time, these systems will need to be designed to support governance, security, audit trails, approval processes and internal monitoring for data integrity and quality. That means they will be more complex than the simple proof-of-concept projects that are common today. They will need to rely on validated analytic frameworks and look much more like modern mission-critical IT systems. They should, as high quality healthcare is indeed critical to all of us.
Register or login for access to this item and much more
All Health Data Management content is archived after seven days.
Community members receive:
- All recent and archived articles
- Conference offers and updates
- A full menu of enewsletter options
- Web seminars, white papers, ebooks
Already have an account? Log In
Don't have an account? Register for Free Unlimited Access