Much like the paradigm of clinical versus financial data, many of the gospels associated with data analytics in healthcare are little more than malingering relics, holdovers from the manufactured narrative that grew up around early and inadequate rules-based offerings. Many of these products were successfully passed off as “big data” solutions for hospitals and health systems–which is almost staggering given that purely rules-based analytics lacks all mechanisms necessary to work with massive data-sets.
Indeed–whether in healthcare or any other economic sector–enterprise-level analytics hinges on data science. Nothing else can handle the volume, complexity, or variety of the data-sets.
Limitations of Rules-Based Analytics
Rules-based analytics is good at capturing the low-hanging fruit. It can run simple reports that flag obvious errors, and make simple decisions that for some basic business applications might prove perfectly adequate.
Simple? Basic? Hey, sounds like healthcare.
No, it doesn’t. Which is why many hospitals today are still bumping up against the limitations of rules-based analytics, and doing so on multiple fronts.
At facilities where rules-based revenue-integrity analytics have long been deployed, charge capture leakage continues more or less unabated; it’s just less obvious because you no longer see obvious instances.
Cut to population health. Clinical and financial leaders who try to pin down the drivers of unplanned readmissions instead end up with a jumbled list of one-offs and false positives. Why? It’s because their rules-based clinical analytics can’t uncover meaningful correlations across disparate data sets.
Over in patient collections: legacy rules-based analytics built around the data point of credit score are not only outdated, they are poised to tip off the cliff into oblivion thanks to new rules for reporting medical debt.
Rules-based software is always limited, sometimes wrong, and never uncertain. It is linear, manually created, and static; it will happily make the same limited set of decisions into perpetuity, with no ability to perceive when those decisions have ceased to be appropriate. Its only hope is that next great rule, the one that will surely solve the whole puzzle. Yet no matter how gifted or prolific the rule-writer, manual iterations and additions will never uncover the patterns and hidden anomalies that elude human detection–because humans themselves pre-limit and prejudice the parameters for discovery by telling the software exactly what to look for.
Machine Learning and Data Mining
Artificial intelligence, often shortened to AI or referred to as machine learning, is non-linear, continuously self-refining, and dynamic.
Where rules-based models lack the capabilities necessary to distinguish between near-same instances, especially within large datasets, machine learning models are designed to make thousands of rapid-fire decisions and improve the quality of those decisions over time by sharpening their understanding of non-obvious distinctions and anomalies.
Rules-based analytics is good at flagging a claim for pacemaker surgery that doesn’t include a code for the actual pacemaker or the record of a patient who has missed two cancer-screening appointments in a row. Predictive analytics, on the other hand, is suited to things like learning physician behavior–so that it knows exactly which implants and devices each individual spine surgeon prefers to use for a given procedure and doesn’t incorrectly flag a missing charge for the surgeon who always uses two screws and a plate rather than a plate and three screws.
Why It Matters Now
Across the clinical and financial worlds of healthcare, and across all provider types, the demand for and demands on healthcare data is growing. Providers are now expected to share data with each other and with patients, and at every turn they’re being pushed to make data-driven clinical and operational decisions. If they don’t–or if they can’t–penalties loom. Healthcare data is also growing organically due to the rise of telemedicine and other forms of HIPAA-compliant electronic communication–and, well, the government-mandated to switch to electronic health records might have had an impact too.
Providers’ ability to navigate this new environment will depend in part on how rapidly they stop expecting rules-based analytics to deliver what it never will, and how quickly they shift instead to technology that is non-linear enough to uncover pockets of opportunity that no one knew were there.
None of us can predict exactly what lies ahead for healthcare–and that’s a sobering thought when you realize how many hospitals still rely on analytics that wait around for instruction.
Kim Labow is ZirMed’s VP of Marketing.