What Is Data Mining in Healthcare?
Data mining holds great potential for the healthcare industry to enable health systems to systematically use data and analytics to identify inefficiencies and best practices that improve care and reduce costs. Some experts believe the opportunities to improve care and reduce costs concurrently could apply to as much as 30% of overall healthcare spending. This could be a win/win overall. But due to the complexity of healthcare and a slower rate of technology adoption, our industry lags behind these others in implementing effective data mining and analytic strategies.
Like analytics and business intelligence, the term data mining can mean different things to different people. The most basic definition of data mining is the analysis of large data sets to discover patterns and use those patterns to forecast or predict the likelihood of future events.
That said, not all analyses of large quantities of data constitute data mining. We generally categorize analytics as follows:
- Descriptive analytics—Describing what has happened
- Predictive analytics—Predicting what will happen
- Prescriptive analytics—Determining what to do about it
It is to the middle category—predictive analytics—that data mining applies. Data mining involves uncovering patterns from vast data stores and using that information to build predictive models.
Many industries successfully use data mining. It helps the retail industry model customer response. It helps banks predict customer profitability. It serves similar use cases in telecom, manufacturing, the automotive industry, higher education, life sciences, and more.
However, data mining in healthcare today remains, for the most part, an academic exercise with only a few pragmatic success stories. Academicians are using data-mining approaches like decision trees, clusters, neural networks, and time series to publish research. Healthcare, however, has always been slow to incorporate the latest research into everyday practice.
The question that leading warehouse practitioners are asking themselves is this: how do we narrow the adoption time from the bench (research) to the bedside (pragmatic quality improvement) and affect outcomes?
The Three Systems Approach
The most effective strategy for taking data mining beyond the realm of academic research is the three systems approach. Implementing all three systems is the key to driving real-world improvement with any analytics initiative in healthcare. Unfortunately, very few healthcare organizations implement all three of these systems.
The three systems are:
- The analytics system. This system includes the technology and the expertise to gather data, make sense of it and standardize measurements. Aggregating clinical, financial, patient satisfaction, and other data into an enterprise data warehouse (EDW) is the foundational piece of this system.
- The best practice system. The best practice system involves standardizing knowledge work—systematically applying evidence-based best practices to care delivery. Researchers make significant findings each year about clinical best practices, but, as I mentioned previously, it takes years for these findings to be incorporated into clinical practice. A strong best practice system enables organizations to put the latest medical evidence into practice quickly.
- The adoption system. This system involves driving change management through new organizational structures. In particular, it involves implementing team structures that will enable consistent, enterprise-wide adoption of best practices. This system is by no means easy to implement. It requires real organizational change to drive adoption of best practices throughout an organization.
If a data mining initiative doesn’t involve all three of these systems, the chances are good that it will remain a purely academic exercise and never leave the laboratory of published papers. Implementing all three enables a healthcare organization to pragmatically apply data mining to everyday clinical practice.
Pragmatic Application of Data Mining in Healthcare—Today
When these principles are in place, we have seen clients make some very energizing progress. Once they implement the analytics foundation to mine the data and they have the best practices and organizational systems in place to make data mining insights actionable, they are now ready to use predictive analytics in new and innovative ways.
One client is a health system trying to succeed in risk-based contracts while still performing well under the fee-for-service reimbursement model. The transition to value-based purchasing is a slow one. Until the flip is switched all the way, health systems have to design processes that enable them to straddle both models. This client is using data mining to lower its census for patients under risk contracts, while at the same time keeping its patient volume steady for patients not included in these contracts. We are mining the data to predict what the volumes will be for each category of patient. Then, the health system develops processes to make sure these patients receive the appropriate care at the right place and at the right time. This would include care management outreach for high-risk patients.
With another client, we are mining data to predict 30-day readmissions based on census. We apply