This article originally appeared on the BeyeNETWORK
Organizations just beginning the business intelligence journey often face the same fundamental question of where to begin. While this is the question for organizations in all industries, it is especially true right now in healthcare. There is currently intense pressure on organizations across the industry—for providers, payers, purchasers and suppliers—to perform better, faster and cheaper. The situation is only going to get worse as the population ages and requires more medical care. Because of this, shortages of doctors, nurses and other healthcare professionals will become more acute.
The answer to the question of “where to begin” lies in the analyses that are being performed right now in your organization and the reasons behind them. These reasons could be clinical or business in nature. These reasons could also be driven by research needs to provide evidence for your decisions.
Regardless of the driving forces behind the questions, the solution with the most promise is the same. Data. And the data you need to address key problems is probably data you already own. Examples of such data include quality demands, performance contract demands, ensuring patient safety and improving patient satisfaction. Getting value from that data requires managing data effectively to slice, dice, sort, sum, combine and distill it. This can be done for various decision-making purposes, both inside and outside your organization.
But where do you begin? I offer five steps, along with examples, to help healthcare organizations begin the business intelligence journey below. These five steps are prioritized in order to help you and your organization deliver value early and consistently.
Step 1: Publish Your Performance and Quality Measures
Hundreds of healthcare collaboratives are beginning to form throughout the country. Many organizations, like the Wisconsin Collaborative for Healthcare Quality, offer opportunities to publish your performance and quality measures. While some people believe that a public display of your performance and quality measures is a negative thing, I view it as an opportunity. There are several reasons why performance and quality measures should be seen as an opportunity:
- Cooperation and Competition. Most members of these healthcare collaboratives band together because they do not share markets. Because of this, they can use the collaborative entities to share ideas and engage in friendly competition, by ranking themselves against each other.
- Focus. It focuses your entire organization on what is important to your patients, purchasers and the governmental agencies that keep an eye on the healthcare industry.
- Foresight. Today, publication of your performance and quality measures on collaborative websites (and with regional health information organizations) is a new phenomenon. In the near future, though, it will be a requirement. By managing your intelligence data now, you can better prepare for it in the future.
- Clinical and Business Value. Publication of your performance shows your commitment to improving clinical excellence, as well as a commitment to delivering value to your stakeholders.
Most of the measures used have been developed by quality accreditation organizations (e.g. NCQA, JCAHO), governmental agencies (e.g. Centers for Medicare and Medicaid) or healthcare purchaser groups (e.g. Leapfrog). Right now, most of these measures revolve around clinical processes and outcomes, or on costs. In the future, measures are likely to focus on the total value of health services. Even the measures currently being used can provide value to your organization if you apply business intelligence practices. This is evident in the example that follows in step 2.
Step 2: Squeeze Value from Your Structured Data
It is estimated that structured data (i.e. quantifiable) represents 10-20% of the knowledge in an organization in any industry, whereas unstructured data (i.e. free-flowing text) makes up the other 80-90%. This appears especially true in healthcare, where practitioners make extensive use of clinical notes to record their observations, activities and results.
But not all of the data is unstructured text. Numerical measures, percentages, counts, categorizations and other structured data points abound in the healthcare field. Moreover, this structured data is greatly undervalued by many such organizations, especially in terms of analyzing trends and patterns. This is true on the clinical, business and research side of the organization.
Here is one example of a measure that shows how extensive structured data is in healthcare provider organizations. Note that this is all structured data.
- Let’s say a patient is a member of the organization’s diabetic population of 7800.
- Her blood sugar (hemoglobin A1c) level is 6.4%.
- This puts her into the Optimal Control category because her A1c level is below 7.0%. The other three categories are near optimal control (7.0 to 9.0%), poor control (above 9.0%) and not tested.
- The organization is currently reporting 56% of its patients are in optimal control, which translates into 4368 patients. In addition, this organization reports that 9% of its diabetic patients are not tested (i.e. 702).
- Using its diabetes patient registry, the provider learns that 137 of these not tested patients belong to the East Street Clinic, but only 3 of the 702 patients belong to the North Street Clinic.
- In addition, Dr. Jones has 53 patients who are not tested. Dr. Smith has none in the not tested category.
- Furthermore, Dr. Green has 82% of her patients in optimal control, while Dr. Brown has 26% of his in optimal control.
- What is the revenue for a diabetes-focused visit and what does that translate into as lost revenue for the 702 patients who are not tested?
- For every one percent drop in A1c, the chance of a four-day hospitalization is reduced by 14-20%.
- The organization’s weighted average A1c level across its entire population is 7.2%. By reducing this average to 6.2%, the organization will have prevented a potential 6240 hospital days. And what is the cost of a hospital stay per day to payers and purchasers?
All of this data is structured (quantifiable). This is just one example of a measure out of over 700 that are currently collected by the Agency for Healthcare Research and Quality. Virtually every type of clinical and business performance and quality measure, germane to healthcare, has an organization championing and defining these measures.
Using these measures effectively is up to you. They are an important step in your business intelligence journey.
Step 3: Translate Measures for Different Decisions and Decision-Makers
Look back at the example above. Notice how the unit of measure changed during the example. It began as a scientific percentage (A1c of 6.4%), moved to a population count and a population percentage (56%), then to patient counts and finally to days and dollars. This was deliberate and based on real-world experience.
To get the maximum value from your data, you need to translate measures (often the same measure) for different audiences. This is essential, yet often overlooked.
Who might have an interest in these various measures?
- Diabetic Population Count. Your Chief Medical Officer and medical management would be interested, as would your quality, marketing and operations administration people.
- A1c Percentage. This would interest physicians, nurses, dieticians, diabetes educators and certainly patients.
- Untested Patient Count. Health plans are measured on the percentage of members who have had a blood sugar screening. Thus, this is a key measure for collaboration between providers and payers.
- Lost Revenue. Your CFO might be interested in this.
It obviously makes sense to use the structured data that you already own and translate it into value for the different decision-makers, both inside and outside your organization.
Step 4: Aim for Predictive Analytics
Predictive analytics is one of the hottest topics in business intelligence today. Predictive analytics represents a strategic opportunity with great potential value for both healthcare and non-healthcare organizations. Along with this value, however, comes a need for data. This is why it is Step 4, instead of Step 1.
You must have a good handle on your data before applying predictive analytics tools. It is challenging enough to get people to make reliable, consistent decisions at human speed using the often confusing and massive amounts of clinical, business and research data. It would be a monumental mistake to rush into making predictions in these areas at computer speed.
Your organization, however, must aim for predictive analytics. After all, evidence-based medicine is based on the assumption that you can predict a patient outcome by studying the patterns of interventions and outcomes from past clinical encounters. The same is true of wider decisions regarding healthcare effectiveness, efficiency, safety, access and cost-effectiveness.
Once you have a sufficiently deep evidence base about your organization and its activities, then you can begin to apply predictive analytics. Here is a brief list of profitable applications:
- Demand and Revenue Forecasting. This allows for better allocation of resources.
- Patient Outcomes Prediction. This is the essence of evidence-based medicine. Using various patient characteristics, the guidelines used in their care and adherence to care protocols leads to predictions of how quickly and completely they can recover, return to full functionality, return to work, etc.
- Risk Prediction. There are numerous examples including surgery risk, infection rates and severity, hospitalization, mortality, etc.
Prediction is also a key element in many healthcare initiatives, such as disease management, case management and, of course, epidemic forecasting. Clearly, it pays to be ahead of the curve.
Step 5: Dig Into Your Unstructured Data
If 80-90% of your organization’s knowledge exists as unstructured data (emails, documents, web pages, etc.), why not start here, instead of with the 10-20% of data that is structured? The answer is faster payback from an overlooked source of value, as described above.
However, your organization can now do two things to position itself for effectively using unstructured data. First, you must define key applications where your hidden data will be critical. Some examples of this include:
- Hypothesis Discovery. One organization found a strong correlation between certain types of liver conditions and skin cancer using data mining. This is an example of using unstructured data that you already own to find and focus research projects.
- Satisfaction Surveys. Discover what is really driving satisfaction and dissatisfaction among your patients and staff, leading to improved care services.
- Finding Co-morbidities. What patients report and clinicians’ record can help you find patients who are at risk of more than one chronic condition (e.g. diabetes and depression).
Second, you can get your organization ready to use data mining results. Sometimes stories can be helpful. Here is an actual example:
A clinic group scanned their clinical notes and found a strong association between good diabetes control and vehicles. In turns out nurses were scheduling rides to get their patients in for tests and focused visits, and therefore were in control. Words like car, ride, taxi, etc. repeatedly showed up in the scans. They then contracted with a local limo service to put the clinic’s name on the side of dedicated minivans. They increased revenue, improved their community reputation and received a positive write-up in the local paper.
Data is your ally if you use it wisely. But you must know where to begin. You should first examine the data that you already own and may be neglecting.
Each of the five steps presented above could actually be giant leaps for some organizations, especially culturally. Once you start down the business intelligence road, you can never go back. But given the tremendous potential for clinical, business and scientific benefits, you may not want to go back.
Thanks for reading. I look forward to your comments.