The Expertise that Powers Health Intelligence
What is Health Intelligence? Health Intelligence means moving beyond simple analytics by providing insights into data that can then drive action.
The line between “analytics” and “intelligence” is a gray one, with no clear cut division between the two. I like to think of it as a spectrum, which ranges from simple mathematics and data presentation all the way to prescriptive models that can point someone in the recommended direction. The further we move across the spectrum, from simple analytics to true Health Intelligence, the more important it becomes to integrate expertise into your methods and models.
I’ve spent over 20 years working in the healthcare information field, and have seen time and time again the pitfalls of cutting corners on the depth of clinical, data science, and industry expertise in designing solutions. Investing in such expertise can be expensive, but the ramifications of not doing so are even more costly in the long run.
Let’s Examine a “Simple” Example
People unfamiliar with the US healthcare system and/or healthcare administrative data might think that it can’t possibly be that difficult to glean information out of such data. After all, the data is standardized and plentiful - the approaches used to extract meaningful information shouldn’t be that hard, right? I’d like to share a very simple example that illustrates why that assumption is wrong.
Suppose a benefits manager, who has access to their population’s administrative healthcare data, wants to answer the question, “How many diabetic members are in my plan, and how much of my healthcare spending is going toward diabetes”?
At first glance, this seems relatively straightforward. Most people with a basic understanding of claims data know that each medical claim includes financial data linked to ICD-10 diagnosis codes. These diagnosis codes provide insight into what conditions were involved in providing the healthcare service. A Google search can tell you the ICD-10 diagnosis codes that represent diabetes. Can’t you just find all the claims that have these codes and add up the payments? A novice user might try something just like that, looking at claims incurred over some timeframe that included a diagnosis code for diabetes. And the answer obtained might be “good enough” for the purposes of that user.
When Simple Isn’t Simple
But this simple approach belies the complexity that lurks within claims data. Even to execute the plan above, certain questions must be answered, such as:
- Which types of diabetes am I interested in? Type 1 and 2 only? Gestational diabetes? Drug-induced diabetes?
- How much claims history is sufficient to review in order to determine if someone is a diabetic? Six months? One year? Two years? What if the member has not been enrolled long enough to determine their health status?
- How many times does one have to see a code for a diabetes diagnosis in the patient’s history to be sure they really have diabetes? And to ensure that a rogue diagnosis code doesn’t represent a “rule out” blood test for diabetes (which can contain the same diagnosis codes)?
- What about members that don’t have any medical services for their diabetes, but are taking a diabetic medication such as Metformin or Trulicity? Could they be taking that for other reasons or does this mean they are diabetic?
- What costs are you interested in capturing pertaining to diabetes? The medical costs or the drug costs as well? Or are you interested in the total costs for all conditions incurred by a diabetic?
- If you want to include prescription drug costs, which generally represent the majority of diabetic costs, how do you find the appropriate claims? There are no diagnosis codes on pharmacy claims, instead, they each contain a National Drug Code (NDC). What is the best way to determine (of the nearly 200,000 NDCs) which ones represent anti-diabetic treatments?
- Medical claims can contain multiple diagnosis codes for a given record - are you interested in the cost of any claim that has any indication of diabetes, or only those where diabetes was the primary reason for treatment?
We extracted a sample of about 20,000 members and their claims from the Springbuk database, representing commercially insured members enrolled in December of 2019, determined who appeared to be a diabetic, and calculated their costs for diabetes during 2019, using different approaches to answering the bullet points above. Depending on the measurement approach, the diabetic population makes up anywhere from 6.0 to 7.4% of the population (1,175 to 1,456 individuals), and the average total (medical and drug) diabetic cost per person measured anywhere from $3,700 to $7,200. This is not insignificant variation! And this is exactly why expertise is so critical to answer even the simplest of healthcare data questions. Expertise in claims coding practices, clinical understanding, and data are all required.
Let’s Take Another Look
So what is the “right” answer to the initial question posed above? It depends. We need to understand the business need behind the question to know which set of assumptions are best to apply. Albert Einstein is credited as saying, “If I had an hour to solve a problem and my life depended on the solution, I would spend the first 55 minutes determining the proper question to ask.”
In this example, if the goal were to “cast a wide net” and identify anyone who might be diabetic in order to facilitate a low-cost intervention, such as mailing educational information, then the largest number of diabetics might be the most appropriate answer. Suppose you are instead using this as the basis for measuring the quality of diabetes care being delivered. In that case, we likely want to limit the members to those who almost certainly have diabetes, which would require a more conservative approach.
As for those diabetic costs, rather than measuring total costs, it likely makes more sense to provide better information on what types of costs are incurred. Do these costs represent hospitalizations? Emergency room visits? Routine blood work? And which drugs are being used by your diabetics - are they the best choices to ensure clinical effectiveness while containing cost?
And maybe most importantly, how do you know where there is room for improvement?
Quite Simply, Giving You Direction
At Springbuk, deep expertise is the foundation of our ability to answer the right business questions in the most appropriate way, delivering value to our customers. Clinicians, data scientists, and healthcare industry experts work together to provide the most appropriate answer for a given business problem, accommodating all the various idiosyncrasies that exist in healthcare data. We address the “simple” questions (like the example provided above) and the extremely complex questions with a thoughtful, deliberate approach. Our consistent goal is to answer the business question - to provide direction, not just data.
It’s easy to get an answer. It’s much more challenging to get the best answer to the specific business question. Addressing this challenge is what takes us from simple analytics to true Health Intelligence.
Meet the Author: Anne Fischer, VP, Data Science and Methods
With more than 20 years of experience, Anne Fischer has held multiple leadership roles in the Healthcare Information Technology industry and garnered extensive experience in healthcare analytics, with a focus on complex algorithms such as episode groupers and risk models. Prior to joining Springbuk, Anne spent the bulk of her career at Truven Health Analytics, eventually leading the Value-Based Care Emerging Analytics team after Truven’s acquisition by IBM Watson Health. Anne joined the Springbuk Health Intelligence team in 2019 as the Sr. Director of Data Science and Methods to spearhead the enhancement and development of analytics that drive value to healthcare payers and their members.