Remember the book Freakonomics? Authors Steven Levitt and Stephen Dubner chronicled very interesting insights revealed by microeconomic researchers; insights that macroeconomics would not necessarily identify. The book went on to be an international best-seller and was followed up by a sequel that also did very well.

A similar pattern is emerging in the area of healthcare predictive analytics and especially in population health. The availability of millions of patient records contained in different healthcare data sources such as claims, EMR (electronic medical records), self-reported surveys, along with consumer data on social determinants has made it possible to reveal interesting insights. In this blog series, my team and I will share some of these interesting insights.

Affluent and Ailing

Consider the relationship between income levels and health outcomes. Numerous previous studies have reported that increasing income levels correlate with better health outcomes. The reasoning behind this seems obvious. For one, higher income levels are (usually) associated with higher education. Higher incomes also allow better and healthier lifestyles with a focus on preventive care. Together, this translates to better health outcomes.

To analyze this relationship further, we looked at clinical risk as a measure of health outcomes. For each member, the VitreosHealth Clinical Risk Score is a composite of that individual’s disease specific risk scores (which could be specific to diabetes, CHF, CHD among others depending on that member’s chronic conditions). Each individual’s risk indicates the risk of complications due to their chronic conditions and comorbidities. We studied a cohort of patients that had the same set of diseases but were at different income levels. We saw a very interesting trend.

As income levels increase, the clinical risk decreases. The risk keeps decreasing as income levels go from $25,000 to $50,000 to $100,000. That’s not surprising. Clinical risk somewhat flattens out once household income level increases beyond $200,000. But once income crosses $500,000, an interesting thing happens: the risk levels reverse course and start rising again. For example, highly affluent cancer patients had higher ER visits and admissions post chemotherapy than patients that had similar chronic conditions and cancers but with a moderate income.

How can you interpret this? One (untested) theory is that professionals and business owners making upwards of half a million dollars a year lead extremely hectic and busy lives, and deal with crushing pressures and stresses, all of which causes overall health to deteriorate. How can you approach this problem? Many of them don’t have the time to sit in clinic waiting rooms, but are wealthy enough for concierge services. Would such personalized service improve their outcomes? The underlying reason for the oncology patients I described above was determined to be family composition; most of these affluent members were living alone, and side effects from chemo were impossible to handle without in-home support. When these affluent members were given access to home health nurses, their costs due to emergent visits dropped drastically. Learn more about our work predictive analytics for oncology here.

Cost Versus Quality for ACOs

In the world of population health management, there is a lot of focus on quality scores. Higher scores can mean bonuses for health plans and physicians. With the new requirements around MACRA, even independent and small physician groups are now required to report quality scores. But do higher quality scores necessarily correlate with better health outcomes and lower costs? Specifically, do Accountable Health Organizations (ACOs) who perform well on quality also perform well on reducing costs? The findings really don’t seem to suggest this.

In general, ACOs have performed better on quality measures than on cost reductions. For example, Pioneer ACOs improved on 28 of the 33 quality measures in the first year. However, the cost results were mixed for the first 2 years. In the first year, they ranged from a 7% decrease to a 5% increase; in the second year, they ranged from a 5.4% decrease to a 5.6% increase. Why would that be the case?

One reason that has been advocated and has a strong buy-in is that there is a lag effect between quality and cost. The reasoning is that quality scores, especially clinical quality measures are tied to gaps-in-care, and reducing those gaps would have a longer-term effect on cost. While this theory makes sense, we are now in the 4th year on some of the ACOs, and we still have a mixed bag of success stories. The recent FierceHealthcare article “ACO savings estimate misses mark by $2B, but risk-based tracks on the rise” brings to light some of the challenges. What is REALLY happening here?

At Vitreos, we have had a chance to work with many MSSP and Pioneer ACOs. In the case of MSSPs, we have worked for Track 1 (upside risk only) and Track 2 (upside and downside) organizations. What we have seen is that often, hitting a quality target does not reduce costs in the short term or long term. The reason is actually very straightforward.

Let me illustrate this with a common gap for diabetes – HbA1C. As a plan or a physician group, you want to make sure that at least 90% of your diabetics are getting an A1C test done every 6 months. You would expect that hitting 90% would ensure that your diabetics would (ultimately) control their blood sugars and avoid unnecessary ER visits and admissions due to diabetes complications. What we saw is that simply getting 90% of your diabetics to come in for A1C was not enough. What was important was which 90% of those diabetics were targeted. In many cases, the remaining 10% who were not targeted were accounting for 60-70% of the overall diabetes costs. These were often patients with multiple co-morbidities, who were home-bound, with no caretakers, in a low economic stratum, and had significant access-to-care issues. These are the members that are very difficult to reach and take up too much of your resources if you want to have a quick ROI on your quality score maximization programs. So, reaching out to these 10% would perhaps have had a better impact on both quality, costs and patient experience. How do you identify which are the 10% most vulnerable patients? You utilize AI-driven predictive models that are trained to identify these risky patients. This is a non-intuitive problem that can solved by predictive modeling and optimization algorithms.

The Most Valuable Gap

Talking about gaps-in-care, has anyone done a study to determine which gaps have a higher impact on costs? If a diabetic patient has multiple gaps open, which gaps should you close first? Does an eye or foot exam prevent future cost increases? Or should you simply focus on A1C? What about if the patient also has heart disease? Should you first focus on getting the patient to complete his cardiology appointment? Or is it a basic access-to-care issue that could address chronic care compliance guidelines by bringing in the patient to the clinic? To the best of our knowledge, there is no research done on this. Thus, we at VitreosHealth started studying this.

To do this in a statistically rigorous way is tough and time consuming, but after completing our first phase (where we studied the impact of individual gap closures), we have some interesting results. For example, for Medicaid populations, we saw that on an average, hypertensive patients who do their blood pressure tests regularly reduce their monthly costs by $25 over the next 12 months compared with similar patients who do not do regular BP tests. If the patient is also undergoing cholesterol tests regularly, this reduction increases just slightly to $28. If the patient happens to be a high cost patient (with monthly costs between $150 – $700), then this reduction doubles to $56. What does this mean? If you have to focus on one gap to have the maximum impact, then target regular BP tests to control hypertension. This might sound like common sense, but the answers are different for different members depending on their different chronic conditions and comorbidities. This is what we call Precision Analytics at a member level.


We will continue sharing these interesting nuggets with you in future blogs. In the meantime, feel free to check out our webinar on using AI for more effective population health and start gathering your own Predictonomics insights.

 

Predictonomics – Insights from Years of Experience in Healthcare Predictive Analytics

One thought on “Predictonomics – Insights from Years of Experience in Healthcare Predictive Analytics

Leave a Reply

Your email address will not be published. Required fields are marked *