Next-generation analytics meets first principles

| Article

Achieving lasting improvements in long-term care cost and quality

In the United States, providing care to individuals with special or supportive care needs1 costs roughly $800 billion annually—a sum greater than the economies of some G20 nations. Although these individuals comprise less than 20% of the US population, they account for more, perhaps far more, than 35% of total health expenditures.

Three groups of individuals typically have special or supportive care needs: those with behavioral health conditions, including substance abuse; those with intellectual or developmental disabilities; and those requiring long-term services and supports because of chronic, complicated medical conditions or physical disabilities. Most are Medicaid and/or Medicare beneficiaries.2

Because of the complexity of their needs, these populations can be challenging to manage effectively. Social determinants—including poverty, housing insecurity, and lack of transportation—often increase the disease burden these individuals face. The array of services they use is often poorly coordinated.

Another important barrier to effective management is the lack of integrated information. Many of these individuals are dually eligible for Medicaid and Medicare or have service coverage through multiple Medicaid programs. Until recently, information about these individuals—such as claims and needs-assessment data—has been fragmented across these various programs.

As we work to address these challenges, we can derive important lessons from the first principles espoused by Hippocrates, the founder of Western medicine. Many of his writings, which date back to 400 BCE, are directly applicable to the social and clinical issues affecting the special needs populations.

As the services these populations require get “carved into” integrated programs, information aggregation is increasing. By building capabilities for improved data definition and data manipulation, we can apply Hippocrates’s first principles to better serve these individuals.

Understanding the person

“It is far more important to know what person the disease has than what disease the person has.”—Hippocrates

How can we use the expanded data sets now available from integrated programs to determine “what person the disease has”? A first step is to look at spending on care needs.

For example, among the 5% of Americans with the highest healthcare spending, the average annual cost is about $47,455.3 Data analytics can help us better understand their needs and the ways in which they use care services. However, we need to be asking the right questions, including:

  • What medical services do they use?
  • What non-medical services do they use?
  • What are their diagnoses?
  • Who provides the services?
  • Where are the services provided?

Querying claims data to get answers to these simple, intuitive questions has been constrained by the lack of a consistent and sufficiently detailed data dictionary for long-term care services. So, McKinsey built one.

Building a data dictionary

To build it, we aggregated and organized the universe of long-term care claim codes into service categories, types, and locations (Exhibit 1).


Why this matters: Let’s say you wanted to understand rising home healthcare costs. You might review a few codes typically used to bill for home healthcare and find nothing amiss. Unless you looked comprehensively at all categories in the service matrix, you would miss other “hotspots,” such as rising costs for durable medical equipment.

Looking comprehensively across service types and locations allows for clearer, apples-to-apples comparisons across markets, regions, and member cohorts. With this, we can create granular profiles of cost and care use, which allow us to identify the root causes of high care utilization and tailor interventions appropriately. For example, within the subset of “high utilizers” in one state, we found a nearly threefold difference in supportive care costs and a nearly sevenfold difference in medical care costs, depending on the diagnostic condition (Exhibit 2).


Predictive modeling

“He will manage the cure best who has foreseen what is to happen from the present state of matters.”—Hippocrates

Modeling tools facilitate both retrospective diagnosis and prospective prognosis. A retrospective review of data can enable some improvements, such as instituting policies to manage the overuse of care that may have been observed in the prior year’s data. Prospective analytics can be used to better predict adverse events or healthcare costs. By adding a range of demographic- and claims-based variables, we have been able to improve significantly the ability to predict high care costs.

An accelerated journey

Since we began building this analytical infrastructure, we have been on an accelerated journey to apply Hippocrates’ first principles to better serve those most in need of care. This effort has resulted in better targeting of care management programs and resources, more tailored strategies to engage members, and an enhanced ability to predict and manage risk factors and cost.

The future will be brighter for these vulnerable individuals if the industry can gain even greater analytic sophistication. When it does, it can mine and manipulate an even wider universe of data—and use it to better address the social determinants of health.