We Already Have Health-Risk Scores. Now Let's Use Them.
Most Americans know they have a personal credit score, and many know where to find it. Few know they also have a personal health-risk score. If these were better known, and better constructed, health insurance markets in the U.S. would work more smoothly.
Commercial health insurance plans, as well as Medicare, Medicaid and other government programs, generate risk scores every year for most of the people they cover. These scores are estimates of each person’s cost of care, compared with the average costs in a large population. And they play a big role in health insurance; they’re often used, for example, to determine how much more insurers are paid for sicker beneficiaries.
Why don’t you know about your score? Because it belongs to your insurance provider, and it doesn’t travel with you when you switch plans. A portable, higher-quality health-risk score could help insurance function better. And, in time, generating more accurate scores for everyone shouldn’t be that difficult.
A portable, reliable health-risk score would enable insurers to better predict their costs. This could be especially helpful for the Affordable Care Act’s public exchanges, which have encountered significant headwinds in large part because actuaries have not accurately foreseen the costs of enrollees. It is well known that predicting costs in large, stable populations is fairly straightforward, but in small or changing ones (such as new buyers of an exchange plan), it’s notoriously difficult. Having access to accurate health-risk scores would make it much easier.
The government already uses health-risk scores to set prices for insurance companies on the Obamacare exchanges and in the private component of Medicare. With more accurate scores, though, it would become less important for the exchanges to attract a great number of young and healthy customers to make up for those who need more care. Similarly, more accurate risk adjustment would obviate the lingering debate over whether Medicare Advantage pays insurers more than what it would cost to cover the same people under traditional Medicare. And it would make it easier to manage alternative payment models such as bundled payments and accountable care organizations.
Existing scores, including those used for the public exchanges, are better than nothing, but they’re not as reliable as they could be. This is in large part because they are calculated from data on medical claims. Unfortunately, using claims for this purpose doesn’t work well enough to account for most of the variance in health-care spending. We could do a lot better.
Risk scores would be much more accurate if they were derived from clinical health data, such as that contained in electronic health records. To see the difference, imagine a woman who suffers a heart attack. The claims data will show that she was taken in an ambulance to a hospital, was admitted through the emergency room, had an angioplasty procedure, and ultimately was discharged. But these data will indicate little or nothing about her health thereafter. Electronic clinical data include the woman’s blood pressure, cholesterol, weight, diabetes indicators and other details about her health. These could be used to generate a much more accurate risk score.
Using clinical data provides substantial additional power to explain variations in people’s health-care spending. Adding variables based on a patient’s socioeconomic status, even just the person’s census tract, improves the score even more.
With more than 85 percent of U.S. physicians now using electronic health records, it is time we started putting all that information to work making the health system more efficient -- and risk scores are an important use for the data.
Several challenges remain, though. The first is that few people (certainly neither of us) would want to turn their clinical data over to the federal government or anyone else to generate risk scores. One way around this obstacle is to develop algorithms that could be run on the data where they are, generating risk scores without allowing external access to the clinical information.
Risk scores themselves contain no information about people’s health conditions. They are a mechanism for estimating future costs, not for gaming the system or discriminating against people based on their pre-existing medical conditions, though current protections against such abuses must continue to be enforced.
A second challenge, and one that’s harder to address, is that beneficiaries’ clinical data, even when stored electronically, are often scattered across multiple health-care systems and providers. As the payment system evolves so that a single provider becomes responsible for all costs associated with a given beneficiary, and as efforts progress to improve interoperability across data systems, this problem will diminish.
As things stand, claims data is the best data available for generating health-risk scores. For the next few years, such scores should continue to be used to improve the functioning of insurance markets. The more they are allowed to move with people when they switch insurers, the better.
In five to 10 years, it should be possible to use more accurate scores, based on clinical data. These will go a long way toward ensuring fair payment for health-care services and better functioning insurance exchanges.
This column does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners.
To contact the editor responsible for this story:
Mary Duenwald at email@example.com