04 4 / 2014
Today, the Woodrow Wilson School hosted a fascinating conference called Big Data and Health: Implications for New Jersey’s Health Care System. Dr. Jeffrey Brenner, founder of the Camden Coalition, subject of a New Yorker profile, and recipient of a MacArthur Genius Grant, spoke on one of the panels. Halfway through, he reeled off a list of areas where Big Data comes up short and where the industry misuses it. Here they are:
1. We’re obsessed with prediction instead of surveillance
In other words, we focus on guessing rare events that will happen in the future instead of what is happening right now and why.
2. Stratification instead of segmentation
Brenner gave the somewhat facetious example of lining up people from tallest to shortest instead of segmenting the group to make sure they get what they want, like smart commercial businesses do.
3. Centralizing data instead of decentralizing it
We make it hard for researchers and consumers to access vital data.
4. We’re obsessed with averages
"Epidemiologists have ruined data sets by cutting off outliers," Brenner said. Outliers can actually be the most important part of the data set. Super-utilizers are the best people to tell us where our health care system is failing
5. We focus on numerators over denominators
He referred to the distinction, in other words, between an event and the absence of an event.
6. We focus too much on seeking out conclusions instead of looking for better hypotheses.
Correlation is beneficial insofar as it can generate hypotheses. We can’t necessarily infer causation from correlation, because causation can go in a number of different directions. Big data sets should just tell us, “go in that direction.”
11 3 / 2014
Recently, a WSJ contributor claimed that the ACA was responsible for the “rapid spread” in high-deductible plans. His argument ignored the history of high deductibles, which arose last decade due to a number of factors. For more, check out my post over at The Incidental Economist.
12 2 / 2014
(2/14 update: I elaborated on this post over at The Incidental Economist)
Below is the new table, presenting (1) the proportion of consumers who have enrolled with financial assistance out of total enrollees, (2) the proportion of theoretically subsidy-eligible consumers out of the total population who could buy coverage on a state’s exchange, and (3) the absolute difference between the two columns.
(This time I’ve bolded SBM state names, with all states relying on healthcare.gov—whether FFM or Partnership—appearing as normal).
Compare the final columns in the two tables to see how things have changed in a short period of time. As you can see, the previous months have seen a flood of subsidy-eligible enrollees relative to other enrollees in many FFM states. Texas, Ohio, and Virginia went from way in the red to substantially in the black.
In December, most states weren’t going a good job of attracting subsidy-eligible consumers—they were generally far behind the overall new market proportion, hence all the red text in the last column.
Now, as you’ll see below, most states are enrolling a greater proportion of subsidy-eligible consumers than contained in their overall populations of people who could theoretically buy on the exchange. In other words, they are reaching people in need, with FFMs and Partnership states now outpacing most SBMs on that account. ASPE reported that the overall subsidized numbers are 79% for SBMs and 83% for FFMs (page 9 here.)
Keep in mind that these numbers don’t reflect total enrollments, merely the proportion of people receiving tax credits. This is just one way out of many to judge the success of a state’s exchange.
(One caveat to this analysis: In December, ASPE reported the number of people found eligible for subsidies out of everyone who filled out an application. We don’t know how many of them actually selected a plan at the time. Now, ASPE reports the number of people eligible for subsidies out of people who selected a plan, so it’s not a perfect comparison, unfortunately.)
Update: smart people on Twitter have raised some good questions about these data.
(3) @mannymr pointed out that Minnesota data may not include MinnesotaCare, which is a program for people between 138-200% FPL in the state. Although ASPE did not provide subsidized MN data in the recent release, this still serves as an important reminder. MinnesotaCare is siphoning off consumers who would have otherwise acquired subsidized QHP coverage on the MNsure exchange. As a result, we won’t be able to compare Minnesota to other states in an apples-to-apples sort of way. More on MinnesotaCare here and here.
17 12 / 2013
Over at Project Millennial I wrote an expanded version of my previous post here on Tumblr.
Check it out, and please leave a comment if you have any feedback: http://projectmillennial.org/2013/12/17/how-should-we-judge-subsidized-enrollment-numbers/
13 12 / 2013
With the recent release of enrollment data, substantial attention in the media has focused on estimates made by the Congressional Budget Office in early 2013. In this document, CBO predicts that 7 million people would enroll in the first year, with 6 million subsidy-eligible and 1 million ineligible (for an eligibility rate of 86%).
But is it right to focus on those projections? It may not be as useful as many people think.
It isn’t a goal
CBO’s first-year enrollment projection of 7 million is merely an estimate, not a goal. Similarly, the 86% number is nothing more than a guess of who may enroll when. It’s not a normative judgment of how many subsidy-eligible people should be enrolling, or how much of that population the Administration wants to enroll. Additionally, because CBO did not divulge its methodology, it isn’t even clear what factors are considered in the 86% estimate. As far as I know, it has no implication for the sick/healthy mix of enrollees, the most important factor for sustainability of the exchanges.
Despite all these questions, the projections consistently pop up in the media.
For all the unwarranted stock put into the first year enrollment projection, the subsidy estimate is even flimsier. To accurately predict subsidy eligibility, you first have to nail the overall enrollment. And that’s not a given. Getting the subsidy estimate right is doubly hard.
There are big state differences
Another problem with the 86% number: it’s an estimate for the country overall, rather than broken out by state. But the reality is that we see tremendous variation among states, in terms of eligible population, socioeconomic breakdown, enrollment capacity, and more. Furthermore, states surely have different goals for themselves. With all of that in mind, a national average becomes less useful.
With healthcare.gov troubles and delayed marketing pushes, the enrollment pool for the first year could end up looking much different than many expected. Without actually seeing CBO’s model, I can say with confidence that it did not consider technical issues with the website when making its estimate. And, get this—that might be totally fine! Coming up short of their estimate is not bad, in and of itself. It might not be bad at all.
When talking about “low enrollment” among consumers who qualify for subsidies, we need to think about enrollment relative to the overall population of people eligible for a marketplace plan.
There are two potential reasons to worry about this type of low enrollment for subsidy-eligible consumers. One is the balance of the risk pool. Maybe we think that people ineligible for subsidies are more likely to be sick, meaning that high rates of such enrollment might be bad news for insurers. However, I know of absolutely no evidence to suggest that sick people are less likely to qualify for subsidies. Unless someone can come up with a different theory, we should cross this off of the list of possible worries.
The second reason pertains to social justice. The subsidies exist to enable people with financial need to afford health insurance. If they aren’t enrolling in the first place, that’s bad for the law and for our health system in general. If they are enrolling but not taking advantage of the generous financial assistance available to them, that also would be really unfortunate. It is a legitimate concern. And it is exactly why comparing current enrollment numbers to the overall subsidy-eligible market is much more useful than comparing to CBO’s opaque first-year projection. Using KFF’s national eligibility estimate of 60% is good. Looking at individual states to see how well they are reaching residents in need is even better.
11 12 / 2013
With the new enrollment numbers out today, I decided to take a closer look at numbers around financial assistance. In the table below, columns two and three, representing the number of applicants determined to be eligible for a Marketplace plan and then the subset of applicants eligible for tax credits, come from this ASPE report on enrollment. Column four is my own calculation, dividing the preceding two columns. Columns five and six are from this November KFF analysis, which used Census data to make eligibility estimates. Column seven is my own calculation, taking the proportion of five and six, and in column eight I subtract column seven from column four. My apologies for the small size of the text in the table.
I think the final column is the most interesting one. It presents how a state’s current applicant composition looks relative to its total estimated market (see page 2 of KFF report for their method of estimation). In many states, represented by the red text, a low proportion of people who qualify for financial assistance have applied so far, relative to the total population in the state that would be eligible if they applied (per Census data).
As you can see, many state-based marketplaces are succeeding in getting people who need financial assistance to apply. In fact only SBMs have applicant pools that are seemingly lower-income than their overall populations that are theoretically eligible for a plan in the new Marketplace. Maryland, Oregon, Connecticut, Rhode Island, and Minnesota are all above the line. Some SBMs such as Vermont and Hawaii are struggling, but most states deep in the red defaulted to a federal marketplace.
For all states, but especially states in the red, three things could be going wrong: 1) That people who would be eligible for tax credits aren’t applying 2) That such people are applying but misstating their income for whatever reason 3) That such people are applying but the system is erring in making eligibility determinations. It’s also conceivable that the system is determining people to be eligible who are not.
1 2 3 4 5 6 7 8
12/13 update: I wrote a follow-up Tumblr post discussing state estimates and CBO’s enrollment projection
12/17 update: I expanded on my follow-up post over at Project Millennial
10 12 / 2013
Last month I did a post at Project Millennial about actuarial value and metal levels. I’ve seen too many newspaper articles refer to a Bronze plan as covering 60% of one’s costs, a Silver plan covering 70%, etc. The reality is that actuarial value is about an average amount, across a standard group containing both sick and healthy people.
Here’s the post:
06 9 / 2013
Today, Minnesotans learned that they will pay some of the lowest insurance rates in the country. The state’s insurance exchange, MNsure, unveiled the costs and networks associated with the 78 new individual plans, which consumers will begin purchasing on October 1. Coverage will go into effect starting January 1.
The low rates partly stem from active rate review, conducted by the state’s Department of Commerce and arguably the most robust in the country. Two years ago, Minnesota beefed up its rate review process, which was already strong, through a $3.9 million grant from HHS. According to today’s announcement, rate review “resulted in premiums for Minnesotans four to 37 percent lower than had originally been filed.”
Competition also helped to keep rates low. As the map below shows, consumers will have choice, with at least two issuers offering plans in every county. Across most of the state, in fact, consumers will have at least three issuers to choose from, and even more plans.
The table below presents average statewide costs for each metal level. Within each age group, there is surprisingly little variation between the metal levels, relative to other states.
The numbers compare very favorably to the rates released by other state-based exchanges. Minnesota is on the left side of this graph, with average rates lower than even in Oregon, lauded for affordable costs and high levels of competition on its own exchange.
With subsidies, many consumers will pay far less than the sticker price. In the Minneapolis-St. Paul metro area, for instance, a non-smoking 25-year-old making $22,000 will pay between $21 and $50 for a Silver plan, taking the subsidies into account. A young person earning minimum wage and working a typical week will not pay one cent for insurance.
For the same plan, a 40-year-old making $22,000 will also pay between $21 and $50. A 40-year-old making twice that income will still only pay around $150 per month. A minimum-wage worker will most likely pay nothing.
13 4 / 2013
Oregon’s Medicaid gamble continues to fascinate me. Yesterday the Times published a piece on the development of Oregon’s coordinated care organizations (CCOs), fifteen of them across the state. After having its waiver approved by CMS last summer, Oregon now has five years to show the federal government that it can slow Medicaid spending and improve health outcomes for beneficiaries, relative to the rest of the country. If unsuccessful, the state could face millions of dollars in fines each year.
A new venture like this comes down to a calculated risk. In accountable care organizations (ACOs), health care providers must be willing to accept financial responsibility for their patients’ health. A recent report released by the United Hospital Fund of New York described ACOs in this way: “The ACO program… [attributes] a defined population of Medicare fee-for-service beneficiaries to an organized group of providers (based on the beneficiaries’ historical patterns of utilization) and holding the group accountable for the quality of care, the patient experience of care, and the total costs of care generated by that population.”
The localized CCO model is similar, although ACOs seem to face more scrutiny and risk. Oregon appears to offer only positive financial incentives to its CCOs, according to the Times, ”for meeting goals like rates of adolescent well-care visits and colorectal cancer screening.”
Local leadership is what really set CCOs apart. Community Advisory Councils set local goals, based on the most pressing needs of those particular areas. Many of Oregon’s fifteen CCOs are stationed in rural, relatively conservative areas, far from the “joiner” mentality of urban Portland described by the Times article. As such, we will see a number of different approaches to the provision of effective care. If CCOs are to succeed across the state, they will require the buy-in of people of all political stripes.