Editor’s note: This post is part of a Health Affairs Blog symposium stemming from “The New Health Care Industry: Integration, Consolidation, Competition in the Wake of the Affordable Care Act,” a conference held recently at Yale Law School’s Solomon Center for Health Law and Policy. Links to all posts in the symposium will be added to Abbe Gluck’s introductory post as they appear, and you can access a full list of symposium pieces here or by clicking on the “Yale Health Care Industry Symposium” tag at the bottom of any symposium post.
As insurers embark on a new wave of mergers and acquisitions, lawmakers in the House and Senate, regulators, consumer advocates, and health care providers are asking tough questions about the likely impact of these combinations. Much has been written about how health insurance mergers raise premiums even as they reduce prices paid to providers. But very little has been said about the impact of consolidation on insurance innovation.
In part this oversight is the result of decades of inattention to insurance innovation. After all, insurance plans today can be described using virtually the same terms as 35 years ago: deductibles, coinsurance rates, and limitations on certain services. Interest in a slightly different model—the classical, tightly-managed health maintenance organization (HMO)—has waxed and waned, and even those models have adopted the standard financial characteristics of other health insurance arrangements.
Variation In Network Breadth Is A Measure Of Product Innovation In Health Insurance Markets
The “innovation” of today—“limited,” “narrow,” or “high-performance” networks—is actually a variant on the well-worn strategy of “selective contracting.” Identifying the most effective and efficient providers, negotiating favorable terms with these providers, and providing financial incentives to patients to utilize these providers (e.g., by including only selected providers in-network), are tried and tested tactics in insurers’ toolkits. However, insurers pursued selective contracting with gusto for products offered on health insurance marketplaces in 2014 and 2015. (There are signs that these products are now displacing plans with broader networks on the off-exchange market as well, at least for individuals and small groups.)
McKinsey reports that 92 percent of the population had access to a “narrow network” product in 2014 via the public health insurance marketplaces (HIMs), where “narrow” is defined as a product with under 70 percent of local hospitals in-network. Roughly half of the networks offered on the HIMs were narrow by this definition, and subsequent analyses show the prevalence of narrow networks has only increased over time.
Reducing network size is neither inherently good nor inherently bad. A limited network is not strictly worse for all consumers, both because providers could be selected for their quality and efficiency and because limited networks may be jointly offered with other services that improve quality and/or efficiency of care while facilitating in-network utilization. At a minimum, a greater variety in networks—holding all else constant—expands the set of options available to consumers and can be construed as a marker of greater innovation in the insurance marketplace (Note 1).
We assembled a dataset to explore the link between insurance market concentration and the variety of provider networks available. We use the state-level private health insurance Herfindahl–Hirschman Index (HHI) reported in the 2014 edition of the “Competition in Health Insurance” series issued by the American Medical Association. We study physician rather than hospital networks, as the hospital sector is more consolidated and insurers may therefore have less flexibility in choosing how many and which hospitals to include in a given area. (For example, if there is only one hospital in a state that provides cardiac surgery, insurers will need to include that hospital in-network to satisfy regulatory requirements. Relatedly, hospital systems may insist that insurers take all system members if they wish to include one).
Collecting data on each insurer’s network of physicians is a daunting task. Fortunately, the University of Pennsylvania’s Leonard Davis Institute of Health Economics (LDI) and the Robert Wood Johnson Foundation (RWJF) have jointly undertaken this effort and posted information on the physician networks of every silver plan offered on every HIM in 2014. In total, the data represent 394 provider networks offered by all 267 insurers participating in the marketplaces.
For each physician network in each state, LDI calculated the share of practicing physicians who are in-network (in the “rating areas” within the state where the network is offered). We censor both 5 percent tails of network breadth. For each state, we calculate the standard deviation of network breadth (counting each network once) (Note 2).
The Relationship Between Innovation And Market Concentration On The Public Exchanges
Figure 1 charts the relationship between market concentration (measured by the state private health insurance HHIs) and innovation (measured by the standard deviation of network breadth in each state). We weight each observation by the population of each state, as reflected in the size of the data points on the graph. The correlation coefficient is statistically indistinguishable from zero: there is no evidence of greater innovation in more concentrated markets.
This analysis has an obvious limitation: there is a “mechanical” link between the number of networks in a state and the standard deviation of network breadth. For example, the standard deviation of network breadth is tautologically zero in a state with only one network. States with more insurers will automatically have more networks (so long as insurers don’t exactly copy one another) and there may be an “artificial” relationship between the number of insurers and our measure of innovation.
To address this concern, we performed simulations to determine how the standard deviation of network breadth varies with the number of insurers (Note 3). We then constructed an “adjusted network breadth deviation” by taking the ratio of our original measure from Figure 1 and the expected standard deviation of network breadth given the number of insurers in the state.
Figure 2 presents the relationship between market concentration and the standard deviation of network breadth relative to what would be expected based on the number of insurers in the state. Numbers larger than one indicate that the networks in the state are more dispersed than we would expect based on chance. The preponderance of points less than one reveals that networks in most states are less dispersed than we would expect based on chance. We continue to find a negative rather than a positive association between insurance concentration and innovation—if anything, innovation is lower in states with more consolidated insurance markets (a one-sided t-test yields a p-value of 0.17).
Putting It All Together: What Are The Implications For Insurance Mergers?
The bottom line is there is no evidence of greater product innovation in more concentrated insurance markets. As with most cross-sectional analyses, the data we present is suggestive rather than definitive. There are potential omitted factors that could explain our result.
For example, residents of states with large insurers may have a stronger preference for broad networks, leading insurers to offer a smaller range of plan breadths (i.e., very broad networks) in those states. Of course, it is also possible—and perhaps more plausible—that insurers in more concentrated markets are less motivated to innovate because it isn’t necessary to retain customers.
Given the history of slow innovation in health insurance, one could argue that a change in insurance markets is necessary to stimulate more. To date, there is no evidence that consolidation will be that catalyst.
To be more precise, we are suggesting that a mean preserving spread of network breadth – i.e., an increase in product variety – represents more innovation.
We cannot construct weighted average values because enrollment data have not been released in most states, including all federally facilitated HIMs.
We performed three steps:
- For each state, we constructed a synthetic sample of insurers from the set of insurer x state observations, maintaining the state’s actual number of insurance participants in the 2014 exchange. We then calculated the mean and standard deviation of network breadth for each synthetic sample.
- We repeated step one 10,000 times. This resulted in 10,000 simulations of the mean and standard deviation of network breadth for each state.
- For each state, we retained the 500 simulations with an average network breadth closest to the actual average network breadth of the state. We calculated the mean of the standard deviations for these 500 estimates.