David H. Howard, PhD; Brad Herring, PhD; John Graves, PhD; and Erin Trish, PhD
Health policy experts have touted the benefits of integration between insurers and providers for over 40 years,1
but the model has been slow to catch on. Commercial insurers and Blue Cross Blue Shield plans have dominated the health insurance market. The federal government has tried to promote provider–insurer integration via accountable care organizations (ACOs). Although ACOs face limited downside risk, policy makers clearly hope to nudge the healthcare system to be more like Kaiser Permanente,2,3
a fully integrated delivery system that is often held up as a model of efficient, high-quality healthcare delivery.
In this paper, we describe the number and geographic reach of provider-owned insurers in the individual insurance market, and we compare premiums between provider-owned and traditional insurers. The individual market is of special concern for policy makers, given the exits of some prominent insurers from the exchange markets and subsequent reports of rising premiums. Previous analyses have focused on the role of provider-owned insurers in the Medicare Advantage market and the insurance market generally. Twenty percent of Medicare Advantage enrollees were in plans operated by provider-owned insurers in 2016, and 11 of 19 new Medicare Advantage insurers were provider owned.4
In 2013, there were 107 providers selling insurance that covered 18 million enrollees, including Medicaid managed care and Medicare Advantage enrollees.5
Historically, it has been difficult to evaluate the claim that Kaiser Permanente6
and other provider-owned insurers have lower costs. Insurance premiums for employer-based insurance policies are not publicly available, and the lack of standardization of benefits makes it difficult to compare premiums even when they are available. Insurance claims data are an important source of information about healthcare spending generally, but either integrated systems do not release claims data or the data do not have encounter-level transaction prices.
The RAND Health Insurance Experiment remains the most rigorous comparison of costs between provider-owned and traditional insurers. Subjects were randomized to different tiers of fee-for-service health plans or Group Healthcare Cooperative of Puget Sound, a health maintenance organization. Costs were 28% lower in the Group Healthcare arm.7
Frakt et al8
found that provider-owned insurers had higher premiums in the Medicare Advantage market in 2009. However, premiums in the Medicare Advantage market are highly regulated. If a plan submits a bid below the benchmark, the plan receives a rebate that it must use to increase benefits. Half of the plans in the sample used by Frakt et al had $0 premiums.
The degree of standardization in individual policies brought about by the Affordable Care Act (ACA) offers a novel opportunity to compare premiums between provider-owned and traditional insurers. Premiums are publicly available, insurance products are standardized, and premiums depend on consumer characteristics to a limited degree and in ways that are observable. For plans sold on the exchanges, robust risk adjustment weakens the close correspondence between insurers’ profits and enrollee characteristics that exists in most other insurance markets.
Blumberg et al,9
using 2016 data on premiums for plans sold on the exchanges, found that the minimum Silver plan premium in a market was $12 lower if 1 of the insurers participating in the market was provider owned. Their analysis had 1 observation per market.
La Forgia et al10
used 2016 data for plans sold on the exchanges in 30 states to compare premiums between provider-owned insurers and 5 categories of traditional insurers. They found that premiums for Blue Cross Blue Shield plans were similar to those for provider-owned insurers, but premiums for national commercial plans were $260 higher. Our study differs from theirs in a number of respects. We (1) use data covering all 50 states and Washington, DC; (2) compare premiums between provider-owned insurers and all other insurers, which we believe is a more interesting, policy-relevant comparison; (3) include market, versus state, fixed effects; (4) adjust standard errors for clustering of plans by market and by insurer; and (5) use 2017 data.
We used the Robert Wood Johnson Foundation’s 2017 HIX Compare data set (the February 2, 2018, version), compiled from insurers’ mandatory submissions to the CMS Health Insurance Oversight System. The data cover every plan sold on the exchanges, including plans sold on federally facilitated, state partnership, federally supported, and state-based Marketplaces, as well as off-exchange, ACA-qualified plans. The data cover every plan’s characteristics but do not include information on the number and characteristics of enrollees. Observations are at the plan-market level. Insurers sell multiple plans across and within markets.
We classified an insurer as provider owned if the insurer was owned by or shared a common owner with a hospital system or multispecialty physician clinic. (We hypothesized that insurers that owned only primary care clinics would not have sufficient leverage or control over the delivery of healthcare to substantially reduce costs.)
We identified provider-owned insurers by examining the About Us section of each insurer’s website. We looked for statements indicating that (1) the insurer was owned by a health system or (2) the insurer owned and operated hospitals and clinics. We then examined network directories to determine if the network included hospitals or physician clinics affiliated with the insurer. A research assistant classified each insurer, and then one of the authors reviewed each classification, using the same methods, to determine its accuracy. eAppendix Table 1
(eAppendix available at ajmc.com
) contains a list of these insurers and their classifications.
We classified observations at the insurer rather than plan or product level. For example, UnitedHealthcare is a traditional insurer but had a recently closed subsidiary, Harken Health, that operated its own physician clinics. We classified both UnitedHealthcare and Harken Health as traditional insurers because only a small share of the United plans were sold under the Harken brand and because United may have used the profits from Harken to cross-subsidize its other products or vice versa.
Corporate names are used inconsistently in the HIX Compare data (for example, some plans sold by Harvard Pilgrim are associated with “Harvard Pilgrim”; others, with “Harvard Pilgrim Health Care”). We therefore recoded the carrier variable in the HIX Compare data so that all plans sold by an insurer had a common code. This step was necessary to correctly adjust standard errors for clustering of plans by insurer. We grouped all of the Blue Cross Blue Shield plans owned by Anthem, Inc, together. Plans sold by non-Anthem Blue Cross Blue Shield insurers were assigned distinct codes.
We assessed the presence and number of provider-owned and traditional insurers in each market (as defined by exchange rating areas). We merged the HIX Compare data with the Health Resources and Services Administration’s Area Resource File to determine the characteristics of the markets in which provider-owned insurers operate. We measured the association between market characteristics and the presence of at least 1 provider-owned insurer in the market using logistic regression.
We estimated differences in premiums between plans sold by provider-owned and traditional insurers using a correlated random-effects model.11,12
We restricted attention to each insurer’s lowest-cost Silver plan for an individual aged 50 years in each market. There is 1 observation per insurer per market.
Similar to a fixed-effects model, the correlated random-effects model estimates within-market differences in premiums. However, similar to a mixed-effects model, it also permits adjustment of standard errors for clustering at the market and insurer levels. Plans are nested in markets and in insurers, but neither markets nor insurers are nested in each other (they are “crossed” effects). A traditional fixed-effects model cannot accommodate the complex error structure. The model is of the following form:
yijm = β1POIj + β2POIm + αj + μm + εijm,
indexes plans, j
represents insurers, and m
means markets. POI
indicates whether the insurer is provider owned. Controlling for the market-level average of the share of provider-owned insurers, POI
, ensures that the coefficient on the provider-owned insurer indicator, β1
, is equivalent to an estimate from a fixed-effects model. The variables αj
represent insurer and market random effects, respectively.
Availability of Provider-Owned Insurance Plans
There were 149 insurers that sold ACA-compliant plans in 2017, of which 51 were provider owned based on our criteria. Provider-owned insurers operated in 208 of the 503 markets. Figure 1
displays the number of provider-owned insurers in each market. Provider-owned insurers operate in geographically diverse areas in the United States, including the upper Midwest, the coastal West, and the mid-Atlantic. Five providers (CHRISTUS Health, Covenant Health, Memorial Hermann, Prominence Health, and Baylor Scott & White Health) sell insurance in Texas, a state not normally associated with delivery system innovation.
Between 2016 and 2017, 4 provider-owned insurers exited the ACA-compliant market entirely, 5 entered (for a net gain of 1), and 46 participated throughout both years. In the same time frame, 9 traditional insurers exited, 3 entered (for a net loss of 6), and 95 participated throughout both years. The mean number of insurers (of both types) per rating area in the ACA-compliant market decreased from 6.1 in 2016 to 4.2 in 2017. In most markets in which provider-owned insurers sold plans, there were usually just 1 or 2 provider-owned insurers per market.
displays the change in the availability of plans sold by provider-owned and traditional insurers by population. The x-axis represents the number of carriers per market of each type operating in 2016, and the y-axis represents the number per market operating in 2017. For example, the large dot at the x = 1, y = 1 position in the “Provider-Owned Insurers” panel indicates that there are more than 50 million individuals who live in markets where exactly 1 provider-owned insurer sold plans in 2016 and 2017.
Large regions of the United States have no provider-owned insurers, but many states are sparsely populated. About 62% of US residents (more than 170 million people) live in a rating area in which a provider-owned insurer sells plans.
Six percent of the population live in areas in which the number of provider-owned insurers increased between 2016 and 2017 (represented by the lighter circles in Figure 2). Eleven percent of the population live in areas in which the number of provider-owned insurers decreased (the darker circles in Figure 2), and 51% live in areas in which the number of insurers stayed the same (the medium circles in Figure 2). The remaining 32% of the population live in areas in which no provider-owned plan was available in either 2016 or 2017.
One percent of the US population live in areas in which the number of traditional insurers increased, 56% live in areas in which the number of insurers decreased, and 43% live in areas in which the number of insurers stayed the same.
summarizes market characteristics and shows the association between market characteristics and the presence of a provider-owned insurer in the 503 rating areas. Only per capita income was significantly associated with the presence of a provider-owned insurer in a market. An increase in income of 1 SD ($8700) is associated with a 14-percentage-point increase in the likelihood that there is a provider-owned insurer in a market.
presents summary statistics at the insurer-market level for each insurer’s least expensive Silver plan in the markets in which they offer at least 1 plan. Provider-owned insurers sell 30% of the plans in the sample. Seventy-eight percent of plans were classified as health maintenance organizations or exclusive provider organizations. Thirty-five percent of all plans in the sample were available in the off-exchange market only.
Provider-owned insurers sold 1396 Silver plans in total, of which 156 (10%) were available off-exchange only. Traditional insurers that operated in the same markets as provider-owned plans sold 3868 plans in those markets, of which 502 (14%) were available in the off-exchange market only. On average, provider-owned insurers sell on-exchange plans in 80% of the markets in which they operate. Traditional insurers sell on-exchange plans in 90% of the markets in which they operate (considering only markets in which provider-owned insurers sell plans, to facilitate comparison).
shows estimates of differences in premiums between provider-owned and traditional insurers’ plans from 6 mixed-effects logistic regression models estimated on the sample described in Table 2 (n = 970). Full estimates are presented in eAppendix Table 2
. The first model includes a random effect for market but not insurer. The second, our preferred specification, adds a random effect for insurer (insurers sell multiple plans across markets). The model indicates that premiums for plans sold by provider-owned insurers are lower, on average, but not statistically different from premiums for plans sold by traditional insurers.
The third model includes plans sold by multistate insurers (n = 533), and the fourth includes plans sold by all insurers other than Kaiser Permanente (n = 923). The fifth and sixth add controls for plan type and the cost-sharing parameters described in Table 2. There are 190 variables in the HIX Compare data that describe different plans’ cost-sharing requirements. We did not control for these in the baseline model because benefit levels in the exchange are tightly regulated. Silver plans must have an actuarial value of at least 70%. A plan that requires lower cost sharing along one dimension will have higher cost-sharing in another. We nevertheless estimated a model in which we control for several especially salient cost-sharing parameters, such as whether the plan has a deductible for in-network care, the size of the in-network deductible, and the maximum out-of-pocket payment for in-network care.6,13
The results are consistent across models: Point estimates are negative (indicating that provider-owned plans have lower premiums), and the CIs are wide. Results are similar when we re-estimate the model on plans in different metal tiers (Bronze and Gold), using premiums for different policy holder types (individuals aged 27 years and couples both aged 30 years), and on plans sold in the exchange market only.
When we re-estimate the second model (our preferred specification with market- and insurer-level random effects) using 2016 data, the coefficient on provider-owned plans is –15 (95% CI, –39 to 9). Premiums for provider-owned plans were lower but not statistically different from premiums for plans sold by traditional insurers. If we omit carrier-level random effects, the coefficient is –18 (95% CI, –28 to –8).
Provider-owned insurers play an important but underappreciated role in the individual insurance market. About one-third of the insurers who sell plans in the ACA-qualified individual market are provider owned. Two-thirds of the population live in areas where at least 1 provider-owned insurer sells plans. Provider-owned insurers are located on the West Coast and in the upper Midwest, some Western states, and Texas.
We find that premiums are similar between provider-owned and traditional insurers. The exchange market has a number of features, including standardization of actuarial values and benefits and market-level risk adjustment, that facilitate direct comparisons of premiums between insurers. Although risk adjustment is imperfect, risk adjustment provisions of the ACA entail substantial transfers to and between plans. Risk adjustment significantly diminishes selection-driven differences in plans’ profits.14,15
As a result, plans ought to set premiums based on the average level of risk in the market rather than the level of risk among likely enrollees. To the extent that there is residual selection, the direction of the bias is unclear. Healthier enrollees are probably more willing to trade off lower premiums for stricter limits on provider choice in provider-owned plans, but participants in the associated health system form a natural customer base for provider-owned insurers.16
Unlike La Forgia et al,10
we use data covering the entire United States, our data are recent (from 2017), and we adjust standard errors for clustering by insurer. Adjusting standard errors for clustering has a large effect on the size of CIs.
Our results may be biased by differences in plan quality between provider-owned and traditional insurers. Johnson et al17
and Frakt et al8
find that provider-owned insurers in the Medicare Advantage market score higher on a composite quality measure.
Providers that want to sell insurance face a number of obstacles.18
They must employ a sufficiently large number of physicians or contract with unaffiliated providers, pay an external vendor for claims processing, and attract enrollees from outside their patient population. They must also maintain their pre-existing relations with commercial insurers. Unlike Kaiser Permanente, most providers that own insurers treat patients with other sources of insurance coverage, including traditional commercial plans.
Despite these challenges, a number of providers have successfully entered the individual insurance market and are able to sell competitively priced plans, according to our data. One-third of the insurers selling ACA-compliant plans are provider owned. Lacking data on enrollment, we cannot determine if the number of enrollees in provider-owned insurers has increased over time. However, the ability of provider-owned insurers to wring cost savings out of the system may improve as they gain experience and market share.