The Old and the Beautiful

How age and gender affect costs and premiums in commercial health care

Doug Norris, Hans Leida, Erica Rode and Travis (T.J.) Gray

We generally consider living a long life an important goal, and it certainly does beat the alternative. Living longer affords us the opportunity to see more, do more and experience more. We get to spend more time with friends and family. Living longer is such a great thing that more and more Americans are doing it—between the years of 1960 and 2010, the life expectancy of the average American male has increased from 66.6 years to 75.7 years, and the life expectancy of the average American female has increased from 73.1 years to 80.8 years. 1

One side effect of getting older is that, as we age, we typically acquire additional acute and chronic medical conditions. Some conditions that likely would have killed us in the past are now treatable (but perhaps not curable). As we age, the prevalence of many common chronic medical conditions increases significantly. Figure 1 shows the prevalence of four common conditions—hypertension, hypercholesterolemia, type 2 diabetes and asthma—by age for a large population of commercially insured adults in the 2014 health insurance market.2

Figure 1: Disease Prevalence by Age, Commercially Insured Adults

Source: Milliman Internal Research Database

For some types of insurance (such as whole life), an enrollee’s policy remains in force for that person’s entire lifetime. Premiums are fixed at issue age, and remain the same for the life of the policy, with the enrollee essentially overfunding the policy in the earlier years to account for higher expected policy benefits in later years. It would be quite difficult to price a similar policy in the commercial health insurance market, considering how the insurable event is defined and how well the benefit amount can be predicted. Although “death” is a well-defined term that has not changed for some time, the concept of “health” changes constantly, as do prices of ever-evolving care practices.

Nevertheless, some health insurance coverages do employ issue age rating, including certain Medicare Supplement products. Many major medical plans offered by employers use tiered rating instead of age rating, with rates varying by family structure but not by age. In contrast, the vast majority of individual major medical health insurance policies historically have been rated on an attained age basis, where the premium rates automatically increase as the insured gets older. Under the Patient Protection and Affordable Care Act (ACA), attained age rating is required for individual major medical insurance, but the degree of rate variation by age that is allowed has been prescribed.

WHAT IS THE NEED FOR AGE/GENDER RESTRICTIONS IN HEALTH CARE PRICING?

Actuarial Standard of Practice (ASOP) No. 12, “Risk Classification,” 3 states that risk characteristics should be related to expected outcomes. Moreover, “rates within a risk classification system would be considered equitable if differences in rates reflect material differences in expected cost for risk characteristics.” To say it another way, to the extent that underlying costs of commercial health care differ by age and gender, it would be considered equitable to offer rates that vary by age and gender.

However, by the same rationale, ASOP No. 12 would consider premium variation related to a member’s underlying health conditions at time of issue to be appropriate. This practice of underwriting is common in commercial large group health insurance (where members’ expected costs are attributed to the group as a whole), and was common in the commercial individual and small group markets prior to the full implementation of the ACA in 2014. One downside of health underwriting is that applicants with high-cost conditions are often unable to afford underwritten coverage (and may not be offered coverage at any rate). Is it possible that rating health coverage by age and gender could lead to a similar issue, where older individuals are unable to afford appropriate coverage?

All types of insurance represent some form of cross-subsidization, in the sense that policyholders with higher than expected costs are generally subsidized by those with lower than expected costs. In the case of health care, those who are healthy help to cover future costs for those who are sick (while those who are healthy insulate against future risk)—at least this is true to the extent that rates are not allowed to vary by health status. Societies often have additional goals for insurance (and regulate it to that effect). One of these goals is to spread risk more broadly and keep coverage affordable for higher-cost cohorts, such as older enrollees. Keeping coverage affordable for those in the higher-cost cohorts can require additional prospective subsidization, either directly from society (through taxation) or from other risk pool participants (in the form of rating restrictions where policies are not rated solely based on their expected costs).

Societies may also decide that certain forms of rating are unacceptable for other reasons that go beyond actuarial concerns. In other words, a society may decide that what is “equitable” from a public policy standpoint is different from what is “equitable” in the narrow sense of ASOP No. 12. Some examples of risk characteristics that are related to expected outcomes, and therefore “equitable” according to ASOP No. 12, but are sometimes viewed as less acceptable from a public policy standpoint, include income level, race, disability, and gender. ASOP No. 12 acknowledges this situation by noting that when establishing risk classes, the actuary should:

  • Comply with applicable law;
  • Consider industry practices for that type of financial or personal security system as known to the actuary; and
  • Consider limitations created by business practices of the financial or personal security system as known to the actuary.4

However, ASOP No. 12 has some words of caution when choosing to impose such additional rating restrictions. “If the variation in expected outcomes within a risk class is too great, adverse selection is likely to occur. To the extent practical, the actuary should establish risk classes such that each has sufficient homogeneity with respect to expected outcomes.”5 It also notes that “adverse selection can potentially threaten the long-term viability of a financial or personal security system.”6 In other words, if we compress the age/gender rating curve too much, adverse selection (among the less-costly younger potential enrollees) represents a danger to the market as a whole, as well as to particular insurers participating in the market. One of the challenges the ACA has presented is that it has been difficult to convince younger and healthier individuals to purchase coverage.7 There is natural tension between the policy goals of making coverage more affordable for older people (with higher average costs) and the goal of encouraging younger people (with lower average costs) to purchase coverage.

In the ACA markets, all carriers currently are required to use the same age slope when pricing, and the risk adjustment mechanism attempts to lessen the impact of selection against any particular carrier. As a result, any selection that occurs is against the market in general (that is, individuals choosing to remain uninsured rather than purchase coverage at rates higher than their actuarial risk would indicate). Although such selection effects can be cause for concern, they are generally smaller than the selection that would occur if only one carrier in the market chose to artificially compress rates while its competitors did not.

WHAT IS THE VARIATION IN UNDERLYING COSTS FOR ENROLLEES OF DIFFERENT AGE AND GENDER?

Prior to the full implementation of the ACA in 2014, it was generally up to the states’ authority to establish and enforce demographic limitations in the individual and small group markets. As of 2011, 43 states either had no restriction on age rating in the individual market, or had restrictions that were more relaxed than the ACA’s 3:1 requirement. Moreover, only 14 states required carriers to rate men and women (of the same age) equally.8

Using data for adults under age 65 from Milliman’s 2017 Health Cost Guidelines—Commercial,9 it is evident that, at most ages, females incur greater health care expenses than males on average (even when maternity costs are excluded). As we see in Figure 2, it is not until the age 60 to 64 cohort that the cost of male enrollees exceeds the cost of female enrollees. However, commercial premiums are not set strictly in accordance with expected variation in claim charges, and many other factors can drive the relativity of premiums by age in a competitive marketplace. For example, how a carrier allocates administrative costs in premium development could vary by age, and market pressures could lead a carrier to modify the premium relativity by age group to be inconsistent with expected claim variation, while still maintaining sufficient premium in total to cover all expected health care and administrative costs.

Figure 2: Allowed Claim Cost Relativities by Age/Gender, Commercially Insured Adults

Source: Milliman 2017 Health Cost Guidelines—Commercial

The shapes of the curves in Figure 2 roughly mimic those previously researched in the 2011 Milliman paper “The Young Are the Restless.”10

The analysis in Figure 2 focused on an estimate of allowed cost differences by age and gender in the commercial market. However, differences in member cost-sharing (by age and gender) also are important considerations when determining expected premium relativities. As a benefit plan’s member cost sharing (primarily, the deductible) increases, the ratio between the costs for the oldest cohorts and the youngest cohorts increases significantly. With the high deductibles common to many ACA plans, the expected average paid claims for many of the youngest male cohorts approaches zero. The maximum out-of-pocket (MOOP) prominently included in ACA plans also serves to increase the ratio of costs between older and younger enrollees. By capping the amount all members can pay, this feature will take effect more in older cohorts (because older individuals have greater allowed costs on average, more older adults will hit their MOOPs).

In total, adult females incur approximately 33 percent more in total health care costs than their adult male counterparts. This difference is exaggerated at younger age levels, and is not entirely accounted for by maternity costs. Indeed, only 6 percent of adult female medical costs are from direct maternity care.11 Overall, if we were to apply a pure actuarial ratio to account for underlying health care cost differences in age and gender, we would expect to see a difference of approximately 6.9:1 between the highest-cost age/gender category (older men and women) in the commercial market and the lowest (younger men). These ratios represent allowed cost differentials, and as noted previously, the ratios would increase as the actuarial value of the plan decreased (going from platinum to bronze, for instance).

If one were to mandate unisex rates, as is done in the ACA rating rules (where males and females of a given age are charged identically), the variation by demographic cell is smaller, as the lowest-cost cohort (males age 18 to 24) and the highest-cost cohort (males age 60 to 64) are each blended with their female counterparts. Combining male and female data by age cohort, we find from the Health Cost Guidelines that the ratio of average allowed claim costs between the highest and lowest cells would be approximately 4.2:1.12

THE ACA’S RATING RESTRICTIONS AND THEIR INTERACTIONS WITH RISK ADJUSTMENT

Effective for individual and small group market plan years beginning in calendar year 2014, two key provisions of the ACA eliminated rating differences based upon gender and mandated a difference between premium rates for the oldest adult members and the youngest adult members to be no greater than 3:1. This represented the widest age-rating variation allowed, and states with more stringent restrictions (such as New York, where age rating is prohibited) continued to enforce their own restrictions.

The risk adjustment transfer program implemented as part of the ACA aims to mitigate the impact of selection effects related to some of these restrictions within the market. However, because transfers among insurers under the risk adjustment program sum to zero across each state and market, risk adjustment cannot protect the market as a whole from adverse selection effects.

Here’s how it works. Under the federal default ACA risk adjustment program, transfers among insurers in each risk pool are based upon an estimate of each insurer’s expected premium requirement without rating restrictions (which—in theory—includes health status and unlimited age and gender variation in costs) and the expected premium the insurers were allowed to charge, given market rules limiting how premium rates can vary. At the heart of the program is a risk adjustment model, which assigns each individual a risk score representing that person’s expected relative cost to insure.

The model assigns older enrollees a larger demographic factor than younger enrollees, and older enrollees also are more likely to have claims documenting conditions that contribute to a higher risk score. Therefore, an insurer can expect to see higher risk scores for older enrollees than for younger enrollees. The question is how closely these differences in risk scores and transfers align with differences in actual underlying costs. In a perfect world, the ACA risk adjustment program would make insurers indifferent as to whether they enroll a disproportionate share of insureds in higher- or lower-cost demographic cells. Figure 3 shows risk scores and paid claim costs (relative to adult males age 18 to 24) for a large database of ACA individual market insureds enrolled in silver plans.13 The scores used are silver plan scores from the 2016 U.S. Department of Health and Human Services Hierarchical Condition Categories (HHS-HCC) model, and exclude the impact of cost-sharing reductions.

Before examining Figure 3, it is important to note that when demographic pricing restrictions are in place (as they are under the ACA), we would expect the observed claim cost distribution to be flatter than in a commercially insured population without overt restrictions. Among younger adults (particularly males), the ones more likely to purchase coverage under demographic restrictions will be the ones who expect to receive greater benefits than the average person of their age. This is somewhat counteracted by the ACA’s Advanced Premium Tax Credit (APTC) feature. These premium subsidies limit premium costs to a fixed percentage of income for lower income households and provide significant incentives for healthier and younger individuals to purchase coverage.

Figure 3: Average Risk Score and Cost Relativity by Age/Gender, 2015 Individual Market

Source: Milliman Internal Research Database

We do see somewhat of a flatter cost curve in Figure 3, which exhibits a unisex ratio of 3.2:1 for the ACA individual market silver members. This is in contrast to Figure 2, which shows a 4.2:1 unisex ratio when considering data from Milliman’s 2017 Health Cost Guidelines—Commercial.

The results in Figure 3 represent the version of the ACA risk adjustment model that will be used for transfers in 2016, although the underlying claim data was incurred in 2015. For both genders, the curve for risk scores is steeper than for claim costs. In composite, these risk scores exhibit a 4.7:1 ratio between members age 60 to 64 and adult members age 18 to 24.14 Although there are other potential confounding variables that may need to be accounted for in order to reach a fully rigorous conclusion, these results suggest that actual costs by age and gender in the individual ACA markets differ from those projected by the 2016 HHS-HCC model. In fact, the silver HHS-HCC model age/gender curves more closely resemble those in the commercial large group market, which is unsurprising given that the federal government has to this point used large group data to develop and calibrate the model. This also suggests that major changes in average risk score by age and gender may result if the government moves forward with its stated intent to recalibrate the model using ACA data for the 2019 benefit year.15

The government already has made changes to the risk adjustment model, is implementing other changes for the 2017 benefit year, and has proposed changes for future benefit years. These adjustments may alter the relative average score by demographic cell.

WHAT’S COMING NEXT?

Proposed in early 2017 as a replacement to the ACA, the American Health Care Act of 2017 (AHCA) included a number of changes to the commercial individual and small group health care markets. One change, with a proposed implementation in the 2018 benefit year, would have relaxed the ACA’s 3:1 age-rating limitation; instead, rates would have been allowed to vary by 5:1 between older and younger individuals (with states still being able to apply more rigorous limitations at their discretion).

Although the initial AHCA was withdrawn from consideration on March 24, 2017, a revised version was passed by the House of Representatives on May 4, 2017 (including the 5:1 rating limitation). Regardless of what happens to this legislation during the next few months, the AHCA still serves as a useful guide to what alternative policies may be on the minds of the new administration and congressional leadership.

In January 2017, a Milliman research report16 discussed the projected impact of switching the age-rating limitation from the current 3:1 level to a wider 5:1 level. This report modeled the overall effect the regulation change would have on the ACA marketplace in isolation (it did not simultaneously model proposed subsidy changes, for instance).

Under the assumptions made in their model, the authors found increased enrollment from younger cohorts (with 2 percent to 4 percent increased enrollment for ages through 49), and small decreases in enrollment for older age groups after implementing a 5:1 allowable rate variation due to age (in isolation). Although premiums increase significantly for older individuals, the authors note that reaction to these increases would be mitigated by both older enrollees’ greater recognition for the need for health insurance and by the rate increase protection built into the current ACA subsidy calculations. The study also explored the potential impact of the likely movement among benefit plan levels that might result from changes in allowable rate variation. Overall, the study projected a per capita increase in premium of about $7 per member per month (PMPM), or about 1 percent, resulting from a relaxed 5:1 age curve, assuming that all other current ACA provisions (such as premium subsidies) remain.

THE MARKET IMPACT OF AGE/GENDER RESTRICTION CHANGES IS NOT OBVIOUS

In the absence of premium subsidies, if a greater proportion of young people enter the market, then the overall risk and claims within the market will likely decrease. This is not necessarily the case when restrictions are placed on the variation in premiums that are allowed due to age and gender.

Restricting the allowable variation in premiums by age causes younger people in the market to subsidize the costs of older enrollees, because the overall cost of the population must still be covered. As premiums increase for young people (because of this subsidization), fewer young people will purchase coverage in the market (and potentially, additional older people will purchase coverage). The rating rules also will likely affect the level of coverage purchased by consumers. For example, if coverage for younger insureds is more expensive because of cross-subsidization, more may choose to purchase bronze plans than otherwise would have. Conversely, loosening age-rating restrictions would tend to have the opposite effect. Consider a hypothetical (and greatly simplified) scenario where the composition of the commercial market in the presence of a 3:1 age-rating restriction is as shown Figure 4, along with the relative costs associated with these cohorts.

Figure 4: Hypothetical Enrollment and Costs
Age Bucket Low Cost Medium Cost High Cost Total
Hypothetical Commercial Market Enrollment
50 and over 24% 12% 4% 40%
49 and younger 36% 18% 6% 60%
Hypothetical Costs Relative to Commercial Market
50 and over 0.78 1.56 6.26 1.56
49 and younger 0.31 0.63 2.51 0.63

Starting from this hypothetical scenario, assume that the allowable age band ratio is expanded to a 5:1 ratio and, as a result, 15 percent of the population (taken solely from the older cohort) decides to not purchase the more expensive coverage that is offered to them. If these individuals’ costs are distributed proportionally across the entire older cohort, then the overall market costs will decrease by 3.6 percent.

However, it is more likely that older individuals will not disenroll in a uniform fashion; instead, this market shift will be disproportionally borne by the healthier older members. If the entire 15 percent reduction in the older population comes from the low-cost cohort, the overall market costs would not decrease, and would instead increase by 1.4 percent.

This example illustrates how the directional result of these changes is not obvious. In reality, there are many other factors that need to be considered when estimating the market impact in response to a change in rating structure. As we discussed, the connection from cost levels to premium levels is not a straightforward one.

Premium subsidies available in the ACA market represent another significant factor that has a profound impact on the market behavior of enrollees, because these subsidies have a direct impact on the net cost to a member when purchasing a product. For those considering a purchase in this market, it is the “net cost” (premium less subsidies received) that drives their decisions. As currently structured, the ACA scales the subsidy based upon the second-lowest-cost silver plan available to the purchaser. The amount of the subsidy is affected by the income level of the potential enrollee, to limit the cost of the second-lowest silver plan to a certain percentage of the enrollee’s income. To the extent that the structure of subsidies is changed in future legislation, the interactions between subsidies and rating restrictions could likewise change significantly, potentially greatly altering consumer purchasing decisions.

WHAT’S THE ANSWER?

For those who have read this far and are hoping for a unified solution to this issue, we have some bad news. Age/gender rating is an area in which actuarial considerations (such as those expressed in ASOP No. 12) are often in direct tension with social or public policy considerations, and for those nonactuarial considerations, there isn’t a result that will please everyone in the market. We can model the effect of purchaser decisions; in general, rational purchasers compare their perceived value with an assessment of expected costs, with varying accuracy, and make their decisions accordingly (also factoring in their risk tolerances). However, purchasers frequently make decisions based on subjective impressions rather than purely on dollars and cents, and without access to perfect information.

As much as it makes things easier to analyze actuarially, changes to market rules do not happen in isolation. Most likely, a change in age/gender rating requirements will be accompanied by other changes to the coverage mandate, and rules related to preexisting condition coverage, subsidies and the like. This is the first in a series of papers written for the Society of Actuaries (SOA) to discuss commercial health care considerations, and these papers should be taken as a whole when considering what may happen in the commercial marketplace going forward.

Doug Norris, FSA, MAAA, Ph.D., is a principal and consulting actuary with the Denver office of Milliman. He focuses on commercial health care considerations, including health care reform strategy and tactics, risk adjustment and predictive modeling, plan pricing, medical underwriting, and interactions with managed Medicaid plans. He is a past chair of the Society of Actuaries’ Predictive Analytics and Futurism section.
Hans Leida, FSA, MAAA, Ph.D., is a principal and consulting actuary with the Minneapolis office of Milliman. He works on a wide variety of projects related to commercial health insurance and Medicare Advantage products. He has particular areas of emphasis in health care reform, predictive modeling, and risk adjustment. Leida is co-author of the textbook Individual Health Insurance and his work was cited in a majority opinion (King v. Burwell) by the Chief Justice of the U.S. Supreme Court.
Erica Rode, ASA, MAAA, Ph.D., is an associate actuary with the Minneapolis office of Milliman. She works in health insurance, with a focus on risk adjustment and predictive modeling of health care costs. She joined the firm in 2013. Prior to joining the firm, Rode taught mathematics while pursuing her Ph.D.
Travis (T.J.) Gray, FSA, MAAA, is a consulting actuary with the Denver office of Milliman. He has experience in product development, pricing, rate filing and exchange strategy considerations in the commercial health insurance market. Gray works with local, regional and nationwide carriers, and is familiar with the intricacies of operating in a wide variety of environments and the diversity of challenges that carriers face.

This paper reflects the opinions of the authors, and not those of Milliman or the SOA. Our information is current as of the time of this writing; however, future events will likely affect the conclusions presented here.

References:

  1. 1. 2011 Risks and Process of Retirement Survey Report: Longevity. Society of Actuaries. June 2012. Accessed April 9, 2017.
  2. 2. Condition flags assigned using Milliman Advanced Risk Adjusters (MARA) clinical grouping algorithm on a sample of approximately 7 million adult lives from Milliman’s internal research database.
  3. 3. “Actuarial Standard of Practice No. 12: Risk Classification (for All Practice Areas).” Actuarial Standards Board. May 1, 2011. Accessed April 9, 2017.
  4. 4. ASOP No. 12, Ibid.
  5. 5. ASOP No. 12, Ibid.
  6. 6. ASOP No. 12, Ibid.
  7. 7. Herman, Bob. “What, Me Buy Insurance? How Slow Uptake by ‘Young Invincibles’ is Driving the ACA’s Exchange Rates Higher.” Modern Healthcare. May 14, 2016. Accessed April 9, 2017.
  8. 8. Norris, Doug and Mary van der Heijde. “The Young Are the Restless: Demographic Changes Under Health Reform.” Milliman. August 30, 2011. Accessed April 9, 2017.
  9. 9. The commercial Health Cost Guidelines are intended to be representative of individuals and groups who are subject to little or no medical underwriting. Claim costs represent nationwide average billed charge levels for a comprehensive major medical plan with nationwide average discount levels applied (and are representative of an allowed cost basis here).
  10. 10. Norris and van der Heijde, Ibid.
  11. 11. We note that it is difficult to fairly allocate delivery costs between the neonate and the mother, and also difficult to quantify indirect costs associated with maternity.
  12. 12. Females alone have an adult allowed cost relativity of approximately 3.0:1, whereas males have an adult allowed cost relativity of the aforementioned 6.9:1.
  13. 13. Based on a sample of approximately 1.5 million members enrolled in a silver plan in the individual market in 2015 from Milliman’s internal research database. For the purposes of this illustration, we did not make any adjustments for differences in area mix by age, or for other potential confounding factors.
  14. 14. Females alone have an adult cost relativity of approximately 4.0:1, and males have an adult cost relativity of 5.8:1.
  15. 15. Patient Protection and Affordable Care Act; HHS Notice of Benefit and Payment Parameters for 2018; Proposed Rule. Federal Register. Vol. 81, Issue 172. September 6, 2016. Accessed April 11, 2017.
  16. 16. Fontana, Joanne, Thomas Murawski, and Sean Hilton. Impact of Changing ACA Age Rating Structure: An Analysis of Premiums and Enrollment by Age Band. Milliman Research Report. January 31, 2017. Accessed April 11, 2017.