BACKGROUND: There is significant concern about the financial burdens of new insurance plan designs on families, particularly families with children and youth with special health care needs (CYSHCN). With value-based insurance design (VBID) plans growing in popularity, this study examined the implications of selected VBID cost-sharing features on children.
METHODS: We studied children’s health care spending patterns in 2 data sets that include high deductible and narrow network plans among others. Medical Expenditure Panel Survey data from 2007 to 2013 on 22 392 children were used to study out-of-pocket (OOP) costs according to CYSHCN, family income, and spending. MarketScan large employer insurance claims data from 2007 to 2014 (N = 4 263 452) were used to test for differences in mean total payments and OOP costs across various health plans.
RESULTS: Across the data sets, we found that existing health plans place significant financial burdens on families, particularly lower income households and families with CYSHCN; individuals among the top 10% of OOP spending averaged more than $2000 per child. Although high deductible and consumer-driven plans impose substantial OOP costs on children, they do not significantly reduce spending, whereas health maintenance organizations that use network restrictions and tighter management do.
CONCLUSIONS: Our results do not support the conclusion that high cost-sharing features that are common in VBID plans will significantly reduce health care spending on children.
- CDHP —
- consumer-driven health plan
- CYSHCN —
- children and youth with special heath care needs
- EPO —
- exclusive provider organization
- HDHP —
- high deductible health plan
- HMO —
- health maintenance organization
- MEPS —
- Medical Expenditure Panel Survey
- OOP —
- POS —
- VBID —
- value-based insurance design
Insured Americans spend a significant portion of their income on out-of-pocket (OOP) health care expenses such as deductibles, copays, and coinsurance, or for services not covered by their health insurance benefit package.1 The term “underinsurance” has been used to describe the situation in which health insurance imposes significant financial burdens on, and jeopardizes the provision of adequate health care to, individuals and families.2,3 Schoen et al4 defined underinsurance specifically as 1 of 3 indicators of financial exposure relative to income: (1) OOP medical expenses for care amounting to ≥10% of income; (2) among low-income adults (below 200% of the federal poverty level), medical expenses amounting to at least 5% of income; or (3) deductibles ≥5% of income. Children are particularly vulnerable to underinsurance: the number of children who are underinsured (22.7% of all continuously covered children)5 exceeds the number who are uninsured (∼6% in 2009, fewer today).6
Financial burdens are of particular concern for families with 1 or more children and youth with special health care needs (CYSHCN), who comprise ∼11.2 million or 15% of all children aged 0 to 18 years.7 By definition, CYSHCN use more health services than other children8 and are thus particularly at risk for underinsurance. Although much of the attention on CYSHCN has focused on children with Medicaid or Children’s Health Insurance Program coverage, the majority of CYSHCN have private, employer-sponsored health insurance, and a growing number of these private health plans are associated with underinsurance.9 The 2009/2010 National Survey of Children with Special Health Care Needs found that >34% of CYSHCN are underinsured and that the prevalence of underinsurance is greater in some subgroups (children with complex health needs, older children, and Hispanic children).7 Families raising CYSHCN who have income that is marginally higher than their state’s income eligibility limit for Medicaid are particularly vulnerable for underinsurance and financial hardship. More than 20% of US families of CYSHCN report having financial problems as a result of their child’s health condition.10
There is currently significant interest in value-based insurance design (VBID), a mechanism that promotes improved health care decision-making by consumers and providers by linking levels of enrollee copayments to the clinical value of the services provided.11 There are multiple dimensions of VBID, but 4 hallmark features are the promotion of prevention over curative treatment (through free or low-cost preventive care), increased consumer information (in support of better informed choices), increased consumer cost sharing (so that families will have stronger incentives to choose lower cost providers), and a restricted provider panel or at least steeply tiered provider networks and benefit designs (so that consumers are incentivized to choose among only low-cost and/or presumably high-quality providers and services).12 There is wide variation in the way that VBID programs have been implemented across the country.13 In some instances, VBID may impose sizeable OOP burdens on consumers if they do not choose the lower cost or preferred providers. These additional OOP payments can lead to underinsurance and, ultimately, family financial hardship. Whether this predictable risk of increased financial burdens is offset by meaningful cost savings from VBID is largely unknown for children, especially for CYSHCN.
The present article examines the patterns of OOP spending among the privately insured in 2 data sets to improve our understanding of the likely impact of VBID on children overall and on CYSHCN in particular. Our goal was to answer 3 questions of interest to policy makers related to VBID. First, how significant are the financial burdens in high-deductible plans and narrow network plans that share some features with VBID plans? Second, what is the relationship between these burdens and other dimensions, such as household income and age, and for families with and without CYSHCN? Third, is there evidence that plans with higher cost-sharing burdens actually save money overall by reducing total payments for health care on children?
Study Population and Settings
Data from 2 secondary sources were used for this study. We first used the Agency for Healthcare Research and Quality Medical Expenditure Panel Survey (MEPS) data from 2007 to 2013.14 The MEPS is a nationally representative subsample of the families who were involved in the previous year’s National Health Interview Survey. Each family is followed up for 2 calendar years, thus generating an overlapping panel structure for the MEPS. A significant advantage of the MEPS insurance component data over other claims data sets is the availability of income measures. Because the primary focus of VBID to date has been on the privately insured population, we focused our analysis on the subset of children and youths aged 1 to 20 years with private health insurance. Children who were simultaneously under both private and public health insurance, including Medicaid, TRICARE, and the State Children’s Health Insurance Program, were dropped from the sample for clarity. Also excluded were children who were not in the sample for an entire year; this method was used because the expenditure estimates in MEPS are annual figures, and thus it was important to ensure that differences in expenditures were not due to differences in length of time in the sample. Finally, to capture families with private insurance over the range in which OOP burdens are of greatest relevance, the MEPS sample was restricted to children with family income between $10 000 and $150 000; family income comprised annual earnings, business and farm gains and losses, interest and dividends, employment-related compensation, private pensions, private cash transfers, welfare payments, and estate gains or losses but excluded tax refunds and capital gains.15
Our second data source was the IBM Watson/Truven Analytics MarketScan Private Claims and Encounters data, over a 1-year longer sample period, from 2007 to 2014. This insurance claims data set contains information on >300 million person-years of utilization, cost, and eligibility records, and it has been widely used for statistical analysis (including by the US Department of Health and Human Services for the Health Insurance Marketplace model calibration) because it encompasses 40 to 50 million privately insured individuals each year, which is more than one quarter of the privately insured US population aged <65 years.16 Because we were interested in being able to uniquely assign each person to a health plan, we included only those individuals who could be assigned to any of the 73 large employers (averaging >11 000 children in each plan-year) that were present in the MarketScan data for at least 3 consecutive years. Altogether, data were included on 4 263 452 children aged 1 to 20 years. We eliminated people not eligible for 12 complete calendar months in a given year (including most deaths and births), required 3 years of eligibility, and included people who changed plans only on January 1, but dropped people who switch between ≥2 health plans within a calendar year. We included both single and family coverage enrollees, and children aged 1 to 20 years. Newborns (age 0) were excluded because most are eligible for coverage for <12 full months, and cost patterns were distinct for this group, with choices tied to the mother’s health care needs. As with the MEPS data, we used children’s ages as of the end of the calendar year.
The primary health care measures of interest in both samples were the total spending and OOP spending on health care for children. Because individuals in health plans that are considered VBID were not identifiable in our MEPS or MarketScan samples, we explored cost distributions in the existing health coverage plans, many of which also contained high deductibles and copayments. Spending in VBID could be higher or lower than in plans such as high deductible health plans (HDHPs) or consumer-driven health plans (CDHPs), but it should be informative to look at spending patterns in these 2 newer plan innovations. Moreover, many CDHPs and HDHPs, as well as point-of-service (POS) plans, contain tiered networks, with some providers given lower cost sharing. Although the criteria used to select preferred providers and services in these plans may differ from VBID, we do not see why fundamentally the pattern of spending, as distinct from the quality achieved, in VBID plans need differ from CDHPs and HDHPs. The MEPS data were used to study burdens across income levels, and the larger MarketScan data were used to examine spending according to age and across plan types that varied in their use of cost sharing and provider breadth of choice.
An important feature of US health plans is that they tend to attract people with varying health status, and this variation can create differences in payments that dominate the actual incentives of the plan type itself. To measure and control for this selection on health status into different plan types, the Verisk Health (Waltham, MA) DxCG Risk Solutions (Version 4.21) prospective risk scores were used. These risk scores are an enhancement on the hierarchical condition category risk adjustment classification system used for both the Medicare Advantage risk adjustment and the Patient Protection and Affordable Care Act’s health insurance exchanges. The scores, used in numerous earlier studies, predict average total spending by using age, sex, and diagnoses and were normalized by dividing by a constant so that the mean risk score was 1 in the current sample of children. Because the relative risk scores control for observable health status as measured by diagnoses, we attributed remaining variation in spending to health plan design effects.
For these data, we did not have available the plan premiums or cost-sharing features (eg, deductibles, copayments, OOP maximums), and we therefore simply reported observed health plan OOP cost and inferred the effects of these cost-sharing features. To remove inflation, all health care costs were inflated into 2014 dollars by using the Personal Health Care Index,17 and income measures were inflated into 2014 dollars by using the Consumer Price Index.18 No adjustments or exclusions were made for high-cost outliers in either sample.
The statistical methods differed slightly for the MEPS and MarketScan data, reflecting sample sizes and how the data were generated. For the MEPS data, to create smooth patterns of health care spending across income levels, Stata’s local second-order polynomial regression method was used with the default Epanechnikov kernel function and rule-of-thumb bandwidth estimator.19 This estimation method selects a neighborhood around each family income level and runs a second-order regression of spending on family income over this range. Means were calculated for the grand mean of the sample (including zero spending observations), for the top quartile of spending, and for the top decile of the sample. Person weights from MEPS were used to reflect national averages, but scaling multipliers were used to give each cross-section an equal weight in the aggregate sample.
Because the MarketScan data set was much larger than the MEPS data, we did not use kernel estimation to smooth the MarketScan graphs. Results are presented by using the simple average for each 1-year age group according to plan type, with and without risk adjusting spending within each age. To test for differences across plan types, plan type averages were calculated by weighting the spending of each age (1–20 years) in each plan by the full sample shares of enrollees at each age to remove variation across plans due to age. We also calculated risk-adjusted averages, which deflated the age-adjusted spending by the average risk score for that group. These risk scores adjust for sex and diagnoses across plan types. We used 100 bootstrapped values of each mean according to plan type to calculate standard errors and confidence intervals.
The MEPS data contain a binary variable flagging CYSHCN for children aged <18 years, which we used on this subsample of the total study sample. The MEPS used a series of questions and determined whether each child aged 0 to 17 years had physical limitations or required more health care services than average; these findings were used on this age subsample to identify CYSHCN.15 We did not create a similar CYSHCN variable using the MarketScan data.
Using the MEPS data, we focused on results from our 7-year panel of children with private insurance, enrolled for all 12 months. The final sample size was 22 392 children aged 1 to 20 years, inclusive of those with and without special needs. On average, after an initial decline as household income increased from $10 000 to $30 000, total spending and OOP spending per child increased modestly across income groups (Fig 1A) before leveling off at incomes above $70 000. Figure 1B highlights that although average OOP costs were low, on average, in the full sample, the top quartile average OOP spending was substantial at approximately $1500. Among the top 10% of spenders on child health care, the average OOP expense exceeded $2000 across family incomes. Furthermore, spending for the top 10% and 25% was slightly more evenly distributed across income groups. Thus, although the absolute burdens are similar, the relative financial burden is higher for families earning less income. We quantified the financial burden of child health care spending as the ratio of OOP spending to family income (Fig 2). The financial burden of child health services was greatest for those in the low and lower–middle income groups.
The sample was further split by grouping individuals according to financial burden. Among individuals in the top 10% of financial burden, the lowest income families had the greatest proportion of OOP spending, averaging as much as 10% of family income. Based on the criteria for underinsurance discussed in the present article’s introduction, this magnitude of financial burden would qualify as underinsurance.
CYSHCN were similar to other children in terms of family size, composition, and income. The main difference between the 2 groups was in OOP health care spending, with spending on CYSHCN >3 times that for other children (Table 1).
We then used the much larger sample of privately insured MarketScan enrollees to examine spending patterns according to 1-year age increments for children aged 1 to 20 years in 7 different plan types. The results shown in the 4 panels of Fig 3 reveal several patterns. Fig 3A shows that total spending declined steadily as children aged from 1 to 4 years but then increased modestly every year in all plans except health maintenance organization (HMO) plans until age 20 years. Total nonrisk-adjusted spending on children in HMOs continued to decrease through age 9 years, and from age 6 to 20 years, total spending was noticeably lower compared with children insured in other plans.
Fig 3B reveals that although total spending patterns are similar, there are sharp differences across health plan types in their OOP spending burdens, with HMOs and exclusive provider organization (EPOs) having the lowest OOP, and CDHPs and HDHPs having the highest average OOP spending. The average OOP of preferred provider organizations and POS plans, both of which use network tiering, were not meaningfully different from the traditional comprehensive health plan, which imposes modest cost sharing and little care management.
Figure 3C shows that there are meaningful differences of 10% to 20% in the average risk scores within an age group. If this variation is not taken into account, plans that enroll sicker children (ie, children with higher mean risk scores) will then appear to incur higher health care costs. The age pattern evident in the first panel is also evident in the average risk scores.
Adjustment of health care spending for risk scores reveals a different pattern of health care spending (Fig 3D). The greatest but expected difference is that risk adjustment substantially reduced the age-dependent variability in total spending. Notably, even after risk adjustment, HMO plans still exhibit the lowest child health care spending. After controlling for health status using risk adjustment, there is no apparent savings in total spending across the age spectrum from 1 to 20 years for children in HDHPs and CDHPs relative to preferred provider organizations and POS plans.
The results shown graphically according to age group in Fig 3 are summarized for all children in Fig 4. Weighted averages across age according to plan type are shown for total spending and for OOP spending. The bottom panel in each section illustrates the effect of risk adjustment for diagnoses on total and OOP spending according to health plan type. Average OOP spending was not meaningfully affected by risk adjustment and was highest in HDHPs followed by CDHPs. HMO and EPO plans had the lowest OOP spending.
HMO plans exhibited the lowest total spending in both nonrisk-adjusted and risk-adjusted analyses. HDHPs and CDHPs had total spending slightly lower than the average in both analyses but shifted burden to families through high OOP expenses. Risk adjustment of total spending removed much of the apparent differences across plan types. Significantly, most of the apparent reduction in total spending from imposing cost sharing disappeared when samples were risk- adjusted to control for the age and diagnoses present on health insurance claims. Narrow panel health plans such as HMOs were more successful at reducing spending than plans with higher deductibles and cost sharing. We have no explanation for why the narrow panel EPO plans are not more similar to HMOs in this sample.
The present article focused on the distributions of total and OOP health care spending for children in various existing health plans. The goal was to understand their potential implications for VBID plans, even though we could not identify enrollees specifically in VBID plans in these data sets. Our results showed that differences across plans in OOP costs were sizeable and that differences in total spending across health plans which differ dramatically in their OOP cost burdens were modest.
Our results show that total spending per child broadly increased across family income, ranging from $10 000 to $50 000, before remaining nearly constant at higher income levels. Private health plans imposed significant risk of underinsurance on many families with less than $50 000 in household income, particularly among households with CYSHCN. We also found that although OOP differences across plan types were large, differences in total spending were modest. This finding is consistent with extensive literature which finds that although health care spending overall responds to cost sharing, spending on children is much less responsive than spending on adults.20,21
These findings suggest that the increased reliance on VBID plans will impose significant further increases in OOP spending but may have little effect overall on controlling spending on children. Our analysis suggests that selecting narrow networks of providers has better promise for controlling costs among children than raising cost sharing, but given the important role and small numbers of highly specialized children’s hospitals, the VBID approach may create incentives for plans to avoid or underpay these crucial facilities from their favored networks because of their higher cost.
An important limitation is that although we wish to make statements about the effects of VBID plans, our sample does not actually allow us to distinguish and separately analyze these new forms of insurance contracts. Instead, we are only able to comment on the effects of narrow networks, as reflected in HMO and EPO health plans, POS plans (which use network tiers), and on high deductible health plans, which include both CDHPs (which use health savings accounts) and HDHPs (which according to MarketScan definitions, use only health reimbursement accounts).
The MEPS data do not contain a sufficient number of observations to make precise estimates of changes over time or across health plans, and thus we present only the pooled results. Because average plan features were changing, and the data reflect the time period during which the Affordable Care Act was first implemented, we did not attempt to study changes over time in our 2 samples.
Our analysis of the MarketScan data included only claims for large employers with relatively generous coverage and is not generalizable to all children or even all privately insured children. Future research could examine differences in rates of spending among enrollees covered by such large employers and those in smaller firms, or in the Marketplace insurance exchange plans, as well as differences between privately and publicly insured children.
In 2007, 19% of the US population had trouble paying medical bills, up from 15% in 2003.1,22,23 The implications of underinsurance are cumulative and may affect the well-being of CYSHCN and their families over the life course. Medical debt has been shown to be a strong predictor of delayed or relinquished health care.22,24 For CYSHCN, we have documented that cost burdens were unsurprisingly even greater. Previous studies linked underinsurance to other diminished outcomes in domains such as satisfaction with care, family partnership in medical decision-making, access to a medical home, accessible community-based service systems, and availability of services for transition from pediatric to adult care.2 Parents in lower income families raising CYSHCN were more likely to reduce or cease employment to provide needed care for their child9 and thus incur greater medical debt.
The finding of significant financial burdens of existing health plans on families with children, and in particularly on families with CYSHCN, is confirmed in the unpublished work by Angela Walter and colleagues. However, our hope is that the evidence presented here will help refine understanding of the implications of high cost sharing and narrow networks of providers such as are favored with VBID plans. Cost-sharing burdens among the privately insured are growing over time,25 and we present evidence that OOP burdens for many families with CYSHCN are already significant. We also found no evidence of significant cost saving overall on children despite these high OOP costs. Our analysis suggests that narrow networks, and not higher cost sharing, was more effective at controlling costs. Planning for equitable and efficient access to care will require careful planning going forward with VBID plans for children.
The authors gratefully acknowledge the helpful review of Angela Wangari Walter, PhD, and Wenjia Zhu, MA.
- Accepted January 12, 2017.
- Address correspondence to Randall P. Ellis, MS, PhD, Department of Economics, Boston University, 270 Bay State Rd, Boston, MA 02215. E-mail:
FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.
FUNDING: Drs Ellis and Bachman were supported by the Catalyst Center, the National Center for Health Insurance and Financing for Children and Youth with Special Health Care Needs, which is supported by the Health Resources and Services Administration of the US Department of Health and Human Services under grant number U41MC13618 ($473 000). This information or content and conclusions are those of the Catalyst Center staff and should not be construed as the official position or policy of, nor should any endorsements be inferred by, the Health Resources and Services Administration, the US Department of Health and Human Services, or the US Government. Mr Tan received no funding for this work.
POTENTIAL CONFLICT OF INTEREST: The authors have indicated they have no potential conflicts of interest to disclose.
- Gabel JR,
- McDevitt R,
- Lore R,
- Pickreign J,
- Whitmore H,
- Ding T
- Schoen C,
- Collins SR,
- Kriss JL,
- Doty MM
- National Survey of Children with Special Health Care Needs
- McPherson M,
- Arango P,
- Fox H, et al
- Data Resource Center for Child and Adolescent Health
- Maciejewski ML,
- Wansink D,
- Lindquist JH,
- Parker JC,
- Farley JF
- Choudry NK,
- Fischer MA,
- Smith BF, et al
- Agency for Healthcare Research and Quality
- Agency for Healthcare Research and Quality
- The Henry J. Kaiser Family Foundation Health Insurance Coverage of the Total Population
- Agency for Healthcare Research and Quality
- Bureau of Labor Statistics
- Leibowitz A,
- Manning WG Jr,
- Keeler EB,
- Duan N,
- Lohr KN,
- Newhouse JP
- Ellis RP,
- Zhu W
- Collins SR,
- Rasmussen PW,
- Beutel S,
- Doty MM
- Copyright © 2017 by the American Academy of Pediatrics