Abstract
Objective:
Estimate balance-billing for out-of-network behavioral health claims and describe subscriber characteristics associated with higher balance-billing.
Methods:
Using a national managed behavioral health organization’s employer-sponsored behavioral health insurance claims (2011–2014) (n=196,034 family-years with out-of-network behavioral health utilization claims), we calculated inflation-adjusted annual balance-billing amounts. Balance-billing equals the “submitted amount” (the amount charged by the provider), minus the “allowed amount” (the amount the insurer agreed to pay plus patient cost-sharing) and any discounts offered by the provider. Using a two-part model to account for family-years without balance-billing, for family-years linked to sociodemographic data (N=68,659), regressions modeled balance-billing as a function of plan and provider supply, subscriber and family-year, and employer characteristics.
Results:
Among the 50% of family-years that had balance-billing, the mean balance-billing value was $861±$3500 (Median: $175, 90th percentile: $1684). The adjusted analysis found balance-billing was higher for carve-out ($523 95%CI: $340, $705) and HMO enrollees ($156 95%CI: $75,$237) as well as for subscribers with a bachelor’s degree (between $172 95%CI: $228, $116 and $224 95%CI: $284,$163, compared to Associate’s degree and high school or lower, respectively) and who were 45–54 years old (between $57 95%CI: $103,$10 and $290 95%CI:$398,$183, compared to 35–44 years and 18–24 years, respectively). Balance-billing was lower for families in states with more in-network providers per-capita in the state ($8 95%CI: 10,−5).
Conclusions:
When families submitting out-of-network behavioral health claims face balance-billing, the resulting charges may be burdensome for many families. Expanded networks for behavioral health may improve access to behavioral healthcare services.
INTRODUCTION
Balance-billing occurs when patients receive a bill to cover the difference between what the provider charged and what private insurers and patients are contractually obliged to pay.1 This practice has been documented for medical care.2,3,4–6 The practice of balance-billing can be financially devastating to patients, however, little is known about balance-billing in the context of behavioral health (e.g. specialty mental health services, addiction treatment, etc.).
The literature points to two scenarios involving behavioral health balance-billing. Recent evidence identifies high rates of inaccurate network status listings on insurance physician directory pages,7 which may lead to patients unknowingly getting care from out-of-network providers. Alternatively, since narrow networks (less than 25% of providers in a market) are more common for behavioral health than for primary care,8 patients are more likely to go out-of-network when using behavioral health providers than when using general healthcare providers9 and thus more likely to have balance-billing.
Although estimates of behavioral health balance-billing have not been published, existing evidence points to high out-of-network cost-sharing for behavioral health services. Researchers observe that patients with chronic mental health conditions had significantly higher average out-of-network cost-sharing than patients with congestive health failure and patients with diabetes.10 Other work observes that prices (what insurers pay) and cost-sharing (e.g., patient copayments, coinsurance and deductibles) for out-of-network psychotherapy11 and other behavioral health services12 have increased over this period, even as cost-sharing for many other out-of-network services decrease.13 The present study builds on this literature by adding calculations of out-of-network behavioral health balance-billing amounts --which is one component of total out-of-network cost-sharing-- from a national sample of commercial claims.
Here, balance-billing is calculated as the difference between the provider charge amount and the amount allowed by the insurer plus patient cost-sharing. This provides a measure of what patients were potentially charged as balance-billing. Our analyses also contribute to the literature by aggregating claims into membership units (called “families”) and examining differences in behavioral health balance-billing levels across family characteristics. We answer two research questions: 1. How were balance-billing levels for behavioral health care distributed among a commercially insured sample? 2. Do subscribers with greater financial resources and education have higher levels of balance-billing compared to other family-years, controlling for plan characteristics and provider supply?
METHODS
Data
We leveraged a database linking employer-sponsored behavioral health claims and enrollment data from 2011–2014 as well as plan, employer, and provider network information, provided by a national managed behavioral health organization (representing all 50 U.S. states and territories)14,15 with self-reported subscriber demographic data provided by Optum Insight® and state inpatient supply from the Area Health Resource File provided by the Health Resources and Services Administration.16
The claims data included records from carve-in plans (i.e., plans that administer both medical and behavioral health benefits) and carve-out plans (i.e., plans that administer behavioral health benefits separately from the medical benefits, which are covered through a separate contract with the employer, often through a different vendor). They provided information on patient utilization (with indicators of whether services were provided in-network or out-of-network) and diagnoses. Claims data also included the amount charged by the provider, the amount the plan ultimately paid and the amount the patient owed through their cost-sharing (copayment, deductibles, coinsurance, etc.). These data were used to create the study outcomes.
A study ID in the enrollment data allowed us to map individual enrollees to a subscriber. This established membership within family units in a given year (creating a unit called “family-year” hereafter), as well as provided information about subscriber age, relationship status, dependents, and behavioral health diagnoses across family members.
We linked a sub-set of the claims to subscriber socio-economic characteristics drawn from Optum consumer marketing data. These include subscribers’ highest level of educational attainment, income/net-worth, and race/ethnicity/language. The unit of analysis is the familyyear.
Study samples
The study sample was selected from among 363,048 family-years who had any out-of-network behavioral health expenditures recorded in their employer sponsored insurance claims in a given year. Exclusion criteria are detailed in Web Appendix Figure 1. In brief, since family-year healthcare expenditure-related outcomes (such as balance-billing) were expected to be dependent, in part, on household composition, the first set of exclusion criteria dropped family-years with undetermined, not continuously enrolled, or changing family composition during a year. We also excluded families with any members residing outside of the 50 U.S. states. Sample families were excluded if they were enrolled in plans that did not cover behavioral health services, were retiree/supplemental plans, or were not a calendar year plan. Finally, families were excluded if any member had negative expenditures, if their only behavioral health diagnosis was dementia, or they were missing claims data. Because 35% of the remaining 196,034 family-years had complete data (due to missing Optum Insight data), we used two samples in our analysis. We examined unadjusted balance-billing among both the 196,034 family-year sample (hereafter called the “full sample”) and the 68,659 family-year sample with non-missing socioeconomic data (hereafter called the “SES sample”). We examined adjusted balance-billing among the SES sample.
Outcomes
Our measure of balance-billing used claims expenditure fields recording how much was paid (or not paid) to the billing provider for out-of-network behavioral health services. Balance-billing was calculated by identifying the “submitted amount” (i.e., the total dollar amount that the provider charged) and subtracting out the “allowed amount” (i.e., the amount that the insurer and the patient agree to pay), as well as any discounts made available to the insurer by the provider (observed in less than 5% of the sample). Balance-billing amounts were adjusted using 2014 inflation factors. The outcomes variables measured balance-billing in three ways: 1. The total level of balance-billing per family-year 2. The level of balance-billing per family member; and 3. The total level of balance-billing per family-year as a proportion of their total out-of-network, out-of-pocket expenditures.
Family-year characteristics
To test the study hypotheses, we used Optum Insight® data variables measuring the subscribers’ income/net-worth, educational attainment, and race/ethnicity/language. The categories reported in this analysis were created by Optum Insight® to preserve statistical de-identification. We also controlled for subscriber age and other family-level characteristics using demographic information derived from the claims data including: Subscriber relationship status; Number of dependents; Number of behavioral health diagnoses in the family, and the presence of sixteen behavioral health diagnosis categories.
Plan and employer characteristics
The plan-level covariate indicated whether plans were “carve-out” (versus “carve-in”) or an HMO (versus a PPO). State-level measures of total inpatient beds (from the Area Health Resource File)16 and state-level measures of the MBHO behavioral health provider network (aggregating PhDs, MDs, MSWs and RNs) per 10,000 people were also used as covariates. The following employer-level covariates are described elsewhere17: Size; US Census region of employer’s headquarters; and employer industry, defined by NAIC codes.
Statistical Analysis
We report the distribution of family-year characteristics among both the full and SES samples. We also describe descriptive statistics (mean, standard deviation, median, 75th percentile and 90th percentile) of balance-billing outcomes, overall, and among family-years with any balance-billing, and in both the full and SES samples.
Our multivariate analysis was conducted among the SES sample. We used a two-part model framework to: 1. Calculate the probability of having any balance-billing (using logistic regression) among all family-years, 2. Calculate the level of balance-billing among those who have any balance-billing (using a generalized linear model gamma regression), 3. Calculate the level of balance-billing among all family-years18. We report the marginal effects, calculated using the margins command in STATA, to indicate the average difference in probability or level of in balance-billing between a given subscriber or family-level characteristic value, compared to the reference value. 95% confidence intervals are reported.
Sensitivity analyses
To account for potential multiplicative effects of having more than one diagnosis on balance-billing, we repeated the main multivariate analysis using a model that distinguished between having a particular diagnosis by itself from having that particular diagnosis together with at least one other diagnosis. To test the possibility that our balance-billing estimates were driven by extreme values, we repeated all of our main analyses within samples excluding family-years with balance-billing values in the top 99.5th percentile of the distributions for inpatient, intermediate, and outpatient care. This resulted in a modified full sample of 190,072 family-years and a modified SES sample of 66,648 family-years. For family-years with additional plan information (57%), we repeated analyses among the 104,134 family-years for whom plan OON coverage could be confirmed. Finally, we conducted a sensitivity analysis to determine how well the results in the SES sample generalize to the full sample. We did this by comparing the results from a parsimonious model (excluding the SES variables) in the SES sample to the results of the same model in the full sample. This work was judged exempt by the XX Institutional Review Board.
RESULTS
Distribution of family-year characteristics
Table 1 reports distributions of plan, state provider supply, and family-year characteristics among the 196,034 family-years in the full sample and the 68,659 family-years in the SES sample. Both samples had higher enrollment in carve-in plans and in HMOs. State provider supply was similar across the two samples. In both samples, most subscribers were middle aged, married to a different gendered spouse, and had dependents (63%, data not shown in Table 1). Also in both samples, most family-years had more than 1 behavioral health diagnosis (62%) (where multiple diagnoses could be associated with the same or different family members; 88% of family-years have 1 person with a diagnosis, 10% have 2 people with a diagnosis, and 2% have 3+ people with a diagnosis) and just under half of the family-years had a member diagnosed with depression. Other diagnoses that were common in both samples were adjustment disorder (37%), generalized anxiety disorder (37%), and bi-polar disorder (25%).
Table 1.
Descriptive statistics of plan characteristics, provider supply, and family-year characteristics in full and SES1 samples
| Characteristics | Full Sample (n= 196,034) | SES sample (n=68,659) | ||
|---|---|---|---|---|
| Plan characteristics | N | % | N | % |
| Carve-out status | 55,773 | 28% | 5032 | 7% |
| Carve-in status | 140,261 | 72% | 63,627 | 93% |
| HMO | 138,020 | 70% | 61,000 | 89% |
| Non-HMO | 58,014 | 29% | 7659 | 11% |
| Provider supply | Mean | SD | Mean | SD |
| Short-term hospital beds per 10K people | 23 | 5 | 23 | 4 |
| Psychiatric hospitals per 10K people | 0.02 | 0.009 | 0.02 | 0.01 |
| In-network behavioral health providers (MD, PhD, MSW, RN) per 10K people | 7 | 8 | 7 | 7 |
| Subscriber age | N | % | N | % |
| 18–24 | 1404 | 1% | 461 | 1% |
| 25–34 | 25,167 | 13% | 9657 | 14% |
| 35–44 | 55,148 | 28% | 20,497 | 30% |
| 45–54 | 68,313 | 35% | 23,727 | 35% |
| 55–64 | 39,300 | 20% | 12,117 | 18% |
| 65+ | 6702 | 3% | 2200 | 3% |
| Subscriber relationship status | ||||
| Single | 68,037 | 35% | 23,972 | 35% |
| Domestic partner, different gender | 1749 | 1% | 808 | 1% |
| Domestic partner, same gender | 1336 | 1% | 563 | 1% |
| Spouse, different gender | 124,271 | 63% | 43,172 | 63% |
| Spouse, same gender | 641 | 0.3% | 144 | 0.2% |
| Number of unique diagnoses in family 2 | ||||
| 1 | 74,031 | 38% | 25,968 | 38% |
| 2 | 47,317 | 24% | 16,485 | 24% |
| 3 | 29,907 | 15% | 10,290 | 15% |
| 4+ | 44,779 | 23% | 15,916 | 23% |
| Diagnoses | ||||
| Adjustment disorder | 72,890 | 37% | 25,893 | 38% |
| PTSD | 11,439 | 6% | 3983 | 6% |
| Generalized anxiety disorder | 72,880 | 37% | 26,038 | 38% |
| Obsessive compulsive disorder | 7327 | 4% | 2514 | 4% |
| Panic disorder | 10,817 | 6% | 3844 | 6% |
| Phobia disorders | 7067 | 4% | 2527 | 4% |
| Attention deficit hyperactivity disorder | 32,910 | 17% | 11,646 | 17% |
| Other child behavioral health disorders | 39,800 | 20% | 14,037 | 20% |
| Pervasive developmental disorder | 6559 | 3% | 2253 | 3% |
| Bipolar disorder | 49,412 | 25% | 17,186 | 25% |
| Depression | 93,539 | 48% | 32,081 | 47% |
| Personality disorder | 3468 | 2% | 1201 | 2% |
| Schizophrenia | 10,595 | 5% | 4139 | 6% |
| Alcohol use disorder | 14,943 | 8% | 5393 | 8% |
| Drug use disorder | 20,260 | 10% | 7062 | 10% |
| Other behavioral health disorders | 34,116 | 17% | 12,000 | 17% |
| Number of dependents, by dependent age groups3 | Mean | SD | Mean | SD |
| < 5 years old | 0.2 | 0.5 | 0.2 | 0.5 |
| 6–11 years old | 0.3 | 0.7 | 0.3 | 0.7 |
| 12–17 years old | 0.4 | 0.7 | 0.4 | 0.7 |
| 18–25 years old | 0.4 | 0.7 | 0.4 | 0.7 |
| Subscriber level of educational attainment | N | % | ||
| High school or lower | 8770 | 13% | ||
| Some college | 22,541 | 33% | ||
| Associates degree | 7749 | 11% | ||
| Bachelor’s degree or higher | 29,599 | 43% | ||
| Subscriber income/net-worth | ||||
| Income: <75K & net-worth: <= 25K | 4949 | 7% | ||
| Income: <75K & net-worth: > 25K & < 100K | 3360 | 5% | ||
| Income: < 75K & net-worth: >100K | 5541 | 8% | ||
| Income: 75K–150K & net-worth: <100K | 4269 | 6% | ||
| Income: 75K–150K & net-worth: 150K-250K | 5571 | 8% | ||
| Income: 75K–150K & net-worth: >250K | 10,155 | 15% | ||
| Income: >150K & net-worth: < 500K | 7591 | 11% | ||
| Income: >150K & net-worth: >=500K | 13,522 | 20% | ||
| Income: unknown & net-worth: <150K | 4941 | 7% | ||
| Income: unknown & net-worth: 25K-100K | 8760 | 13% | ||
| Subscriber race/ethnicity/language | ||||
| Asian, English | 1281 | 2% | ||
| Asian, Other language | 877 | 1% | ||
| Black, Any language | 3294 | 5% | ||
| Hispanic, English | 2385 | 3% | ||
| Hispanic, Other language | 2074 | 3% | ||
| Other, Any language | 3438 | 5% | ||
| White, Any language | 55,310 | 81% | ||
SES: Socio-economic status; PTSD: Post-traumatic stress disorder; SD: Standard Deviation
: The SES sample includes family-years with non-missing socio-economic information.
: If a family-year had multiple instances of the same diagnostic category, only one additional diagnosis was counted. For example, a family-year with multiple claims with depression, but no other claims, had a count of 1 diagnosis. Also, multiple diagnoses can be associated with either the same person or different people in the family.
: Dependents were counted here if they were not the subscriber, and they are not the subscribers’ spouse or domestic partner. If a family-year had children in multiple age groups, only the children in the stated age-group were counted Page 15 of 18 towards the mean on a particular row. For example, if a family had a child who was 2 and a child who was 10, the 2year-old counted towards the mean on the row labeled “<5 years old”, and the 10-year-old counted towards the mean on the row labeled “6–11 years old”.
In the SES sample, over half of family-years had a subscriber with a college degree (Table 1). One third of subscribers earned between $75,000 and $150,000 and another third earned over $150,000. A substantial proportion of family-years had a subscriber who was Hispanic, Black, or Asian.
Unadjusted balance-billing levels
Table 2 describes the distribution of balance-billing among all family-years and among family-years with balance-billing over the four study years, for the full sample. Among the 97,979 (50%) of family-years with any balance-billing, mean total family-year balance-billing was $861±$3500 (median $175); mean balance-billing per family member: $381±$1723 (median $64); balance-billing as a proportion of family-year out-of-network out-of-pocket expenditures: 33%±26% (median 28%). Balance-billing values in the SES sample were similar (Web Appendix Table 2).
Table 2.
Summary of balance-billing1 levels in full sample
| Mean | SD | Median | 75th percentile | 90th percentile | |
|---|---|---|---|---|---|
| Among all family-years (n=196,034) | |||||
| Total Balance-billing for family-year | $430 | $2460 | $0 | $175 | $802 |
| Balance-billing per member in family-year | $191 | $1233 | $0 | $64 | $323 |
| Balance-billing as a proportion of family out-of-network out-of-pocket expenditures | 17% | 25% | 0% | 29% | 55% |
| Among family-years with any balance-billing (n=97,979) | |||||
| Total Balance-billing for family-year | $861 | $3500 | $175 | $600 | $1684 |
| Balance-billing per member in family-year | $381 | $1723 | $64 | $236 | $738 |
| Balance-billing as a proportion of family out-of-network out-of-pocket expenditures | 33% | 26% | 28% | 49% | 70% |
SD: Standard Deviation
: Dollar amounts were adjusted for inflation to 2014 values.
Differences in adjusted balance-billing levels by family-year characteristics
Table 3 reports the adjusted average differences in balance-billing for each predictor in the model, relative to the reference group. Total family-year balance-billing was substantially higher for families enrolled in carve-out plans ($523; 95%CI: $340, $705) and for families enrolled in HMO plans ($156; 95%CI: $75, $237) (Columns 6 & 7). Both differences were driven by differences in the probability of any balance-billing (Columns 2 & 3), and the differences for carve-out enrollment was additionally driven by the level in balance-billing among those with any balance-billing (Columns 4 & 5).
Table 3.
Adjusted differences in probability of any balance-billing, and differences in level of total balance-billing among those with any balance-billing and among all family-years, by plan characteristics, provider supply, and subscriber and family characteristics
| Difference in the probability of any balance-billing (n=68,659)1,2 | Difference in the level of balance-billing among family-years with any balance-billing1,3 (n=32,777) | Difference in level of balance-billing among all family-years1,4 (n=68,659) | ||||
|---|---|---|---|---|---|---|
| Percentage points | 95% CI | $ | 95% CI | $ | 95% CI | |
| Plan characteristics | ||||||
| Carve-out status (versus Carve-in) | 27* | (25, 29) | 430* | (169, 691) | 523* | (340, 705) |
| HMO (versus non-HMO) | 15* | (13, 17) | 46 | (−120, 212) | 156* | (75, 237) |
| Provider supply | ||||||
| Short-term hospital beds per 10K people | 0.0003 | (−0.1, 0.1) | 16* | (2, 30) | 8* | (.7, 15) |
| Psychiatric hospitals per 10K people | 1.2 | (−47, 49) | −2004 | (−13, 166) | −947 | (−3128, 1234) |
| In-network behavioral health providers (MD, PhD, MSW, RN) per 10K people | 0.01 | (−0.05, 0.08) | −16* | (−21, −11) | −8* | (−10, −5) |
| Subscriber highest level of educational attainment (ref= Bachelor’s degree or higher) | ||||||
| High school or lower | −4.6* | (−6.0, −3.2) | −380* | (−503, −257) | −224* | (−284, −163) |
| Some college | −4.5* | (−5.5, −3.6) | −304* | (−400, −207) | −188* | (−236, −139) |
| Associates degree | −3.7* | (−4.9, −2.4) | −285* | (−398, −172) | −172* | (−228, −116) |
| Subscriber income/net-worth (ref= Income: <75K & net-worth: <= 25K) | ||||||
| Income: <75K & net-worth: 25K-100K | −1.7 | (−3.8, 0.5) | −24 | (−224, 175) | −26 | (−124, 70) |
| Income: < 75K & net-worth: >=100K | −1.5 | (−3.4, 0.5) | 70 | (−121, 261) | 20 | (−73,114) |
| Income: 75K-150K & net-worth: <100K | −4.1* | (−6.0, −1.9) | 65 | (−133, 263) | −7 | (−102, 87) |
| Income: 75K-150K & net-worth: 100K-250K | −1.8 | (−3.7, 0.1) | −71 | (−244, 102) | −49 | (−134, 35) |
| Income: 75K-150K & net-worth: >250K | −1.9* | (−3.7, −0.1) | −71 | (−234, 91) | −50 | (−130, 29) |
| Income: >150K & net-worth: < 500K | −2.0* | (−3.9, −0.1) | 23 | (−154, 199) | −7 | (−93, 80) |
| Income: >150K & net-worth: >= 500K | 1.2 | (−0.7, 3.1) | 21 | (−150, 192) | 21 | (−64, 107) |
| Subscriber race/ethnicity/language (ref=white, any language) | ||||||
| Asian, English | −1.5 | (−4.1, 1.2) | 26 | (−234, 286) | −0.7 | (−124, 123) |
| Asian, other language | −1.7 | (−4.9, 1.5) | 124 | (−214, 461) | 42 | (−117, 202) |
| Black, any language | −1.4 | (−3.2, −0.4) | 31 | (−140,203) | 3 | (−79, 84) |
| Hispanic, English | −1.2 | (−3.2, 0.8) | −8 | (−193, 177) | −14 | (−102, 74) |
| Hispanic, other language | −0.7 | (−2.9, 1.5) | 172 | (−62, 406) | 75 | (−38, 188) |
| Other, any language | 0.7 | (−1.0, 2.4) | 29 | (−125, 183) | 21 | (−56, 97) |
| Subscriber age (ref=45–54 years old) | ||||||
| 18–24 | −3.2 | (−8.7, 0.4) | −557* | (−789, −325) | −290* | (−398, 183) |
| 25–34 | −3.2* | (−4.5, −1.9) | −258* | (−367, −148) | −148* | (−201, −96) |
| 35–44 | −1.1* | (−2.1, −0.009) | −96* | (−191, −2) | −57* | (−103, −10) |
| 55–64 | −0.2 | (−1.3, 0.9) | −53 | (−158, 51) | −28 | (−79, −24) |
| 65+ | −8.3* | (−10.6, −6.0) | 6 | (−263, 274) | −78 | (−192, −36) |
| Subscriber relationship (ref=single) | ||||||
| Domestic partner, different gender | 0.9 | (−2.5, 4.3) | −112 | (−409, 185) | −45 | (−191, 100) |
| Domestic partner, same gender | 0.02 | (−4.0, 4.1) | 259 | (−231, 750) | 123 | (−114, 359) |
| Spouse, different gender | 1.0* | (0.07, 1.9) | −51 | (−135, −34) | −16 | (−56, 25) |
| Spouse, same gender | −3.3 | (−11.2, 4.5) | −316 | (−852, 219) | −170 | (−412, 71) |
| Number of dependents (ref=0) | ||||||
| 1 | 1.5* | (1.0, 2.2) | 44 | (−28, 116) | 34 | (−1, 69) |
| 2 | 1.8* | (1.2, 2.4) | 7 | (−48, 61) | 19 | (−8, 46) |
| 3 | 1.3* | (0.8, 1.9) | −37 | (−86, 12) | −6 | (−30, 18) |
| 4+ | −0.04 | (−0.6, 0.5) | −45 | (−96, 6) | −22 | (−47, 3) |
| Diagnosis | ||||||
| Adjustment disorder | 4.0* | (3.1, 4.8) | 20* | (−54, 94) | 45* | (8, 82) |
| PTSD | −0.01 | (−1.7, 1.7) | 267* | (83, 450) | 127* | (38, 217) |
| Generalized anxiety disorder | 3.4* | (2.6, 4.2) | 186* | (111, 262) | 120* | (83, 158) |
| Obsessive compulsive disorder | 7.7* | (5.7, 9.7) | 167* | (−22, 357) | 158* | (53, 263) |
| Panic disorder | −0.2 | (−1.9, 1.5) | −103* | (−242, −37) | −51 | (−118, 17) |
| Phobia disorders | 5.6* | (3.6, 7.6) | 191 | (−14, 395) | 149* | (40, 259) |
| Attention deficit hyperactivity disorder | 3.3* | (1.1, 5.4) | −109 | (−280, 62) | −25 | (−112, 61) |
| Other child behavioral health disorders | 5.4* | (3.3, 7.4) | 14 | (−162, 190) | 54 | (−37, 146) |
| Pervasive developmental disorder | 8.0* | (5.9, 10.1) | 1111* | (738, 1485) | 678* | (469, 887) |
| Bipolar disorder | 1.8* | (0.8, 2.7) | 83 | (−4, 170) | 56* | 12, 99) |
| Depression | 3.6* | (3.0, 4.6) | 119* | (45, 194) | 89* | (52, 125) |
| Personality disorder | 1.0 | (−1.8, 3.9) | 299 | (−18, 617) | 154 | (−4, 312) |
| Schizophrenia | −9.1* | (−11.0, −7.9) | −162* | (−296, −27) | −148* | (−205, −91) |
| Alcohol use disorder | −0.3 | (−1.3, 1.9) | 930* | (709, 1150) | 447* | (339, 555) |
| Drug use disorder | 6.3* | (4.8, 7.8) | 2182* | (1849, 2516) | 1190* | (1010, 1370) |
| Other behavioral health disorders | 3.0* | (2.0, 4.0) | 157* | (59, 254) | 104* | (54, 153) |
SES: Socioeconomic status; ref: Reference group
P<0.05
: All models also controlled for subscriber employer characteristics (size, industry, region, etc.) whether or not plans are an HMO, unknown income and net-worth < $150K and unknown income and net-worth >= 150K, and calendar year. Regressions used a two-part model for total family-year balance-billing and marginal effects were generated using the margins command in STATA.
: Part 1 of the two-part model used logistic regression to determine the probability of having any balance-billing.
: Part 2 of the two-part model used a gamma regression to determine the average difference in the level of balance-billing among SES sample family-years with any balance-billing.
: The combined parts 1 and 2 report the average difference in the level of balance-billing among all family-years in the SES sample, unconditional on whether or not they had any balance-billing.
Looking at the state provider-supply predictors, higher numbers of short-term hospital beds was associated with significantly higher balance-billing ($8, 95%CI: $0.7, $15) while higher numbers of in-network behavioral health providers were associated with significantly lower balance-billing ($8; 95%CI: -$10, -$5) levels (Columns 6 & 7).
Several family-year characteristics were significant as well. Balance-billing was between $172 (95%CI: $−228, -$116) and $224 (95%CI: -$284, -$163) lower when the subscriber did not have a bachelor’s degree (Table 3, Columns 6 & 7) compared to when they did have a bachelor’s degree or a higher level of educational attainment. This was due to both lower probability of any balance-billing (Columns 2 & 3) and to lower levels among those with any balance-billing (Columns 4 & 5). Table 3 also reports that total family-year balance-billing was significantly lower (between $28; 95%CI: -$79, -$24; and $290; 95%CI: -$398, -$183) for family-years with subscribers in all age groups, compared to those with subscribers aged 45–54 years old.
Table 3 also shows associations between behavioral health diagnoses and balance-billing. All but four diagnoses (panic disorders, attention deficit and hyperactivity disorder, other childhood disorders and personality disorders) were significantly associated with balance-billing. Among those with significant associations, all but one was associated with a higher level of balance-billing compared to not having the diagnosis. Adjusted differences in balance-billing ranged from $45 (Adjustment disorder, 95%CI: $8, $82) to $447 (Alcohol use disorder, 95%CI: $339, $555) and $1190 (Drug use disorder, 95%CI: $1010, $1370). Web Appendix Table 3 shows that differences in balance-billing per family member were similar to differences in total family-year balance-billing.
Sensitivity analyses
Controlling for whether members in a family-year had each of the 16 diagnoses by itself or together with at least one of the other 15 diagnoses did not change the magnitude, direction or significance of most model predictors (Web Appendix Table 4). Excluding family-years with extreme balance-billing values resulted in mean conditional balance-billing of $741 in the modified full sample (±$2710) (Web Appendix 5a), and the direction and significance of most predictors in the multivariate models aligned with the main results (Web Appendix 6–8). Among family-years with confirmed OON coverage status, mean conditional balance-billing was $777 (±$3099), about $100 lower than in the full sample, and median conditional balance-billing was $166, about $10 lower than in the full sample. The multivariate results among this sample closely resembled the main model results. Finally, the multivariate results of the parsimonious model were nearly identical between the full sample and the SES sample.
DISCUSSION
This analysis documents additional cost-sharing burden (i.e., balance-billing) for behavioral health services in a national sample of commercially insured families who used out-of-network behavioral health services. It finds: 1) half of families submitting a claim for out-of-network behavioral health services in a given year had any balance billing; 2) among those with balance-billing, half of the families had annual balance-billing levels over $175; and, 3) higher total family-year balance-billing was experienced by families enrolled in carve-out plans as well as those enrolled in HMOs, compared to other families.
Several limitations require mention. First, administrative data may include data entry errors. Excluding family-years with the highest balance-billing values did not find substantially different results, but underestimates are also possible. Second, we are missing SES data on a substantial portion of families with out-of-network behavioral health use. The SES and full sample have similar distributions of predictor variables and similar regression results for a parsimonious model, but we cannot confirm whether the SES data are missing at random. Third, this paper focuses on balance-billing among providers who submit claims to the insurer, and does not capture the full out-of-pocket burden by those seeing behavioral health providers who do not accept any insurance. Fourth, state-level provider supply measures may obscure the actual supply in a particular region in the state.
With respect to external validity, our data represent claims for one large national managed behavioral health organization (MBHO) and might not generalize to other organizations offering employer-sponsored insurance or other forms of insurance. Differences in plan generosity or network size could impact balance-billing. Previous literature describes financial requirements (copayments, coinsurance etc.) for these and other plans offered by the MBHO, and can help readers’ assess the generalizability of this MBHO’s plan generosity.19,20 However, due to the wide reach of the study MBHO, data on this MBHO alone merits inquiry. The data’s age may also reduce external validity.
Our study found that one in two sample families were asked to cover the difference between what their provider charged and what the plan agreed to pay and what the family owed as cost-sharing. Among them, the families in the top 25th percentile had annual balance-billing values over $500, even after excluding the most extreme balance-billing values. Those facing the most extreme values (n=9322 sample family-years), saw balance-billing levels nearly three times as high.
A national survey of household well-being, fielded by the Federal Reserve, suggested that for many families, an unexpected bill of this size would involve financial hardship, including increased debt, deferred necessities, etc.21 As noted above in the limitations, this financial burden may be further compounded by bills from providers who do not accept insurance; these bills cannot be measured using insurance claims data. It is also important to note that the families who did not have balance-billing may have had higher cost-sharing for out-of-network behavioral health care via coinsurance and deductibles, etc., although characterizing this cost-sharing is beyond the scope of the current analysis.
Carve-out plans are associated with higher balance-billing than carve-in plans. Carve-out plans have become less ubiquitous following the Mental Health Parity and Addiction Equity Act (MHPAEA) (prior to the current analysis’ study period).14 Given the sample’s heavy concentration of family-years enrolled in HMOs, which typically have narrower networks22 and often do not reimburse for out-of-network care, the finding that only half of family-years had any balance-billing may be counterintuitive. However, the adjusted analysis found that HMOs were associated with substantially higher out-of-network balance-billing. Additionally, a post-hoc analysis found that 75% of claims for HMO plans in our sample had a non-zero plan-pay amount. Higher balance-billing in HMO plans was consistent with the finding that balance-billing was higher for families in states with smaller networks (i.e. smaller provider supply). Policies that increase the supply of providers who accept insurance and, simultaneously, disincentivize narrow insurance networks, may indirectly reduce balance-billing burden on households while also improving access to behavioral health care.
CONCLUSION
This is the first national study of balance-billing in out-of-network behavioral health claims, a topic of interest given the frequent need to use out-of-network providers to obtain behavioral health care. The analysis found that about half of family-years with employer-sponsored insurance claims for out-of-network behavioral health services had some balance-billing amount. Balance-billing levels, which were substantially higher for families enrolled in carve-out and HMO plans, may be burdensome for many families and may reduce use of needed behavioral health services.
Supplementary Material
Highlights:
About half of study families using out-of-network behavioral health services in a given year were responsible for out-of-network provider charges (i.e., balance-billing), beyond what the insurer and patient were contractually required to pay.
Among family-years with these additional behavioral health expenditures, half of the families had annual balance-billing levels over $175.
Higher balance-billing was experienced by families enrolled in carve-out plans and in HMOs.
Disclosures and acknowledgements:
Dr. Azocar is an employee of OptumHealth Behavioral Solutions and as such, she receives compensation in the form of salary and stock. None of the other co-authors have any conflicts of interest to report. This work was supported by a grant from the National Institutes of Mental Health (R01MH117013-01).
References
- 1.Hyman DA, Ippolito B, Silver C. Surprise Medical Bills: How to Protect Patients and Make Care More Affordable. Georgetown Law J. 2020;108:24. [Google Scholar]
- 2.Cooper Z, Nguyen H, Shekita N, Morton FS. Out-Of-Network Billing And Negotiated Payments For Hospital-Based Physicians. Health Aff Proj Hope. 2020;39(1):24–32. doi: 10.1377/hlthaff.2019.00507 [DOI] [PubMed] [Google Scholar]
- 3.Garmon C, Chartock B. One In Five Inpatient Emergency Department Cases May Lead To Surprise Bills. Health Aff (Millwood). 2017;36(1):177–181. doi: 10.1377/hlthaff.2016.0970 [DOI] [PubMed] [Google Scholar]
- 4.Sun EC, Mello MM, Moshfegh J, Baker LC. Assessment of Out-of-Network Billing for Privately Insured Patients Receiving Care in In-Network Hospitals. JAMA Intern Med. Published online August 12, 2019. doi: 10.1001/jamainternmed.2019.3451 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Chhabra KR, Sheetz KH, Nuliyalu U, Dekhne MS, Ryan AM, Dimick JB. Out-of-Network Bills for Privately Insured Patients Undergoing Elective Surgery With In-Network Primary Surgeons and Facilities. JAMA. 2020;323(6):538–547. doi: 10.1001/jama.2019.21463 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Fuse Brown EC, Trish E, Ly B, Hall M, Adler L. Out-of-Network Air Ambulance Bills: Prevalence, Magnitude, and Policy Solutions. Milbank Q. 2020;98(3):747–774. doi: 10.1111/1468-0009.12464 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Busch SH, Kyanko KA. Incorrect Provider Directories Associated With Out-Of-Network Mental Health Care And Outpatient Surprise Bills. Health Aff (Millwood). 2020;39(6):975983. doi: 10.1377/hlthaff.2019.01501 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Zhu JM, Zhang Y, Polsky D. Networks In ACA Marketplaces Are Narrower For Mental Health Care Than For Primary Care. Health Aff (Millwood). 2017;36(9):1624–1631. doi: 10.1377/hlthaff.2017.0325 [DOI] [PubMed] [Google Scholar]
- 9.Kyanko KA, Curry LA, Busch SH. Out-of-Network Provider Use More Likely in Mental Health than General Health Care Among Privately Insured. Med Care. 2013;51(8):699–705. doi: 10.1097/MLR.0b013e31829a4f73 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Xu WY, Song C, Li Y, Retchin SM. Cost-Sharing Disparities for Out-of-Network Care for Adults With Behavioral Health Conditions. JAMA Netw Open. 2019;2(11):e1914554. doi: 10.1001/jamanetworkopen.2019.14554 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Benson NM, Song Z. Prices And Cost Sharing For Psychotherapy In Network Versus Out Of Network In The United States. Health Aff (Millwood). 2020;39(7):1210–1218. doi: 10.1377/hlthaff.2019.01468 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Song Z, Lillehaugen T, Busch SH, Benson NM, Wallace J. Out-of-Network Spending on Behavioral Health, 2008–2016. J Gen Intern Med. 2021;36(1):232–234. doi: 10.1007/s11606-020-05665-w [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Song Z, Johnson W, Kennedy K, Biniek JF, Wallace J. Out-Of-Network Spending Mostly Declined In Privately Insured Populations With A Few Notable Exceptions From 2008 To 2016. Health Aff Proj Hope. 2020;39(6):1032–1041. doi: 10.1377/hlthaff.2019.01776 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Ettner SL, Xu H, Azocar F. What Happens When Employers Switch from a “Carve-Out” to a “Carve-In” Model of Managed Behavioral Health? J Ment Health Policy Econ. 2019;22(3):85–94. [PubMed] [Google Scholar]
- 15.Thalmayer AG, Friedman SA, Azocar F, Harwood JM, Ettner SL. The Mental Health Parity and Addiction Equity Act (MHPAEA) Evaluation Study: Impact on Quantitative Treatment Limits. Psychiatr Serv Wash DC. 2017;68(5):435–442. doi: 10.1176/appi.ps.201600110 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Health Resources and Services Administration, Bureau of Health Workforce, National Center for Health Workforce Analysis. Area Health Resources Files.; 2018. https://data.hrsa.gov/data/download
- 17.Ettner S, Harwood J, Thalmayer A, et al. The Mental Health Parity and Addiction Equity Act evaluation study: Impact on specialty behavioral health utilization and expenditures among “carve-out” enrollees. J Health Econ. 2016;50:131–143. doi: 10.1016/j.jhealeco.2016.09.009 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Belotti F, Deb P, Manning WG, Norton EC. Twopm: Two-Part Models. Stata J. 2015;15(1):3–20. doi: 10.1177/1536867X1501500102 [DOI] [Google Scholar]
- 19.Friedman SA, Thalmayer AG, Azocar F, et al. The Mental Health Parity and Addiction Equity Act Evaluation Study: Impact on Mental Health Financial Requirements among Commercial “Carve-In” Plans. Health Serv Res. 2018;53(1):366–388. doi: 10.1111/14756773.12614 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Friedman SA, Azocar F, Xu H, Ettner SL. The Mental Health Parity and Addiction Equity Act (MHPAEA) evaluation study: Did parity differentially affect substance use disorder and mental health benefits offered by behavioral healthcare carve-out and carve-in plans? Drug Alcohol Depend. 2018;190:151–158. doi: 10.1016/j.drugalcdep.2018.06.008 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Board of Governers of the Federal Reserve. Report on the Economic Well-Being of U.S. Households in 2020. Federal Reserve; 2021. https://www.federalreserve.gov/publications/files/2020-report-economic-well-being-ushouseholds-202105.pdf [Google Scholar]
- 22.Polsky D, Weiner J. The Skinny on Narrow Networks in Health Insurance Marketplace Plans. Data Briefs. Published online June 23, 2015. https://repository.upenn.edu/ldi_databriefs/11 [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.