Total Direct Medical Expenses and Characteristics of Privately Insured Adolescents Who Incur High Costs

Importance  Accountable care payment models aim to reduce total direct medical expenses for high-cost patients through improved quality of care and preventive health services. Little is known about health care expenditures of privately insured adolescents, especially those who incur high costs.

Objectives  To assess health care expenditures for high-cost adolescents and to describe the patient characteristics associated with high medical costs.

Design, Setting, and Participants  A retrospective cohort analysis was conducted of data from January 1 to December 31, 2012, of 13 103 privately insured adolescents aged 13 to 21 years (mean [SD] age, 16.3 [2.4] years; 6764 [51.6%] males) at 82 independent pediatric primary care practices in Massachusetts. Analysis was conducted from April 1, 2014, to April 1, 2015.

Main Outcomes and Measures  We compared demographic (age, sex, median income by zip code) and clinical (obesity, behavioral health problem, complex chronic condition) characteristics between high-cost (top 1%) and non–high-cost adolescents. We assigned high-cost adolescents to clinical categories using software from the Agency for Healthcare Research and Quality to describe clinically relevant patterns of spending.

Results  Total direct medical expenses were $41.2 million for the entire cohort and a median $1167 per patient. A total of 132 (1.0%) patients with the highest costs accounted for 23.6% of expenses of the cohort, with a median $52 577 per patient. Mental health disorders were the most common diagnosis in high-cost patients; 78 (59.1%) of these patients had at least 1 behavioral health diagnosis. Pharmacy costs accounted for 28.4% of total direct medical expenses of high-cost patients; primary care accounted for 1.0%. Characteristics associated with being a high-cost patient included having 1 complex chronic condition (relative risk [RR], 6.5; 95% CI, 4.7-9.0), having 2 or more complex chronic conditions (RR, 23.5; 95% CI, 14.2-39.1), having any behavioral health diagnosis (RR, 3.6; 95% CI, 2.6-5.1), and obesity (RR, 2.0; 95% CI, 1.3-3.0).

Conclusions and Relevance  Total direct medical expenses for privately insured high-cost adolescents are associated with medical complexity, mental health conditions, and obesity. Cost reduction strategies in similar populations should be tailored to these cost drivers.

Health care expenditures were 17.9% of the US gross domestic product in 2011 and are projected to rise to 19.6% by 2021.¹ Children and adolescents generally have lower health care costs than adults, but little is known about how specific conditions among adolescents are associated with health services that increase costs. Understanding adolescents’ health care use and costs is important because many chronic conditions, including obesity and mental health disorders, that originate in adolescence are associated with long-term increased morbidity and costs in adulthood.^2-5 The influence of these chronic health conditions relative to other health care costs for adolescents remains unknown.

With the development of new health care payment models, there is a need to devise strategies to maintain or reduce costs while meeting quality and performance standards.⁶ In accountable care models, physicians receive information about their patients’ clinical care and costs from payers. Physicians also receive incentives for meeting quality benchmarks, and they share savings for stabilizing or reducing costs.⁶ Massachusetts was one of the first states to implement accountable care payment reform on a large scale.⁷

It is now well known that a small number of patients—both children and adults—consume a disproportionate amount of health care resources and generate a disproportionate amount of cost.^8-15 To our knowledge, little has been published about what characteristics or clinical conditions in privately insured adolescents drive high health care costs. In this study, we report the findings from 82 pediatric primary care practices in Massachusetts that entered an accountable care–style contract with a private insurer, which allowed us to undertake this analysis. Our objectives were to characterize total direct medical expenses (TDME) for a primary care population of privately insured adolescents, and then identify the clinical diagnoses and health care services that contribute the most to high-cost health care.

Box Section Ref ID

We analyzed data from paid medical and pharmacy claims from January 1 to December 31, 2012, for all members enrolled for 9 or more months in a large, nonprofit commercial health insurance plan who were aged 13 to 21 years (the upper age limit for many practices) as of January 1, 2012. They were patients of primary care providers (PCPs) from the Pediatric Physicians’ Organization at Children’s, an independent association affiliated with Boston Children’s Hospital. Data usage agreements prohibit naming the insurance plan in this article. The association consists of 82 pediatric primary care practices ranging from 1 to 23 physicians. Although they are independently owned, the practices collaborate in quality improvement. This project was approved by the Boston Children’s Hospital Institutional Review Board. Analysis was conducted from April 1, 2014, to April 1, 2015.

We categorized TDME using Current Procedural Technology codes (eTable 1 in the Supplement). Total direct medical expenses include insurer costs and family costs, such as copayments, but do not include indirect costs, such as lost wages, parking, and out-of-pocket charges. In our organization’s contract, behavioral health claims are not separated from medical claims (“carved out”) as they are in some other contracts. We reviewed the distribution of TDME by cost per member per month and defined the top 1.0% of patients (n = 132) with the highest medical costs as high cost.

We compared high-cost and non–high-cost adolescents by age, sex, and neighborhood income level. We did not have income data for individual patients but used their zip codes to assess whether median income in the neighborhood was 200% or more of the federal poverty level in 2012.^16,17 Race and ethnicity data were not available. Our choice of clinical characteristics was opportunistic and limited to data available in a billing database, but based on prior studies suggesting risk factors for high expenses in both children and adults, as well as a pilot study in an academic adolescent clinic.^{2,8,10,14,15,18} eTable 2 in the Supplement contains a description of the clinical categorization. We used Clinical Classifications Software developed at the Agency for Healthcare Research and Quality (AHRQ) to assign each patient to 1 of 16 mutually exclusive groups.² For each high-cost patient, we identified the proportion of TDME associated with each of the patient’s primary billing diagnoses. If 1 diagnosis was associated with most of the patient’s TDME, we assigned the patient to an AHRQ clinical category based on that diagnosis. Two pediatricians (S.H.G. and L.V.) reviewed all the claims to determine what condition was most clearly associated with a patient incurring high costs. In 19 of 132 patients, the AHRQ category was reassigned based on clinical judgment. As an example, a patient with a congenital brain anomaly was hospitalized multiple times with pneumonia. Most of the claims were for pneumonia, and the patient was initially categorized as having a respiratory condition, but we thought that the true cost driver was the brain anomaly and thus changed the categorization to congenital.

We used χ² analysis to compare all binary variables—neighborhood income level (≥200% or <200% of the federal poverty level), sex, presence of behavioral health diagnosis, obesity, chronic complex conditions (CCC) (none, 1, or >1)—as well as calculating relative risk and 95% CIs of being in the top 1.0% of expenses. The one continuous variable considered, mean age distribution, was analyzed using a t test. To assess risk factors for being a high-cost patient, we used logistic regression to obtain unadjusted and adjusted relative risk.¹⁹ Mean TDME for each predictor was modeled using analysis of variance. Statistical analyses were performed with SAS, version 9.3 (SAS Institute Inc). Statistical significance was set at P < .05.

There were 13 103 adolescents in the final cohort, with a mean (SD) age of 16.3 (2.4) years; 6764 (51.6%) were male. More than 97% lived in a zip code with median income greater than or equal to 200% of the federal poverty level. A total of 2768 patients (21.1%) had a behavioral health diagnosis; 680 (5.2%) had an obesity diagnosis, 1401 (10.7%) had at least 1 CCC, and 138 (1.1%) had 2 or more CCCs. Total direct medical expenses for the cohort were $41.2 million.

The median annual TDME per patient was $1167. High-cost patients had a median annual TDME of $52 577 compared with $1151 for the rest of the cohort.

The top 1.0% of adolescents with the highest medical costs accounted for 23.6% of total spending ($9.7 million) for the cohort. Inpatient care accounted for the largest single portion of spending (38.8%) for high-cost patients, as shown in the Figure. More than two-thirds (90 [68.2%]) of high-cost patients used inpatient hospital services. Inpatient expenses were highest for adolescents with AHRQ categories of congenital and mental illness (Table 1). Medications accounted for the next largest percentage of expenditures (28.4%) in high-cost patients. When we reviewed pharmaceutical costs more closely, we found that 8 orphan drugs accounted for $1.5 million of spending, 15.5% of the TDME for the 132 high-cost adolescents and 3.6% of the TDME for the entire cohort of 13 103 patients. This amount spent on orphan drugs was greater than all pharmacy costs for prescriptions written by PCPs for all adolescents in the entire cohort.

Primary care services accounted for 1.0% of spending for high-cost adolescents. Although 92.4% of high-cost patients had at least 1 visit with a PCP, the mean cost per patient was $777. Eighty-nine high-cost adolescents (67.4%) had at least 1 emergency department visit, which accounted for 3.1% of TDME. One hundred six high-cost adolescents (80.3%) had at least 1 radiology study and 94.7% had at least 1 laboratory study ordered. Radiology and laboratory services accounted for 2.8% and 2.6%, respectively, of TDME.

For non–high-cost adolescents, 47.0% of TDME was for non-PCP specialist outpatient services and prescriptions; 8556 of these adolescents (66.0%) were seen by a non-PCP specialist. Primary care provider outpatient services and PCP-derived pharmacy costs represented 22.9% of TDME, as shown in the Figure. Most non–high-cost adolescents (10 756 [82.9%]) had at least 1 PCP visit, and 4460 (34.4%) received at least 1 prescription from a PCP. Few (264 [2.0%]) non–high-cost patients were hospitalized, accounting for 5.6% of TDME; 2273 (17.5%) had at least 1 emergency department visit, accounting for 5.7% of TDME; 3720 (28.7%) had at least 1 radiology service, accounting for 6.6% of TDME; and 7493 (57.8%) had at least 1 laboratory test, accounting for 5.7% of TDME. A total of 2492 (19.2%) of these adolescents had no expenses for the year.

Age, sex, and neighborhood income level were not significantly different between high-cost and non–high-cost patients in this cohort, as shown in Table 2. High-cost patients were more than twice as likely to have a behavioral health diagnosis, be obese, or have 1 or more CCCs compared with non–high-cost patients. Many high-cost patients (44.7%) did not have a CCC. Those without a CCC fit largely into 2 categories: those with chronic conditions not included in Feudtner et al’s²⁰ categorization scheme (eg, mental health disorders, hemophilia, short stature), and those with acute or time-limited but expensive conditions ranging from complex fractures and surgical emergencies to pregnancy. Table 3 shows the results of a multivariable model of predictors of being a high-cost adolescent. After adjustment for age, sex, and median income by zip code, having 2 or more CCCs conferred a 23.5-fold risk of being a high-cost patient (95% CI, 14.2-39.1); having 1 CCC, a 6.5-fold risk (95% CI, 4.7-9.0); having a behavioral health diagnosis, a 3.6-fold risk (95% CI, 2.6-5.1); and being obese, a 2.0-fold risk (95% CI, 1.3-3.0).

The AHRQ clinical categories associated with the most TDME among the high-cost adolescent cohort were congenital ($1.8 million); endocrine, metabolic, and immunity ($1.7 million); and mental illness ($1.2 million) (Table 1). The most common conditions within the congenital category were congenital heart disease, Down syndrome, and spina bifida. Inpatient hospitalizations were the greatest driver of expense within the congenital category (Table 1). The next most expensive AHRQ category was endocrine, metabolic, and immunity, with the most common specific condition being short stature (n = 10). Pharmaceutical costs within this category accounted for $1.3 million, 77.0% of the total cost for this category and more than the total amount spent on the next most expensive AHRQ category, mental illness, with total expenses of $1.2 million. Mental illness was the category with the largest number of patients in the high-cost cohort (n = 22); the most common specific condition among these patients was mood disorders (n = 14). The most expensive AHRQ clinical category per individual patient was hematology, with an average TDME of more than $230 000 per patient (Table 1). Most (89.1%) of these costs were pharmacy costs.

In this analysis of TDMEs of a large population of privately insured adolescent patients, we found that the 1.0% of patients who incurred the highest charges accounted for almost one-fourth of costs, a skewed distribution similar to other pediatric studies of both privately and publicly insured patients.^10,12-14 The source of costs varied widely by clinical condition, with hospitalization and pharmacy costs driving expense for the highest-cost adolescents. Orphan drugs in particular contributed disproportionately to these costs. Having 2 or more CCCs was the strongest predictor of being a high-cost patient, although almost half of the high-cost patients in our cohort did not have a CCC.

Understanding predictors of high-cost health care is increasingly important owing to shifts in insurance benefit design that increase the cost-sharing burden to families. Stiles et al recently outlined the need for “predictive models to identify children with complex medical conditions whose care may be ‘impactable’ by improving care quality and reducing costs.”^21(p204) Increasing quality and decreasing expense for adolescents will require tailored approaches. The skewed distribution of costs makes the idea of care coordination for a small number of adolescents attractive, but in many cases, high costs may not be modifiable unless pharmacy costs can be decreased. Pediatric populations may be particularly vulnerable to high pharmacy costs because of orphan drugs. A drug is eligible for orphan drug designation from the US Food and Drug Administration if it is intended to treat a condition that affects fewer than 200 000 people in the United States or there is “no reasonable expectation that the cost of developing and making available a drug for the disease or condition will be recovered from sales of the drug.”²² These drugs have benefited children with rare disorders but account for a disproportionate amount of pharmacy costs. As of July 2014, orphan drugs are specifically excluded from the 340B discount drug pricing program for children’s hospitals and safety-net hospitals, so public policy or private insurer negotiation with individual pharmaceutical companies is potentially the only means of lowering these expenses for pediatric patients.²³

By contrast, patients with expenses incurred by frequent hospitalization and non-PCP outpatient visits could be modifiable via care coordination. Patients with congenital conditions and mental illness had the highest hospitalization costs in our cohort. Patients with congenital conditions also had the highest non-PCP outpatient costs, followed by those with neoplasms. Our finding that primary care services accounted for only a tiny fraction of the TDME of the highest-cost adolescents is consistent with a recent study demonstrating that primary care services represented only 2.2% of spending for Medicaid-insured medically complex pediatric patients.¹¹ Our data suggest that PCPs do not currently have a large role in providing care coordination for the highest-cost adolescents (or at least are not reimbursed for it), but it is unclear whether PCP coordination would reduce expenses in patients with rare diseases using orphan drugs or patients without CCC who experience acute injury, illness, or pregnancy.

The fact that mental health conditions were prevalent among high-cost adolescents suggests that mental health care professionals must be included in case management models to reduce medical expenses. There is significant unmet need for care coordination for children with behavioral health diagnoses.²⁴ Some studies suggest that intensive case management improves clinical outcomes and reduces hospitalization for adults with severe mental illness, and some smaller studies demonstrate cost-effectiveness of integrated behavioral health care.^25-27 New care models that integrate behavioral health care professionals into pediatric primary care practices have the potential to demonstrate significant cost-effectiveness if hospitalizations can be reduced.

Our data suggest that some strategies for cost reduction may yield only low savings in practices with high numbers of privately insured adolescents, such as initiatives centered on primary care prescribing, which accounted for the lowest percentage of spending for high-cost and non–high-cost patients. Relatively little TDME in this cohort of adolescents was related to the use of emergency services, suggesting that efforts to reduce emergency department use may also yield only low savings for some patient populations. This finding stands in contrast to a recent study showing frequent emergency department use in publicly insured children and adolescents, although adolescents used emergency services significantly less than did younger children.²⁸ Because the patterns of health care use appear to be significantly different between younger children and adolescents and between privately and publicly insured adolescents, investigators should consider stratified or separate analyses for these groups. These kinds of analyses will be vital for health care professionals in pediatric accountable care models. It is also worth noting that for overall cost containment, adolescent health care expenses are dwarfed by those of neonates and the elderly in the last year of life.¹³

This study had several limitations. The cohort was limited to adolescents in pediatric primary care practices who were largely well. They live in a geographic area with access to a high concentration of pediatric subspecialists, including a large freestanding children’s hospital and many academic institutions with pediatric inpatient and specialty programs, which may increase costs and may not be generalizable to other areas. It is also probable that children with special health care needs and CCCs are underrepresented because their conditions may make them eligible for Medicaid. The data were derived from a billing database with limited demographic and clinical information, so we had little information about variables that are important determinants of health. For example, only 5.2% of our cohort had a claim including a diagnosis of obesity. Since the proportion of adolescents known to be obese is substantially higher than that in our study (9.9% in Massachusetts in 2011),²⁹ it is possible that our use of claims data underestimated the true rate of obesity in our population, although obesity is more prevalent in low-income families. Supplementing administrative data with clinical data in future studies may be a way to reduce such potential misclassifications. Another limitation is the identification of youth with special health care needs using coding data when there is no universal definition of special needs.³⁰ We chose to use the definition of CCCs used by Feudtner et al because this model has demonstrated utility in predicting morbidity and mortality and because we wanted to provide a model that was clinically relevant as well as cost informed.^20,31,32 For future studies of adolescents’ health care use and costs, investigators could consider a “CCC+” model that would add hemophilia, pregnancy, and mental health disorders to the categories proposed by Feudtner et al.

Future studies could take several directions. The group of adolescents with CCCs who were not in the highest TDME group warrants further study in searching for cost reduction strategies. Interestingly, while adolescents with 2 or more CCCs are overrepresented in the top 1.0% of the cohort, many of these young people are actually not in the top 1.0% of spending. We stand to learn from what is going well in terms of cost containment for these complex patients. Another next step will be to study longer-term persistence of these health care costs. Many adolescents’ conditions are self-limited by their nature, such as acute injuries, pregnancy, and short stature.

There is an urgent need for health care professionals to understand and proactively identify high-cost patients for clinical interventions. For privately insured adolescents, potential strategies for cost reduction include care coordination for those with CCCs and mental health diagnoses, integrated behavioral health care, and negotiation to reduce pharmacy costs, especially for orphan drugs.

Accepted for Publication: August 4, 2015.

Corresponding Author: Susan H. Gray, MD, Division of Adolescent and Young Adult Medicine, Boston Children’s Hospital, 333 Longwood Ave, Fifth Floor, Boston, MA 02115 (susan.gray2@childrens.harvard.edu).

Published Online: October 5, 2015. doi:10.1001/jamapediatrics.2015.2682.

Author Contributions: Dr Gray had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Gray, Emans, Woods, Berry, Vernacchio.

Acquisition, analysis, or interpretation of data: Gray, Trudell, Emans, Vernacchio.

Drafting of manuscript: Gray, Vernacchio.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Trudell, Vernacchio.

Study supervision: Gray, Emans, Woods, Berry, Vernacchio.

Conflict of Interest Disclosures: None reported.

Funding/Support: This study was partly supported by Leadership Education in Adolescent Health Training grant T71 MC00009 from the Maternal and Child Health Bureau, Health Resources and Services Administration (Drs Emans and Woods) and grant R21 HS023092 from the Agency for Healthcare Research and Quality (Dr Berry).

Role of the Funder/Sponsor: The funding source had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Additional Contributions: Barry Zallen, MD, Ann Suny, BA, MBA, and Nora Boukus, MA, Children’s Hospital Integrated Care Organization, and Urmi Bhaumik, MBBS, MS, ScD, Division of Adolescent and Young Adult Medicine and Office of Community Health consulted with us on the design of the pilot project that served as the basis of this investigation. They were not compensated for their contribution.