Abstract
BACKGROUND: Claims-based measures of “low-value” pediatric services could facilitate the implementation of interventions to reduce the provision of potentially harmful services to children. However, few such measures have been developed.
METHODS: We developed claims-based measures of 20 services that typically do not improve child health according to evidence-based guidelines (eg, cough and cold medicines). Using these measures and claims from 4.4 million commercially insured US children in the 2014 Truven MarketScan Commercial Claims and Encounters database, we calculated the proportion of children who received at least 1 low-value pediatric service during the year, as well as total and out-of-pocket spending on these services. We report estimates based on "narrow" measures designed to only capture instances of service use that were low-value. To assess the sensitivity of results to measure specification, we also reported estimates based on "broad measures" designed to capture most instances of service use that were low-value.
RESULTS: According to the narrow measures, 9.6% of children in our sample received at least 1 of the 20 low-value services during the year, resulting in $27.0 million in spending, of which $9.2 million was paid out-of-pocket (33.9%). According to the broad measures, 14.0% of children in our sample received at least 1 of the 20 low-value services during the year.
CONCLUSIONS: According to a novel set of claims-based measures, at least 1 in 10 children in our sample received low-value pediatric services during 2014. Estimates of low-value pediatric service use may vary substantially with measure specification.
- ACO —
- accountable care organization
- ICD-9 —
- International Classification of Diseases, Ninth Revision
What’s Known on This Subject:
Claims-based measures of “low-value” pediatric services could facilitate the implementation of interventions to reduce the provision of unnecessary care to children. Few such measures have been developed. The extent and financial burden of low-value pediatric service use are unknown.
What This Study Adds:
Using novel claims-based measures of 20 low-value pediatric services and claims from 4.4 million commercially insured children, at least 1 in 10 children in our sample received a low-value pediatric service during 2014.
Recent high-profile initiatives such as Choosing Wisely have highlighted the importance of avoiding “low-value” pediatric health care services that typically do not improve child health.1 Reducing the use of these services could prevent iatrogenic harm (eg, radiation exposure from imaging tests)2 while decreasing unnecessary financial burden on families and the health care system.3–11 Improving the efficiency of pediatric care is an increasingly important policy goal given the rapid rise in pediatric health care expenditures12 as well as the proliferation of alternative payment models such as pediatric accountable care organizations (ACOs), which incentivize providers to judiciously use resources and enhance the quality of care.13–18
Large-scale analyses of overuse can be facilitated by measures that directly assess low-value pediatric service use in insurance claims databases.4–8 These measures exploit the relative strengths of claims data, including their large sample sizes, longitudinal nature, widespread availability, and inclusion of utilization across multiple providers and care settings.3,4 Using 1 such set of measures, for example, Schwartz et al4 found that 25% to 42% of elderly Medicare beneficiaries received at least 1 of 26 low-value services in 2009, resulting in $1.9 billion to $8.5 billion in spending. A subsequent study used these measures to report reduced use of certain low-value services after implementation of the Medicare Pioneer ACO model.6
To date, however, few claims-based measures of low-value pediatric services have been developed. Furthermore, the extent and financial burden of low-value pediatric service use are unknown. To address these gaps, we developed a novel set of claims-based measures for 20 low-value pediatric services that occur across a variety of conditions and settings. We used these measures and a large claims database to estimate use of and spending on the 20 services among 4.4 million commercially insured US children in 2014.
Methods
Construction of Low-Value Service Measures
In fall 2015, members of our research team of pediatricians and health services researchers compiled a candidate list of low-value pediatric services by reviewing >400 recommendations from the Choosing Wisely initiative, recommendations from the US Preventive Services Task Force and the United Kingdom’s National Institute for Health and Care Excellence, child-focused evidence reports from the Cochrane Collaboration, clinical practice guidelines published by US medical specialty societies such as the Infectious Diseases Society of America, and peer-reviewed literature.19–56 Based on this review, we identified several hundred low-value pediatric services, including services that cause more harm than benefit (eg, cough and cold medicines for young children) and services that typically do not improve child health (eg, Papanicolaou tests). From this list, we excluded services that could not be easily identified as low-value in claims due to the lack of necessary clinical information (eg, head imaging for minor head trauma), as well as services that were likely to be infrequent among children (eg, electroencephalograms for headache).
We ultimately selected 20 pediatric services that could be identified as low-value in claims data, including 6 diagnostic tests, 5 imaging tests, and 9 prescription drugs. We constructed claims-based measures of these 20 services based on data elements that are typically contained in US insurance claims data, including: International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnosis and procedure codes; Current Procedural Terminology codes; and demographic information. Our measures assessed low-value service use in primary care offices, hospital outpatient departments, urgent care centers, retail clinics, emergency departments, community hospitals, and academic children’s hospitals.
For the main analysis, we constructed “narrow measures” that included multiple restrictions to only capture instances of service use that were low-value, potentially at the expense of missing some instances of low-value service use. For many of the narrow measures, we employed a modified version of a widely used administrative algorithm to exclude services received by children with a “complex chronic condition” (eg, congenital anomalies, dependence on technology, cancer).57 These children are excluded from many clinical practice guidelines, and assessment of care appropriateness for these children may be difficult without detailed clinical information. Other restrictions were based on relevant studies, guidelines, and reports identifying each service as low-value (Table 1).19–56 The codes used in each measure are presented in Supplemental Information.
Claims-Based Measures of 20 Low-Value Pediatric Services
To assess whether the amount of detected low-value service use varies with measure specification, we also created broad versions of our 20 measures, following previously published studies in the adult population.4 These “broad measures” contained minimal restrictions and were designed to capture most instances of service use that were low-value, potentially at the expense of misclassifying some instances of appropriate service use as low-value.
For each measure, we defined a denominator population of children who could potentially receive the service (eg, children with a diagnosis of bronchiolitis for the low-value bronchiolitis measures). For denominator populations based on age cutoffs (eg, children aged <3 years), we used age as of January 1, 2014. In a sensitivity analysis, results were not substantially altered when using age as of December 31, 2014 (Supplemental Information).
Data Source and Study Design
We conducted a cross-sectional analysis of the 2014 Truven MarketScan Commercial Claims and Encounters database, a convenience sample of claims data from >47 million US residents aged 0 to 64 years with employer-sponsored private health insurance. Because many measures excluded children with specific diagnosis or procedure codes in previous claims, we used the 2013 MarketScan database as a “look-back period” for the 2014 analyses.
Study Population
There were 12.2 million children aged 0 to 18 years in the 2014 MarketScan database. To ensure a sufficiently long look-back period for each instance of service use, we limited the sample to children who were born before 2013 and continuously enrolled for 365 days in 2013 and 2014, or born in 2013 or 2014 and continuously enrolled until the end of 2014. Of the resulting 6 374 551 million children, we excluded 1 336 938 million children whose prescription drug claims were not included in the MarketScan database and an additional 632 843 children who were enrolled in capitated plans in any month during 2014 (because claims submitted by these plans may not reliably report payment).58–60 The final sample included 4 404 770 children, representing 5.6% of the 78.1 million US children in 2014 and 11.9% of the 36.9 million US children with employer-sponsored private insurance in 2014.61
Statistical Analyses
Summary statistics were calculated by using the demographic information available in the MarketScan enrollment file. Using the narrow and broad measures, we calculated the percentage of eligible children in the denominator population who received each service at least once during the year, the number of services received per 100 eligible children, and the percentage of children in the overall sample who received at least 1 of the 20 low-value services during the year. For each service and across all services, we also calculated out-of-pocket spending (the sum of coinsurance, copays, and deductibles) and total spending (allowed charges). To provide context for spending estimates, we limited the sample to children who received each service at least once in 2014, then calculated the proportion of annual health care spending among these individuals that was due to the service.
A single instance of service use can be represented by multiple claims in the MarketScan database due to separate billing of facility and professional components, claims adjustment, or billing error.58 For analyses of service use, we considered multiple instances of a service on the same day for a given individual to represent a single instance of service use. For spending analyses, we summed payment variables across all claims for a service on the same day for a given individual. This method of collapsing payment to the “claim-day” has been used in previous MarketScan analyses, including an Institute of Medicine report on geographic variation in spending among the privately insured US population.59,60
Analyses were performed by using SAS version 9.4 (SAS Institute, Inc, Cary, NC). The institutional review board of the University of Chicago exempted this study from review.
Results
Of the 4.4 million children in the sample, 29.3% were aged 0 to 5 years, 35.6% were aged 6 to 12 years, and 35.1% were aged 13 to 18 years; 51.1% were male; and 85.2% lived in an urban area. Our sample included children from all US states, Washington, DC, and Puerto Rico.
Table 2 and Fig 1 report the percentage of overall service counts, out-of-pocket spending, and total spending constituted by each category of low-value service use (diagnostic tests, imaging tests, and prescription drugs). Table 2 also reports the percentage of children in the sample who received at least 1 service in these categories during the year.
Use of and Spending on Low-Value Diagnostic Tests, Imaging Tests, and Prescription Drugs
Distribution of: A, low-value service counts, B, out-of-pocket spending on low-value services, and C, total spending on low-value services. Estimates are based on the narrow measures and represent the proportion of sample totals accounted for by low-value diagnostic tests, imaging tests, and prescription drugs.
According to the narrow measures, 37.9%, 2.8%, and 59.4% of all low-value services in our sample were for diagnostic tests, imaging tests, and prescription drugs, respectively (Table 2). Low-value imaging occurred relatively infrequently but accounted for 26.7% of all out-of-pocket spending and 33.7% of all spending on low-value services detected by using our measures. During the year, 3.9%, 0.4%, and 6.1% of children in the sample received at least 1 low-value diagnostic test, imaging test, and prescription drug. Overall, 9.6% of children in the sample received at least 1 of the 20 low-value services during the year. Total spending across all 20 services was $27.0 million, of which $9.2 million was paid out-of-pocket (33.9%).
According to the broad measures, 14.0% of children in the sample received at least 1 of the 20 low-value services during the year (Table 2). Total spending was $53.0 million, of which $16.3 million was paid out-of-pocket (30.8%).
Table 3 displays use and spending estimates for each of the 20 low-value pediatric services. Some services were received by a low percentage of eligible children but a large number of children overall due to the size of the denominator (eg, vitamin D testing), whereas other services were received by a high percentage of eligible children but a small number of children overall because the condition is rare (eg, ultrasound for cryptorchidism). Estimates of low-value service use varied significantly between the narrow and broad measures for some measures (eg, allergy testing for eczema) but not for others (eg, cervical cancer screening). Among the major categories of low-value pediatric services, we observed the highest rates of use for prescription drugs. Spending on each low-value service represented small proportions of annual health care spending among children who received the service during the year (median among 20 services, 0.7% for broad measures and 0.5% for narrow measures).
Use of and Spending on 20 Low-Value Pediatric Services
Due to the large sample size of our study, our estimates had very small 95% confidence intervals.62 For example, estimates of the percentage of children who received at least 1 low-value pediatric service during the year had 95% confidence intervals with a width <0.08 percentage point.
Discussion
Even though most children are healthy and have a limited number of interactions with the health care system, at least 9.6% of children in our sample received at least 1 of 20 low-value pediatric services during 2014, consistent with the notion that waste in pediatrics may be widespread. Although previous literature has documented significant underuse of high-value pediatric services such as immunizations,63 our findings suggest that overuse of low-value services may be an equally pressing deficiency in pediatric care quality. Our findings are similar to those of previous studies showing frequent use of low-value services among adults,3–11 which suggests that interventions to improve care efficiency should include patients across the entire age spectrum.
This study highlights the clinical and policy importance of reducing the use of low-value pediatric services. Most importantly, the services assessed in our study can directly harm children both in the short term (eg, side effects from antibiotics) and the long term (eg, increased lifetime incidence of cancer from unnecessary imaging). Furthermore, one-third of all spending on low-value pediatric services in our study was paid out-of-pocket, suggesting that use of these services could lead to unnecessary financial burden for families exposed to cost-sharing. Finally, our findings suggest that reducing use of low-value pediatric services could substantially decrease health care spending.10 Annual spending on low-value pediatric services totaled $27.0 million in our sample, which included 11.9% of all commercially insured children in the United States.61 Under the strong assumption that our sample is representative of this population, total annual spending on the 20 low-value pediatric services is roughly $227 million just for commercially insured children alone.
Our use of direct claims-based measures differs from the approach of previous pediatric overuse studies, most of which have indirectly assessed waste by documenting unexplained variation in care patterns between regions and hospitals.64–66 This variation-based approach is useful for setting target rates of utilization based on a percentile rate, or “achievable benchmark of care.”66 However, it is difficult to determine appropriateness based on relative rates of utilization, especially if case mix differences between comparison groups cannot be adjusted for with available data or if waste frequently occurs even among providers with the lowest utilization rates.3,4 In contrast, direct approaches produce estimates of the absolute level of low-value service use, a quantity that is easier to interpret in isolation.
Conversely, a potential challenge of using claims-based measures to assess overuse may be the sensitivity of estimates to measure specification. In our study, we frequently observed a wide range of estimates when using narrow and broad versions of measures, illustrating a fundamental tradeoff when measuring overuse in claims: overly narrow measures may only capture a small fraction of all service use that is low-value, whereas overly broad measures may capture a significant amount of service use that is high-value.4 This scenario suggests that for each application of claims-based measures of low-value services, organizations and researchers must carefully tailor the measures to their measurement goals.
Despite this challenge, claims-based measures of low-value services could be essential tools in several types of pediatric quality improvement efforts,3 including pay-for-performance initiatives. For example, Massachusetts provider groups participating in the Alternative Quality Contract received a global budget and additional financial payments based on improvements in performance on several claims-based quality measures, including 1 measure assessing low-value antibiotic prescriptions.14 However, although payment could be carefully linked to aggregate performance on claims-based measures at the level of large provider groups, we would caution against using these measures to summarily deny payment for individual instances of apparent low-value service use, as claims may be unable to fully account for the heterogeneous and idiosyncratic clinical circumstances leading to service use in each health care encounter.67
Our study has a number of limitations. First, as with any administrative data analysis, we did not have access to potentially important clinical information that might influence assessments of appropriateness, which could lead to potential misclassification of services as low-value even when using the narrow measures. Although future efforts such as chart reviews may help quantify the reliability of claims-based measures, we also note that in many research settings, the impact of any misclassification bias can be attenuated by including additional restrictions or by using strong study designs. For example, 1 study used a difference-in-differences design and claims-based measures to evaluate the impact of the Medicare Pioneer ACO model on low-value service use.6 In this study, any significant differential changes in low-value service use were unlikely to be driven by misclassification bias unless the change in this bias before and after implementation of the ACO model differed in magnitude between the non-ACO and ACO groups. Analyses of trends in low-value pediatric service use will also not be affected by any misclassification bias unless the magnitude of this bias changes over time.
Second, it is unclear whether our results generalize to other pediatric populations, including publicly insured children. It is possible that publicly insured children receive fewer low-value services due to access barriers that result in less care overall; however, it is also possible that these children receive more low-value services because of systematic disparities in the quality of their care. Third, condition-based measures that use diagnosis codes as inclusion criteria may miss instances of low-value service use if the condition was not appropriately coded in claims (eg, if acute otitis media with effusion was coded as acute otitis media, not otherwise specified). Fourth, although we relied on recommendations from several high-profile organizations to classify a service as low-value, we acknowledge that some providers may have different perceptions of the utility of some of the services we assessed.
Finally, we only assessed 20 services, which we carefully selected from a large list of candidates primarily on the basis of whether they could be identified as low-value in claims. As such, our findings undoubtedly underestimate use of and spending on low-value pediatric services, including other services that could potentially be measured with additional claims-based measures as well as services that cannot be easily classified as low value by using claims data. We also note that our spending estimates do not account for any downstream events associated with low-value service use, including immediate events (eg, follow-up testing for false-positive initial results) and events that may occur much later (eg, use of broad-spectrum antibiotics due to antimicrobial resistance from antibiotic overuse).68–70 To more fully capture the scope of low-value service use in pediatrics, future studies should quantify these downstream costs, identify costly low-value interventions for children with complex chronic conditions,57 and assess low-value applications of lucrative elective procedures that may not have been included by medical specialty societies participating in Choosing Wisely.71
Conclusions
Overuse of low-value services may be widespread in pediatrics. As health care systems increasingly incorporate payment and delivery models that prioritize value, claims-based measures of low-value pediatric services could facilitate the implementation of interventions to reduce the provision of potentially harmful services to children.
Acknowledgment
The authors thank Mona Sharifi, MD, MPH, for her helpful comments on the manuscript.
Footnotes
- Accepted September 27, 2016.
- Address correspondence to Kao-Ping Chua, MD, PhD, 5841 S. Maryland Ave, AMB 241 W241, MC 2000, Chicago, IL 60637. E-mail: kchua{at}peds.bsd.uchicago.edu
FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.
FUNDING: Dr Schwartz is supported by a training grant from the National Institute on Aging (F30 AG044106). Dr Huang is supported by a grant from the National Institute of Diabetes and Digestive and Kidney Diseases (K24 DK105340). Supported by the National Institutes of Health (NIH).
POTENTIAL CONFLICT OF INTEREST: Dr Schwartz reports consulting on quality measurement for Nuna Inc; the other authors have indicated they have no potential conflicts of interest to disclose.
COMPANION PAPER: A companion to this article can be found online at www.pediatrics.org/cgi/doi/10.1542/peds.2016-3228.
References
- Copyright © 2016 by the American Academy of Pediatrics