BACKGROUND: Although children with medical complexity have high health care needs, little is known about the variation in care provided between centers. This information may be particularly useful in identifying opportunities to improve quality and reduce costs.
METHODS: We conducted a retrospective population-based observational cohort study using all payer claims databases for children aged 30 days to <18 years residing in Maine, New Hampshire, and Vermont from 2007 to 2010. We identified hospital-affiliated cohorts (n = 6) of patients (n = 8216) with medical complexity by using diagnostic codes from both inpatient and outpatient claims. Children were assigned to the hospital where they received the most inpatient days, or their outpatient visits if no hospitalization occurred. Outcomes of interest included patient encounters, medical imaging, and diagnostic testing. Adjusted relative rates were calculated with overdispersed Poisson regression models.
RESULTS: Adjusting for patient characteristics, the number of inpatient (relative rate 0.84 vs 2.28) and intensive care days (relative rate 0.45 vs 1.28) varied by more than twofold, whereas office (relative rate 0.77 vs 1.12) and emergency department visits (relative rate 0.71 vs 1.37) varied to a lesser extent. There was also marked variation in the use of imaging, and other diagnostic tests, with particularly high variation in electrocardiography (relative rate 0.35 vs 2.81) and head MRI (relative rate 0.72 vs 2.12).
CONCLUSIONS: Depending on where they receive care, children with medical complexity experience widely different patterns of utilization. These findings indicate the need for identifying best practices for this growing patient population.
- CMC —
- children with medical complexity
- CPT —
- Current Procedural Terminology
- CT —
- computed tomography
- ED —
- emergency department
- HSA —
- hospital service area
What’s Known on This Subject:
Children with medical complexity require a disproportionate amount of health services due to a multitude of chronic severe illness, and their impact on the health care system appears to be increasing.
What This Study Adds:
This study provides one of the first comparisons of health care utilization patterns for children with medical complexity between medical centers in a population-based cohort.
Children with life-altering chronic illnesses represent a small percentage of the pediatric population but account for a large proportion of health care utilization, including as much as 40% of pediatric hospital charges.1 These children are often referred to as children with medical complexity (CMC),2 or children with complex chronic conditions,3 and they make up an important subset of children with special health care needs.4 National data suggest that this cohort of children has grown substantially in recent decades, likely due to increased survival.5 Programs for coordinating the care of these children are increasingly common, although data on the effect of such programs on health care utilization is just beginning to be systematically evaluated.6
Conceptualizing and defining this cohort has been challenging because these children generally have a wide variety of diagnoses, functional limitations, or technology dependence. The initial conceptual definition included children with “any medical condition that can be reasonably expected to last at least 12 months (unless death intervenes) and to involve either several different organ systems or one system severely enough to require specialty pediatric care and probably some period of hospitalization in a tertiary care center.”3 The most commonly used operational definition emerged from a population-based study of childhood deaths in Washington.3 This definition was further refined and applied to the inpatient setting by using a large national database of children’s hospital discharges.2,5
Our current understanding of the health care utilization patterns of children with medical complexity is limited in 2 important ways. The first is that, apart from a few studies,7–9 most have sought to characterize the utilization of inpatient services. Although hospital care represents the most costly care provided to CMC, it fails to capture the entire spectrum of care received by children and their families. Nor are studies of hospitalized cohorts able to identify upstream services that might contribute to, or, alternatively, help prevent hospital admissions, such as diagnostic testing. What has generally been missing is a population-based view across the full continuum of care. Additionally, there has been little effort to study variation in the care received by this cohort across health systems to compare the quality, costs, and intensity of services.
The aim of the current study was to describe patterns of care for a population-based cohort of children with medical complexity in northern New England. In this study, we identify and characterize CMC from the resident pediatric population of Maine, Vermont, and New Hampshire and present utilization rates for children receiving care in 4 academic children’s hospitals and 2 regional hospitals.
Study Design and Data Sources
This was a retrospective observational cohort study using an all payer claims database for children aged 30 days to <18 completed years of age residing in Maine, New Hampshire, and Vermont for the period from 2007 to 2010. The database drew from 6 sources: the commercial all payer claims database and the Medicaid claims databases from each of the 3 states, which included children with 1 or more months of enrollment in a plan meeting state reporting requirements. Reporting requirements differed by states for commercial insurance plans. For example, Maine exempted plans with <50 covered lives, and New Hampshire required reporting only by insurance plans licensed in the state. Maine Medicaid data were available only for 2007 to 2009. Also, ∼10% of children in the region were uninsured.
This cohort of children with medical complexity was identified beginning with the definition of Feudtner as refined by Simon et al.1 This definition was based on International Classification of Diseases, Ninth Revision (ICD-9) codes (Supplemental Table 3) and technology dependence codes for tracheostomy, gastrostomy, or ventriculoperitoneal shunt listed on insurance claims. Patients were eligible for our study if they had claims submitted for any hospitalization or at least 2 separate outpatient visits for these codes during the 4-year study period. Based on initial review of the cohort, we further refined our definition by excluding the codes for isolated atrial and ventricular septal defect, isolated hypercholesterolemia without an inborn error of metabolism, and codes for a single bone and joint anomaly, as these codes alone were common and were not felt to adequately represent medical complexity. To associate each patient with a diagnostic code, the codes for technology dependence alone were not considered qualifying diagnoses to enter the cohort, but were considered comorbid conditions for the purposes of adjustment. Finally, certain specialized procedures were available only at a single hospital in the region; therefore, we excluded from our cohort patients undergoing solid organ or bone marrow transplantation, or repair of congenital heart disease. These exclusions were chosen to ensure hospitals were analyzed on services for which care was available equivalently. Because of the centralized nature of children’s services in the region, we analyzed only health care utilization associated with children’s and regional hospitals with >500 cohort members. Hospitals with a full range of subspecialty services (Supplemental Table 4) and who were members of the Children’s Hospital Association were classified as children’s hospitals.
Children could enter the cohort based on diagnoses assigned in the first month of life, but we did not measure utilization until the second month of life.
Patients residing in the 3 study states were attributed to hospitals based on where they received most of their care after initial diagnosis. Those with inpatient claims were assigned to the hospital with the most inpatient days. Patients with only outpatient claims were assigned to the hospital of the provider hospital service area (HSA) where they had the most number of physician visits claims, with ties going to the provider HSA of the most recent claim. HSAs are relatively self-contained geographic health care markets in which the resident population receives most of their inpatient care from within area hospitals. They were initially defined using Medicare data from adults, but have been demonstrated to be useful for the pediatric population, with localization indices generally comparable with that of the Medicare population.10 Localization indices are calculated as the percentage of health care events for children assigned to a particular HSA provided by clinicians practicing within that same HSA. For this study, localization measures, reflecting the loyalty of CMC to the attributed hospital, were calculated for hospitalizations, outpatient visits, and emergency department (ED) visits. We report on 1 freestanding children’s hospital, 3 children’s hospitals within larger academic medical centers, and 2 regional hospitals. For the outpatient claims attributed to the freestanding children’s hospital, we further refined the HSA method to include only the zip code in which the outpatient claim occurred due to the presence of multiple competing specialized pediatric programs within that HSA. Children meeting the cohort definition who were attributed to community hospitals in the region were assigned to a residual group and were excluded from our analysis of utilization rates; however, a sensitivity analysis also was performed to assess the effects of this exclusion.
We analyzed utilization for the CMC population residing in the 3 study states by hospital events, physician events, medical imaging, and diagnostic testing. Events of interest were defined by using Current Procedural Terminology (CPT) codes as reported in the claims databases. Inpatient days included patients classified as inpatient or observation status. Months of observation used for rate denominators were calculated as the month of medically complex diagnosis and all following months of enrollment until December 2010 or when patients turned 18. Neonatal intensive care CPT codes were not included because we were unable to distinguish hospitalizations due to prematurity from those due to medical complexity; and, utilization of studies included under bundled intensive care payments also were not included.
Patients were assigned to diagnostic categories based on their longest length of stay with a medically complex diagnosis. In the absence of hospitalization, we used the most frequent diagnosis from outpatient visits. The categories chosen were malignancy, neuromuscular conditions, cardiovascular, respiratory, congenital anomalies, and other. In the case of ties, preference went to malignancy, neuromuscular diagnoses, cardiovascular, respiratory, and congenital anomalies in that order. Median household income by zip code was provided by the 2009 US Census.
Statistical significance of differences in patient characteristics was tested with the Pearson χ2 test. We then calculated utilization crude rates by the assigned hospitals. Adjusted relative rates were estimated with overdispersed Poisson regression models, adjusted for differences in age, gender, payer (commercial or Medicaid), median household income, primary condition, the presence of multiple conditions, and the presence of technology-dependent procedures. Reference groups were chosen for each event based on the hospital whose rate of usage most closely matched the overall rate of usage across all 6 hospitals; 95th percentile confidence limits were calculated for all variables. Analysis was conducted by using SAS version 9.3 (SAS Institute, Inc, Cary, NC).
We calculated travel time measures for each zip code of patient residence to the zip code of the hospital assigned. The origins and destinations were the postal delivery centroids. The quickest road-based route was calculated by using ESRI (Redlands, CA) Network Analyst function within ArcGIS. Median travel times are presented for patients attributed to each hospital.
This study was approved by the Dartmouth College Committee for the Protection of Human Subjects.
Table 1 presents the demographic information and diagnostic categories for the entire cohort of CMC (n = 11 804) and by each of the 6 hospitals (n = 8216) primarily serving the study population in the region. The residual category includes patients (n = 3588) who were assigned to community hospitals with limited pediatric services and <500 cohort members. Hospital A was a quaternary referral center for the region and was the only hospital analyzed that provided a comprehensive management program for children with medical complexity, so it is possible that their patient cohort includes more severely affected children even after adjustment. As for all hospitals, it should be noted that the rates and relative rates for the hospital A cohort only reflect the utilization of children (n = 885) attributed to this hospital from the 3-state study region.
Patient characteristics differed across the children’s hospitals (Table 1) (note that these are the characteristics of study children only, not all CMC at each hospital). For example, children assigned to hospital A were more likely to reside in zip codes with higher median household income. Certain diagnostic categories were more likely in patients assigned to specific hospitals, such as neuromuscular diagnoses at hospital C and cardiovascular diagnoses at hospital A. In our analyses of utilization rates, these characteristics were used to control for severity of illness, including the presence of multiple conditions and technology dependence.
Table 2 presents localization indices for CMC assigned to each hospital by discharges, office visits, and ED visits. These measures indicate the proportion of utilization that occurred at the hospital to which the patient was assigned. These were quite high in general, although ED visit localization for hospitals C and A were lower than hospitalization and office visits, reflecting that the populations attributed to these hospitals had farther to travel (Table 1) and were likely to have sought a greater percentage of their urgent care closer to home.
We found marked variation in crude and adjusted utilization rates (Figs 1, 2, and 3). Generally, the adjusted rates did not alter the rank order of the rates by hospital and the sensitivity analysis did not alter the rank order nor the magnitude of the variation (Supplemental Tables 5, 6, and 7). Figure 1 presents hospital utilization and outpatient visits. In this analysis, we note higher utilization of inpatient days for patients assigned to the quaternary referral center, hospital A, and for hospital F, a regional facility. Also of note, despite the hospital C cohort’s rate of inpatient days falling close to and between hospitals B and D, its use of ICU days was approximately twice as high, appearing more similar to hospital A’s cohort.
Figure 2 presents plain radiograph, computed tomography (CT), and MRI utilization, and Fig 3 presents a sample of other testing. Plain radiographs, specifically chest radiography, demonstrated the lowest rates of variation. Head MRI and chest and abdominal CT usage demonstrated the widest variation, with 3 to 4 times as many imaging procedures obtained for the cohort assigned at the highest utilization site as the lowest sites. Of note, the hospital A cohort, despite being primarily cared for at the quaternary referral center, did not necessarily have the highest rates of utilization of imaging or testing. For instance, hospital C’s cohort received more than twice the number of head MRIs as other hospital cohorts, whereas the hospital B cohort used more head CTs than most other hospitals. Hospital A’s cohort, the only center doing the full range of cardiac surgery, received the highest rates of cardiac testing even after our exclusions and adjustment for the volume of cardiac diagnoses. However, renal ultrasound and EEG usage were more than twice as common in the hospital C cohort.
Our study represents the first analysis that we know of to examine between-center variation in the care for children with medical complexity. These markedly different patterns are unlikely to be primarily explained by population differences in the cohort. Unlike much of the United States, northern New England is strikingly homogeneous in its ethnicity and socioeconomic population characteristics. Although 1 of the 4 children’s hospitals has a higher range of services than the other 3, we limited the cohort to diagnoses and services for which all 4 children’s hospitals offered services and included covariates in the models to adjust for illness severity and diagnosis. Additionally, as noted in our results, hospital A was not a consistently positive outlier. Moreover, the children’s hospital cohorts did not consistently have higher adjusted utilization measures compared with regional hospital cohorts. Although we cannot exclude differential selection of hospitals by parents, we note that travel times are significant in this region and each of the academic medical centers are often the only provider in the entire state for a significant percentage of pediatric subspecialty care. Even if parents were willing to travel farther to choose a center perceived as better, the geography is such that travel times could often double and there are potential insurance barriers to seeking out-of-state care. Finally, the adjusted rates generally differed little from the crude rates, suggesting that population differences are likely to be limited.
Although there is no single explanation for the magnitude of observed variation in health care utilization, previous studies have found that uncertainty in medical evidence, a lack of professional consensus in the value of diagnostic and treatment procedures, and local capacity all contribute to differing practice styles.11,12 These possibilities are consistent with the patterns observed in our study. One might note the very high rates of head MRI and EEG usage at hospital C even after adjusting for neuromuscular diagnoses and conclude that this hospital must be a center of excellence for this type of care. A plausible, although unproven, explanation is that these hospitals have differing thresholds for ordering tests on similar patients. Such hospital-specific patterns of testing are seen in studies of adult Medicare patients and are often related to the local supply of subspecialty care, imaging capabilities, and local physician practice patterns.11 The lower rate of neurologic testing rates at hospital A, the quaternary referral center, to hospital C highlights the unexpectedly high rate of this particular type of testing at hospital C and supports the thesis that such testing is discretionary.
Our study is not designed to test for the potential causes of unwarranted variation in pediatrics, but there are some intriguing patterns. For instance, the utilization of chest radiography demonstrates tighter clustering around a central tendency suggesting greater certainty in clinical indication for the test than in the use of head CT or MRI. Additionally, the actual magnitude of variation may be greater than reported in our adjusted measures. Studies have demonstrated that patients living in areas of high care intensity are seen by their physicians more often and tend to accumulate more diagnoses despite comparable health status.13 If this form of coding, or observational intensity, bias were operating in our cohort, then our adjustments would minimize the true underlying variation in utilization.
Ideally, data such as these could be used to evaluate directed management programs for this cohort of high resource using patients. Several groups have described their experience improving the management of such patients clustered around specialized children’s hospitals, each reporting some measure of decreased resource utilization associated with their program.14–19 One group further described their experience of moving the locus of care into community hospitals, achieving further cost savings.19 As these programs disseminate, the ability to compare their approaches, outcomes, and resource utilization between centers using population-based data would help every center identify best practices sooner. Furthermore, new payment models, such as Accountable Care Organizations, developed and tested in the adult population are increasingly being extended to pediatrics,20 although our field has lagged in our ability to accurately track the medical care received by pediatric patients at the population level. We also lack validated measures on which to assess improved patient outcomes for this population.
As with any claims-based analysis, our study has limitations in the information to classify severity of illness and the potential for coding and selection bias. Analyses using registry data may more accurately capture illness severity; however, the inclusion criteria for most registries are condition-dependent and thus prevent a population-based study. In addition, our methods of adjustment for factors other than the diagnosis may have been inadequate. Other methods of identifying and classifying children with medical complexity have been developed, including a newly developed algorithm designed to better stratify by severity.21 There are also proprietary software algorithms available.22,23 Our method of assignment to hospitals may have misattributed some patients, although we did note high localization indices. Furthermore, our analysis does not distinguish whether the variation in care is driven by inpatient, outpatient, or emergency care and cannot link the variation encountered to patient outcomes. Finally, our analysis was limited to children residing in 3 states in northern New England. Although the high degree population homogeneity is useful for internal validity, it limits the generalizability of our findings.
Hospital-specific resource utilization in cohorts of children with medical complexity exhibits a high degree of variation, particularly around high-cost, discretionary services. These findings indicate the need for identifying best practices for this growing patient population.
- Accepted August 4, 2015.
- Address correspondence to Shawn L. Ralston, MD, Children’s Hospital at Dartmouth, 1 Medical Dr, Lebanon, NH 03745. E-mail:
Dr Ralston designed the study, and drafted the initial manuscript; Mr Harrison critically reviewed all analyses, prepared all tables and figures, and reviewed and revised the manuscript; Mr Wasserman designed the data collection instruments, performed the analysis, and critically reviewed the manuscript; Dr Goodman conceptualized the original cohort, revised the study design and data collection, supervised the analysis, and reviewed and revised all portions of the work; and all authors approved the final manuscript as submitted.
FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.
FUNDING: This study was funded by the Charles H. Hood Foundation and the Robert Wood Johnson Foundation.
POTENTIAL CONFLICT OF INTEREST: The authors have indicated they have no potential conflicts of interest to disclose.
COMPANION PAPER: A companion to this article can be found on pages 974, and online at www.pediatrics.org/cgi/doi/10.1542/peds.2015-0440.
- Simon TD,
- Berry J,
- Feudtner C,
- et al
- Cohen E,
- Kuo DZ,
- Agrawal R,
- et al
- Feudtner C,
- Christakis DA,
- Connell FA
- McPherson M,
- Arango P,
- Fox H,
- et al
- Burns KH,
- Casey PH,
- Lyle RE,
- Bird TM,
- Fussell JJ,
- Robbins JM
- Berry JG,
- Hall M,
- Neff J,
- et al
- Goodman DC,
- Morden NE,
- Ralston SL,
- Chang CH,
- Parker DM,
- Weinsten SJ
- Wennberg JE
- ↵American Academy of Pediatrics. Accountable care organizations (ACOs) and pediatricians: evaluation and engagement. Available at: https://www.aap.org/en-us/professional-resources/practice-support/Pages/Accountable-Care-Organizations-and-Pediatricians-Evaluation-and-Engagement.aspx. Accessed March 19, 2015.
- ↵Simon TD, Cawthon ML, Stanford S, et al. Pediatric medical complexity algorithm: a new method to stratify children by medical complexity. Pediatrics. 2014;133(6). Available at: www.pediatrics.org/cgi/content/full/133/6/e1647.
- Copyright © 2015 by the American Academy of Pediatrics