Variation in the structures, processes, and organization of healthcare services in the United States has been studied for decades. As the field of hospital medicine grows, contributions from hospitalists in the fields of health services research and quality improvement have been instrumental in advancing our understanding of the drivers of variation and the impact of this variation on the cost, quality, and outcomes of care delivered to hospitalized children.

The first population-based study of variation in pediatric healthcare was published by British researcher J. Allison Glover in 1938.¹ Starting early in the 20th century, rates of tonsillectomies among school children in England and Wales rose dramatically. By 1931, the surgery made up three-quarters of all procedures performed in children in the United Kingdom. Across boroughs, however, the proportion of children having their tonsils removed varied in some cases by tenfold, even among areas with similar populations. Seeking to understand why these rates were so variable, Glover closely examined the hospital and education health records from several neighboring boroughs and found significant disparities in the rates of tonsillectomy. For example, “well-to-do” vs. poor children were more likely to enter secondary school without their tonsils. However, neither socioeconomic status nor any other health service area or population characteristics explained why a child living in one borough had a higher probability of getting their tonsils removed than a similar child in a different borough. Put simply, Glover concluded that the variation “defie[d] explanation.”

In the 1960s, John Wennberg—then a researcher at the Dartmouth School of Medicine—would refer to this type of unexplained geographic variation as unwarranted, or variation in the utilization of health care services that cannot be explained by patient illness, patient preferences, or evidence, but rather indicates differences in health system performance. Taking advantage of newly implemented health data systems in Vermont and other areas of New England, Wennberg and his colleagues at Dartmouth and Johns Hopkins used advanced statistical methods to analyze patterns in the healthcare utilization among neighboring hospital service areas.^2,3 In what was a highly influential series of publications, the Dartmouth group reported wide geographic variation in hospital admissions, diagnostic testing, and surgical procedures—including a tenfold difference in rates of tonsillectomies. All findings were still highly significant after case-mix adjustments, including prevalence of disease, patient demographics, socioeconomic factors, and medical practice differences.²

Stemming from these early studies, Wennberg and colleagues would go on to found the Dartmouth Atlas Project (http://www.dartmouthatlas.org), a publically available series of reports providing comprehensive information and analysis about how healthcare resources (and healthcare quality) vary, even within very small areas, across the United States. The Atlas publications have played a significant role in moving forward the agenda to standardize the care delivered and improve the overall value of healthcare. Perhaps the most significant Dartmouth Atlas finding has been that greater healthcare spending does not always correlate with higher quality of care, and in some cases may be associated with care that is suboptimal.⁴

In many ways, the Atlas has formed the foundation for our understanding of variation in healthcare, but until recently had remained largely limited to analyses of Medicare data from the adult population. The Atlas Project recently investigated geographic variation in pediatric health care across Maine, New Hampshire, and Vermont using data from commercial insurers and Medicaid. The Dartmouth Atlas of Children’s Health Care in Northern New England⁵ mimicked the adult analyses, reporting rates of utilization and spending for primary care, inpatient care, emergency room care, advanced imaging, surgical care, and medication use across regions and hospitals.

In parallel with the growth of the Atlas, pioneers in hospital medicine have utilized health data systems in Boston, Rochester, and other US states to uncover wide variation in the management of inpatient conditions such as asthma, gastroenteritis, and lower respiratory tract infections.^6-8 More recently, the broader availability of patient data from sources such as the Health Care Utilization Project (HCUP) Kids’ Inpatient Database (KID), and the Pediatric Health Information Systems Database have provided researchers with the opportunity to examine variation from the perspective of nationally representative pediatric patient samples.^9-12

The challenge of much of the work surrounding practice variation lies in defining what constitutes “unwarranted” differences in care. Variation in practice can occur for many legitimate reasons, leading some to argue that variation is only unwarranted if it deviates from evidence-based standards of care or has a significantly negative impact on outcomes. By these definitions, much of the observed variation in healthcare could be considered unexplained as opposed to unwarranted. This is especially the case for variation in the care provided to children, since there are few conditions with indisputable standards of care. Regardless of whether it is considered unwarranted or unexplained, a general framework for understanding variation and generating hypotheses about its sources and significance has emerged from the literature, and involves grouping care into three major categories: (1) supply-sensitive, (2) preference-sensitive, and (3) effective.

Supply-sensitive care refers to healthcare services that appear to be utilized at higher rates in areas where there is greater availability of certain resources. In Wennberg’s early studies, this was demonstrated by observations that patients living in areas with greater bed capacity were more likely to be hospitalized, patients living in areas with more surgeons were more likely to have a surgical procedure, and patients living in areas with more radiology services were more likely to get x-rays or Computerized Tomography Scans (CTs).¹³ The primary driver of this type of variation is felt to be the subtle influence of service availability on clinician decision-making in cases where the tradeoffs are not as well defined and choices to admit or test are at the discretion of the physician.

An obvious negative effect of supply-sensitive variation is that it leads to greater resource utilization and higher per capita healthcare expenditures. Unfortunately, this high utilization has not been shown to lead to better outcomes or patient satisfaction, and often represents waste of healthcare dollars and lost opportunity costs. Many believe that with rapid growth of the healthcare system, greater system capacity will lead to rampant overuse of diagnostic and therapeutic services and may in fact be causing significant and as of yet under-measured patient harm.^14,15

Fee-for-service payment systems are felt to be largely responsible for what has been unchecked overuse of supply-sensitive care. Accordingly, proposed strategies for reducing variation and lowering these costs involve payment model reforms such as transitioning to bundled payment systems, pay-for-performance, and implementing penalties for high utilizers of discretionary services. Accountable care organizations (ACOs) also show great potential for reducing variation in the use of hospital services. In the ACO model, healthcare organizations contract to provide comprehensive care to a population of patients for a fixed per-member price, and payments for each hospitalization are shared across all members of the hospital system’s medical staff, including providers responsible for the outpatient care of the child. ACOs are paid on a capitated basis and/or for achieving certain quality/cost targets, and thus are incentivized to more carefully manage the quality and utilization of care. In their early stages of development, some ACOs appear to be successful at reducing cost while improving quality, but more needs to be learned about which organizational factors most contribute to a successful ACO model.¹⁶