Variation and Drivers of Costs for Emergency Department Visits Among Children in 8 States Free

We performed a cross-sectional study of EDs from 2014 to 2018 in Arizona, Florida, Iowa, Maryland, Nevada, New Jersey, New York, and Wisconsin. We excluded hospitals that were nonresponders to the American Hospital Association (AHA) survey for all study years and those that self-identified as a facility solely dedicated to the care of subspecialty, surgical, long-term care, or rehabilitation patients in 1 or more data years. We also excluded hospitals that had fewer than 50 ED encounters for children per year. Within each hospital, we included all patients under 18 years. Encounters resulting in transfer from the ED (1.3%), discharge against medical advice (0.6%), or death (0.02%) were dropped because a complete episode of care did not occur. This study was exempted from review by the local Institutional Review Board.

ED diagnosis, charge data, and hospital cost-to-charge ratios were obtained from the Agency for Health Care Research and Quality’s Health Care Cost and Utilization Project State Inpatient and ED Databases (SID and SEDD). Hospital characteristics were obtained from the 2014 to 2018 AHA Annual Survey. The AHA survey is a self-report survey sent annually to all US hospitals and has a high response rate; 88% of hospitals had at least 1 response during the study period. The AHA survey is linkable to the SID and SEDD. Regional population counts were obtained from the 2018 5-year American Community Survey.

The main exposures were each ED’s region (the market) and state (the policy environment). We defined regions based on hospital referral regions from the Dartmouth Atlas of Health Care.²⁰  Although hospital referral regions can cross state borders, they are primarily based in a single state, so we treated a given ED’s state as that of its region. Thus, for model specification, hospitals were nested within regions, and regions were nested within states.

The primary outcome was the ED costs determined as the ED-level mean inflation-adjusted ED costs for children. Inflation adjustments were drawn from the Center for Medicare and Medicaid Services National Health Care Expenditures price index inflation values.²¹ 

Underlying our analysis was a conceptual model that divided potential cost drivers as operational or contextual.²²  Operational factors were nonmodifiable drivers of costs, and contextual factors were potentially modifiable cost drivers. Operational factors included ED case mix of children and wage index. Case mix is a measure of complexity, severity, and resource needs of the particular patient population for an ED. To measure case mix, we used the Hospitalization Resource Intensity Score for Kids (H-RISK).²³  The score assigns a relative weight based on demographics, diagnoses, and procedures codes for an encounter. H-RISK was assigned to each encounter and a hospital-specific H-RISK was calculated as the mean across encounters. An H-RISK score of 1.0 represents the average cost for an inpatient discharge across all encounters. The other operational factor was wage index, which measures the relative cost for paying staff wages in the region, which is the largest aggregated cost in providing health care.²⁴  Regional health care wage indices were obtained from the Center for Medicare and Medicaid Services.²¹  A regional wage index of 1.0 represents the average for hourly health care wages across the entire country.

Contextual factors included pediatric volume (mean annual ED visits by children under 18 years), ownership status (for-profit, not-for-profit, nonfederal government), critical access designation, trauma level (as defined by the American College of Surgeons), presence of a pediatric ED, inpatient pediatric bed count quartiles, overall Medicaid share (for adults and children), number of regional hospitals per capita, and teaching hospital status (defined by membership in the Council of Teaching Hospitals or both of the following: association with a medical school and resident full time equivalent to inpatient bed ratio > 0.25).²⁵  We also evaluated regional market concentration using the Herfindahl-Hirschman index,²⁶  which estimates local market competitiveness, with smaller Herfindahl-Hirschman index values indicating an even distribution of encounters to hospitals throughout the region.

To evaluate variability in costs, we determined the median and interquartile range (IQR) of EDs’ costs by region and state. The regions and EDs with the highest and lowest costs were determined.

We then created 2 linear mixed models to evaluate variability and drivers of ED costs: an operational model including only the 2 operational factors as fixed effects, and a contextual model including both operational and contextual factors as fixed effects. The dependent variable for both models was the log of hospital-level mean ED costs. Log costs were used because raw costs were right skewed, but log costs were normally distributed (Kolmogorov-Smirnov P = .55). Random intercepts at the region and state levels were included in both models.

To evaluate sources of cost variability, we examined the proportion of cost variance attributable to each of the 3 levels (hospitals, regions, and states) from the contextual model. To determine hospital- and region-level characteristics associated with costs, we report the fixed effects from the contextual model. Effect size for each variable was quantified using mean percent change.

We used the operational model’s fixed effects to determine expected ED costs for each hospital on the basis of case mix and wage index. For each hospital, we calculated observed:expected (O:E) ratios (actual cost divided by model-predicted cost) to evaluate the extent of apparent over- and under-charging. O:E ratios by state and for each contextual factor were determined using the geometric mean ratio for each subgroup. Finally, we summarized operations-adjusted costs for each contextual characteristic; for instance, we reported the spread in case mix- and wage-adjusted costs by hospital ownership.

We examined the impact of outliers on our determination of each ED’s costs. The contextual model was repeated using the outcome of ED-level log median costs instead of log mean costs. Similar results would indicate outliers were not influential.

Significance was defined as a 2-sided P < .05. Statistical analysis was completed using Stata 17 (StataCorp, College Station, TX) and data preparation with R version 4.2.0 (R Foundation, Vienna, Austria).

We included 713 acute care hospitals encompassing 22.9 million ED encounters. Most EDs were in nonteaching (67.9%) and nonprofit (68.5%) hospitals and did not have a pediatric ED (70.7%) (Table 1). Cost-to-charge ratios were present for 99.4% of ED encounters.

The median hospital had an ED-level mean cost of $269 (IQR $211–$355, range $99–$1863). Costs by region are displayed in Fig 1. The 3 most expensive regions were St. Petersburg, FL; Ridgewood, NJ; and Hackensack, NJ (medians of $605, $482, and $446, respectively). The 3 least expensive regions were Ocala, FL; Bradenton, FL; and Panama City, FL (medians of $119, $146, and $159, respectively). Thus, there was an 18.9-fold difference between the most and least expensive EDs and a 5.1-fold difference between the most and least expensive regions.

Based on observed variance in the random intercepts, there was variation in ED-level costs within and between states. Variance attributable to the state level was 5.1%, and 94.9% of variance remained unexplained and is presumed to be attributable to the ED level. Regional random intercepts were not significant.

The operational and contextual models accounted for 20.6% and 50.2%, respectively, of cost variation. In the contextual model, ED-level costs were associated with case mix index (adjusted costs increased 38% per 10% increase, 95% confidence interval [CI] 30 to 47). (Table 2). Wage mix had a more modest overall effect size in predicting ED-level costs in our contextual model (adjusted costs increased 7% per 10% increase, 95% CI 5 to 9). Aside from case mix and wage index, critical access status (adjusted costs, +24%, 95% CI 13 to 35), for profit status (−20%, 95% CI −26 to −14), lowest level trauma designation (+17%, 95% CI 5 to 30), highest amount of inpatient pediatrics beds (+13%, 95% CI 4 to 23), and increasing Medicaid share (Quartile 3: -13%, 95% CI −19 to −7) had the largest effect sizes.

Hospital O:E cost ratios ranged from 0.51 to 3.70. Within hospital ownership groups, for-profit, nonprofit, and nonfederal government hospitals had mean O:E cost ratios of 0.87, 1.02, and 1.02 respectively (Table 2). Level 4 trauma centers and critical access hospitals had the highest overall O:E ratios of 1.20 and 1.15 respectively.

Modeling median hospital ED costs instead of mean hospital ED costs did not change the most or least expensive states. The use of median costs did not alter the interpretation of the contextual model, except that the highest Medicaid share or teaching hospital status were no longer associated with costs (Supplemental Table 3).

Across 22.9 million childhood ED encounters in 713 acute care hospitals in 8 states, we found substantial variability in costs for pediatric emergency care. There was an 18.9-fold difference between the mean costs of the most and least expensive EDs and a 5.1-fold difference between the most and least expensive regions. States explained 5.1% of cost variability, whereas ED-level factors explained 94.9% of cost variability. The factors most associated with increased costs were regional wages, case mix, critical access status, lowest-level trauma status, and highest number of pediatric inpatient beds. For profit status, highest number of annual pediatric ED visits, and increased Medicaid share were most associated with decreased costs. Taken together, our findings indicate that differences in patients and regional wages (the 2 main costs of providing health care²⁴ ) are the primary drivers of ED costs for children but other drivers contribute substantially to observed variability.

ED costs for children varied both within and between regions, consistent with adult ED care.¹¹  Case mix, which quantifies the complexity and severity of patients’ illnesses, was a key driver of costs. This was unsurprising, as children with higher-severity illness generally require more resources and more intense care, which drives costs.²³  For-profit hospital ownership was strongly associated with lower costs. This is different than prior observations of increased costs associated with for profit institutions in other settings.^13,14,27  The cause of this discrepancy is unclear, and it is possible there is residual confounding by the type of patients visiting each, or it is possible that in the pediatric space, for-profit hospitals are able to be more efficient. EDs in hospitals with higher numbers of annual child ED visits had lower costs, suggesting possibly an efficiency of scale or increasing pediatric experience leading to lower costs. However, EDs in hospitals with higher numbers of inpatient beds had higher costs. Although it is possible that hospitals with larger admitting services receive incoming transfers that require complex work ups or other evaluations, such patients were excluded from our analysis. These hospitals are also already known to charge more for inpatient care than hospitals with smaller pediatric units.²⁸  Cost of ED care may be inflated by availability of advanced diagnostics or specialty providers for children not available in hospitals with fewer pediatric resources. Hospitals with the lowest-level trauma designation had higher costs, which could be related to necessary fixed costs for additional resources necessary and available for the sickest patients. Critical access hospitals were associated with the largest increase in costs in our study, speaking to the razor thin margins and financial pressures inherent in running such an institution. Finally, hospitals with larger Medicaid shares had lower costs, potentially reflecting differences in reimbursement that could affect what a hospital recoups.^29,30 

State was a modestly important factor driving variability. This could have been because of the state policy environment. Medicaid expansion has been associated with increased Medicaid-paid emergency department visits and fewer self-pay visits, although the relationship is complex.^19,31,32  In our study of 8 states, 6 of the 8 states (AZ, IA, MD, NJ, NV, and NY) adopted and implemented Medicaid expansion when first allowed in 2014, whereas Florida and Wisconsin have not.³³  Although it might be expected that Medicaid expansion would have a blunted effect on pediatric emergency care since many children had Medicaid or other public payer options before Medicaid expansion, the rate of Medicaid among children has been increasing.¹⁹ 

This study raises concerns both for families and policymakers. As health care costs rise, it is important to understand the drivers. During a medical emergency, families generally visit the nearest ED without regard to differences in cost. Yet, in many cases, this decision will be associated with a cost that could vary by hundreds or thousands of dollars. This is particularly concerning because our study suggests a portion of cost variation is explainable by contextual factors not directly related to patient care. It is not clear whether these factors provide value or improve patient outcomes. We believe addressing such low-value cost variation is a policy challenge but also should be a priority.

Our study highlights potential areas of focus for policymakers, such as the significant differences in ED charges between ownership models. It is unclear why not for profit hospitals cost more for pediatric emergency care, after adjusting for the basic cost of providing care. Visible disclosure of prices also might improve patient choice and result in lower prices.³⁴  Alternative payment models that limit overcharging relative to costs could also decrease the large observed variability in ED costs.

Our study has several limitations. First, case mix was measured using a system designed to evaluate inpatient case mix so it may not produce perfect estimates of emergency department case mix. Second, retrospective database studies based on diagnosis coding are imperfect at capturing true complexity and severity. Third, these results are based on only a subset of states and may not completely generalize to the rest of the country. Finally, these results are based on precoronavirus disease 2019 pandemic data, and it is difficult to the assess the impact of the coronavirus disease 2019 pandemic on hospital costs.

Costs for emergency care for children varied substantially within and between hospitals, states, and regions. Labor costs and patient illness severity were the largest drivers of costs. Low-level trauma status, critical access status, and increasing numbers of inpatient pediatric beds were associated with higher costs. For profit status, larger number of annual pediatric ED visits, and larger Medicaid share were associated with decreases in ED costs. These results can inform efforts to reign in outlier ED costs while maintaining quality of care.