Trends in Pediatric Emergency Department Use After the Affordable Care Act

We performed a cross-sectional, retrospective study of national ED visit rates among children aged 0 to 17 from 2009 to 2016 using the American Community Survey (ACS) and Nationwide Emergency Department Sample (NEDS). The Boston Children’s Hospital institutional review board deemed the protocol exempt.

The ACS, sponsored by the US Census Bureau, is an annual survey that includes over 700 000 children each year and provides weighted, nationally representative estimates of demographic information including age, sex, health insurance status, and geographic location.²² NEDS²³ is a publicly available database sponsored by the Agency for Healthcare Research and Quality’s Healthcare Cost and Utilization Project. NEDS includes more than 28 million ED visits annually and uses a deliberate sampling strategy stratified by geographic region, hospital teaching status, trauma center, urban or rural location, and hospital ownership, allowing for accurate national population-based estimates. As NEDS is an “event-level” database, each ED visit is treated independently.²⁴ 

From the ACS, we extracted estimates for the total number of children living in each census region (Northeast, Midwest, South, or West) each year from 2009 to 2016, stratified by sex (boy or girl), insurance status (insured or uninsured), and age (either 0–5 or 6–17, the most granular categorization provided by the ACS within strata of the other factors).

From NEDS, we examined ED visits for encounters from pediatric patients (ages 0–17 years old) across the United States from 2009 to 2016. We then estimated the nationally representative counts of pediatric ED visits in each stratum as defined above. All analyses were conducted within the context of the survey design characteristics of NEDS, with the ED encounter used as the unit of analysis. We specified the primary sampling units, patient visit sampling weights, and the stratum identifiers as implemented in Stata’s svyset command (version 13.1; StataCorp, College Station, TX), allowing for the calculation of nationally representative estimates of ED visit frequencies. We then divided the number of pediatric ED visits from NEDS by the population totals from the ACS to generate the ED visit rate per 100 individuals per year, stratified by age, sex, census region, and insurance status.

Using an interrupted time series (ITS) analysis,²⁵ we tested whether ED visit rate levels and slopes (with slopes representing the rate of change in ED visit rates over time) were significantly different before (2009–2013) versus after (2014–2016) the full implementation of the ACA in 2014, controlling for age, sex, and census region. (We reserved the use of insurance status as an additional control variable for sensitivity analysis, as described below.) Specifically, we performed a negative binomial regression in which the unit of analysis was defined by the cross-stratification of year, age group, sex, and census region. We used the estimated count of ED visits as the dependent variable and the log of the population counts as the offset (coefficient constrained to 1). This ITS model compared the pre-2014 to the post-2014 slopes as well as the pre- versus post-2014 intercepts (ie, the level change). The independent variables were ACA status (a binary variable indicating whether a visit occurred before or after 2014, when the ACA took full effect), time (an integer count indexing calendar year within period defined by ACA status), and the ACA status-by-time interaction.

A significant ACA status effect would indicate a level change (ie, a difference of intercepts between the pre- and post-2014 periods). A significant time effect would demonstrate that ED visit rates were changing over time pre-2014 (in other words, the time effect measures the pre-2014 slope). A significant interaction term would indicate that the pre- and post-2014 slopes differed significantly.²⁰ 

In our secondary analyses, we analyzed the same ITS model separately in each of the 4 census regions to assess whether 1 region might be driving any apparently national trends (eg, whether 1 region had much higher insurance uptake after the ACA and thus saw more significant changes in ED visit rates). Finally, we ran our national model again, this time controlling for insurance status (insured versus uninsured) to examine whether any effects noted were influenced by the effect of increased insurance coverage.

In agreement with previous reported findings,²⁰ pediatric uninsurance rates fell from 7.1% in 2013 to 6.0% in 2014, 4.8% in 2015, and 4.5% in 2016. Supplemental Table 4 displays changes in insurance rates by region.

As displayed in Table 1, there were an estimated 129 899 936 pediatric ED visits nationally from 2009 to 2013 (based on 30 230 961 observed ED encounters) and an estimated 80 602 176 visits from 2014 to 2016 (based on 18 425 633 observed encounters). As shown in Fig 1, the crude visit rate per 100 children fell initially from 36.6 in 2009 to a low of 33.0 in 2010 (−9.8%) before rising again to 37.9 in 2016 (+14.8%). The results of negative binomial ITS analyses are shown in Table 2. Patients aged 6 years or older were less likely to use the ED. The Northeast had the highest ED visit rate, followed by the Midwest and South, and finally the West.

We did not detect evidence of a significant level change (incidence rate ratio [IRR] 0.97, 95% confidence interval [CI] 0.85–1.12, P = .71); that is, the ED visit rate in 2009 was not significantly different from that in 2014. Controlling for demographic characteristics, there was no evidence that ED use was changing before 2014 (IRR 1.01, 95% CI 0.99–1.04, P = .32). Table 3 shows that from 2014 to 2016, after the ACA was fully implemented, ED use increased by 9.7% per year (IRR 1.10, 95% CI 1.04–1.16, P < .001). The slope of ED visit rates over time was significantly greater in 2014–2016 compared with 2009–2013 (IRR 1.09, 95% CI 1.03–1.15, P = .005).

In secondary analyses stratified by region, we found an initial 2014 decrease in the annual ED visit rate in the South (IRR 0.82, 95% CI 0.70–0.96, P = .01), no change in the Northeast (IRR 0.81, 95% CI 0.57–1.17, P = .27) and Midwest (IRR 1.01, 95% CI 0.81–1.27, P = .91), and an increase in the West (IRR 1.34, 95% CI 1.09–1.64, P < .01). As displayed in Table 3, we detected a significant increase in the slope of ED visit rates before and after the full implementation of the ACA in 2014 in the South (IRR 1.20, 95% CI 1.12–1.28, P < .01). We found no difference in the Northeast (IRR 1.10, 95% CI 0.95–1.29, P = .20), Midwest (IRR 1.03, 95% CI 0.93–1.12, P = .58), or West (IRR 1.02, 95% CI 0.93–1.11, P = .73).

When our nationwide model was run while controlling for insurance, the difference in the pre- versus post-2014 slopes remained statistically significant (IRR 1.08, 95% CI 1.03–1.13, P < .01).

In our study of the effects of the ACA and increased insurance coverage on ED use in children, we found no immediate 2014 change in pediatric ED visit rates. Our analyses did, however, find that the growth rate of ED use was significantly higher in 2014–2016 compared with the pre-2014 period. In secondary analyses, we found a significant 2014 increase in visit rates in the West and a decrease in the South. The South had an increasing rate trend after 2014, whereas none of the other 3 regions displayed a statistically significant change in rate trends. Finally, we found that the growth in ED visit rate trends after 2014 did not vary significantly depending on the inclusion or exclusion of insurance as a control variable, indicating that insurance did not seem to be driving trends in visit rates.

Policymakers often hope that improving insurance coverage, by improving access to lower-cost health care settings, can decrease ED visit rates and thus reduce spending. Our findings suggest that this may not be a successful strategy in pediatric populations. We found no immediate decrease in ED visit rates, and, contrary to hopes that ED use might gradually decrease with increased insurance coverage, our ITS analysis revealed a significantly higher slope in visit rates over time after the ACA’s full implementation in 2014. These findings represent an early suggestion, because we used only the first 3 years’ worth of postimplementation data, against the deceleration over time that would likely be seen if insurance decreased ED visits by gradually improving patient health.

Although insured children did tend to use the ED less than uninsured children, our findings suggest that this may be driven more by background demographics than the actual effect of insurance coverage because including insurance as a control variable had no impact on our model’s findings regarding post-2014 changes in visit rate trends. It may be that the factors that would drive newly insured patients to have increased ED visit rates (“pent-up” demand in the setting of chronic illness or the removal of financial barriers to outpatient care) are less applicable to children, who tend to have fewer chronic illnesses than adults. Alternatively, those factors could apply but be cancelled out by countervailing trends such as improved access to primary care leading to an overall improvement in health.

Indeed, studies in the adult population have revealed that the impact of increased insurance coverage on ED visit rates varies significantly depending on context.^1,4,–18 For example, in 2014, Smulowitz et al¹⁴ found that regions of Massachusetts that had larger increases in insurance coverage after its 2006 health reform had increased ED visit rates, perhaps due to inadequate access to primary care. In contrast, Miller¹⁶ found that Massachusetts overall saw a slight decrease in nonurgent ED visits. In a landmark experiment, Oregon’s randomized and controlled Medicaid expansion did increase office visits but also increased ED visit rates by an astonishing 40% in the first 15 months.¹⁸ This effect did not abate in a follow-up study of the second year after expansion.¹⁷ In sum, the effect of insurance on ED visit rates is likely not universal and probably depends on a multitude of interacting factors such as availability of alternative sources of care and chronic disease burden.^1,16 

In our analysis, regional trends were not definitively driven by the fact that the states in each region varied in their approach to Medicaid expansion, which is consistent with our national finding that controlling for insurance did not alter our results. The West, which experienced the greatest decrease in pediatric uninsurance, demonstrated an initial increase in visit rates but not a sustained trend difference thereafter. The South demonstrated an initial decrease in visit rates followed by an increasing rate trend, perhaps returning to its baseline. These secondary analyses may suggest regional differences in the timing with which changes were taking place in response to the ACA. Notably, none of the other regional analyses demonstrated significant initial changes or changed rate trends, and further study is merited.

Our study must be considered in the light of methodologic limitations. First, we are unable to draw conclusions about patients at the individual level with our ecologic study design; specifically, we are unable to test the association between any one ED visit for an individual and changes in insurance status over time with our data. Second, we cannot parse the respective roles of increased insurance coverage, other changes associated with the ACA, or other national trends that occurred during the study period. Theoretically, there may be countervailing trends that are cancelling each other out at the national level. Third, we are limited by the fact that only 3 years of post-2014 data are available at this point. Follow-up studies in which additional years of data are used will be critical to confirming this trend.

Further study regarding the causes of this increase, if sustained, would be valuable. In particular, past studies have revealed that ∼13% of pediatric ED visits are for ambulatory-sensitive conditions.²⁶ If these visits comprise a disproportionate share of the post-2014 increase in ED use, improved primary care access may help to slow that increase. Individual-level, prospective studies of primary care access and ED use for varying conditions among children are needed. Additionally, although we found no differences in models that do or do not control for insurance status, the ACS is only able to provide information regarding insured versus uninsured patients when stratified by sex, age, and region. Other data sets that permit more precise identification of the type of insurance involved (public versus private, for example) may be able to explore whether the increase in ED visits is disproportionately concentrated among children with a particular form of coverage.

Finally, our data set is unable to capture many of the factors that might contribute to varying responses to increased insurance coverage across regions. Other literature has revealed, for example, that the density of federally qualified health clinics significantly affected ED visit rates in children with Medicaid,²⁷ but primary care density is not available in either NEDS or the ACS. NEDS does not include information regarding patient race, and neither data set is meant to capture epidemiological trends in infectious disease severity or incidence, which can also drive ED use. Additionally, the geographic information available in NEDS is limited to census region. Although those data allow us to describe changes regionally rather than just at a national level, the lack of state-by-state data obscures within-region variability in Medicaid expansion, in which many states did not expand Medicaid at all and several others used waivers to implement slightly different versions of the Medicaid expansion.^28,29 

Expanded pediatric insurance coverage remains a worthy goal. We urge, however, that policymakers not expect that ED use will fall as a result of improved access and that they budget appropriately for possible increases in visits over time if the increased growth rate from 2014 to 2016 continues. Further study is needed to specify the exact factors that are driving this increase, to examine whether this trend continues over the next several years of data, and to link any possible changes in use to clinically relevant health outcomes in children.

 
• ACA

  Affordable Care Act

 
• ACS

  American Community Survey

 
• CI

  confidence interval

 
• ED

  emergency department

 
• IRR

  incidence rate ratio

 
• ITS

  interrupted time series

 
• NEDS

  Nationwide Emergency Department Sample