Abstract
This article describes trends in nonemergent emergency department (ED) visits by insurance type, using the 2000-2009 National Hospital Ambulatory Medical Care Survey and Current Population Survey. We analyzed trends in the probability that an ED visit is nonemergent and in nonemergent ED visit rates per person. We found that visits for Medicare enrollees were least likely to be for nonemergent reasons, while uninsured visits were most likely to be nonemergent. When we accounted for total visits and population size by insurance group, we found nonemergent ED visit rates per person were largest among Medicaid enrollees. Trends in nonemergent ED visit rates were stable for all insurance groups. The findings suggest a reliance on the ED for nonemergent care by the Medicaid population. It will be important to continue to track patterns of nonemergent ED utilization after Medicaid expansions under the Affordable Care Act.
Introduction
Emergency department (ED) visits have been on the rise in the United States over the past decade, leading to concerns about costs and potential overcrowding of EDs (Tang, Stein, Hsia, Maselli, & Gonzales, 2010). The increase in demand for EDs, coupled with inadequate inpatient bed capacity, has contributed to longer ED wait times and more patients leaving EDs without being seen (Government Accountability Office, 2003; Hoot & Aronsky, 2008; Wilper et al., 2008). In response, discussions by policymakers have focused on ways to reduce inappropriate ED use. These discussions have largely concentrated on the potential effects of the growth of safety net populations given federal legislation requiring that patients receive emergency services regardless of ability to pay and the potential for increased reliance on the ED as a safety net provider. The uninsured and underinsured patient populations often lack access to primary care and may have greater need for emergency care or may disproportionately use the ED for care that could be provided in alternate, less costly ambulatory settings (Mehrotra et al., 2009).
Previous research shows that while the number of uninsured has increased over time, this population has not accounted for the disproportionate amount of the increase in ED use. In fact, after accounting for population size, ED visit rates for the uninsured have declined. Instead, adult Medicaid enrollees account for most of the growth in ED visits (Tang et al., 2010). Adults covered by Medicaid have also been shown to be more frequent users of the ED than the uninsured (Zuckerman & Shen, 2004). These findings complement other work using different data that find Medicaid and SCHIP (State Children’s Health Insurance Program) patients made up the majority of ED visits among low-income adults—38% of ED visits, compared with 20% by the privately insured and 20% by the uninsured (Cunningham, 2006). Because of the substantial expected increase in the adult Medicaid population under the Patient Protection and Affordable Care Act in 2014 (Congressional Budget Office, 2010), it is important to understand how different patient populations use the ED and implications of the growing reliance on the ED by adult Medicaid patients.
One explanation for differences in ED use across insurance categories is relative access to primary care, which varies considerably by insurance status. Though rates of ED visits by the uninsured have remained stable over the previous decade, earlier research has shown that the uninsured have more limited access to primary care and rely more on EDs for ambulatory care than the insured. The ED was the setting for 25% of uninsured ambulatory care visits and for 18% of Medicaid ambulatory care visits according to one study. In contrast, ED visit rates for ambulatory care for the privately insured and Medicare populations were only about 8% (Cunningham & May, 2003). Furthermore, reports of limited provider participation in Medicaid and associated access problems are common (Bisgaier & Rhodes, 2011; Boukus, Cassil, & O’Malley, 2009; Government Accountability Office, 2011).
New Contribution
There has been little previous evidence of how nonemergent ED use varies by insurance status and over time. Policy should aim to improve access to emergent care through EDs and decrease ED use for nonemergent visits. This requires understanding how different patient populations use the ED. This study investigated the use of the ED for nonemergent care by insurance type and over time. Results regarding the uninsured and Medicaid populations may have important implications for ED use as these populations change and Medicaid enrollment grows. Understanding patterns of nonemergent ED use will also inform discussions regarding the availability of primary care services for different insurance groups, as well as other efforts to optimize health care provision in appropriate settings.
Data and Methods
We used a nationally representative sample of hospital-based ED visits, combining repeated cross-sections of the 2000-2009 National Hospital Ambulatory Medical Care Survey (NHAMCS). Response rates of hospitals sampled in the NHAMCS were similar over the survey years; in 2009, the weighted response rate was 89.8% of hospitals sampled. Patient visit-level information was collected by hospital staff for each patient sampled during a randomly selected 4-week period each year. The NHAMCS data contained patient demographic and visit characteristics. The sample included visits to noninstitutional, general, and short-stay hospitals located in the 50 states and the District of Columbia; federal, military, and Veterans Administration hospitals were not sampled in the survey. This analysis used patient demographic data on age, sex, and race; geographic region; and visit data on physician diagnosis codes and expected primary source of payment. For race categories, we used White and Black race as collected in the NHAMCS and collapsed Asian, Native Hawaiian/Other Pacific Islander, American Indian/Alaska Native, and More than One Race Reported groups into a single race category for Other.
We used information on expected source of payment to create categories for insurance status: privately insured, Medicare insured, Medicaid/SCHIP insured, and uninsured (including self-pay and charity). The NHAMCS reported a single primary payer source in the data from 2000-2004, which did not permit identification of individuals dually enrolled in both Medicare and Medicaid. Beginning in 2005, the NHAMCS collected data on multiple payer sources, allowing for the identification of dually enrolled individuals. We classified visits in the 2005-2009 NHAMCS that reported both Medicare and Medicaid payer sources as Medicare as the primary payer since emergency department visits for these patients are more likely to be covered under Medicare (Sonnenfeld, Decker, & Schappert, 2011). This classification scheme assumed that Medicare was reported as the primary payer for dually enrolled patients in the 2000-2004 NHAMCS. However, because categories of insurance status were the main variables in our analysis, we also constructed two other versions of insurance status categories in order to test the sensitivity of our results to the insurance classification. First, we restricted the sample to non-Medicare visits by the younger-than-65-years population to minimize uncertainty in categorizing dually enrolled patients. Second, we reclassified visits in the 2005-2009 NHAMCS with Medicare and Medicaid as payer sources to Medicaid as the primary payer because dually enrolled patients are both poorer and sicker and may be more similar to the Medicaid population (Tang et al., 2010).
Next, we classified ED visits into emergent and nonemergent categories by applying the New York University (NYU) ED algorithm to the physician primary diagnosis codes in the NHAMCS. Specifically, the algorithm assigned a probability of each ED visit being (1) nonemergent (NE), (2) emergent but primary care treatable (E-PCT), (3) emergent but preventable or avoidable if appropriate ambulatory care had been received (E-PA), and (4) emergent, ED care required and not preventable or avoidable (E-NPA; NYU Center for Health and Public Service Research, 2011). We created a dichotomous variable based on these probabilities and classified ED visits as nonemergent when the sum of probabilities of NE and E-PCT are greater than 50% and as emergent when the sum of probabilities of E-PA and E-NPA are greater than or equal to 50%. This measure of emergent (vs. nonemergent) ED visits has been validated in the NHAMCS as well as in data with private and Medicare patients and has been shown to be a strong predictor of mortality and hospital admission (Ballard et al., 2010; Gandhi & Sabik, 2014). Residual ED visits that could not be classified by the algorithm or were classified as injury, mental health, or alcohol- or drug-related by the algorithm were not included in our final analytic sample. In total, we observed 185,331 ED visits from the surveys, representing 630.3 million ED visits in 2000-2009.
Statistical Analysis
We used linear probability models accounting for NHAMCS complex survey weights and adjusting for patient age, sex, race, geographic region, and survey year to estimate the association of insurance type with whether an ED visit was classified as nonemergent. We included insurance status–specific time trends to the model by including insurance–year interactions in order to test for differential trends by insurance type over time. We used the regression results to compute predicted probabilities of nonemergent use by year for ease in interpreting trends in utilization. We used the method of “recycled predictions” (Korn & Graubard, 1999) to calculate the predicted probability of nonemergent use in each year based on the regression results under different hypothetical scenarios: that all individuals in the sample were uninsured, that all were enrolled in Medicaid, that all were enrolled in Medicare, and that all were privately insured. For each year and for each of the insurance categories, the indicators for the given year and insurance category were fixed at one, and the original values for all other covariates were used to generate predicted probabilities based on the regression results. The predictions were then averaged over the entire sample. We also constructed bootstrapped 95% confidence intervals (CIs) for the predicted probabilities.
Next, we computed estimates of the total number of ED visits for our sample in the NHAMCS and multiplied the totals by the predicted probabilities from our regression results to generate an estimate of the age-, sex-, race-, and region-adjusted number of nonemergent visits for each insurance category and year. Using the Current Population Survey (CPS) March Annual Social and Economic Supplement, we created unduplicated counts of individuals in each of the four insurance categories. Because individuals can report more than one form of insurance coverage in the CPS, we categorized those who reported Medicare coverage as Medicare, those who reported Medicaid but not Medicare as Medicaid, and those who reported private coverage but not Medicaid or Medicare as private. Individuals were categorized as uninsured if they reported no form of coverage. Using this hierarchy, we were able to assign each observation to a single insurance category in order to avoid duplication across categories (as in published Census Bureau estimates by category). We categorized those with Medicare and another form of coverage (such as Medicaid or private) as Medicare to match the categorization we used in the NHAMCS data. We then used the unduplicated population estimates from the CPS to create an estimated nonemergent visit rate per 1,000 individuals, adjusted for age, race, sex, and region, by year and insurance status. We also created these estimates for the younger-than-65-years population (excluding visits for those with Medicare coverage), as well as for the adult (age >18 years) and children (age ≤18 years) subpopulations. These measures captured differences in the number of nonemergent visits per person by insurance group and year not driven by demographic differences across groups or over time.
Results
Table 1 presents descriptive statistics of the full sample of patients with classifiable emergent and nonemergent ED visits by primary payer. On average, ED patients in the sample are 37 years old and less likely to be male than female. Adults represent nearly three fourths of our sample. The majority of the sample is White (72%) and about a quarter of the sample is Black. The sample includes individuals covered by private insurance (38%), Medicare (19%), Medicaid/SCHIP (25%), and no insurance (17%). Comparing across insurance groups, Medicare enrollees are oldest on average and Medicaid enrollees youngest. Medicaid enrollees are less likely than the population average to be male while uninsured individuals are more likely to be male. Medicaid enrollees and the uninsured are more likely to be Black and are more highly concentrated in the Southern region of the United States.
Descriptive Statistics for NHAMCS ED Visits by Insurance Coverage Group.
Note. NHAMCS ED = National Hospital Ambulatory Medical Care Survey; ED = emergency department. Includes visits classifiable as emergent or nonemergent using New York University ED algorithm. Categories may not sum to 100% because of rounding.
Figure 1 shows average predicted probabilities that an insurance visit is classified as nonemergent for each insurance group in each year based on the regression results and illustrates insurance specific trends, showing that Medicare enrollees had the lowest proportion of nonemergent visits, though the rate was slightly increasing over this period from 0.66 (95% CI = 0.64-0.69) in 2000 to 0.69 (95% CI = 0.67-0.71) in 2009. (See Tables A1 and A2 in the appendix for regression results and predicted probabilities.) Predicted probabilities of a visit being nonemergent among the uninsured on the other hand were stable, ranging from 0.76 (95% CI = 0.74-0.78) in 2000 to 0.77 (95% CI = 0.75-0.79) in 2009 but higher than any other insurance group. We found that Medicare visits were significantly less likely to be nonemergent, compared with the privately insured in both models. Estimates of nonemergent use were highest among uninsured visits and significantly higher than the privately insured reference group. Results also indicated that the probability that an ED visit is nonemergent is not increasing over time, on average.
Regression-adjusted trends in the probability that an emergency department (ED) visit is nonemergent.
Figure 2 shows average predicted probabilities that a visit is classified as nonemergent for each insurance group in each year for the subsample of visits by patients younger than 65 years. (See Table A3 in the appendix for predicted probabilities.) The proportion of nonemergent visits was higher on average compared with the full sample, but the trends were similar to those in the full sample. The proportion of visits that were nonemergent was highest among the uninsured and stable over the time period, increasing from 0.80 (95% CI = 0.78-0.82) in 2000 to 0.81 (95% CI = 0.79-0.83) in 2009. Results were similar when we classify dual eligibles as Medicaid. (See Figure A1 and Table A4 in the appendix.)
Regression-adjusted trends in the probability that an emergency department (ED) visit is nonemergent, younger than 65 years sample.
Figure 3 displays the adjusted nonemergent visit rates per 1,000 individuals by year and insurance category for the total population. These rates reflect the nonemergent visit burden on EDs by insurance group, taking into account population size. We found that the largest number of nonemergent visits per person occurred among Medicaid enrollees, with visits per 1,000 individuals ranging from 364.3 (95% CI = 301.1-427.6) nonemergent ED visits in 2000 to 387.7 (95% CI = 309.2-466.2) in 2009. (See Table A5 in the appendix.) Additional analyses of visit rates indicated that the highest rates of nonemergent visits in the Medicaid population were among adults, with nonemergent visits rates per 1,000 individuals ranging from 470.6 (95% CI = 384.2-557.0) in 2000 to 515.1 (95% CI = 405.1-625.1) in 2009. The highest growth in the full sample occurred among Medicare enrollees, increasing from 190.0 (95% CI = 155.8-224.3) to 227.2 (95% CI = 183.4-271.0), but the increase was not statistically significant. In comparison, nonemergent visit rates for the other insurance groups were relatively steady or slightly decreasing over this period.
Estimated adjusted nonemergent visit rates per 1,000 by year.
Figure 4 presents results limiting the sample to individuals younger than 65 years, and compares the privately insured, Medicaid only (not including dual eligibles), and uninsured populations. (See Table A6 in the appendix.) Visit rates tend to be stable within group over time. As in the full sample, Medicaid enrollees have substantially more nonemergent ED visits than either the privately insured or uninsured. We generally found similar patterns of nonemergent visits per person when we classified dual eligibles as Medicaid, with the exception that this categorization results in a slight decrease in the number of visits among Medicare enrollees and an increase in visits among Medicaid enrollees, though these changes were not statistically significant. We do observe a statistically significant increase in the visit rate for Medicaid adults under this classification. (See Figure A2 and Table A7 in the appendix.)
Estimated adjusted nonemergent visit rates per 1,000 by year, younger than 65 years sample.
Discussion
This study considers trends in the proportion of ED visits for nonemergent conditions and nonemergent visit rates by insurance status. We found that the percentage of ED visits that were for nonemergent reasons did not change significantly over this time period. In addition, on average, visits among Medicare enrollees were less likely to be for nonemergent conditions, while Medicaid and uninsured visits were more likely than privately insured visits to be nonemergent. When we accounted for total visits and changes in population by insurance group, we found that nonemergent ED visit rates per person were relatively stable within each insurance group, but that Medicaid enrollees had far more nonemergent visits per person than the privately insured, Medicare enrollees, or the uninsured.
This study is the first to examine differences across insurance groups in nonemergent ED use and trends in nonemergent use over time using the NYU ED algorithm and a national sample of ED visits. There are some limitations to this analysis. First, the NHAMCS surveys EDs, so we were only able to observe information on individuals who present at an ED and are not able to examine rates of overall ED use or other ambulatory care use using the NHAMCS data alone. Because of the detailed visit information and the large, nationally representative sample of ED visits, these are the best available data for studying nonemergent use. We combine these data with population estimates from the CPS to address the question of population visit rates. Second, the data consist of repeated cross-sectional surveys of ED visits, thus unobserved changes in the demographic composition of insurance groups could bias our results. We control for individual-level demographic factors to account for any changes in populations over this time period. Third, the ability to define ED visits as nonemergent has been a methodological challenge in the literature. The algorithm produces a probabilistic assessment of whether the visit is emergent, and there may be some error in this outcome measure. However, the categorization of visits as nonemergent using the NYU ED Algorithm has been validated in the literature (Ballard et al., 2010; Gandhi & Sabik, 2014), and using a consistent measure over time is paramount in assessing trends in ED visits. In general, there is no clear objective assessment of what qualifies as an emergent or appropriate visit. Patients may not be able, nor can be required, to perfectly preassess their conditions as emergent or not before visiting the ED.
We do not find evidence of significant changes over time in nonemergent visit rates within insurance groups. While earlier research found overall increases in ED visits, and particularly among Medicaid enrollees (Tang et al., 2010), we do not find a significant increase in nonemergent visits among any of the insurance groups we consider. Our trends analysis is somewhat sensitive to the classification of dual eligibles, though. If we classify all individuals who are dually eligible for Medicare and Medicaid as Medicaid (rather than Medicare, as in our main analysis) we do find evidence of a significant increase in the number of nonemergent visits among adults in Medicaid. Given that Medicare is likely to serve as the primary payer for dual eligibles, we believe it is more appropriate to categorize them as Medicare, though arguments can be made in either direction (Sonnenfeld, Decker, & Schappert, 2011). The sensitivity to classification that we encountered may also affect the interpretation of other results based on NHAMCS data (Tang et al., 2010).
Our findings regarding the uninsured and Medicaid populations in particular may have important implications for health reform, under which many of the uninsured will become eligible for Medicaid. We found uninsured visits are most likely to be nonemergent. While other work shows that the uninsured have the lowest number of overall ED visits and fewer visits per person than Medicaid enrollees (Tang et al., 2010), when the uninsured do use the ED those visits are more likely to be for nonemergent reasons than among covered populations. When we account for the lower number of overall visits among the uninsured and size of the uninsured population, we found that they have fewer nonemergent visits per individual than Medicaid enrollees. Although the probability that a given visit is nonemergent is lower among the Medicaid population than among the uninsured, the large number of overall visits among the Medicaid population translates into substantially higher rates of nonemergent visits per Medicaid enrollee than among any other insurance group.
Recent research from the Oregon Health Insurance Experiment, which randomly assigned individuals to Medicaid eligibility through a lottery, found that those who gained access to Medicaid through the lottery had significantly increased use of the ED after gaining coverage compared with the control group. Increases were found across a range of types of visits, including those that could be treated in primary care settings (Taubman, Allen, Wright, Baicker, & Finkelstein, 2014). Other qualitative research among the population enrolled under the Oregon Medicaid expansion found that new Medicaid enrollees did not have a clear understanding of their coverage and associated benefits, had poor experiences with the health care system, or faced barriers such as long waits or transportation difficulties (Allen, Wright, & Baicker, 2014). All these factors could play a role in explaining the pattern we see nationally of high ED use among Medicaid patients for conditions that in many cases could be more appropriately treated in other ambulatory care settings. Further research is needed to assess factors that may be driving the use of the ED by Medicaid enrollees, particularly for nonemergent conditions, and to track ED use as health reform is implemented. As Medicaid expands under the Affordable Care Act, work must be done to ensure access to ambulatory care for Medicaid enrollees and curb use of the ED for conditions that could be treated in other settings.
Footnotes
Appendix
Adjusted Nonemergent Visit Rates per 1,000, Duals Classified as Medicaid.
| Private | 95% CI | Medicare | 95% CI | Medicaid | 95% CI | Uninsured | 95% CI | |
|---|---|---|---|---|---|---|---|---|
| Full sample | ||||||||
| 2000 | 100.5 | (84.1, 116.9) | 217.3 | (178.2, 256.5) | 303.3 | (250.6, 356.0) | 222.5 | (180.7, 264.2) |
| 2001 | 102.6 | (85.5, 119.8) | 206.5 | (169.0, 244.0) | 306.8 | (248.4, 365.1) | 192.1 | (152.4, 231.9) |
| 2002 | 103.0 | (85.1, 120.9) | 216.5 | (178.3, 254.8) | 337.8 | (265.4, 410.2) | 185.0 | (148.1, 221.9) |
| 2003 | 101.6 | (85.0, 118.1) | 236.3 | (198.3, 274.4) | 344.4 | (288.7, 400.1) | 181.3 | (147.9, 214.8) |
| 2004 | 95.3 | (78.5, 112.1) | 221.2 | (181.5, 260.8) | 308.7 | (252.4, 365.0) | 193.2 | (151.0, 235.5) |
| 2005 | 94.1 | (78.8, 109.4) | 204.4 | (168.5, 240.2) | 365.0 | (294.3, 435.7) | 199.8 | (159.2, 240.3) |
| 2006 | 100.2 | (84.1, 116.4) | 221.5 | (181.6, 261.3) | 387.0 | (315.1, 458.9) | 219.1 | (173.3, 264.8) |
| 2007 | 90.9 | (72.8, 108.9) | 198.8 | (161.7, 235.8) | 347.0 | (277.6, 416.4) | 179.5 | (144.0, 215.0) |
| 2008 | 104.7 | (86.2, 123.2) | 213.1 | (174.9, 251.3) | 332.6 | (273.2, 392.0) | 191.6 | (153.6, 229.6) |
| 2009 | 105.6 | (85.8, 125.4) | 208.8 | (168.0, 249.6) | 381.1 | (304.9, 457.3) | 191.3 | (148.3, 234.4) |
| Adults | ||||||||
| 2000 | 97.3 | (81.2, 113.5) | 211.2 | (172.7, 249.7) | 305.7 | (249.6, 361.8) | 210.1 | (170.2, 250.0) |
| 2001 | 96.6 | (80.2, 113.0) | 199.7 | (163.5, 236.0) | 336.2 | (272.5, 399.9) | 190.9 | (150.9, 230.8) |
| 2002 | 99.4 | (81.8, 117.0) | 209.3 | (172.7, 245.9) | 356.1 | (281.7, 430.4) | 175.8 | (140.8, 210.7) |
| 2003 | 94.5 | (78.6, 110.4) | 228.2 | (191.4, 265.1) | 365.8 | (305.0, 426.7) | 173.5 | (140.4, 206.7) |
| 2004 | 92.4 | (75.8, 109.0) | 216.1 | (177.3, 255.0) | 328.1 | (267.9, 388.2) | 187.2 | (144.9, 229.4) |
| 2005 | 90.7 | (75.8, 105.6) | 198.9 | (164.2, 233.5) | 404.3 | (331.3, 477.2) | 192.3 | (153.6, 230.9) |
| 2006 | 101.1 | (84.3, 117.9) | 215.1 | (176.5, 253.8) | 462.1 | (372.8, 551.4) | 221.1 | (172.7, 269.4) |
| 2007 | 89.9 | (72.2, 107.5) | 193.9 | (157.9, 229.9) | 401.5 | (325.0, 478.1) | 180.2 | (144.3, 216.0) |
| 2008 | 103.0 | (84.6, 121.3) | 208.1 | (170.8, 245.5) | 405.9 | (332.3, 479.4) | 185.3 | (148.1, 222.5) |
| 2009 | 104.2 | (84.0, 124.4) | 203.1 | (163.1, 243.0) | 467.1 | (369.4, 564.9) | 189.4 | (145.5, 233.3) |
| Children | ||||||||
| 2000 | 107.3 | (86.7, 127.8) | 332.5 | (259.6, 405.4) | 243.8 | (187.6, 300.0) | ||
| 2001 | 117.5 | (95.0, 140.1) | 310.3 | (243.7, 376.8) | 168.1 | (129.1, 207.1) | ||
| 2002 | 110.3 | (85.8, 134.8) | 356.7 | (259.1, 454.3) | 192.9 | (141.9, 243.8) | ||
| 2003 | 120.2 | (98.0, 142.3) | 360.9 | (292.5, 429.2) | 183.7 | (143.3, 224.1) | ||
| 2004 | 100.6 | (79.9, 121.3) | 317.8 | (250.4, 385.2) | 183.0 | (136.4, 229.6) | ||
| 2005 | 100.9 | (81.7, 120.0) | 359.2 | (271.4, 447.0) | 195.2 | (142.4, 248.0) | ||
| 2006 | 92.7 | (75.4, 110.0) | 346.2 | (277.5, 414.9) | 166.2 | (128.2, 204.3) | ||
| 2007 | 88.8 | (67.0, 110.7) | 326.8 | (244.4, 409.1) | 138.7 | (105.3, 172.1) | ||
| 2008 | 105.5 | (80.1, 131.0) | 291.6 | (229.9, 353.2) | 177.9 | (134.9, 220.8) | ||
| 2009 | 104.7 | (82.3, 127.0) | 331.9 | (257.1, 406.7) | 146.4 | (110.2, 182.6) | ||
Note. CI = confidence interval.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.