DataWatch Excess Cost Of Emergency Department Visits For Nonurgent Care by Laurence C. Baker and Linda Schuurman Baker Abstract: After examining data for patients with selected conditions and statistically adjusting for patient, diagnosis, and treatment characteristics, this DataWatch finds that charges for emergency department visits were two to three times more than charges for visits in other settings. Large differences persist when conditions are examined individually and when total episode charges are examined. Based on our findings, a rough estimate of nationwide excess charges is $5-$7 billion for 1993. Significant attention has been paid in recent years to persons who use the emergency departments of the nation's hospitals for their primary care. The fact that many patients obtain nonurgent care in emergency departments has been well documented. 1 In fact, according to the National Ambulatory Medical Care Survey, more than half of the 89.8 million emergency department visits made in 1992 were for nonurgent care. 2 Such use of expensive emergency services is often pointed to as a source of excess health spending. It has been argued that significant savings, as well as increased quality of care, could be achieved if these patients used appropriate sources of care. However, although some "back-of-the-envelope" estimates of the ratio of emergency department to non-emergency department costs have been generated over the past fifteen years, broad-based, careful estimates of the excess cost of these visits have not been available. 4 In this DataWatch we use data from the 1987 National Medical Expenditure Survey (NMES) to study the difference between the charge for an emergency department visit and the charge for a visit to another setting for patients with several conditions that are not likely to require urgent attention. Data And Methods Data sources. We use data from the 1987 NMES household survey file, the ambulatory visit file, and the health status questionnaire. 5 NMES is a Laurence Baker is assistant professor of health research and policy at the Stanford University School of Medicine in Stanford, California, and is a faculty research fellow at the National Bureau of Economic Research. Linda Baker is a research analyst for The David and Lucile Packard Foundation in Los Altos, California.
DATAWATCH 163 national survey of roughly 35,000 persons in 4,000 U.S. households. Survey items cover health expenditures by or on behalf of families and individuals, the financing of those expenditures, and each person's use of health care services. In the ambulatory care supplement, data are collected from persons on all contacts with the ambulatory medical care system, including visits to medical provider offices, hospital outpatient units, and emergency departments. Total charges for each visit, procedures performed, and payers for the visit also are included. The household survey file provides basic demographics (age, race, sex, income, and so on), and the health status questionnaire provides information on medical conditions, self-perceived health status, and other health-related variables. The key variable of interest here is visit charges. This variable represents actual charges for the services performed, except when charges were reduced to amounts allowed by third-party payers. In these cases, the data contained the reduced amount. 6 Since NMES expresses charges in 1987 dollars, the values we present are also in 1987 dollars, except where noted. Construction and selection of episodes. We examined charges for visits and episodes of care. Episodes were constructed from the NMES visit file by grouping visits made by an individual to any provider for the same International Classification of Diseases, Ninth Revision (ICD-9) diagnosis code. Sequences of visits occurring before a break in service of four months or more were considered to be in the same episode. 7 We identified a sample of episodes for analysis using a three-step process. First, we included only episodes for a select set of conditions that are not expected to often require urgent care. NMES provides the ICD-9 code for the patient's condition. Because coding is retrospective and may not capture all available information, the selection of codes to be designated as nonurgent is difficult. Although previous literature on nonurgent care in emergency departments has not often used the ICD-9 codes as the basis of categorization, one retrospective chart review did incorporate diagnosis. 8 Here, a number of conditions were considered nonurgent, including superficial injuries, uncomplicated fractures of any digit or metacarpal, pharyngitis (sore throat), upper respiratory tract infections, and gastroenteritis. 9 We included episodes for the treatment of these conditions along with the treatment of skin disorders (for example, diaper rash, eczema, corns, callouses, and ingrown nails), head- and backaches, and miscellaneous symptoms (for example, conjunctivitis, earwax, hiccoughs, and heartburn). 10 Second, since our estimates of charge differences may be biased if the illnesses or injuries treated in emergency departments are more severe than those treated in the offices of medical providers, we excluded from consideration episodes with characteristics that we expect to be correlated with higher severity. Specifically, we excluded episodes for which the patient
164 HEALTH AFFAIRS Winter 1994 had visits in which he or she arrived in the emergency room by ambulance, received stitches, was subsequently admitted to the hospital, had more than fifteen visits in the episode, or was diagnosed with more than one condition. Episodes related to pregnancy also were excluded. Finally, since NMES only includes visits that occurred during 1987, our data on episodes that actually began in late 1986 would be inaccurate. We therefore excluded from our analysis of first visit charges all episodes beginning before March 1987 for which the first visit was to the office of a medical provider. Examination of the distribution of episodes indicated that episodes for which the first visit was to an emergency department are not likely to be censored. In all, after excluding episodes with missing data, our sample consisted of 3,305 episodes beginning in non-emergency department settings and 779 episodes beginning in an emergency department. We used only condition and severity in the selection process, excluding information related to visit appropriateness (for example, day of the week). We do not wish to consider access issues here, but rather to compare costs with those that would be incurred if all nonurgent cases were cared for in non-emergency department settings. Our sample of emergency department-initiated episodes is about 10 percent of the total number of emergency department-initiated episodes constructed from the NMES visit file, much fewer than the 50 percent of emergency department visits that are widely agreed to be nonurgent. In constructing our sample, we considered including episodes for treatment of additional conditions so that the number of episodes in our sample might be closer to 50 percent. However, since we must rely on ICD-9 codes to identify episodes, including additional categories would have exacerbated the potential for differences in severity to bias our findings. For example, we could have included lacerations, since some lacerations are nonurgent, but in so doing, we might also have included truly urgent lacerations that are also more likely to present in an emergency department. This would have made it more difficult to compare charges. Faced with this trade-off, we chose the set of conditions discussed above in hopes of examining a range of conditions while minimizing the risk of improper comparisons. Methods. We examined charges for visits and for episodes of care, focusing on the mean charge for emergency department and nonemergency department visits and, in the case of episodes, on the mean charge for episodes that were initiated in the two settings. To control for variation in patient characteristics and injury severity, we used regression adjustment techniques. Conceptually, these techniques use the experiences of patients who went to a non-emergency department provider to predict what the patients in our emergency department sample would have been charged for their visits if they had gone to other providers
DATAWATCH 165 instead. To implement the procedure, we used ordinary least squares to regress the first visit charges incurred by patients in the non-emergency department sample on an extensive set of variables describing the characteristics of their visits (condition, treatments received, source of payment, total number of visits in the episode), health status, health insurance status, regular source of health care, and demographic characteristics. 11 In some regressions we also controlled for medical history. We then used the regression results to predict the charges that would have been incurred by the emergency department patients if they had gone elsewhere. Using the projected charges, we can examine the difference between the charges for care in the different settings by looking at the difference between the amount that the emergency department sample was actually charged and their projected non-emergency department charge. Note that in the regression analysis, we control for the number and types of treatments received. This may cause us to understate the true underlying charge difference, since treatment difference may reflect differences in practice style. However, treatment difference also may reflect differences in severity. Including treatment in the regression will yield conservative estimates of charge differences. Study Results First visit charges. We examined differences in the charges for first visits in emergency department and other settings (Exhibit 1). We focus on first visits since they are likely to have similar content in both settings, while visits at other points in an episode may differ. The first two columns present the distribution of the episodes over condition types. The last three columns present the unadjusted mean charge for the first visit from episodes beginning in various settings. All charges are presented in 1987 dollars, except where noted. The differences are striking. Overall, the mean first visit charge in a non-emergency department setting is $43, while the mean first visit charge in an emergency department is $144. Large differences between the two settings persist within the condition categories. In all cases, the differences are highly statistically significant. Regression-adjusted estimates of mean charges for first visits are presented in the top section of Exhibit 2. The first column presents the actual mean charge for emergency department patients; the second column presents the charge we project that these patients would have incurred with a non-emergency department provider. For all condition categories combined, the projected non-emergency department charge is $50, while the actual charge of $144 was nearly three times higher. For conditions examined separately, large differences also persist. The actual emergency depart-
166 HEALTH AFFAIRS Winter 1994 Exhibit 1 Sample Sizes And Mean First Visit Charges For Emergency Department And Non-Emergency Department Episodes, By Condition, 1987 Condition Upper respiratory infection Number (percent) of episodes Non-ED ED 3,305(100.0%) 331 (10.0) 1,332 (40.3) 779(100.0%) 40(5.1) 158 (20.3) Mean first visit charges Non-ED $42.68 34.10 39.05 ED $143.72 121.33 113.02 Difference $101.74 a 87.23 a 73.97 a 642(19.4) 183 (5.5) 435(13.2) 56(7.2) 63 (8.1) 109(14.0) 49.32 57.20 38.03 120.88 143.24 147.35 71.56 a 86.04 b 109.32 a Gastroenteritis Minor fracture 82 (2.5) 35(1.1) 265 (8.0) 19 (2.4) 61 (7.8) 273 (35.0) 45.20 70.51 48.71 177.79 193.53 154.62 132.59 a 123.02 a 105.91 a Source: Authors' analysis of 1987 National Medical Expenditure Survey (NMES) data. Notes: Percentages may not sum to 100 because of rounding. Dollars presented are 1987 dollars. a Denotes significance in t-tests at the 0.001 level. Denotes significance in t-tests at the 0.01 level. ment charges are about three times the projected non-emergency department charges. Episodes for the treatment of fractures and gastroenteritis were not separately examined since there were too few visits to permit reliable regression analysis. Although we carefully selected the episodes analyzed here to minimize the possibility that severity bias would cloud our analysis, it is possible to make additional exclusions from the sample to reduce this risk even further. In doing so, however, we run the opposite risk of removing cases that should be examined those for which no severity bias exists. Nonetheless, to provide a conservative estimate of the difference, the bottom section of Exhibit 2 presents actual and projected emergency department charges for a subset of our original sample that excludes episodes in which patients received x-rays, scans, or surgical procedures. As expected, the mean actual first visit charge is notably lower in this sample than in the whole sample; the projected charges are also lower. Since both actual and projected charges fall by about one-third, the ratio of actual emergency department to projected non-emergency department charges remains about the same emergency department visit charges were about three times as much as non-emergency department visit charges. However, the absolute difference is smaller in this sample than in the overall sample. For all conditions, emergency department visit charges were about $65 more than other visit charges. Since this sample is likely to exclude some valid episodes, we interpret this as a conservative estimate of the charge difference. It is important to note that the regression techniques used can only
DATAWATCH 167 Exhibit 2 Mean Actual And Projected First Visit Charges For Emergency Department Patients, By Condition, 1987 Condition Actual ED charge Projected Non-ED charge Difference Ratio All episodes Upper respiratory infection $143.72 121.33 113.02 $49.87 35.96 40.82 $93.85 85.37 72.20 2.882 3.374 2.769 120.88 143.24 42.00 46.41 78.88 96.83 2.878 3.086 147.35 154.62 44.41 59.74 102.94 94.88 3.318 2.588 Restricted sample episodes Upper respiratory infection 100.06 100.76 90.06 34.80 33.77 33.50 65.26 66.99 56.56 2.875 2.984 2.688 98.51 99.02 36.87 33.10 61.64 65.92 2.672 2.991 110.86 101.08 37.42 33.34 73.44 67.74 2.963 3.032 Source: Authors' analysis of 1987 National Medical Expenditure Survey (NMES) data. Note: Dollars presented are 1987 dollars. control for patient and visit characteristics that are recorded. If there are differences in important unrecorded patient characteristics, especially condition severity, our estimates of the difference could overstate its size. Total episode charges. When assessing the overall extent of excess charges, it is important to consider episode charges as well as visit charges, because episodes begun in emergency departments may differ from episodes begun in non-emergency department settings. One important possibility is that the length of episodes may differ. For example, if office-based providers are more likely than emergency department providers to recommend follow-up visits, episodes begun in a non-emergency department setting may be longer. If this is the case, the differences observed in first visit charges may overstate the differences in total episode charges. To examine this issue, we first looked directly at the number of visits in episodes. To reduce the possibility of truncation at the end of the NMES sample period, we examined only episodes that ended in or before October 1987. Thus, the sample used here is somewhat smaller than that used above. Exhibit 3 presents the mean number of visits in episodes that began in the two settings. Contrary to expectation, episodes beginning in emer-
168 HEALTH AFFAIRS Winter 1994 Exhibit 3 Mean Number Of Visits Per Episode, Non-Emergency Department And Emergency Department,1987 Condition Upper respiratory infection Gastroenteritis Non-ED episodes 1.376(1.039) 1.195(0.541) 1.264(0.836) 1.516(1.146) 1.553(1.438) 1.318(0.821) 1.717(1.728) 1.479(1.367) ED episodes 1.514 a (1.148) 1.588(1.760) 1.670 b (1.219) 1.277 b (0.498) 1.523 (0.976) 1.538(1.492) 1.583 (0.793) 1.309(0.825) Source: Authors' analysis of 1987 National Medical Expenditure Survey (NMES) data. Note: Standard errors in parentheses. a Denotes significantly different from non-ed value at the 0.05 level. Denotes significantly different from non-ed value at the 0.01 level. gency departments had more visits per episode than those beginning elsewhere, on average, although the differences are not large. We then looked at total episode charges directly, using the regression adjustment technique described earlier (Exhibit 4)- 12 The top panel presents results from the analysis of all episodes. The estimates presented in the Exhibit 4 Mean Actual And Projected Total Episode Charges For Emergency Department Patients, For All Episodes, By Condition, 1987 Condition Actual ED charges Projected non-ed charge Difference Ratio Upper respiratory infection $174.09 150.18 153.91 $63.63 52.52 38.93 $110.46 97.66 114.98 2.736 2.859 3.953 132.89 170.49 57.05 74.78 75.84 95.71 2.330 2.280 204.55 162.48 62.95 70.51 141.60 91.97 3.250 2.304 Restricted sample episodes Upper respiratory infection 116.41 113.82 106.70 42.04 37.62 40.00 74.37 76.20 66.70 2.769 3.025 2.667 115.07 121.11 48.45 43.43 66.62 77.68 2.375 2.788 153.96 98.57 41.10 37.63 112.86 60.94 3.746 2.620 Source: Authors' analyses of 1987 National Medical Expenditure Survey (NMES) data. Note: Dollars presented are 1987 dollars.
DATAWATCH 169 bottom section use a conservative sample in which only episodes having no x-rays, scans, or surgical procedures are used. Taking all conditions together, the average emergency department-initiated episode charge was $174, while we project that these episodes would have cost one-third as much, about $63 per episode, if they had begun in another setting. Actual and projected charges for specific conditions also are presented in Exhibit 4. In general, emergency department-initiated episode charges were two to three times more than those for episodes initiated outside an emergency department. Estimates based on the restricted sample are similar, although, as before, the absolute size of the differences is smaller. Estimates of excess charges. Using these results, it is possible to construct rough estimates of the overall excess charges associated with the use of emergency departments for nonurgent care. About 89.8 million emergency department visits took place in 1992. 13 In our analysis of the NMES data, we found that 10 percent of the emergency department-initiated episodes of care were for the treatment of the narrow set of nonurgent conditions on which we focused. We thus suspect that about 10 percent of the 89.8 million emergency department visits, or 8.98 million visits, were for the treatment of these conditions. Applying the per visit charge differential we obtained earlier ($93.85, see Exhibit 2) to these visits, we estimate that the nationwide excess charges for the treatment of only our narrow set of conditions were about $840 million in 1987 dollars and, accounting for inflation in medical costs using the medical Consumer Price Index (CPI), about $1.3 billion in 1993 dollars. Two assumptions underlie these estimates. First, we assume that we have accurately estimated the charge differences for the conditions that we examined. Although we may have overestimated this difference by failing to completely control for severity or other unobserved factors, we hope to have minimized this problem by carefully selecting the episodes examined and by using extensive controls in our regression analysis. The second assumption is that the first visit charge difference is representative of charge differences for other (for example, second) visits. We suspect that this is the case since our analysis of total episode charges produced results similar to those from the first visit analysis. However, $1.3 billion may be a significant understatement of nationwide excess charges for the treatment of all nonurgent emergency department cases. One recent estimate indicated that 55.4 percent of 1992 emergency department visits, or 49.7 million visits, were nonurgent. 14 Under the assumption that the excess charge estimate we derived from the analysis of our set of conditions is applicable to all nonurgent emergency department visits, a rough estimate of overall excess charges is $7.2 billion in 1993 dollars. 15 Unfortunately, we cannot assess the extent to which our findings
170 HEALTH AFFAIRS Winter 1994 may be extrapolated to the broader set of nonurgent emergency department conditions. Because $7.2 billion could be an overestimate, we have computed a more conservative estimate by applying the estimated charge difference obtained from our conservative restricted sample ($65.26, see Exhibit 2). Here, we find estimated excess charges of $5 billion in 1993 dollars. Policy Implications These are rough estimates. But even these rough calculations suggest that there are significant excess expenditures associated with the use of emergency departments for nonurgent care. They also suggest that policies that make alternative, less expensive sources of care available to patients and that encourage patients to use such facilities may achieve savings. Such policies would have to address patients' incentives and knowledge regarding appropriate sources of care. It is important to note, however, that it may still be difficult to achieve all of these savings. Emergency departments probably have substantial fixed costs, which may be difficult to reduce without shutting facilities down. If our estimates of excess charges do not represent excess variable costs, they may not represent achievable savings. There are also significant access issues to be surmounted. In some places and at some times there may be no alternative sources of care available. Much of this work was completed while the authors were at The Robert Wood Johnson Foundation and (Laurence) Princeton University. This paper represents the opinions of the authors. No endorsement by The Robert Wood Johnson Foundation or The David and Lucile Packard Foundation is intended or should be inferred. The authors thank Joel Cantor and Steven Schroeder for helpful comments and discussions. NOTES 1. See, for example, L.F. McCaig, National Hospital Ambulatory Medical Care Survey : 1992 Emergency Department Summary, Advance Data from Vital and Health Statistics, no. 245 (Hyattsville, Md.: National Center for Health Statistics, 1994); L.S. Gage et al., Americas Urban Health Safety Net (Washington: National Association of Public Hospitals, January 1994), 32-39; U.S. General Accounting Office, Emergency Departments: Unevenly Affected by Growth and Change in Patient Use, GAO/HRD-93-4 (Washington: GAO, January 1993); L.J. Kelly and R. Birtwhistle, "Is This Problem Urgent? Attitudes in a Community Hospital Emergency Room," Canadian Family Physician (June 1993): 1345-1352; P.A. Driscoll, C.A. Vincent, and M. Wilkinson, "The Use of the Accident and Emergency Department," Archives of Emergency Medicine 4 (1987): 77-82; and J.S. Gaveler and D.H. Van Thiel, "The Nonemergency in the Emergency Room," Journal of the National Medical Association 72, no. 1 (1980): 33-36. 2. McCaig, "National Hospital Ambulatory Medical Care Survey." 3. See, for example, S.A. Schroeder, "The Increasing Use of Emergency Services: Why Has It Occurred? Is It a Problem?" Western Journal of Medicine (January 1979): 67-69. 4. For example, Schroeder telephoned five emergency departments and asked the price
DATAWATCH 171 of treatment for a sore throat and found that it cost between $63 and $99. Without direct comparison to charges for visits in other settings, he suggested that those costs are high. Schroeder, "The Increasing Use of Emergency Services." The other estimate we are aware of is based on a comparison of the average cost of an office visit and an emergency department visit for any condition, not adjusting for condition, visit, or patient characteristics. S. Rich, "Most Emergency Cases Aren't," The Washington Post, 3 March 1994, A7; and personal communication with Linda McCaig, National Center for Health Statistics, April 1994. 5. Agency for Health Care Policy and Research, National Medical Expenditure Survey, 1987: Household Survey, Population Characteristics and Person-Level Utilization, Rounds 1-4, Public Use Tape 13 (1992); National Medical Expenditure Survey, 1987: Ambulatory Medical Visit Data, Public Use Tape 14.5 (1993); and National Medical Expenditure Survey, 1987: Household Survey, Health Status Questionnaire and Access to Care Supplement, Public Use Tape 9 (1993) (Ann Arbor, Mich.: Inter-University Consortium for Political and Social Research). 6. See documentation for NMES 1987: Ambulatory Medical Visit Data, Public Use Tape 14.5 (1993) for more information. In cases for which visit charges were missing or unavailable, they were assigned using imputation techniques. See B. Hahn and D. Lefkowitz, "Annual Expenses and Sources of Payment for Health Care Services," AHCPR Pub. no. 93-0007 (Rockville, Md.: AHCPR, November 1992). 7. We considered other break lengths and found that they made little difference. 8. R.I. Haddy, M.E. Schmaler, and R.J. Epting, "Nonemergency Emergency Room Use in Patients With and Without Primary Care Physicians," The Journal of Family Practice 24, no. 4 (1987): 389-392. 9. Defined in our sample to also include uncomplicated minor contusions. 10. A list of the specific ICD-9 codes used within each condition category is available from the authors upon request. Contact Laurence Baker, Department of Health Research and Policy, Stanford University, Stanford, CA 94305. 11. The log of charges is the dependent variable since it gave a better regression fit than the untransformed charges. Predicted charges were converted from logarithmic form using both standard estimators and a smearing estimator suggested. See N. Duan, "Smearing Estimate: A Nonparametric Retransformation Method," Journal of the American Statistical Association 78 (1983): 605-610. The different estimators gave highly similar results; estimates based on the standard estimator are presented. We also experimented with reduced sets of independent variables and found that, where there were differences, the results we present are the smaller (more conservative) estimates. Details of the regression analysis are available from the authors upon request. 12. These estimates do not include the total number of visits in the episode as a control variable. Inclusion of this variable would control for difference in episode length that should be left as part of the difference we examined. 13. McCaig, "National Hospital Ambulatory Medical Care Survey." 14. Ibid. 15. $93.85 x 49.7 million visits = $4.7 billion (1987 dollars). The medical CPI indicates 54.8 percent cost growth between 1987 and 1993. We get $4.7 billion x 1.548 = $7.2 billion (1993 dollars).