ACADEMIC EMERGENCY MEDICINE 2011; 18:158–166 © 2011 by the Society for Academic Emergency Medicine
Objectives: The objective was to describe the emergency department (ED) resource burden of the spring 2009 H1N1 influenza pandemic at U.S. children’s hospitals by quantifying observed-to-expected utilization.
Methods: The authors performed an ecologic analysis for April through July 2009 using data from 23 EDs in the Pediatric Health Information System (PHIS), an administrative database of widely distributed U.S. children’s hospitals. All ED visits during the study period were included, and data from the 5 prior years were used for establishing expected values. Primary outcome measures included observed-to-expected ratios for ED visits for all reasons and for influenza-related illness (IRI).
Results: Overall, 390,983 visits, and 88,885 visits for IRI, were included for Calendar Weeks 16 through 29, when 2009 H1N1 influenza was circulating. The subset of 106,330 visits and 31,703 IRI visits made to the 14 hospitals experiencing the authors’ definition of ED surge during Weeks 16 to 29 was also studied. During surge weeks, the 14 EDs experienced 29% more total visits and 51% more IRI visits than expected (p < 0.01 for both comparisons). Of ED IRI visits during surge weeks, only 4.8% were admitted to non–intensive care beds (70% of expected, p < 0.01), 0.19% were admitted to intensive care units (44% of expected, p < 0.01), and 0.01% received mechanical ventilation (5.0% of expected, p < 0.01). Factors associated with more-than-expected visits included ages 2–17 years, payer type, and asthma. No factors were associated with more-than-expected hospitalizations from the ED.
Conclusions: During the spring 2009 H1N1 influenza pandemic, pediatric EDs nationwide experienced a marked increase in visits, with far fewer than expected requiring nonintensive or intensive care hospitalization. The data in this study can be used for future pandemic planning.
On April 21, 2009, the Centers for Disease Control and Prevention (CDC) reported the first cases of novel 2009 influenza A (H1N1) virus (2009 H1N1 influenza) in the United States.1 Unlike prior reported seasonal influenza outbreaks, almost half of patients hospitalized in the first wave of the 2009 pandemic were children.2,3 Due to mild virulence4 and the predominance in children, a disproportionate numeric burden was experienced in emergency department (ED) settings treating large numbers of children.5–7 A CDC analysis estimated that for every patient hospitalized, 100 patients sought outpatient care, including ED visits.8
Although the pandemic disproportionately affected children and outpatient settings, only two articles, each reporting pediatric utilization at two EDs in a single city, have reviewed ED utilization.6,7 Most other published reports regarding the spring 2009 H1N1 influenza have relied on surveillance by the CDC, which focused on hospitalizations, deaths, positive virologic specimens, and the percentage of outpatient visits for influenza-related illness (IRI).9 In addition, regional descriptions of hospitalizations and deaths associated with the April–July 2009 pandemic have been reported in the United States.2,10–12 These descriptions report a small fraction of actual utilization, due to the comparatively mild spectrum of illness caused by the 2009 H1N1 influenza virus, and CDC guidance to restrict testing13 and treatment.14 Because of this, most ED patients were not tested for influenza, most were not hospitalized, and few died. Thus, a gap remains in our understanding of the impact of the pandemic because their visits have gone largely unreported. Consequently, the overall resource burden these visits placed on the nation’s health care systems remains unknown.
To the best of our knowledge, this is the first study to examine the resource burden on pediatric EDs nationally during the spring wave of the 2009 H1N1 influenza pandemic. The objectives of this study were to describe the ED resource burden of the pandemic at tertiary care children’s hospitals in the United States by quantifying observed utilization as a ratio of expected utilization. Because children’s hospitals play an integral role in coordination of health delivery in many regions, quantifying the resource burden placed on children’s hospitals will allow better planning and preparedness for future pandemics.
This was an ecological analysis. The study protocol was approved by the Colorado Multiple Institutional Review Board with a waiver of informed consent.
Study Setting and Population
We used data from the Pediatric Health Information System (PHIS), which contains demographic and resource utilization data from 41 nonprofit tertiary care U.S. children’s hospitals, including 23 hospitals that had ED data available continuously from 2004 to 2009. The Child Health Corporation of America (Shawnee Mission, KS) and participating hospitals jointly validate data quality and reliability.15 Because PHIS is not designed to obtain national estimates, it was beyond the scope of this analysis to report population rates, although this has been estimated by others.5
We included all ED visits during April–July 2009 (Calendar Weeks 13–29) at all of the 23 hospitals with ED data. Weeks 13–15 were included (Figure 1) only to illustrate the weeks immediately preceding circulation of 2009 H1N1 influenza; the study population for all analyses was from Weeks 16–29 only.
Due to nonspecific or delayed symptoms and limited use and accuracy of viral testing,16,17 few persons with IRI receive a confirmed diagnosis. We therefore used the Electronic Surveillance System for the Early Notification of Community-Based Epidemics (ESSENCE) II list of International Classification of Diseases, 9th revision (ICD-9), codes developed for influenza syndromic surveillance in the military18,19 and validated for the 2009 H1N1 influenza pandemic.20 Although pediatric-specific code lists are available,21–24 ESSENCE II is the only set validated against virologic results. Our case definition for IRI included primary and secondary ICD-9 discharge codes (ED and inpatient) from this list and included suspected (487) and identified (488) influenza virus (Data Supplement S1, available as supporting information in the online version of this paper).23,25
Definition of ED Surge Period. To define surges, we first estimated expected weekly ED visits during the study period. For each hospital ED, to adjust for secular trends, we fitted a time series curve using an autoregressive error model with a lag of 3 to the weeks of April–July for years 2004–2008 and forecasted the expected ED visits for each week of April–July 2009. We chose to focus on the April–July first wave of the pandemic because the epidemiology of the virus was unknown at this time, and visits were being driven as much by public concern due to media reports of deaths in Mexico as by actual clinical presentation. This would most closely correspond to the public response to a true pandemic and possibly lethal virus (e.g., severe acute respiratory syndrome [SARS]).
We used data-driven methods to determine an ED surge threshold for each hospital, as several reports have noted the absence of a standard threshold.26–30 For the study period, we defined the upper 90% confidence limit on the forecasted weekly values as the ED-specific surge threshold for each week.
The start of each surge for each hospital was defined by the first of 2 consecutive weeks with ED IRI visits above the modeled, hospital-specific surge threshold. Similarly, the end of each surge period was defined as the next 2 consecutive weeks with ED IRI visits less than the hospital-specific surge threshold. Fourteen hospitals met this definition of surge. We selected the 90% limit (for two consecutive weeks) based on the impression of the authors (ED providers and hospital administrators) of a level/duration that would challenge a typical ED’s capacity to provide an unaltered standard-of-care.30
Influenza Circulation Data. Influenza circulation data were obtained from U.S. World Health Organization (WHO) collaborating laboratories and the National Respiratory and Enteric Virus Surveillance System (NREVSS).9 The onset of the 2009 H1N1 influenza period for each of the 10 Department of Health and Human Services (HHS) regions began with the first week that any specimen tested positive for 2009 H1N1 influenza. We included these data to define the period in which ED surge could be attributable to 2009 H1N1 influenza and to define a period when H1N1 was known to be circulating as a basis for comparing surge (n = 14) and nonsurge (n = 9) hospitals.
Patient Characteristics. We used the age and race/ethnicity categories used in a prior CDC report.31 Race/ethnicity assignment varies across hospitals and can be based on patient or parent report or registration personnel assignment. Insurance was categorized as private, public (Medicare, Medicaid, or a state-sponsored health plan), or none. Although military insurance is publicly funded, we chose to group it with private insurance as it is obtained through an employer. Geographic regions were assigned by grouping PHIS hospitals by HHS region as follows: Middle Atlantic (Region 3; three hospitals), North Central (Region 5; six hospitals), South Atlantic (Region 4; five hospitals), South Central (Regions 6–7; five hospitals), and West (Regions 8–10, four hospitals).
The CDC and the American Academy of Pediatrics have identified high-risk conditions associated with “elevated risk for complications” from 2009 H1N1 influenza.32 We identified corresponding ICD-9 codes (Data Supplement S2, available as supporting information in the online version of this paper), drawing on an earlier list of chronic medical conditions recognized by the Advisory Committee on Immunization Practices as indication for seasonal influenza vaccination.33,34 The resulting tabulation of high-risk conditions contained eight categories: neurologic disorders; asthma; other chronic respiratory conditions; developmental delay; immune deficiencies; cardiovascular disease; endocrine/metabolic conditions; renal, hepatic, or hematologic conditions; chronic aspirin therapy; and pregnancy. We analyzed asthma separately from other chronic respiratory conditions as it is the most common high-risk condition and has had prior separate analysis.3,34
Utilization Measures. For each utilization, severity, and quality outcome measure, we derived the expected value for the 2009 study period using the same methods described for derivation of the expected ED IRI weekly visits (i.e., forecasting from autoregressive error models fit to the weeks of April–July for years 2004–2008). For the weeks of the spring 2009 H1N1 influenza pandemic, we measured the observed value for each variable and calculated the observed-to-expected ratio.
Utilization variables were aggregated across hospitals. For the hospital-level analysis, we reported IRI visits, as these were most indicative of influenza-attributable visits, and all ED visits, as these conveyed the total burden placed on the ED. We also described the use of viral testing and administration of anti-influenza medications.
Severity Measures. We calculated the frequency distribution for each of the four severity levels (minor, moderate, major, and extreme) in the All Patient Refined Diagnosis-Related Groups (APR-DRG) classification for patients admitted to the hospital; this level was not available for patients discharged from the ED.35 This severity-of-illness indicator, based on the specific diagnoses and procedures performed during a patient’s hospitalization, was included to determine whether the severity of patients admitted increased during the 2009 study period at surge hospitals. We also measured the total number of in-hospital deaths, including both ED and inpatient deaths.
Quality Measures. We assessed ED quality of care using ED revisits within seven days, both for all ED patients and for the subset of ED IRI patients.36–38 Return visits were included for any cause and were reported weekly, using expected 2009 weekly return visit rates as a comparison. Of the most common ED quality measures, this is the only measure available in PHIS at a weekly level, other than mortality, which is reported here as a severity measure.
Frequency distributions for patient characteristics for all ED visits with an IRI code were calculated for all 23 hospitals during the period when the CDC surveillance identified circulating 2009 H1N1 influenza (Calendar Weeks 16–29). To describe the hospital-level variation in influenza-related utilization, we first performed hospital-level analyses. We calculated the duration of surge in weeks and the number of surge weeks. We calculated the mean observed-to-expected ratio for both ED and ED IRI weekly visits during surge weeks for the 14 hospitals who met the study’s surge criteria. To determine all expected values, we fit autoregressive error models time series curves with lags of 3 weeks to all weeks for January 2004–March 2008 and defined the expected values for each week of April–July 2009 as the forecasted values. We used the Box-Jenkins approach to identifying and forecast autoregressive integrated moving average (ARIMA) models and made use of the autocorrelation function to determine appropriate autoregressive and moving average orders.39 Parameter estimation was performed using maximum-likelihood estimation. For all 23 hospitals, we also reported the ED and ED IRI weekly visits for the peak week and for Calendar Weeks 16–29 (14 weeks for 23 hospitals = 322 hospital-weeks). These weeks were selected to start with the date of the first positive surveillance samples for 2009 H1N1 in the CDC’s surveillance reports and to conclude with the last week of the study period—at this week, H1N1 flu was still circulating in WHO/NREVSS reports.
For all surge hospital-weeks, we calculated the mean observed-to-expected ratio for all utilization, severity, and quality measures, using the CDC’s methods for comparing mortality data to 5 years of prior data.37 We then calculated mean observed-to-expected ratios for ED visits, ED admission rates, and in-hospital mortality rates for surge hospital-weeks by patient characteristics to identify patients most likely to be admitted and patients at highest risk for each outcome. We then examined temporal data, grouping hospitals by region, for the observed-to-expected ratio for ED visits and ED IRI visits.
All statistical analyses were performed using SAS v.9.2 (SAS Institute, Cary, NC), and p-values of <0.05 were considered statistically significant.
The 23 study hospitals recorded 390,983 total ED visits and 88,885 ED IRI visits during the 14-week study period. Of these visits, 45% of the patients were less than 24 months of age, and 70% were less than 60 months of age (Table 1). The most common high-risk conditions were asthma (10% of all ED IRI visits), neurologic disorders (1.1%), and cardiovascular diseases (0.7%). The regional distribution of visits reflected the number of study hospitals in each of the five regions.
|0 to <6 months||8,680 (9.8)|
|6 to <24 months||31,126 (35.0)|
|24 to <60 months||22,677 (25.5)|
|5 to < 9 yr||13,211 (14.9)|
|9 to <13 yr||7,330 (8.2)|
|13 to <18 yr||5,130 (5.8)|
|18 to <22 yr||731 (0.8)|
|Female sex||40,990 (46.1)|
|White, not Hispanic||18,456 (21.3)|
|African American, not Hispanic||28,459 (32.8)|
|Neurologic disorders||998 (1.1)|
|Other chronic respiratory||528 (0.6)|
|Developmental delay||260 (0.3)|
|Immune deficiencies||520 (0.6)|
|Cardiovascular disease||582 (0.7)|
|Renal, hepatic, hematologic||168 (0.2)|
|Chronic aspirin therapy||30 (0)|
|>1 High-risk condition||975 (1.1)|
|Region (number of study hospitals in region)|
|Middle Atlantic (3)||9,562 (10.8)|
|North Central (6)||28,828 (32.4)|
|South Atlantic (5)||19,041 (21.4)|
|South Central (5)||14,203 (16.0)|
|West (4)||17,251 (19.4)|
When examined temporally, all five geographic regions experienced a peak of ED IRI visits in Week 17 (Figure 1). The Middle Atlantic region started from a low baseline observed-to-expected visit rate, so the Week 17 rise brought the region’s observed-to-expected visit ratio to just above 1.0 (Data Supplement S3, available as supporting information in the online version of this paper).
Fourteen of 23 hospitals (61%) met the study definition of an ED IRI surge during the study period, with a total of 71 surge-level hospital-weeks. The West region had the highest ratio of surge hospitals (4/4), and the Middle Atlantic the least (1/3). Data Supplement S4 (available as supporting information in the online version of this paper) shows the observed ED IRI visit counts at surge and nonsurge hospitals, respectively. For each hospital, the weekly counts are compared with the expected counts (mean and 90% upper confidence limit). Shaded areas indicate the period(s) of epidemic-level circulation of 2009 H1N1 influenza in the hospital’s HHS region based on WHO/NREVSS laboratory reporting. Because the surge started within a week or two of the decline of the seasonal flu season, a few surge hospitals show high activity even before Week 17, and one hospital (South Central 4) showed surge-level activity starting in Week 15. A few nonsurge hospitals (South Central 2 and 3) showed peaks that crossed the surge threshold for 1 week, but were not sustained for the 2 weeks required under the surge hospital criteria.
For surge hospitals, the median surge duration was 5 weeks (interquartile range [IQR] = 2–6.75 weeks). The median hospital observed-to-expected ratios of ED visits and ED IRI visits for the 14 surge hospitals during surge weeks were 1.24 (IQR = 1.17–1.30) and 1.46 (IQR = 1.31–1.68), respectively, with peak week ratios of 1.36 (IQR = 1.28–1.45) and 1.76 (IQR = 1.53–2.24), respectively. Overall, this resulted in 24,200 additional ED visits for the 14 hospitals during the 71 surge-level hospital-weeks, an average of 49 additional ED visits per ED-day (Table 2).
|Total weekly ED visits||82,130||106,330||1.29*|
|Total weekly ED visits with IRI||21,047||31,703||1.51*|
|Disposition (% of ED IRI visits)|
|Admitted to non-ICU bed||6.83||4.79||0.70*|
|Admitted to ICU||0.42||0.19||0.44*|
|Died in ED or hospital†||0.00||0.04||—|
|APR-DRG severity level for IRI patient visits resulting in admission from ED|
|Testing during ED IRI visits|
|Influenza testing done during visits discharged from ED, %||5.00||14.12||2.82*|
|CXR done in ED (or Hospital Day 1), %||1.19||0.82||0.69*|
|Medications/procedures done in ED (or Hospital Day 1), ED IRI visits|
|Anti-influenza medication given, %||0.00||0.53||159.52*|
|Antibacterial medication given||13.87||9.00||0.65*|
|Systemic corticosteroid given||4.03||3.32||0.82*|
|Positive pressure ventilation, %||0.20||0.01||0.05*|
|Quality measures, % of ED visits|
|7-day ED revisits for ED IRI visits||4.95||3.91||0.79*|
|7-day ED revisits for all ED visits||4.27||4.03||0.94‡|
Severity and Utilization Measures
Surge patients with IRI were less ill than expected. Compared with the percentage expected, the percentages admitted from the ED to non-ICU and to ICU beds were 30 and 56% lower, respectively. Among IRI surge patients who were admitted, severity was higher than expected, with fewer minor-to-moderate severity admissions and more major-to-extreme severity admissions than expected. The lower overall acuity was also reflected in other utilization; there was less overall diagnostic testing (including fewer chest radiographs) and less treatment with antibacterials, bronchodilators, and systemic corticosteroids. There was, however, an increase in influenza testing and treatment with anti-influenza medications compared with expected. Neither finding was surprising, as influenza does not traditionally circulate at this time of year. Despite the high surge volumes, there was a decrease in rate of return to the ED within 7 days (Table 2).
Patient Characteristics and Utilization Measures
During surge-level hospital-weeks the observed-to-expected ratios for ED IRI visits varied by patient subgroup (Table 3). Higher-than-expected ED IRI visits were seen for children 2–17 years (highest in the 9- to 17-year-olds), females, Hispanic patients, the insured, and those with asthma. Lower-than-expected ED IRI visits were seen for children less than 2 years of age, the uninsured, those with immune deficiencies and cardiovascular disease, and those in the Middle Atlantic region. No factors were associated with higher-than-expected ED-to-inpatient admission rates. Lower-than-expected admission rates were seen for 13- to 17-year-olds, those in the Middle Atlantic and North Central regions, and those with immune deficiencies.
|ED visits||Proportion of ED visits hospitalized|
|0 to <6 months||0.71*||0.78|
|6 to <24 months||0.86*||0.83|
|24 to <60 months||1.15*||0.74|
|5 to < 9 yr||1.45*||0.70|
|9 to <13 yr||2.08*||0.74|
|13 to <18 yr||2.02*||0.41*|
|18 to <22 yr||1.62||—|
|White, not Hispanic||0.72*||0.88|
|African American, not Hispanic||0.77*||1.08|
|Other chronic respiratory||1.2||1.70|
|Renal, hepatic, hematologic‡||—||—|
|Chronic aspirin therapy‡||—||—|
|>1 high-risk condition||0.92||1.25|
To the best of our knowledge, this is the first study focused on the resource burden and severity of ED visits across a large nationwide sample of children’s hospitals during the spring 2009 H1N1 influenza pandemic. Children’s hospitals form the cornerstone for provision and coordination of health care for children throughout their regions. Quantifying the magnitude of the surge of additional patients and their severity of illness helps planning for the next pandemic in the context of already crowded EDs and a resource-limited health care environment.
Fourteen of 23 hospitals (61%) experienced a surge in ED volume, based on our definition of exceeding the hospital’s own upper 90% confidence limit on forecasted weekly visits for 2 consecutive weeks. For hospitals experiencing a surge, the resource burden was operationally substantial: ED visits were 24% higher than expected during surge weeks and, during peak weeks, were 30% higher than expected. The Middle Atlantic region EDs were largely spared from the higher-than-expected volumes experienced in other regions, particularly in the West. Factors contributing to the interhospital variation may have included local patient characteristics and health care–seeking patterns, variation in disease prevalence, hospital-level factors such as outreach efforts, local school district and health department policies, and the tenor of local media coverage.
The overall acuity remained low, with only a small minority of patients requiring hospitalization (5.0% of children presenting with IRI to the EDs), with far lower than expected rates of ICU (56% lower) and non-ICU admission (30% lower) and positive pressure ventilation (95% lower). These findings are all clinically as well as statistically significant and are consistent with reported local experiences.5–7
The mild presentation of our cohort, illustrated by the fact that only half met previously defined criteria for an encounter necessitating ED-level care and few were admitted, confirms suggestions that many of these patient encounters could have taken place in a less resource-intense ambulatory setting such as an existing clinic7 or an alternative care site set up to treat high volumes of low-acuity patients.38,40
The large influx of low-acuity patients challenged a nationwide health care system whose mainstays of pandemic preparedness focus on stockpiling supplies for the critically ill or injured, as opposed to having broadly distributed surge capacity to meet differing levels of need. This finding is helpful to future surge and health policy planning, as it highlights the need for a better-integrated delivery system that can appropriately match capacity to demand and acuity. Our “expected” measurements were based on observations during typical noninfluenza periods; had the excess influenza-related utilization occurred during the busier winter respiratory season, we may have come closer to exhausting surge capacity.
The increased use of the pediatric EDs for IRI is multifactorial based on previously reported survey data, including parents’ perception of limited access to primary care, parents’ relatively higher confidence in the tertiary care medical center, parents’ fear of adverse outcomes, inaccurate advice from media, and inappropriate referrals from providers, day care centers, and schools.7,41 While most patients could have received appropriate care in primary care settings, it is unclear whether this was feasible, given limited clinic staff, hours, and space.42 The higher utilization among the insured (uninsured patients 16% of expected vs. private 173% and public 129%) is consistent with findings that most children making nonurgent ED visits have insurance.43 Varied ED utilization by insurance could reflect level of fear, return-to-school requirements, or overwhelming of the primary care networks. Findings could help future public health education campaigns target specific populations that are more likely to overuse in IRI epidemics.
First, only 23 children’s hospitals were included. The experience at these hospitals may not be representative of experiences in other settings such as other children’s hospitals or EDs in general hospitals that also provide emergent care to children. Nevertheless, quantifying the burden placed on children’s hospitals during the most recent epidemic can likely inform strategic planning by hospitals caring predominantly for adults in anticipation of future epidemics where adults might be disproportionately affected.
Second, the study findings are specific to 2009 H1N1 influenza. Novel influenza viruses are inherently unpredictable in terms of their clinical severity, pathogenesis, infectivity, and transmission dynamics. It follows that a future pandemic might produce a substantially different health care utilization profile to the one presented in this study.
Third, we used the upper 90% confidence interval of the expected patient volume to define the surge period. As no data exist to support (or refute) this definition, there is potential to underestimate or overestimate both the duration and the magnitude of the surge. We considered higher and lower surge thresholds to explore the potential effect of our specific surge definition on our conclusions; alternate definitions of the surge threshold did not meaningfully alter our results. Fourth, although we found differences between hospitals’ experiences, it was beyond the scope of this analysis to assess predictors of experiencing a surge in ED volume.
Fifth, revisits can be related to diverse factors and do not always reflect the quality of the initial ED visit. Sixth, for our utilization measures, operational variability exists among the hospitals; when aggregated, this leads to heterogeneity, although this heterogeneity was mitigated by indexing utilization measures against the hospital’s own baseline practice using the observed-to-expected ratio. Finally, our use of a discharge diagnosis-based definition of IRI, rather than virologic confirmation of influenza illness, meant that some patients in our study may not have had influenza. The ESSENCE II codes used for our definition of IRI have been validated against virologic data, but the validation sample included only adults; some of the resulting codes, such as otitis media, may have been excluded in a pediatric population. This latter limitation would cause us to overestimate the resource burden attributed to the spring H1N1 influenza pandemic.
During the spring wave of the 2009 H1N1 influenza pandemic, tertiary pediatric EDs nationwide experienced a marked increase in visits, but only a minority of patients with influenza-related illness required hospitalization, and even fewer required intensive care. The data provided in this study can be used for future pandemic planning at hospital, community, and national levels. Future analyses could measure the pandemic’s impact on inpatient capacity and could model what utilization would have been had the H1N1 pandemic been as virulent or as widespread as other recent influenza pandemics.