SEARCH

SEARCH BY CITATION

Abstract

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. Limitations
  7. Conclusions
  8. References
  9. Supporting Information

ACADEMIC EMERGENCY MEDICINE 2011; 18:1330–1338 © 2011 by the Society for Academic Emergency Medicine

Abstract

Objectives:  The authors sought to determine which quality measures of analgesia delivery are most influenced by emergency department (ED) crowding for pediatric patients with long-bone fractures.

Methods:  This cross-sectional, retrospective study included patients 0–21 years seen for acute, isolated long-bone fractures, November 2007 to October 2008, at a children’s hospital ED. Nine quality measures were studied: six were based on the timeliness (1-hour receipt) and effectiveness (receipt/nonreceipt) of three fracture-related processes: pain score, any analgesic, and opioid analgesic administration. Three equity measures were also tested: language, identified primary care provider (PCP), and insurance. The primary independent variable was a crowding measure: ED occupancy. Models were adjusted for age, language, insurance, identified PCP, triage level, ambulance arrival, and time of day. The adjusted risk of each timeliness or effectiveness quality measure was measured at five percentiles of crowding and compared to the risk at the 10th and 90th percentiles. The role of equity measures as moderators of the crowding-quality models was tested.

Results:  The study population included 1,229 patients. Timeliness and effectiveness quality measures showed an inverse association with crowding—an effect not moderated by equity measures. Patients were 4% to 47% less likely to receive timely care and were 3% to 17% less likely to receive effective care when each crowding measure was at the 90th than at the 10th percentile (p < 0.05). For three of the six quality measures, quality declined steeply between the 75th and 90th crowding percentiles.

Conclusions:  Crowding is associated with decreased timeliness and effectiveness, but not equity, of analgesia delivery for children with fracture-related pain.

Pain is the most common reason for seeking care in the emergency department (ED), accounting for up to 78% of visits.1,2 Underuse of analgesics (oligoanalgesia) is common, especially among pediatric patients.3–7 Although the reasons for oligoanalgesia are multiple,5 five prior studies have found that higher crowding levels are associated with both delays of treatment and lack of pain control in patients with hip fracture, abdominal pain, or back pain in adult ED populations.8–12 No prior studies have assessed this association in a pediatric population.

The Institute of Medicine (IOM) and the Robert Wood Johnson Foundation have identified the effects of ED crowding on children as a research priority.13,14 In adult ED populations, crowding is associated with decreased quality across all six IOM quality dimensions: timeliness, effectiveness, equity, patient-centeredness, safety, and efficiency.15 Our investigation addresses this research priority by modeling the crowding-quality association in an ED population of children with acute long-bone fracture-related pain. We address the scarcity of applicable quality measures by developing several measures across three IOM dimensions of quality.

We selected pain related to acute long-bone fractures as the disease model for this study based on the prevalence of the condition, the clinical importance of timely care, and the existence of applicable clinical practice recommendations for pain management from the Joint Commission16 and the American Academy of Pediatrics.17 Extremity fractures are among the most common reasons children seek ED care, resulting in 850,000 ED visits annually nationwide.18

The objective of this study was to measure the association between ED crowding and the quality of pain management for children with long-bone fractures. We explored this objective in three of the six dimensions of quality: effectiveness, timeliness, and equity. Our secondary objective was to measure the dose–response effect of ED crowding on quality by comparing quality across crowding percentiles.

Methods

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. Limitations
  7. Conclusions
  8. References
  9. Supporting Information

Study Design

We performed a cross-sectional study with data collection via retrospective extraction from an electronic medical record (EMR). The Colorado Multiple Institutional Review Board found the study exempt from full human subjects review and informed consent requirements.

Study Setting and Population

Study variables were extracted from the EMR of a tertiary-care pediatric ED located in an academic children’s hospital staffed around-the-clock by attending physicians who are pediatric emergency medicine (PEM) board-certified. The number of ED beds remained fixed at 48 during the study period.

Our population consisted of patients ages 0–21 years treated for an isolated long-bone fracture in the study ED from November 1, 2007, to October 31, 2008. To preserve the independence of observations, we only included each patient’s first fracture-related visit during the study period. We included patients with an International Classification of Diseases, revision 9 (ICD-9), code for an isolated long-bone fracture (812, 813, 818, 820, 821, 823, 824, 827). Exclusion criteria included any ICD-9 code indicating major trauma (830–902, 925–926, 929, 940–954, 957, 958), a second fracture (800–829), or child maltreatment (995.5). We chose to focus on those with acute, isolated, long-bone fractures, so as to avoid patients with complex analgesia needs, such as hypotensive patients with multisystem trauma.

Study Protocol

We proposed quality measures for three of the six IOM quality dimensions (timeliness, effectiveness, and equity) based on their availability in the EMR. Of the three excluded dimensions, we omitted patient-centeredness and safety due to absence of patient-level measures at the study ED and omitted efficiency due to the difficulty of defining “good” quality for the most commonly used measure of efficiency, length of stay.15

We selected three processes to serve as the basis for timeliness and effectiveness measures: assignment of initial pain score, administration of first analgesic, and administration of first opioid analgesic. These processes were selected based on their availability in the EMR and their use in a prior study of crowding and fracture pain analgesia in adults.8 For all ED patients, the triage nurse assigns a pain score, based on the parent’s or patient’s assessment using (for preverbal or uncooperative children) the Faces, Legs, Activity, Cry, and Consolability Behavioral Pain Assessment Scale19 or (for verbal children) the Faces Pain Scale–Revised.20 Both use a 0–10 scale and correlate well to each other.21 The second process, administration of any analgesia, included nonsteroidal anti-inflammatory medications, acetaminophen, opioids, and drugs used for procedural analgesia, including ketamine. Regional analgesic agents were excluded, as lidocaine did not require an EMR order during the study period. However, in the study ED, regional analgesia is rarely, if ever, used as a first analgesic in patients with acute long-bone fracture.

To define for whom each of the three processes of care was indicated, we defined a relevant subgroup for each process, using methods derived from a prior study.22 For each of the three processes, we defined the relevant subpopulation based on the hospital’s pain policy, which requires pain scores for all ED patients. The policy recommends giving analgesia for any pain, which we defined as any nonzero pain score and recommends opioids for patients with moderate to severe pain, defined as a pain score ≥4. Nurses may order and administer acetaminophen and ibuprofen for extremity pain prior to attending provider evaluation, whereas opioids and other analgesics require a physician order. An additional exclusion criterion for the analgesia and opioid subpopulations was the presence of any analgesic identified via a string search of each patient’s EMR fields for prehospital medications, as prearrival analgesia may lead a clinician to delay ED analgesia appropriately.

For each of the three processes of care studied, we defined poor timeliness as a delay of ≥1 hour from the patient’s ED arrival. The 1-hour threshold was selected based on prior studies.8,23 The start time for each process was the patient’s ED arrival time. The end time for medications was the administration time in the EMR’s medication administration record. The end time for the initial pain score was the time stamp of completion in the triage flow sheet. Because indications for administration of the three processes coincided with indications for timely administration, the relevant subgroup for each timeliness measure included patients who did not receive the indicated process (and thus were categorized as not receiving it within 1 hour).

We selected equity variables to reflect both preferred language and access to care. The variable “preferred language” is routinely recorded for all patients by the triaging nurse and is used to help ED providers know by looking at the track board that they must access medical interpretation services for ongoing communication with the patient family. Based on this use of this field in real-time ED clinical operations, patients with a missing “preferred language” field were categorized as English-speaking. Studies of language and ED care have found increased resource utilization for patients with limited English language proficiency.24,25

Access-to-care variables have also been included in studies of ED analgesia;7,26 We included insurance, categorized as three values (no insurance, public insurance, private insurance), and presence of a primary care provider (PCP), which included both individual providers and clinics. Patients whose insurance or PCP status were missing were categorized as “no insurance” and “no PCP,” respectively. As data analysis was finalized more than 12 months after the close of the study period, all visits would have complete collection data, and an absence of insurance would reflect the patient’s true insurance status rather than missing insurance data. The EMR uses PCP identifiers to transmit a record of each ED encounter to the patient’s PCP; thus, this information is requested from all patient families during registration, and missing data are likely to reflect absence of a medical home. Insurance and PCP variables are gathered during the registration process, which, in the study ED, can occur before and/or after a patient is clinically assessed and treated; thus, higher acuity patients are not more likely to be missing insurance or PCP data.

To optimize the accuracy of timeliness measures, an investigator who is PEM board-certified (MRS) reviewed a sample of 20 charts selected from outliers with a time-to-medication greater than 240 minutes. Two types of error were detected. First, for patients with a time of administration more than 10 minutes after the end of the ED stay, chart review showed that 100% had an apparent error of either 12 or 24 hours in the time/date of their medication administration. These were corrected by the appropriate interval, so that the event time fell within the ED length of stay. Second, among patients admitted to the hospital with a time stamp indicating an event time more than 10 minutes past the ED-to-inpatient transfer time, chart review indicated that 100% of events thus time-stamped occurred on the inpatient ward, rather than in the ED. These were corrected by excluding the event from our data set (i.e., they were considered to have not received the process in the ED).

To optimize the accuracy of the effectiveness measures, a PEM board-certified reviewer (MS) and a PEM fellow (MM) performed chart review on 20 randomly selected from the 446 patients who did not receive an indicated opioid and 20 randomly selected from the 655 who did not receive an indicated analgesic of any type. The department’s research coordinator (SD) served as a tie-breaker. In none of the 40 charts was an opioid or other analgesic documented in provider notes that had not already been captured in the orders; thus, no tie-breaking was required and no data corrections were made. The investigators did not examine the quality of the pain score coding and capture, as the EMR contained no measure against which to validate the pain score field.

We used the EMR to retrospectively assign two ED crowding indicators at each study patient’s ED arrival time: ED occupancy (percentage of ED beds filled) and the number of patients waiting to see an attending provider. These were included based on their inclusion in prior studies of ED crowding and analgesia.8,10,12

To assess the reliability of coding and capture of crowding measures, the lead author (MS) performed real-time counts in the ED at 10 specific times; all were identical to counts obtained via the study’s data extraction methods. Observation, trauma, and fast-track patients were included in crowding measures because they are cared for by ED staff in ED bed spaces.

Prior studies of the association between crowding and quality of ED analgesia have adjusted for known confounders of the ED administration of analgesia: demographics (age, sex, and race) and triage level.8–12 We included age and, as noted above, we used preferred language, but our data set lacked reliable race and ethnicity data as described under “Limitations.” To minimize confounding by severity, we adjusted for triage level and arrival by ambulance. The study ED used a four-level triage system, which we dichotomized, combining the two highest (1 and 2) and two lowest (3 and 4) levels.8 We included an indicator of evening time of arrival to adjust for the independent contribution of time-of-day to quality of care.12

Data Analysis

Frequency distributions were calculated for all nine quality measures. Distributions of the crowding measures were examined. The distribution for number waiting for attending was highly skewed and thus was log-transformed. Data are reported comparing quality measures at 10th and 90th percentiles of crowding so as to describe effects over a wide range of crowding. We also included interquartile ranges (IQRs), as in a prior crowding study.27

We then separately modeled the association between the outcomes (the three effectiveness and three timeliness quality measures) and each of the two crowding measures, adjusting for the demographic, history, and severity variables described previously. Logistic regression models with simple linear splines with knots at the 25th, 50th, and 75th percentiles were fit to check the linearity of the relationship between each outcome and crowding measure. The initial model was logit (Outcome) = Intercept + age + preferred language (indicator for not English) + insurance (indicators for any insurance, and public insurance) + absence of a PCP + evening time arrival + triage level + ambulance arrival + crowding (linear spline including slope and changes in slope at 25th, 50th, and 75th percentiles).

For this equation, we first ran a model for each of the six outcomes including both crowding measures and found a correlation between the two crowding measures of 0.9. Due to this multicolinearity, we eliminated number waiting to see attending from further analyses. Of the two measures, we selected ED occupancy as it uses the more widely available and standardized arrival, discharge, and transfer data, whereas the other measure relies on definitions of “waiting” that vary more between institutions.

Retaining ED occupancy as our crowding measure, we ran six models: one for each of the three effectiveness measures (receipt/nonreceipt of initial pain score, analgesic, and opioid analgesic) and one for each of the three timeliness measures (first-hour receipt of pain score, analgesic, and opioid analgesic). Nonsignificant terms of the crowding spline function were dropped from the model using backward selection; all other covariate terms were retained.

We then calculated the adjusted relative risk (ARR) using the method proposed by Kleinman and Norton,28 converting odds ratios to the more intuitive relative risk. For each quality measure, the risk (mean predicted value) at the 10th, 25th, 50th, 75th, and 90th percentiles of ED occupancy were displayed graphically. We then calculated the interdecile ARRs and the ratio of the mean predicted risk for the quality measure at the 90th percentile and the 10th percentile and assessed the significance of this ratio. We also report selected values of the ratio of the interquartile risk ratio and the adjusted risk for the 75th to the 25th percentiles.

Because the equity measures are neither process nor outcome measures of quality, we analyzed them as moderators of the association between crowding and quality measures by adding interaction terms to the above model. We set significance levels to p ≤ 0.01 to partially adjust for multiple comparisons.

All data extractions were performed using Crystal Reports XI (Business Objects, San Jose, CA) to query data in EpicCare (Epic Systems Corporation, Verona, WI), a computerized patient tracking and charting system. In EpicCare, all ED events, including arrival, discharge, clinical assessments, and medication administration, are automatically time stamped. All data were imported into SAS 9.2 (SAS Institute Inc., Cary, NC) for analysis.

Results

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. Limitations
  7. Conclusions
  8. References
  9. Supporting Information

During the 12-month study period, the ED registered 56,900 patients; of these, 1,444 patients met inclusion criteria, and 215 were excluded for prehospital analgesia, resulting in a final sample size of 1,229. Patient characteristics are shown in Table 1. Preferred language was English in 77.9% and Spanish in 21.5%; 0.6% had “other” language and were classified as non-English preference, and 0.6% had no language preference recorded and were excluded. Regarding access variables, 6.6% had no insurance, 56.7% had public insurance, 36.7% had private insurance, and 14.4% had no PCP. Of the 1,229 patients, 18.6% were admitted and 43.4% were categorized in one of the two most severe triage levels. The median pain score was 4 (IQR = 1 to 7).

Table 1.    Patient Characteristics (n = 1,229)
VariableValue
  1. IQR = interquartile range; PCP = primary care provider.

  2. Values are expressed as n (%) unless otherwise noted.

  3. *Evening was defined as 4 PM until midnight.

Demographics
 Age (yr), median (IQR)6 (3–10)
 Female498 (40.5)
 Preferred language English958 (77.9)
 Preferred language Spanish264 (21.5)
 Preferred language other 7 (0.6)
 No insurance 81 (6.6)
 Public insurance 697 (56.7)
Initial history information
 Has PCP 1052 (85.6)
Initial examination information
 Initial pain score, median (IQR)4 (1–7)
 Triage level 1 or 2 (most severe) 534 (43.4)
Disposition
 Admitted228 (18.6)
 Discharged998 (81.2)
 Left after medical screening examination1 (0.1)
 Transfer to another facility2 (0.2)
Other
 Evening* ED arrival time595 (48.4)
 Ambulance arrival131 (10.7)

The nine quality measures, relevant subgroup sizes, and univariate findings are shown in Table 2. Relevant subgroup sizes for the timeliness and effectiveness measures were as follows: 1,229 were in the pain score indicated subgroup, 958 were in the any analgesia indicated subgroup, and 658 were in the opioid indicated subgroup. For the effectiveness measures, we found high levels of assignment of a pain score (97.3%), but low levels of administration of any analgesia (31.6%) or opioid analgesia (32.2%) within relevant subgroups. For the timeliness measures, we found high levels of assignment of a pain score within the first hour (94.9% overall and 97.5% among those getting a pain score), but low levels of administration of any analgesia within 1 hour (12.7% within subgroup and 40.3% of those receiving any analgesia) or opioid analgesia (13.8% within subgroup and 42.9% of those receiving opioid analgesia). The median (and 10th, 25th, 75th, and 90th percentiles) for ED occupancy were 49 (and 23, 34, 66, and 81, respectively). These values correspond to the x-axes in the Figures 1A and 1B.

Table 2.    Quality Measures of ED Fracture-related Analgesia
DimensionQuality MeasuresRelevant Subgroup*n (%)
  1. PCP = primary care provider.

  2. *Numbers in parentheses indicate the number of patients in the denominator relevant to the quality measure.

TimelinessTime to pain score < 1 hr Pain score indicated subgroup (1,229)1,166 (94.9)
Time to any analgesia < 1 hr Any analgesia indicated subgroup (958)122 (12.7)
Time to opioid analgesia < 1 hr Opioid indicated subgroup (658)91 (13.8)
EffectivenessReceipt of pain scorePain score indicated subgroup (1,229)1,196 (97.3)
Receipt of any analgesicAny analgesia indicated subgroup (958)303 (31.6)
Receipt of opioid analgesicOpioid indicated subgroup (658)212 (32.2)
EquityPreferred language EnglishPatients with preferred language recorded (1,222)958 (78.4)
Has PCPPatients with primary care provider status recorded (1,229)1,052 (85.6)
InsuranceAll (1,229)81 (6.6)
No insurance697 (56.7)
Public insurance451 (36.7)
Private insurance
image

Figure 1.  (A) Mean predicted probability for timeliness processes by crowding measure percentile. Each line graph shows the adjusted risk of timeliness processes on the y-axis and the crowding percentiles on the x-axis for ED occupancy. The 95% CI is indicated by y-error bars. The adjusted risk values were adjusted for age, preferred language, insurance, presence of a PCP, triage level, ambulance arrival, and time of day. The boxed numbers correspond to the ratio of the adjusted risk at the 90th compared to the 10th crowding percentiles. (B) Mean predicted probability for effectiveness processes by crowding measure percentile. Each line graph shows the adjusted risk of effectiveness processes on the y-axis and the crowding percentiles on the x-axis for ED occupancy. The 95% CI is indicated by y-error bars. The adjusted risk values were adjusted for age, preferred language, insurance, presence of a primary care provider, triage level, ambulance arrival, and time of day. The boxed numbers correspond to the ratio of the adjusted risk at the 90th compared to the 10th crowding percentiles. PCP = primary care provider.

Download figure to PowerPoint

The associations between crowding percentiles for ED occupancy and the adjusted risk for the three effectiveness and three timeliness measures are shown in Figures 1A and 1B, respectively. These show a significant overall slope (10th to 90th percentile) in five of the six models. Two of the six, timely receipt of any analgesia and opioid analgesia, show a dose-related association with ED occupancy. For the other three timeliness and effectiveness measure lines with a significant overall slope, the effects of ED crowding on analgesia administration are nonlinear. The adjusted risk remains stable, moving from the 10th to the 75th percentile, and then declines more steeply between the 75th and 90th percentiles. (Crowding-quality models with parameter estimates for all variables are provided in Data Supplement S1, available as supporting information in the online version of this paper.)

Table 3 shows the adjusted relative risk for the three effectiveness and three timeliness measures comparing the risk of receiving each process measure at the 90th and 10th percentiles (interdecile comparison) of crowding, and the 75th and 25th percentiles (interquartile comparison). As also shown in Figures 1A and 1B, five of the six interdecile associations were significant. The changes in relative risks were of greater magnitude in the timeliness measures than in the effectiveness measures. Patients were 4% (95% confidence interval [CI] = 2% to 7%) to 47% (95% CI = 13% to 71%) less likely to receive timely care when each crowding measure was at the more-crowded (90th) than at the less-crowded (10th) percentile. They were 3% (95% CI = 1% to 5%) to 17% (95% CI = 2% to 32%) less likely to receive effective care at the 90th than at the 10th percentile. For both timeliness and effectiveness measures, the magnitude of the adjusted association with crowding was lowest with pain score and greatest with medication administration. Across the six interquartile models, two timeliness models (timeliness of any analgesia and of opioid analgesia) and no effectiveness models showed a significant association with crowding (Table 3). The greater number of significant findings in our interdecile ratios reflects the drop-off in quality between the 75th and 90th percentiles shown in Figures 1A and 1B.

Table 3.    Ratio Comparing the Adjusted* Risk for Timeliness and Effectiveness Measures at the 90th and the 10th Percentiles, and the 75th and 25th Percentiles, of ED Occupancy
DimensionRelevant SubgroupProcessAdjusted Relative Risk (95% CI)
Comparing the 90th to 10th ED Occupancy PercentilesComparing the 75th to 25th ED Occupancy Percentiles
  1. PCP = primary care provider.

  2. *The risk ratios were adjusted for age, preferred language, insurance, presence of a PCP, triage level, ambulance arrival, and time of day.

TimelinessPain score indicated subgroup (1,229)Receipt of pain score in 1st hour0.96 (0.93–0.98)1.0 (1.0–1.0)
Any analgesia indicated subgroup (958)Receipt of any analgesia in 1st hour0.62 (0.39–0.95) 0.72 (0.54–0.97)
Opioid indicated subgroup (658)Receipt of opioid analgesia in 1st hour0.53 (0.29–0.87) 0.66 (0.45–0.91)
EffectivenessPain score indicated subgroup (1,229)Receipt of pain score0.97 (0.95–0.99)1.0 (1.0–1.0)
Any analgesia indicated subgroup (958)Receipt of any analgesia0.83 (0.68–0.98)1.0 (1.0–1.0)
Opioid indicated subgroup (658)Receipt of opioid analgesia0.83 (0.61–1.10)0.9 (0.76–1.05)

With regard to covariates, older age and evening arrival were associated with receipt and timely receipt of a pain score. Having no PCP, having a higher severity triage level, arriving by ambulance, and evening arrival were associated with receipt and timely receipt of any analgesic. Having a higher severity triage level and arriving by ambulance were also associated with receipt and timely receipt of opioid analgesia (Data Supplement S1).

In bivariate comparisons of the association between equity variables and timeliness and effectiveness measures, 16 of the 18 comparisons were nonsignificant (Table 4). In these unadjusted analyses, patients with a PCP were more likely to receive any analgesic and to receive any analgesic within the first hour. In adjusted models of these same 18 comparisons, we found no direct effect of the three equity variables on the timeliness and effectiveness measures (Data Supplement S2, available as supporting information in the online version of this paper). Adjusted models also show that none of the equity variables are moderators of the association between crowding and the timeliness and effectiveness measures (Data Supplement S2).

Table 4.    Bivariate Comparison of Proportion Receiving Timely and Effective Care, by Equity Variables: Language, Presence of a PCP, and Insurance
DimensionProcessPreferred Language EnglishHas PCPInsurance*
YesNop-valueYesNop-valueNonePublicPrivatep-value
  1. PCP = primary care provider.

  2. *These columns involve a three-way comparison between no, public, and private insurance. When analyzed as a dichotomous variable, either as uninsured (yes/no) or as public insurance (yes/no), all findings were also nonsignificant.

TimelinessReceipt of pain score in first hour0.950.960.550.960.950.440.980.950.950.51
Receipt of any analgesia in first hour0.130.110.200.180.130.020.100.130.130.84
Receipt of opioid analgesia in first hour0.150.110.200.180.130.170.130.130.160.63
EffectivenessReceipt of pain score0.970.990.160.980.970.380.990.970.980.59
Receipt of any analgesic0.320.310.720.400.300.020.250.300.350.16
Receipt of opioid analgesic0.330.280.250.360.340.340.290.290.380.09

Discussion

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. Limitations
  7. Conclusions
  8. References
  9. Supporting Information

We found that increased crowding levels were independently associated with decreased receipt and less timely delivery of analgesia for pediatric patients with pain related to an acute long-bone fracture in a large, academic ED. This confirms the findings of studies in adult ED populations that demonstrate an association between ED crowding and oligoanalgesia.8–12 We found that crowding can raise the risk of less timely care by as much as 47%, and the risk of less effective care by as much as 17%, based on the interdecile risk ratio. These findings were robust in multivariable analysis after adjusting for other known risk factors for variations in quality of analgesia.

We modeled nine quality measures across three IOM dimensions. Of these nine measures, only the timeliness and effectiveness measures showed an association with crowding, with all significant associations indicative of an inverse relationship. Of these, the measures that had the largest effect size were timeliness measures, with lesser effects in the effectiveness measures. In the prior five studies of crowding and analgesia in adult populations, three found an association with both timeliness and effectiveness measures,8,9,11 and two found association with timeliness but not with effectiveness measures.10,12

The finding that timeliness and effectiveness (receipt or nonreceipt) of pain score administration was least affected by crowding, and opioid timeliness most affected, makes sense given the stepwise nature of the processes. Pain scores are usually assigned prior to administration of analgesia, and nonopioid analgesia is sometimes attempted prior to opioids; the more steps in this sequence, the more opportunities for crowding to delay or deter the later processes in the sequence. Furthermore, pain scores and many nonopioid medications are administered by nurses only without attending involvement, whereas opioid administration requires both nurse and attending involvement. Both types of process (nurse-only and nurse-plus-attending) were affected by crowding, with a greater effect in the latter type, suggesting that both groups of providers are affected by crowding.

We also found that equity measures had no effect on oligoanalgesia across varying degrees of crowding in adjusted models. Our finding is consistent with some prior literature, in that ED studies of analgesia and race and ethnicity have varied,4,29–33 but differs from one study that found lower rates of analgesia among patients with Medicaid insurance.4

An important finding was the nonlinear relationship between crowding and analgesia delivery. The drop-off in quality between 75th and 90th percentiles of crowding in three of the six models shown in Figure 1 indicates a threshold effect. Queuing theory predicts that, as the crowding measures rise above a certain threshold, patient wait times should experience a steeper increase.34 The finding of a threshold effect was consistent across both receipt and nonreceipt (effectiveness measures), as well as in wait times (timeliness measures).

The rate of analgesia administration (31.6% for any analgesia, 32.2% for opioid analgesia) in our population was comparable to those noted in prior studies of ED analgesia for all children presenting with fractures,35,36 although lower than a study that focused on children with fractures in moderate to severe pain (73% for any analgesia, 54% for opioids).29 The difference from the latter study may result from inclusion of patients in less severe pain in the present study.

Limitations

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. Limitations
  7. Conclusions
  8. References
  9. Supporting Information

The setting of this study in a single, academic referral hospital ED affects the generalizability of findings. Similarly, our findings in a freestanding children’s hospital may not be generalizable to a pediatric ED in a general hospital or to an all-ages ED. The EMR’s active patient-tracking system may have resulted in discrepancies between the EMR event time and the actual event time. It is likely that this discrepancy is present more often when the ED is crowded, leading to overestimation of the association between crowding and delay of care.

Our criteria for defining relevant subgroups were based on pain scores assigned at triage. Because we neither assessed patient desire for analgesia nor reassessed pain after triage, we may have over- or underincluded patients in each subgroup.37 This may have altered the receipt/nonreceipt proportions; an increase in subgroup size would increase the proportion that did not receive a process, but should not have affected the proportion with delayed care or the association with crowding.

The absence of race/ethnicity data from our data set is also a limitation, given that some ED analgesia studies have shown an association.4,29–33 Race and ethnicity data are present in less than 40% of patients in the data set and, when present, are assigned to a hospital-defined category by admissions personnel, rather than self-assigned. We elected not to impute missing data based on non–self-designated race or ethnicity.38 Although preferred language is not intended as a proxy for race or ethnicity, in the county in which the study hospital is located, the racial distribution is 78% white, 3% black or African American, 3% Asian, and 35% Hispanic or Latino (of any race).39

Conclusions

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. Limitations
  7. Conclusions
  8. References
  9. Supporting Information

Our study of crowding and three dimensions of quality of ED care for children with fracture-related pain found that both timeliness and effectiveness of care were inversely associated with crowding and that equity of analgesia care was not associated with crowding. The inverse association is not always linear: for some measures, quality declines steeply above the 75th percentile of crowding. Ours is the first study to report the effect of ED crowding on oligoanalgesia in an ED population of children. Efforts to mitigate ED crowding may reduce oligoanalgesia in children with acute fractures. Areas for further study could include use of crowding measures as covariates in the assessment of analgesia-related process improvement interventions and improved research methods to better define the study populations for which analgesia is indicated.

Dr. Sills acknowledges the chart review contributions of Sara Deakyne, MPH, the article reviews performed by the Colorado Health Outcomes Program’s weekly Primary Care Research Conference, and the support of her mentors, John F. Steiner, MD, MPH, Michael G. Kahn, MD, PhD, and Diane Fairclough, DrPH.

References

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. Limitations
  7. Conclusions
  8. References
  9. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. Limitations
  7. Conclusions
  8. References
  9. Supporting Information

Data Supplement S1. Crowding-Quality Models.

Data Supplement S2. The Significance of Equity Variables—Language, Presence of a Primary Care Provider, and Insurance—With Regard to Their Effect, and Their Effect as Modifiers, of the Quality-Crowding Association.

Please note: Wiley Periodicals Inc. are not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.

FilenameFormatSizeDescription
ACEM_1136_sm_DataSupplement1.doc228KSupporting info item
ACEM_1136_sm_DataSupplement2.doc684KSupporting info item

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.