More than 1.7 million children with acute gastroenteritis present for emergency department (ED) care annually in the United States (1). Gastroenteritis treatment regimens have been outlined in guidelines endorsed by the American Academy of Pediatrics (2). A fundamental principle included in these guidelines is the administration of oral rehydration therapy (ORT) to the vast majority of children with gastroenteritis; however, surveys have shown that a gap exists between guidelines and practice (3).
Numerous justifications are provided to explain the over-reliance on intravenous rehydration. These include vomiting—physicians are more likely to use intravenous therapy when vomiting is the major symptom (4); familiarity—unfamiliarity with ORT techniques and a comfort with intravenous rehydration; beliefs—that intravenous rehydration is faster and decreases the length of hospital stay (5,6); expectations—of the family and by the referring physician (5,7); culture—health care providers are trained to not miss a significant illness which leads to overtesting and treating (8); and technology—the use of a more modern approach is favored by many caregivers (9). Although intravenous rehydration can reverse the presence of dehydration at the time of ED presentation, there is limited evidence describing the relation between intravenous rehydration and outcomes following ED discharge; however, because 75% of unscheduled pediatric ED revisits are because symptoms persist or get worse (10), a better understanding of this relation is warranted.
Given that intravenous rehydration does not reduce symptoms following discharge (11), we hypothesized that unscheduled revisits are not reduced by the administration of intravenous rehydration. Our primary objective was to determine whether intravenous fluid administration to children treated in an ED and discharged is independently associated with a reduction in unscheduled ED revisits within the subsequent 7 days.
METHODS
Study Design and Setting
This retrospective observational cohort study is a secondary analysis of a previously described dataset (12). Eligible participants presented to The Hospital for Sick Children, a tertiary care hospital in Toronto, Canada. The ED is staffed 24 hours/day by attending physicians, and approximately 60,000 patients are treated annually. The institutional protocol specifies the initial administration (5 mL) of an electrolyte maintenance solution (ie, Pedialyte, Abbott Laboratories, Abbott Park, IL) every 2 minutes, via a syringe, to all children (age-appropriate alternatives may be used—bottle, sippy cup, cup/straw) with gastroenteritis and mild-to-moderate dehydration before the administration of intravenous rehydration. For those requiring intravenous rehydration, it is performed using a 0.9% normal saline solution with dextrose and potassium added based on the clinical indications. Approval to conduct this study was provided by the hospital's research ethics board.
Objectives
Our primary objective was to determine whether intravenous fluid administration to children with gastroenteritis treated and discharged from an ED is independently associated with a reduction in unscheduled ED revisits within the subsequent 7 days. Secondary objectives were to identify clinical features associated with intravenous rehydration and to evaluate the relations among age, triage assessment, intravenous fluid administration, and revisits.
Time Period, Patients, and Data Sources
This report includes data from a consecutive series of children younger than 18 years who presented between July 1, 2003 and June 30, 2008. Potentially eligible children were identified by searching the ED database for relevant International Classification of Diseases-9-Clinical Modification (2003–2007) and 10 (2007–2008) codes (eAppendix 1, available online only at https://links.lww.com/MPG/A240).
The following eligibility criteria were used: acute gastroenteritis defined by diarrhea or vomiting for <14 days (13), discharged from the ED, and absence of an alternative diagnosis at the index visit (eg, appendicitis, urinary tract infection, inflammatory bowel disease, central nervous system disorder; eAppendix 2, available online only at https://links.lww.com/MPG/A240) (12). All visits associated with a visit by the same child during the preceding 7-day period had the earlier visit coded as the index visit.
Data were abstracted from our electronic patient chart system by 3 data abstractors who were blinded to the study's hypothesis. A standardized instrument was used to record demographics, history and physical examination findings, medication and intravenous fluid administration, disposition, and revisit data. Precise operational definitions were used along with uniform procedures for handling missing, conflicting, or ambiguous data. Weekly meetings occurred during the data abstraction phase to resolve disputes and review coding. Ten percent of randomly selected charts were independently abstracted by 1 of the investigators (M.R.) to evaluate interobserver agreement.
Definitions
Medical history was considered as a 3-level categorical variable: none, mild systemic disease, and severe systemic disease (ie, would likely affect treatments administered). Pediatric Canadian Triage Acuity Scale (PedCTAS) is a 5-level triage scale, which the triage nurse uses to assign a severity score (14). This tool is used in all Canadian pediatric EDs and in other EDs around the world (15,16).
Temperatures were adjusted for location of measurement (17) and fever was defined as an adjusted temperature >38.0 (17). General appearance was classified using a previously described classification method as “well” (“well appearing,” “no apparent distress,” “alert,” “normal mental status,” “interactive,” “smiling”) or unwell (“sick,” “toxic,” “shocky,” “decreased mental status,” “lethargic,” “unresponsive,” “irritable,” “fussy,” “inconsolable,” “not looking well,” “poor or decreased perfusion,” “decreased pulses”) (18).
Analysis
Twenty percent of potentially eligible visits were randomly identified for chart review by S.F. using the RANDOM and SORT functions in Microsoft Excel (version 2007, Microsoft, Redmond, WA). Because it was anticipated that at most 20 variables would be considered in the model, to avoid an overfit model, we aimed to identify a minimum of 10 to 20 events per covariate (19). Thus, we determined that a minimum of 200 outcomes were required to prevent overfitting. Based on knowledge of our revisit rate (12), the data abstraction plan was anticipated to identify a sufficient number of outcomes to conduct the analysis.
Means with standard deviations are provided for continuous variables; frequency counts and percentages are given for discrete variables. Continuous variables were compared using two sample t tests and analysis of variance as appropriate. The χ2 test or Fisher exact test was used to assess association between discrete variables as appropriate. Percentages reported represent the valid percentage for each item (ie, disregarding missing responses). Statistical tests were 2 sided and evaluated at the 5% level of significance. Statistical analysis was conducted with the use of SPSS (Windows version 16.0, SPSS Inc, Chicago, IL) and SAS software (version 9.2, SAS Institute, Cary, NC).
Initial bivariate analyses were conducted based on the outcome (ie, dependent variable) of ED revisits (Table 1). The relations between independent variables and the exposure to intravenous rehydration were explored using bivariate analyses. To further understand the relations among intravenous fluid administration, PedCTAS, and age, the outcome was defined in 3 ways in stratified analyses (ie, any ED revisits, revisits with intravenous rehydration, and revisits with hospitalization). For these analyses, PedCTAS scores were stratified as life-threatening/emergent (scores = 1–2), urgent (score = 3), and semi/nonurgent (score = 4–5) (20). Additionally, because of sample size limitations, statistical calculations involving revisits with intravenous rehydration, and hospitalization were underpowered and must be considered exploratory.
;)
TABLE 1: Baseline demographic characteristics based on unscheduled ED revisit status
To examine the relation between the independent variables and unscheduled ED revisits, candidate features were selected on the basis of previous research and theoretical grounds. Missing data for the candidate covariates were <5% in all of the cases and were not imputed. The demographic and clinical characteristics in Table 1 were selected for inclusion in the regression model if they were a priori identified as clinically important, were associated with the outcome (P < 0.20), and were uncorrelated with each other (ie, Pearson correlation coefficient <0.50). When collinearity was identified, the variable with the greatest clinical relevance was retained. All of the variables selected were entered into a multiple logistic regression model and then removed individually depending on the regression estimate, its significance, and whether removal altered other variables in the models. In the final model, all of the variables were significant at the 5% level, were clinically significant, or were statistically important with its removal, causing other variables to lose significance. The final model was compared with a constant-only model using the Hosmer-Lemeshow goodness of fit test. Because disease severity, as reflected by PedCTAS, was found to be correlated with the administration of intravenous fluids (P = 0.28), a stratified regression analysis was performed.
The stability of the selected risk factors was assessed in a bootstrap analysis—a resampling technique by which a specified number of population samples are drawn from the data and the process of model building is performed a large number of times (21). The number of times each variable was selected at the 5% level gives information about the stability of each variable in the model. We performed 1000 iterations using both backward and stepwise selection procedures, with sample sizes equal to our complete dataset (ie, 2874), to test the stability of the independent variables. Predictors that remained significant in >50% of the iterations were retained (21,22). Lastly, we assessed interobserver agreement for 6 outcome variables (intravenous fluid administration, ondansetron administration, hospitalization, revisit within 7 days, revisit requiring intravenous rehydration, and revisit requiring hospitalization) with the κ statistic.
RESULTS
Characteristics of Study Subjects
There were 22,125 potentially eligible visits identified; 4064 charts of potentially eligible children were reviewed representing 4425 patient visits and 3346 visits were included in our final cohort (Fig. 1). Participants ranged in age from 2 days to 17.9 years (27, 75 IQR: 1.0, 4.1 years). Stool testing for etiologic organisms included the use of electron microscopy (n = 123), culture (n = 124), and ova/parasite (n = 68). The most commonly identified organisms were viruses (n = 50; rotavirus = 23, adenovirus = 14, other = 13) followed by bacteria (n = 26; Salmonella species = 11, Yersinia enterocolitica = 5; other = 10), and parasites (n = 2).
FIGURE 1: Summary of eligible patients and charts selected for review. Patients with repeat visits within 1 week had the initial visit included as the index visit with the subsequent visit identified as a revisit.
Main Results
Overall, 16% (526/3346) of eligible children revisited our institution, with 83% (434/526) of these visits occurring within 3 days. Children with revisits had a greater frequency of vomiting (P < 0.001) and higher heart rate (P < 0.001) and respiratory rate (P < 0.001) at the index visit (Table 1). Children who received intravenous rehydration at the index visit were more likely to have an unscheduled ED revisit within 1 week (P < 0.001).
A total of 497 children were administered intravenous rehydration (15%; Table 2). Children administered intravenous rehydration were older (P < 0.001), more likely to have a significant medical history (P < 0.001), vomited more frequently (P < 0.001), were more likely to be described as appearing unwell (P < 0.001), and to be hospitalized during the week following ED discharge (26/497 [5%] vs 45/2849 (2%); P < 0.001). When stratified by age and intravenous rehydration status (Table 3), ED revisits were most significantly increased among the youngest children (ie, younger than 5 years old).
TABLE 2: Baseline demographic characteristics based on intravenous fluid administration
TABLE 3: Outcome data stratified by fluid administration status at the index visit
Among the 2874 children with complete data, 450 had an unscheduled revisit and 2424 did not. All variables included in the final, primary-outcome, regression model—intravenous fluids administered (odds ratio [OR] 1.76; 95% confidence interval [CI] 1.36–2.26), number of vomiting episodes in preceding 24 hours (OR 1.16; 95% CI 1.04–1.28/5-episode increase), number of diarrhea stools in preceding 24 hours (OR 1.19; 95% CI 1.03–1.38/5-episode increase), days of diarrhea (OR 0.92; 95% CI 0.88–0.97), and age (OR 0.94; 95% CI 0.91–0.98)—made statistically significant contributions to the model's predictive abilities. This model was statistically significant when compared with a constant-only model (Hosmer-Lemeshow χ2 = 23.2; P = 0.003).
Stratified Analysis and Internal Validation of the Final Model Using Bootstrap Techniques
When stratified by PedCTAS score, all outcomes (ie, 1 week revisit, intravenous rehydration and admission) were more common among children administered intravenous rehydration at the index visit, with the exception being among children with PedCTAS scores of 1 to 2; (Fig. 2 A–D). In the stratified regression models, intravenous rehydration was found to be associated with revisits in the PedCTAS level 3 (urgent), and 4 to 5 (semiurgent/nonurgent) categories with odds ratios of 1.70 (95% CI 1.25–2.33) and 1.97 (95% CI 1.17–3.32), respectively. Among the most unwell children (PedCTAS level 1–2), intravenous rehydration was not associated with revisits (OR 0.98; 95% CI 0.39–2.50). A multiple regression model, which forced the inclusion of PedCTAS, similarly found that intravenous rehydration was associated with unscheduled ED revisits (OR 1.67; 95% CI 1.29–2.17).
FIGURE 2: Revisit, revisit requiring intravenous fluids, and revisit requiring hospitalization. The figures describe the overall (A) proportions of these outcomes based on treatment provided at the index visit. B, C, and D stratify these outcomes based on PedCTAS scores at the index visit (1 and 2—B; 3—C; 4 and 5—D). All P values represent χ 2 test comparing IV rehydration versus oral rehydration. A, PedCTAS Scores = 1–5. B, PedCTAS scores = 1–2 (n = 154). C, PedCTAS score = 3 (n = 1794). D, PedCTAS Score = 4–5 (n = 1398). PedCTAS = Pediatric Canadian Triage Acuity Scale; IV = intravenous.
The backward and stepwise regression models identified the following variables a minimum of 50% of the time—intravenous fluids administered (950/1000), age (922/1000), number of vomiting episodes in preceding 24 hours (787/1000), days of diarrhea (739/1000), and number of diarrheal stools in preceding 24 hours (547/1000) (Figure 3).
FIGURE 3: Result of bootstrap analysis. Backward variable selection process conducted 1000 times using association with the dependent variable (ED revisit with 1 week) at a P value of <0.05 as the unit of outcome. # = number; PedCTAS = Pediatric Canadian Triage Acuity Scale; ED = emergency department.
Interobserver Agreement
Interobserver agreement was assessed for 362 patients. Agreement was extremely good, with κ values ranging from 0.82 (95% CI 0.78–0.86) for revisits requiring intravenous rehydration to 0.97 for ondansetron administration (95% CI 0.96–0.97).
DISCUSSION
In the United States in 2002, ED return visits accounted for nearly 3.5 million patient encounters, with children having the highest rate at 138 of every 1000 people (23). Revisits in children with gastroenteritis are frequent—16% of children discharged from our ED represented within 7 days. We consistently identified several risk factors using several different analytic techniques. These included intravenous fluid administration, frequency of vomiting, days and frequency of diarrhea, and age.
Our findings build upon previous research evaluating the relation between intravenous rehydration and ED revisits in children with gastroenteritis. A prospective cohort of 647 children with gastroenteritis in 11 Canadian EDs reported that unscheduled revisit rates vary significantly between institutions (3) and children administered intravenous rehydration are more likely to have unscheduled health care (29% vs 19%) and ED revisits (20% vs. 9%). Another study detected a small reduction in unscheduled ED revisits among children administered ORT (17%), compared with intravenous fluids (24%) (24).
Although clinical outcomes are similar between children rehydrated orally and intravenously (11), the latter is associated with undesirable consequences including phlebitis, longer hospitalization, and increased major adverse events (11). Nonetheless, intravenous rehydration continues to be used in the treatment of children with gastroenteritis (8) likely reflecting the many clinical and social factors that enter into the decision regarding the choice of rehydration route. This decision is critical because it influences the family's perception of illness (eg, “he/she was so sick they gave him/her intravenous fluids”), which may be poorly influenced by objective measures of hydration (eg, urine output, heart rate). It is this lasting impression that we hypothesize is responsible for the association between intravenous fluid administration and repeat visits independent of objective measures of dehydration severity.
Our findings are not unique—data from Europe and Australia have reported that >80% of hospitalized children do not have evidence of dehydration and yet they still receive intravenous therapy (25,26). Among 7391 children admitted with gastroenteritis in Hong Kong, <8% had signs of dehydration, yet 48% received intravenous therapy (27). Thus, ineffective knowledge translation related to the use of ORT and intravenous rehydration appears to be widespread.
Because intravenous rehydration is a proxy measure of disease severity, we employed analytic techniques to adjust for confounding and conducted a bootstrap analysis to assess the stability of our findings. We stratified our analysis by PedCTAS score and included surrogate clinical markers of disease severity (28). A combination of variables, including demographic (age), intervention (intravenous fluid administration), and symptom severity (numbers and duration of diarrheal stools, vomiting episodes), ultimately was the model with the best predictive ability. Our bootstrap model confirmed the stability of our findings and identified intravenous fluid administration as a predictor >90% of the time. Thus, although severity is a risk factor for unscheduled ED revisits (29), that risk is increased by intravenous fluid administration.
Although at the study institution (located in Toronto, Canada), ED physicians are salaried and there is no link between reimbursement and medical complexity, this may be a concern in other locations (30); however, our population is clinically similar to that of other high-income nations (31), making our findings generalizable to such countries. The challenge in such situations is how to translate clinical trial evidence, which has proven that ORT is as effective as intravenous rehydration in children with moderate dehydration (32), into practice to reduce intravenous rehydration use. This is a particularly challenging question because 45% of children deemed to have life-threatening/emergent conditions (PedCTAS scores 1–2) did not receive intravenous rehydration, whereas many children with presumably less severe dehydration were rehydrated intravenously. Although children in the latter group may have been more dehydrated/unwell than those not administered intravenous fluids, this should have been reflected in the PedCTAS triage scores, which include hydration status as a first-order modifier (14) and correlate with dehydration severity (33).
We cannot ignore our finding that PedCTAS, a marker of illness severity, was present in the bootstrap analysis in <50% of the iterations. In previous work, such measures have predicted revisits (34); however, our stratified regression analysis found that within levels of disease severity, intravenous rehydration was consistently associated with revisits in all but the most unwell children. Although unmeasured confounders may in part explain the association between unscheduled revisits and intravenous rehydration, it is hypothesized that psychological (eg parental anxiety), social, and health care infrastructure issues (eg, physician practice) are additional key factors underlying revisits (35). Caregivers of children administered intravenous fluids may have received less thorough ORT teaching and they may be less confident managing them at home after they have received intravenous rehydration.
Given the retrospective nature of our study, no data were available on home ORT, nutritional status, adverse events attributable to the use of intravenous fluid administration, or clinical dehydration scale scores. Clustering by treating physician could also not be adjusted for because there were too many responsible physicians relative to the number of outcomes for it to be included as a variable in our models. We considered using propensity score matching; however, it is recommended that traditional multivariate regression adjustment be used when the sample size is sufficiently large and the outcome of interest is not rare, as was our case (36). In addition, we only tracked revisits to our ED; caregivers, particularly those of children not administered intravenous rehydration, may have sought care elsewhere for ongoing symptoms. On the contrary, in a study at our institution of children transferred from the ED to regional hospitals, we found that 80% of caregivers would come back to our ED should such care be sought again (37).
In our study population, after adjustment for clinical findings, our data suggest that intravenous rehydration is associated with ED revisits among children with lower triage scores (ie, less unwell). Although children administered intravenous rehydration were more unwell, this was adjusted for in our model and several clinical features predicted revisits. Our data do not support the provision of intravenous fluids to prevent unscheduled emergency department revisits in children with mild-to-moderate dehydration. Future multicenter prospective studies focusing on revisits in this population are required.
REFERENCES
1. Freedman SB, Steiner MJ, Chan KJ. Oral ondansetron administration in emergency departments to children with gastroenteritis: an economic analysis.
PLoS Med 2010; 7:e1000350.
2. King CK, Glass R, Bresee JS, Duggan C. Managing acute gastroenteritis among children: oral rehydration, maintenance, and nutritional therapy.
MMWR Recomm Rep 2003; 52 (RR-16):1–16.
3. Freedman SB, Gouin S, Bhatt M, et al. Prospective assessment of practice pattern variations in the treatment of
pediatric gastroenteritis.
Pediatrics 2011; 127:e287–e295.
4. Ozuah PO, Avner JR, Stein RE. Oral rehydration, emergency physicians, and practice parameters: a national survey.
Pediatrics 2002; 109:259–261.
5. Conners GP, Barker WH, Mushlin AI, et al. Oral versus intravenous: rehydration preferences of
pediatric emergency medicine fellowship directors.
Pediatr Emerg Care 2000; 16:335–338.
6. Banks JB, Meadows S. Intravenous fluids for children with gastroenteritis.
Am Fam Physician 2005; 71:121–122.
7. Goepp JG, Katz SA. Oral rehydration therapy.
Am Fam Physician 1993; 47:843–851.
8. Leape LL. Error in medicine.
JAMA 1994; 272:1851–1857.
9. Karpas A, Finkelstein M, Reid S. Parental preference for rehydration method for children in the emergency department.
Pediatr Emerg Care 2009; 25:301–306.
10. Ali AB, Place R, Howell J, et al. Early
pediatric emergency department return visits: a prospective patient-centric assessment.
Clin Pediatr (Phila) 2012; 51:651–658.
11. Hartling L, Bellemare S, Wiebe N, et al. Oral versus intravenous rehydration for treating dehydration due to gastroenteritis in children.
Cochrane Database Syst Rev 2006; 3:CD004390.
12. Freedman SB, Tung C, Cho D, Rumantir M, Chan KJ. Time-series analysis of ondansetron use in
pediatric gastroenteritis.
J Pediatr Gastroenterol Nutr 2012; 54:381–386.
13. Guarino A, Albano F, Ashkenazi S, et al. European Society for Paediatric Gastroenterology, Hepatology, and Nutrition/European Society for Paediatric Infectious Diseases evidence-based guidelines for the management of acute gastroenteritis in children in Europe.
J Pediatr Gastroenterol Nutr 2008; 46 (suppl 2):S81–S122.
14. Warren DW, Jarvis A, LeBlanc L, Gravel J, et al. Revisions to the Canadian Triage and Acuity Scale paediatric guidelines (PaedCTAS).
CJEM 2008; 10:224–243.
15. Goransson K, Ehrenberg A, Marklund B, et al. Accuracy and concordance of nurses in emergency department triage.
Scand J Caring Sci 2005; 19:432–438.
16. Rehmani R, Norain A. Trends in emergency department utilization in a hospital in the Eastern region of Saudi Arabia.
Saudi Med J 2007; 28:236–240.
17. Baltimore: Lippincott Williams & Wilkins, Alpern ER, Henretig FM. Fleisher GR, Ludwg S, Henretig FM. Fever.
Textbook of Pediatric Emergency Medicine 5th ed.2006.
18. Schnadower D, Kuppermann N, Macias CG, et al. Febrile infants with urinary tract infections at very low risk for adverse events and bacteremia.
Pediatrics 2010; 126:1074–1083.
19. Stoltzfus JC. Logistic regression: a brief primer.
Acad Emerg Med 2011; 18:1099–1104.
20. Gravel J, Gouin S, Manzano S, et al. Interrater agreement between nurses for the
Pediatric Canadian Triage and Acuity Scale in a tertiary care center.
Acad Emerg Med 2008; 15:1262–1267.
21. Beyene J, Atenafu EG, Hamid JS, et al. Determining relative importance of variables in developing and validating predictive models.
BMC Med Res Methodol 2009; 9:64.
22. Schumacher M, Hollander N, Sauerbrei W. Resampling and cross-validation techniques: a tool to reduce bias caused by model building?
Stat Med 1997; 16:2813–2827.
23. Adekoya N. Patients seen in emergency departments who had a prior visit within the previous 72 h-National Hospital Ambulatory Medical Care Survey, 2002.
Public Health 2005; 119:914–918.
24. Freedman SB, Powell E, Seshadri R. Predictors of outcomes in
pediatric enteritis: a prospective cohort study.
Pediatrics 2009; 123:e9–e16.
25. Sandhu BK, Isolauri E, Walker-Smith JA, et al. A multicentre study on behalf of the European Society of Paediatric Gastroenterology and Nutrition Working Group on Acute Diarrhoea. Early feeding in childhood gastroenteritis.
J Pediatr Gastroenterol Nutr 1997; 24:522–527.
26. Elliott EJ, Backhouse JA, Leach JW. Pre-admission management of acute gastroenteritis.
J Paediatr Child Health 1996; 32:18–21.
27. Chow CM, Choi K, Nelson EA, et al. Use of intravenous fluids in Hong Kong children hospitalised for diarrhoea and relationship to severity and aetiology.
Vaccine 2009; 27 (suppl 5):F55–F60.
28. Steiner MJ, DeWalt DA, Byerley JS. Is this child dehydrated?
JAMA 2004; 291:2746–2754.
29. Jacobstein CR, Alessandrini EA, Lavelle JM, et al. Unscheduled revisits to a
pediatric emergency department: risk factors for children with fever or infection-related complaints.
Pediatr Emerg Care 2005; 21:816–821.
30. Pitts SR. Higher-complexity ED billing codes—sicker patients, more intensive practice, or improper payments?
N Engl J Med 2012; 367:2465–2467.
31. Levy JA, Bachur RG. Intravenous dextrose during outpatient rehydration in
pediatric gastroenteritis.
Acad Emerg Med 2007; 14:324–330.
32. Spandorfer PR, Alessandrini EA, Joffe MD, et al. Oral versus intravenous rehydration of moderately dehydrated children: a randomized, controlled trial.
Pediatrics 2005; 115:295–301.
33. Bailey B, Gravel J, Goldman RD, et al. External validation of the clinical dehydration scale for children with acute gastroenteritis.
Acad Emerg Med 2010; 17:583–588.
34. Alessandrini EA, Lavelle JM, Grenfell SM, et al. Return visits to a
pediatric emergency department.
Pediatr Emerg Care 2004; 20:166–171.
35. Pham JC, Kirsch TD, Hill PM, et al. Seventy-two-hour returns may not be a good indicator of safety in the emergency department: a national study.
Acad Emerg Med 2011; 18:390–397.
36. Glynn RJ, Schneeweiss S, Sturmer T. Indications for propensity scores and review of their use in pharmacoepidemiology.
Basic Clin Pharmacol Toxicol 2006; 98:253–259.
37. Freedman SB, Thakkar VA. Easing the strain on a
pediatric tertiary care center: use of a redistribution system.
Arch Pediatr Adolesc Med 2007; 161:870–876.
