Bowman SM, Zimmerman FJ, Christakis DA, Sharar SR, Martin DP. Hospital Characteristics Associated With the Management of Pediatric Splenic Injuries. JAMA. 2005;294(20):2611-2617. doi:10.1001/jama.294.20.2611
Author Affiliations: Department of Health Services (Mr Bowman and Drs Martin and Zimmerman), Child Health Institute (Drs Zimmerman and Christakis), Department of Anesthesiology (Dr Sharar), Department of Pediatrics (Dr Christakis), University of Washington, Seattle; and Office of EMS and Trauma, Washington State Department of Health, Olympia (Mr Bowman).
Context Despite evidence that more than 90% of children with traumatic injuries to the spleen can be successfully managed nonoperatively, there is significant variation in the use of splenectomy. As asplenic children are at increased risk of overwhelming postsplenectomy infection, nonoperative management may be considered a quality of care indicator.
Objective To test the hypothesis that children are more likely to undergo splenectomy in general hospitals than in children’s hospitals.
Design Retrospective cohort study using data from the Kid’s Inpatient Database (KID) for the year 2000. Multivariable regression was used to control for patient and hospital characteristics.
Setting and Participants All children aged 0 to 16 years who were hospitalized with a traumatic (noniatrogenic) spleen injury in nonfederal short-stay hospitals in any of the 27 states participating in KID (N = 2851).
Main Outcome Measure Splenectomy performed within 1 day of arrival.
Results A total of 11 children (3%) with splenic injuries receiving care at children’s hospitals underwent splenectomy compared with 383 children (15.4%) cared for at general hospitals (P <.001). After adjusting for patient characteristics, injury severity, and hospital characteristics, splenectomy was more likely among children treated at general hospitals (odds ratio, 5.01; 95% confidence interval, 2.21-11.36) than among children treated at children’s hospitals.
Conclusions There is considerable variation in the management of pediatric splenic injuries, with significantly lower rates of splenectomy at designated children’s hospitals. Quality improvement interventions, including increased education and training for physicians in general hospitals, may be needed to increase the use of spleen-conserving management practices.
Evidence-based practice guidelines for the management of children with traumatic injury to the spleen recommend nonoperative management for hemodynamically stable patients,1 an approach that is successful in more than 90% of cases.2- 5 However, in many centers, the use of splenectomy is much more common, with one recent study reporting as few as 39% of children managed without surgery.6 Nationally, an estimated 1700 splenectomies are performed each year in the 7000 children with spleen injuries.7 Based on these estimates, about 76% of pediatric spleen injuries are managed nonoperatively, and increasing nonoperative management to 90% could mean an annual reduction of up to 1000 splenectomies in children.
The short-term benefits of nonoperative management of injured spleens may include the avoidance of surgical costs, fewer blood transfusions, and shorter hospital lengths of stay.1 Long-term benefits of spleen conservation include the avoidance of severe infectious complications.8 The estimated lifetime risk for developing overwhelming postsplenectomy infection (OPSI) ranges from 1% to 11%, and OPSI is associated with a mortality of up to 50%.9- 13 Based on these estimates, an estimated 10 to 100 cases of OPSI, and up to 5 to 50 associated deaths, could potentially be prevented through the greater use of nonoperative management.
Long-term cost savings of nonoperative management may include decreased prescription drug costs associated with the prophylactic administration of antibiotics to reduce the likelihood of OPSI and a decrease in hospitalizations associated with OPSI. Long-term cost savings of nonoperative management may include decreased prescription drug costs associated with the prophylactic administration of antibiotics to reduce the likelihood of OPSI and a decrease in hospitalizations associated with OPSI.
The identification of factors associated with variability in splenic injury management may be useful in efforts to design and implement targeted quality improvement interventions. Both surgeon6 and hospital characteristics may contribute to variability in the rate of splenectomy. Hospital characteristics may include volume of spleen injuries, location, ownership/control, teaching status, hospital type (general vs pediatric), and level of trauma designation.14- 17
This study was designed to investigate hospital factors that relate to pediatric spleen conservation using a large national database of child hospitalizations and procedures. We tested the hypothesis that children were more likely to undergo splenectomy in general hospitals than in children’s hospitals.
Discharge administrative data from the most recent release (2000) of the Kids’ Inpatient Database (KID) of the Healthcare Cost and Utilization Project, sponsored by the Agency for Healthcare Research and Quality, Rockville, Md, were used for this study. The KID is provided by public and private statewide data organizations as part of a federal-state-private collaboration.18 KID data are collected and compiled in a uniform format from the 27 participating states (Arizona, California, Colorado, Connecticut, Florida, Georgia, Hawaii, Iowa, Kansas, Kentucky, Maine, Maryland, Massachusetts, Missouri, New Jersey, New York, North Carolina, Oregon, Pennsylvania, South Carolina, Tennessee, Texas, Utah, Virginia, Washington, Wisconsin, and West Virginia). The KID contains data from 2784 hospitals and includes more than 2.5 million pediatric discharge records. The sampling frame includes pediatric discharges from nonfederal, short-term hospitals (excluding rehabilitation hospitals and hospital units of institutions, such as prisons, mental institutions, and developmental centers) in the United States. Federal hospitals (Veterans Administration, Department of Defense, and Indian Health Service hospitals) are not included in KID. KID data comprise a 10% sample of uncomplicated births and an 80% sample of other hospitalizations for children aged 20 years and younger from all hospitals in the sampling frame. Data were obtained from the Healthcare Cost and Utilization Project Central Distributor, with approval from the institutional review board at the University of Washington.
The KID includes demographic information (eg, age, sex), admission type and source, diagnostic codes, procedure codes, payer, data and total charges, length of stay, and discharge disposition. A KID hospital data file with facility characteristics (eg, children’s hospital, ownership, number of beds, urban/rural, geographic region, and teaching status) can be linked to the patient core data. The KID does not contain physiological or laboratory patient data.
For this study, eligibility was limited to children 0 to 16 years of age at admission and who were hospitalized with an International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) code for spleen injury (865) in any of the up to 15 diagnoses on the discharge record. Patients who died while in-hospital with pediatric spleen injuries were included.
A patient was defined as undergoing operative management if an ICD-9-CM procedure code of “41.5 splenectomy” was identified in any of the procedures listed in the discharge record and with a time to procedure of less than or equal to 1 day from arrival. Time to procedure was missing in 45 cases, and in these cases, time was imputed using multivariable regression with available patient and injury variables. Hospital type was not included in the imputation. All of the imputed, missing times were estimated to have occurred within 1 day of arrival. We also performed 2 sensitivity analyses: (1) assigning all cases with missing time to the nonoperative management group, and (2) assigning cases with missing time to the splenectomy group for children’s hospitals and to the nonoperative management group for general hospitals. Splenectomies performed at day 2 or later were considered as attempted nonoperative management. Splenorrhaphy, partial splenectomies, and other spleen-conserving procedures were excluded from the splenectomy group and included as nonoperative management. We created a binary outcome variable to indicate whether a splenectomy was performed. For the Cox survival analysis, we defined operative management using the presence of an ICD-9-CM procedure code of 41.5 independent of when it was performed.
Children’s hospitals were defined using the classification of the National Association of Children’s Hospitals and Related Institutions. We imputed missing values for the 2.7% of hospitals with missing hospital type and also conducted sensitivity analyses to model the effect of the missing data.
We selected control and confounding variables for multivariate analytic models based on the presence of significant bivariate associations found when comparing variables by hospital type (children’s vs general) or splenic management (operative vs nonoperative management). We also included additional variables based on an a priori assessment to add precision and for clinical relevance. Indicator variables were created for Medicaid as payer, rural hospital, teaching hospital, and received-in-transfer patient. We defined a control variable for massive spleen disruption (ICD-9-CM diagnostic codes 865.04 or 865.14), blood transfusion (ICD-9-CM procedure code 99.0), volume of pediatric spleen cases, and penetrating injury (E-code for firearms, knives, cuts). Because this administrative database does not include vital signs such as systolic blood pressure, we included receiving a blood transfusion as a surrogate for hypotension. An injury severity score (ISS) was calculated for each record using the ICDMAP-90 software.19 The ISS is widely accepted as a valid measure of injury severity and is used as a control variable in this analysis.20,21 Other control variables included patient-specific variables such as age and sex.
Trauma designation level is not identified in the KID data set but is of interest as a potential confounder. To identify hospital trauma designation level, we used data from the American Trauma Society’s Trauma Information Exchange Program (funded by the Centers for Disease Control and Prevention/National Center for Injury Prevention and Control). The American Trauma Society surveyed all hospitals nationwide to ascertain trauma designation levels. These data include the American Hospital Association’s hospital identification number for each facility that allowed linking to the KID hospital file. In the KID data, 8 states do not provide hospital identifiers. As such, trauma designation is unknown for about 25% of records. We created indicator variables for missing trauma designation, for higher levels of trauma designation (level I and II), and for a combined group of all lower levels of designation (eg, levels III, IV, V).
Univariate and bivariate analyses were completed using t tests for continuous variables and χ2 testing for categorical variables. To identify significant associations between patient or hospital factors and performance of splenectomy, we used logistic regression with splenectomy as the dependent variable and the independent and control variables described above. We adjusted for clustering on hospital identifier to account for correlation of individual outcomes within hospitals using the cluster option in Stata 8.0 (StataCorp LP, College Station, Tex). Explanatory variables are presented as odds ratios (ORs), with P values and 95% confidence intervals (CIs). A P value of <.05 was considered statistically significant.
To assess the potential impact of the state in which the hospital is located, we ran a conditional fixed-effects logistic regression with the splenectomy as the dependent variable and included the independent and control variables from the original model. We used the state in which the hospital was located as the independent unit for this fixed-effects model. To assess hospital effects, we performed another conditional fixed-effects logistic regression with hospital identifier as the independent unit.
We also assessed the relationship of splenectomy and hospital type using multivariable Cox proportional hazard regression analysis. The relative hazard ratio (HR) of splenectomy after splenic injury was compared between general and children’s hospitals. Survival was measured as the time from arrival to splenectomy, with censoring at hospital discharge. The proportional hazards assumption was confirmed by inspection of the Schoenfeld residuals and the log-rank test.
We identified a total of 2851 hospitalizations with spleen injuries among children 0 to 16 years of age. Almost all (333) of the 394 splenectomies performed during these hospitalizations occurred on day 0 or 1, and 16 were performed at day 2 or later.
Table 1 presents the relative frequencies of patient and hospital characteristics and their relationships with splenectomy vs nonoperative management. Children undergoing splenectomy were older than those managed nonoperatively (13.2 vs 11.2 years; P<.001). There was no significant difference in management based on sex, and patients were equally likely to be Medicaid beneficiaries. The mean ISS was significantly higher in children undergoing splenectomies compared with those in the nonoperative management group (25.4 vs 13.1; P<.001). Children undergoing splenectomies were also significantly more likely to have a massive disruption of the spleen and to have received blood products at any point during their hospitalization. We did not observe a difference in the proportion of cases coded as emergency admissions between the 2 management groups. Thirty-seven (1.3%) of the hospitalizations for splenic injury were due to penetrating injuries, and these children were more likely to undergo a splenectomy.
Table 2 presents the relative frequencies of patient and hospital characteristics and their bivariate relationships with general vs children’s hospitals. Most notably, children’s hospitals represented 12.7% of pediatric spleen hospitalizations, but only 2.4% of splenectomies. Children with splenic injuries receiving care at general hospitals underwent splenectomies in 15.4% of cases compared with 3% of cases at children’s hospitals (P<.001). Teaching hospitals cared for 63.6% of the spleen-injured children, but accounted for 47.9% of the splenectomies. Rural hospitals represented 10.1% of spleen hospitalizations and 14.3% of all splenectomies. In the bivariate analysis, a level I or II trauma designation was associated with more nonoperative management, while other levels of designation were associated with more splenectomies. No significant differences in management were observed in the hospitals with missing trauma designation. Children’s hospitals were more likely to be teaching hospitals, to have a level I or II trauma designation, and to have a higher volume of pediatric spleen injury cases.
Table 3 presents the unadjusted and adjusted results of a logistic regression model evaluating the odds of splenectomy based on patient and hospital risk factors identified in Tables 1 and 2. While initially included as a potential confounder, Medicaid payer status was removed from the final model because its effect on the model was insignificant (OR, 1.02; P = .93) and the effect size of the variable of interest (hospital type) was unchanged. The fit of the logistic regression model was good, with an area under the receiver operating characteristic curve of 0.85 and a P value for the Hosmer-Lemeshow test of .72. Table 3 also reports the results of the fixed-effects model conditioning on state.
Children with spleen injuries were significantly more likely to undergo a splenectomy if cared for at a general hospital compared with a children’s hospital (OR, 5.01; 95% CI, 2.21-11.36; P<.001) even when controlling for the state factors (OR, 5.59; 95% CI, 2.58-12.11; P<.001). Other hospital characteristics (eg, urban location, trauma designation) were not significant in this model. We observed an increased risk of splenectomy in older children, in penetrating trauma, in children with massive spleen disruptions, in those with more severe injuries (higher ISS), and in children who received blood products.
We next compared outcomes between children undergoing splenectomies and those receiving nonoperative management. Controlling for the same patient and hospital characteristics as in Table 3, we found children undergoing splenectomy were at increased risk for in-hospital death compared with children receiving nonoperative management (OR, 3.96; 95% CI, 1.99-7.88; P<.001). While length of hospital stay appeared greater among patients undergoing splenectomy in the bivariate analysis (P = .02), no significant difference was observed after controlling for patient and hospital characteristics (P = .51). Mean hospital charges were $39 568 for patients undergoing splenectomy compared with $23 575 for patients receiving nonoperative management (P<.001).
Children cared for at general hospitals were more likely to undergo a splenectomy at any time during their stay than children receiving care at children’s hospitals (HR, 3.16; 95% CI, 1.71-5.85) (Figure). We found that the results for hospital type were robust with respect to inclusion or exclusion of hospital trauma status and injury type and in analyses including children up to 18 years old. We also conducted additional analyses modeling propensity to receive care at a general hospital and observed similar effect sizes and significance (results not shown). We also analyzed the 41 partial splenectomies and 113 patients who underwent splenorrhaphy that were included in the spleen-conserving group and observed comparable rates of partial splenectomies (1.1% in children’s hospitals vs 1.5% in general hospitals), but that splenorrhaphy was more common in general hospitals than in children’s hospitals (4.3% vs 1.7%, respectively; P<.001).
While nonoperative management of hemodynamically stable children with splenic injuries is the standard of care,1 management variation persists across hospitals. To our knowledge, this is the first study to compare risk of splenectomy between children’s and general hospitals, controlling for hospital trauma designation and other observable potential confounders. We found treatment of a pediatric splenic injury at a general hospital was associated with a 5-fold increased risk of splenectomy; this finding was robust after controlling for many patient and hospital characteristics.
Potoka and colleagues22 previously examined 1993-1997 trauma registry data for Pennsylvania and documented more nonoperative management of splenic injuries at pediatric trauma centers. That study was limited to designated trauma services. Most states do not include a separate designation for pediatric centers, and some children’s hospitals may not meet the general designation requirements and may thus not be trauma designated. Mooney and Forbes14 examined variation in the management of pediatric splenic injuries in 6 New England states and observed an increased risk for splenectomy in children cared for by general surgeons compared with children cared for by surgeons with specialty pediatric training. This same study observed an increased risk for splenectomy at trauma centers, but did not include children’s hospital status as a predictor. Our study showed no association between trauma designation and splenectomy.
There are several limitations to our study. First, we cannot rule out the potential for omitted variable bias due to unobserved covariates that might distinguish the group of children receiving care at children’s hospitals from those cared for at general hospitals. While we believe that our model adequately adjusts for patient and hospital characteristics, we cannot rule out the possibility of unmeasured covariates that could further distinguish the treatment groups. Second, we lacked physiological and laboratory data, such as systolic blood pressure and hematocrit. Third, we relied on ICD-9-CM diagnostic codes to control for spleen injury severity, rather than the preferred, but unavailable, spleen injury grade. About 9% of splenectomies occurred in children with injuries coded as unspecified splenic injury (unspecified codes of 865.00, 865.10), with the potential for misclassification. We also used blood transfusion as a surrogate for hypotension and blood loss, but we do not know the extent of transfusions (ie, number of units received) or the time course of transfusions (ie, in the emergency department, in the operating room, or in the intensive care unit or ward setting). While it is unclear if the ICD-9-CM code for transfusions received is an adequate surrogate for hypotension, we also used the ISS to adjust for severity of trauma. Fourth, for patients undergoing splenectomy within 1 day of admission it may be that the initial management plan was nonoperative. It is expected that this bias would occur independent of site management. Furthermore, while time to procedure appears to be a reasonable surrogate for intended management, 11% of cases did not have time to procedure information and those data were imputed. While we used multivariable Cox proportional hazards regression for the imputation of these missing data points, most procedures occurred within 1 day of arrival and all imputed data points were within 1 day. Also, our data included time to procedure in days, and as such, we lacked the ability to detect smaller differences in time to splenectomy (eg, hours). To further evaluate this limitation we also performed a sensitivity analysis assuming that all of these cases were failures of nonoperative management rather than early splenectomies; the results were comparable to the original model with no significant differences.
We cannot rule out the possibility of selection bias in this sample, with seriously injured children potentially being transported more often to pediatric facilities. However, if this bias does exist in these data, we expect that the direction would be toward the null. We also cannot rule out the possibility that the most seriously injured children were too unstable to be transported to a children’s hospital, in which case it might be expected that a lower rate of splenectomy would be found. We believe that controlling for injury severity, massive disruption of the spleen, other injury in the abdomen region, penetrating mechanism, and receipt of blood products should reasonably control for this potential. We also evaluated rates of splenectomy in nontransferred patients only and found this effect to be robust. Finally, while the 2000 KID data set is the most recently available, we did not find any evidence of a secular trend when we repeated the analysis using the 1997 KID data set.
The clinical decision to perform early operative vs nonoperative management of splenic injury may depend on the training and experience of the physician. Pediatric surgeons and/or pediatric intensivists are more likely to be found in children’s hospitals and may possess skill sets and experience that favor nonoperative management. While this analysis does not provide surgeon-level information (eg, general surgeon, pediatric surgeon, intensivist, and the like) focusing quality improvement initiatives on physicians at general hospitals may be an effective approach to decrease the rate of pediatric splenectomy. Additional pediatric education and training for physicians in general hospitals may increase the adoption of nonoperative spleen management guidelines such as those developed by the Eastern Association for the Surgery of Trauma.1 We also cannot rule out differences in care that may be attributable to hospital resources, such as the presence of an intensive care unit or redundancy in blood product availability that may support nonoperative management.
This study demonstrates significant nationwide variability in the use of pediatric spleen conservation after traumatic injury. Focusing educational programs and evidence-based management guidelines on centers with greater use of splenectomy may be a next step to improve the rate of spleen conservation and decrease postsplenectomy morbidity, mortality, and cost.
Corresponding Author: Stephen M. Bowman, MHA, PhC, Office of EMS and Trauma, Washington State Department of Health, PO Box 47853, Olympia, WA 98504 (firstname.lastname@example.org).
Author Contributions: Mr Bowman had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Bowman, Zimmerman, Christakis, Martin.
Acquisition of data: Bowman, Christakis.
Analysis and interpretation of data: Bowman, Zimmerman, Christakis, Sharar.
Drafting of the manuscript: Bowman, Zimmerman.
Critical revision of the manuscript for important intellectual content: Bowman, Zimmerman, Christakis, Sharar, Martin.
Statistical analysis: Bowman, Zimmerman, Christakis, Martin.
Obtained funding: Christakis.
Administrative, technical, or material support: Bowman.
Study supervision: Christakis, Martin.
Financial Disclosures: None reported.
Funding/Support: Mr Bowman is partially supported by funds from the Division of State, Community, and Public Health (DSCPH), Bureau of Health Professions (BHPr), Health Resources and Services Administration (HRSA), Department of Health and Human Services (DHHS) under grant 1 D36 HP 10027, entitled “Enhancing Health Care in Western Washington’s Rural Areas” (total award amount $839 724).
Role of the Sponsor: The DHHS had no direct involvement in this analysis or the decision to publish.
Disclaimer: The information or content and conclusions are those of the authors and should not be construed as the official position or policy of, nor should any endorsements be inferred by the DSCPH, BHPr, HRSA, DHHS, or the US government. The Child Health Institute, Children’s Hospital and Regional Medical Center, University of Washington, purchased the KID data set for use in this project.