 |
|
 |
|
 |
 |
|
 |
 |
 |
 |
 |
 |
 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
* Johns Hopkins School of Medicine, Baltimore, Maryland
Department of Biostatistics, Johns Hopkins Bloomberg School of Public Health, Baltimore, Maryland
 |
ABSTRACT |
|---|
 TOP ABSTRACT METHODSRESULTSDISCUSSIONREFERENCES |
|---|
Methods. Data were analyzed from the Healthcare Cost and Utilization Project Kids' Inpatient Database 2000. Encounters qualified for evaluation when 1 of the top 3 discharge codes was consistent with pneumonia, gastroenteritis, respiratory syncytial virus, dehydration, or asthma. Our outcomes were LOS and total charges per hospital admission; hospitals were categorized as children's hospitals and nonchildren's hospitals. We adjusted for the following potential confounders: number of diagnoses, insurance information, patient age in years, race of patient, admission source, procedures, teaching status of hospital, and hospital location. Because of the right skew of the outcomes, our primary analyses consisted of robust median regression; to support our final models, we also performed sensitivity analyses.
Results. Of 252262 total inpatient encounters, 24322 met the inclusion criteria. There were 3408 encounters from 23 different freestanding children's hospitals and 20914 encounters from 1749 nonchildren's hospitals. Freestanding children's hospitals provided care to a higher risk population with more children transferred from other hospitals, a higher percentage of minorities, increased number of co-diagnoses, and a higher percentage on Medicaid. There was no statistically significant difference in LOS by hospital type. However, there was a significant difference in total costs, with the median cost of an admission at freestanding children's hospitals $1294 more per hospitalization than at nonchildren's hospitals, after adjusting for confounders.
Conclusion. We found no significant difference in median LOS among freestanding children's hospitals and nonchildren's hospitals, but freestanding children's hospitals had higher total charges per admission, even after adjusting for differences in population characteristics. Additional studies are needed to elucidate whether these increased costs result in better health outcomes or are simply attributable to other characteristics of children's hospitals, in which not all patients may benefit.
Key Words: hospital performance • length of stay
Abbreviations: NACHRI, National Association of Children's Hospitals and Related Institutions • HCUP, Healthcare Cost and Utilization Project • The KID, Kids' Inpatient Database 2000 • AHRQ, Agency for Healthcare Research and Quality • ICD-9-CM, International Classification of Diseases, Ninth Revision, Clinical Modifications • CI, confidence interval
The National Association of Children's Hospitals and Related Institutions (NACHRI) has cited that "all children need children's hospitals."1 For some patients and certain procedures, such as subspecialty surgery, special-needs children, oncology regimens, and rare problems, hospitals with higher volumes, which are generally freestanding children's hospitals, are likely much better equipped than other hospitals.2–6 However, for more common admissions such as asthma, dehydration, pneumonia, and diarrhea, it is not known whether children's hospitals are superior to other hospitals in providing care. Recently there has been an impetus, from both the private sector and the government, for hospitals to provide data on length of stay (LOS), total charges, and other patient outcomes to facilitate evidence-based or quality-of-care decisions with regard to which hospitals truly provide superior care.7,8
Advocates of children's hospitals claim that access to more subspecialists, laboratories, radiology equipment, and ancillary services that are designed for children, coupled with the general composition of children's hospitals, makes them more appropriate places for children to be treated. With adults, it has been shown that access to more subspecialty-oriented care results in higher costs, move invasive procedures, and more tests but does not guarantee improved outcomes.9–11 Similarly, it is possible that children's hospitals may feel obliged to order more computed topography scans or keep children in the ICU longer for conditions such as asthma and pneumonia; although more readily available, these practices may not improve care. One recent study demonstrated that ventilated children who were treated outside the ICU faired just as well, in cost and clinical outcome, as those who were treated inside the ICU.12
In 2000, children's hospital admissions accounted for 18% of the total US admissions, 
6.3 million visits.13 Although freestanding children's hospitals comprise only 1% of all US hospitals, they account for 39% of child admissions, 49% of inpatient days, and 59% of costs at $10 billion a year.1 In addition, governmental funding for general medical education at children's hospitals rose from $38 million in 2000 to $221 million in 2001.14 These high societal costs, in both patient care and dollars, exemplify the importance that freestanding children's hospitals play in our health care system.
We propose to compare children's care in freestanding children's hospitals with that of other hospitals, which treat children with common diagnoses, with respect to LOS and total charges. Our hypothesis is that children with similar diagnoses at freestanding children's hospitals will have longer LOSs and higher costs than children at other hospitals.
|
METHODS |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
All diagnoses were selected using the International Classification of Diseases, Ninth Revision, Clinical Modifications (ICD-9-CM). Our inclusion criteria were all diagnosis of pneumonia, ICD-9-CM codes 480 to 486, gastroenteritis 558.9 or 009, respiratory syncytial virus 079, dehydration 276.5, and asthma 493. Patient records were included when any of these diagnoses were coded as the primary or secondary reason for the hospitalization.
In The KID data set, each hospital is categorized as a nonchildren's hospital, children's general hospital, children's specialty hospital, or children's unit in a general hospital, as defined by NACHRI. We excluded the last category to allow a comparison of freestanding children's hospitals (either children's general hospital or children's specialty hospital) versus nonchildren's hospitals. However, as expected as a result of our ICD-9-CM inclusion criteria, no children's specialty hospitals were included in the analysis.
Our dependent variables were LOS and total charges. LOS was measured in days; a value of 0 was possible when a child was discharged on the same day as admission. Total charges were measured in US dollars.
Statistical Analysis
All statistical analysis was conducted using Stata 8.2 statistical software.16 Because The KID is not a random sample of states, all of our primary analyses were performed using the HCUP sample weights, enabling us to produce national estimates. Because total charge data for the state of Texas were available for only half of the year, total charge-weighted data do not include Texas. Multiple encounters per child were possible but not distinguishable in the data set.
Initially, basic descriptive statistics were used to compare the characteristics of children in the 2 groups, children's hospitals and nonchildren's hospitals. To compare baseline characteristics between the 2 types of hospital groups, we used paired t tests for continuous variables and the 
2 statistic for ordinal variables.
The potential confounders included number of diagnoses, insurance type, patient age in years, race of patient, admission source, performance of procedure, teaching status of hospital, and hospital location. The number of diagnoses was provided as a continuous variable, taking values 1 to 25; however, 90% of the encounters were associated with 5 or fewer diagnoses. We categorized this variable as 1, 2, 3, 4, and 5 or more diagnoses; 1 diagnosis served as the reference group. Insurance information was categorized, with Medicaid as the reference group. Age was recorded in years, and ages for patients who were younger than 1 year were converted to fractions; for example, a 6-month-old was considered 0.5 years. Race was examined as white, black, Hispanic, and other, which included Asians, Pacific Islanders, and Native Americans. Admission source was categorized with the emergency department as the reference category. Nonteaching hospitals were the reference group for the dichotomous variable of teaching status. Location of hospital was a dichotomous variable, with rural as the reference category. When total charges were examined, LOS was also used as a continuous variable. In addition, variance inflation factors were obtained to check for multicollinearity between variables.
Bivariate relationships between the outcomes and covariates with hospital type were explored using ordinary least squares linear regression analysis. Both of the outcome variables LOS and total charges were extremely skewed to the right, and Shapiro-Wilk tests confirmed that these variables were not normally distributed. Differences in outcomes by bivariate analyses were assessed using nonparametric tests (Wilcoxon rank sum and Kruskal-Wallis tests). Median regression analyses were performed for multivariable analyses to attenuate the influence of outliers on results.17 In addition, we bootstrapped our models to provide more accurate estimated SDs and confidence intervals (CIs). To check the performance of our models, we ran different sensitivity analyses: logistic regression, robust bootstrap logistic regression, multiple linear regression using log LOS, and multiple linear regression using log total charges.16 Analyses were also repeated after deleting total charges >$100000, $14077, and $7466, respectively, and categorizing age using different cutoffs. Finally, to assess whether there was clustering among hospitals, we fit a random effects model, using generalized estimating equations with a uniform working correlation matrix.18
Because of the large number of encounters, we analyzed a simple random sample of 10% of the visits, leaving us with 252262 inpatient encounters. To generate a subsample of the data set, we randomly assigned to each study participant a number between 0 and 1 generated from a uniform distribution. We excluded from this study all participants whose numbers were >0.10. The statistics presented in this article were estimated using cases with complete covariate data. None of the covariates had >3% missing values, with the exception of age (9%).
|
RESULTS |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
General Characteristics
Table 1 provides a description of patient and hospital characteristics; the distributions were statistically significantly different by hospital type for all characteristics (P < .0001), except for age. There were 3408 encounters from 23 different freestanding children's hospitals and 20914 encounters from 1749 nonchildren's hospitals. By definition, 100% of the children's hospitals were teaching hospitals as compared with 32% of the nonchildren's hospitals. In addition, 100% of the children's hospitals were located in urban areas compared with 76% of the nonchildren's hospitals.
|
It is not possible to adjust for hospital location or teaching status of hospital, because as shown in Table 1, all-freestanding children's hospitals had 100% teaching status and urban location. Unadjusted differences in total charges by hospital type for all covariates are shown in Table 2.
|
|
0, as a result of the large sample size. After adjusting for hospital type, insurance status, patient age, patient race, and admission source, the only statistically significant differences in median LOS were observed when comparing encounters with 1 or more procedures versus none (increase in LOS of 1 day in freestanding children's hospitals) or encounters with 5 or more diagnoses versus 1 diagnosis (increase of LOS of 2 days in freestanding children's hospitals). Numerous sensitivity analyses were performed to support the conclusion from our final model. Because 50% of the encounters were 2 days or less, we dichotomized LOS as >2 days and 2 days or less. Multiple logistic regression analysis indicated that the estimated adjusted odds of a longer LOS were 3% higher in children's hospitals but not statistically significant (95% CI: 0.95–1.13; P = .451). Multiple linear regression analysis of the log LOS resulted in an adjusted coefficient of 0.05 (95% CI: 0.023–0.079; P < .0001). Nonweighted bootstrap median regression resulted in similar findings to our weighted median regression results. Using generalized estimating equations to adjust for potential correlation of outcomes within hospitals showed a statistically significant but small increase in LOS of 0.51 days for children's hospitals (95% CI: 0.20–0.83; P = .001). The last 2 procedures could not be performed incorporating the sample weights. However, the results of our weighted and unweighted analyses seemed robust. Thus, each of these analyses provided little evidence of a clinically significant difference in LOS by hospital type.
Total Charges
Table 3 also displays the distribution of total charges by hospital type. Unadjusted mean (median) total charges in children's hospitals were $12952 ($5843), whereas they were $6476 ($3644) in nonchildren's hospitals. For children's hospitals, 90% of total charges were <$27000; however, 44 encounters had charges >$100000. For nonchildren's hospitals, 90% of encounters were <$13000; however, 67 encounters had costs >$100000.
Table 2 displays both the unadjusted and adjusted differences in median total charges by hospital type and patient characteristics. Higher total charges were associated with increased LOS, Medicare insurance status, minority race, transfer from another hospital, urban location, teaching hospitals, increased age of patient, 1 or more procedure, and 4 or 5 diagnoses. The unadjusted median total charges for children's hospitals were $2199 more than nonchildren's hospitals. After adjustment for LOS and statistically significant confounders, the difference in median total charges decreased to $1294 (95% CI: $1181–$1408; P < .0001; Table 2).
Additional sensitivity analyses were performed to confirm our results. A multiple linear regression analysis of the log total charges yielded a statistically significant difference by hospital type (adjusted difference = 0.25; 95% CI: 0.23–0.28; P < .0001). We also repeated the regression analysis after excluding outliers defined as total charges >$100000, and the results remained robust. As additional sensitivity analyses, we performed the analysis excluding the top 10th percentile (>$14077) and 25th percentile (>$7466) of total charges, respectively; the differences remained statistically significant, although slightly decreasing in magnitude. Thus, each of these analyses revealed statistically significantly higher total charges observed in the children's hospitals.
|
DISCUSSION |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
Previous studies have found differing results, some showing an increase in LOS among teaching hospitals versus nonteaching hospitals, whereas others have found similar LOS among hospital types.19–21 Although our hypothesis regarding increased LOS in children's hospitals could not be supported, we believe that our study provides a more definitive answer when examining freestanding children's hospitals and nonchildren's hospitals with respect to LOS. Our study has the advantage of national scope, ability to case-adjust for multiple factors, weighted data, and a very large sample size. In addition, we performed multiple different sensitivity analyses under varying assumptions, and no meaningful differences in LOS were observed by hospital type.
Alternatively, our other hypothesis was supported in that total charges were statistically significantly higher among freestanding children's hospitals than nonchildren's hospitals. This is consistent with previous observations regarding total charges for adult patients who were admitted to tertiary hospitals.22 We postulated that total charges would increase in children's hospitals because subspecialists, with additional technology more readily available, likely contributed to more of the care. In fact, in our analyses, 34% of encounters at children's hospitals had at least 1 procedure performed compared with 17% in nonchildren's hospitals. However, these greater costs and increased procedures fortunately did not result in longer LOS for children who were admitted to freestanding children's hospitals.
An important caveat to our results is that although we tested very important outcomes, LOS and total charges, data on direct health outcomes were not available. Thus, it is possible that these additional tests and costs resulted in lower readmission rates, better patient and caregiver education, and overall improved health. One study of childbirth demonstrated superior outcomes at academic centers with higher costs.23 In addition, we were unable to examine whether the nurse-to-patient ratio was higher at freestanding children's hospitals; this has been shown to improve patient outcomes and might also explain the higher costs.24,25 For this investigation, we chose common, generally benign conditions. If the increased costs in children's hospitals resulted in better care, then that is commendable, but if they simply resulted in more tests and procedures, then freestanding children's hospitals must examine the care that they provide.
Our analysis is based on coding of selected ICD-9-CM codes. There are clear limitations to the accuracy of this type of data collection, but the professional coding used by The KID strengthens our investigation.26–29 Also, we used multiple variables for case-mix adjustment that resulted in significant changes from the unadjusted data. However, there are many ways to case-adjust, and it is possible that there were differences between the hospital types for which we were not able to control. Although AHRQ does have a case-mix adjustment for HCUP, it is available only for adult patients; in addition, NACHRI's case-mix adjustment is not publicly available.1,30–32 Finally costs and charges are not identical, with costs generally near 50% of actual charges. Currently, formulas are being developed by AHRQ to provide better estimates of total costs and collections. A senior economist at AHRQ told us that, in general, freestanding children's hospital's costs are roughly 42% whereas other hospitals are 46% on the dollar (C. Steiner, MD, MPH, written communication, June 2004). This does not affect the LOS data; furthermore, the differences in total charges are large enough to remain robust.
We have demonstrated that on at least 1 important indicator of quality of care, LOS, freestanding children's hospitals are on par with nonchildren's hospitals. What is most striking is that they are clearly providing care to a more vulnerable population, with much higher percentages of patients who qualify as minorities, more patients transferred from other hospitals, patients with more diagnoses, and more Medicaid patients. Although the increased total charges for freestanding children's hospitals need additional examination, it is possible that these increased costs result in better health outcomes and reduced costs in the long term. These associations must be understood before market forces, patients, physicians, and insurers make uninformed decisions for themselves.
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
Address correspondence to Dan Merenstein, MD, Johns Hopkins Hospital, 600 N Wolfe St, Carnegie 291, Baltimore, MD 21287. E-mail: dmerenstein{at}jhu.edu
No conflict of interest declared.
|
REFERENCES |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
This article has been cited by other articles:
 |
|
 |
|
  S. Shah, S. A Sinclair, G. A Smith, and H. Xiang Pediatric hospitalizations for bicycle-related injuries Inj. Prev., October 1, 2007; 13(5): 316 - 321. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 F. O. Odetola, A. Gebremariam, and G. L. Freed Patient and Hospital Correlates of Clinical Outcomes and Resource Utilization in Severe Pediatric Sepsis Pediatrics, March 1, 2007; 119(3): 487 - 494. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. J. Okumura, A. D. Campbell, S. Z. Nasr, and M. M. Davis Inpatient Health Care Use Among Adult Survivors of Chronic Childhood Illnesses in the United States. Arch Pediatr Adolesc Med, October 1, 2006; 160(10): 1054 - 1060. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. J. Schneier, B. J. Shields, S. G. Hostetler, H. Xiang, and G. A. Smith Incidence of Pediatric Traumatic Brain Injury and Associated Hospital Resource Utilization in the United States Pediatrics, August 1, 2006; 118(2): 483 - 492. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
 |
||
|
||
Home | My Pediatrics | Journal Information | Current Issue | Past Issues | Subscriptions & Services | Contact Us Click here for faster international access © 2005 American Academy of Pediatrics. All rights reserved. |
||
|
||
