Is Hospital Teaching Status a Key Factor in Hospital Charge for Children with Hip Fractures?

Abstract

OBJECTIVE

Proximal femur fractures cause significant pain and economic cost among pediatric patients. The purposes of this study were (a) to evaluate the distribution by hospital type (teaching hospital vs non-teaching hospital) of U.S. pediatric patients aged 1-20 years who were hospitalized with a closed hip fracture and (b) to discern the mean hospital charge and hospital length of stay after employing propensity score to reduce selec-tion bias.

METHODS

The 2006 Healthcare Cost and Uti-lization Project (HCUP) Kids’ Inpatient Database (KID) was queried for children aged up to 20 years that had principle diagnosis of hip fracture injury. Hip fractures were defined by International Classifi-cation of Diseases, 9th Revision, Clinical Modifica-tion codes 820.0, 820.2 and 820.8 under Section “Injury and Poisoning (800-999)” with principle internal fixation procedure codes 78.55, 79.15 and 79.35. Patient demographics and hospital status were presented and analyzed. Differences in mean hospital charge and hospital length of stay by hospital teaching status were assessed via two propensity score based methods.

RESULTS

In total, 1,827 patients were nation-ally included for analysis: 1,392 (76.2%) were treated at a teaching hospital and 435 (23.8%) were treated at a non-teaching hospital. The average age of the patients was 12.88 years old in teaching hospitals vs 14.33 years old in nonteaching hospitals. The propensity score based ad-justment method showed mean hospital charge was $34,779 in teaching hospitals and $32,891 in the non-teaching hospitals, but these differences were not significant (p=0.2940). Likewise, mean length of hospital stay was 4.1 days in teaching hospitals and 3.89 days in non-teaching hospitals, but these differences were also not significant (p=0.4220).

Conclusions

Hospital teaching status did not affect length of stay or total hospital costs in children treated surgically for proximal femur fractures. Future research should be directed at identifying factors associated with variations in hospital charge and length of stay.

Introduction

Pediatric hip fractures are frequent injuries¹. They pose a serious threat to the health and well-being of young people, and can have profound negative consequences for young people in terms of their physical, mental, and emotional health²–⁴. Treatment of these fractures places a burden on the patient’s family, the health care system, and society as a whole⁵.

Much of the previous research on children fractures focused on injury treatments or patterns^1,6-11. Few studies have used national data to compare the hospital charges and length of stay (LOS) by hospital teaching status for closed hip fractures in children and adolescents. A recent paper¹² describing the impact of comorbidities on hospitalization costs following hip fracture, discussed the impact of hospital teaching status on cost, but only with patients greater than 55-years old. Another paper¹³ discussed hospital charge differences by hospital teaching status, but the patient population was asthma-related. Given that pediatric hip fractures are often complex injuries, understanding the impact of hospital teaching status on inpatient costs and length of stay may elucidate the optimal treatment location for these patients.

Thus in this study, we sought to examine length of stay and cost patterns according to teaching status. We hypothesized that mean hospital charge is higher and hospital length of stay is longer in teaching hospitals compared with non-teaching hospitals.

Methods

Data Source

This study used data from the 2006 Healthcare Cost and Utilization Project (HCUP) Kids’ Inpatient Database (KID)¹⁴ produced by the Agency for Healthcare Research and Quality (AHRQ), which is the only national dataset on hospital use, outcomes, and charges designed to study children’s use of hospital services in the United States. This KID database includes a sampling of all hospital discharges where the patient was age 20 or less at admission during the year 2006. It contains approximately 3.1 million pediatric discharges from 3,739 communities, non-rehabilitation hospitals in 38 states representing all 4 geographic census regions (Northeast, Midwest, West, and South). The database also allows for extrapolation to a national estimation of 7.6 million pediatric hospital discharges. Patient demographic variables include age at time of admission, sex, race, and median household income quartiles based on the ZIP code of the family’s residence. Hospitalization variables include admission month and source, diagnostic and procedure codes, duration of stay, total charges, expected payer, waiting days from admission to procedure and discharge disposition. Hospitals included in this database are divided into strata using 6 characteristics: ownership/control, bed size, teaching status (teaching vs non-teaching), rural/urban location, US region, and hospital type (pediatric vs other). Bed capacity is categorized into small, medium, or large, and varied in specific bed capacity depending on whether the hospital was located in a rural area or was an urban non-teaching or urban teaching hospital.

Study of Population

Patients in the 2006 KID were selected by principal ICD-9-CM¹⁵ (International Classification of Diseases, Ninth Revision, Clinical Modification) diagnosis code for a closed fracture of the hip, 820.0 (closed transcervical fracture), 820.2 (closed pertrochanteric fracture), and 820.8 (unspecified closed fracture of femoral neck). Open fractures were not included in this analysis. Also patients received principal ICD-9-CM internal fixation procedure codes of 79.15, 79.35 and 78.55 designating treatment type.

After strict inclusion criteria, 1,107 discharges records were obtained. After applying sampling weights provided by HCUP to extrapolate national estimate, the final data represented 1,827 discharges nationwide.

Data Analysis

Descriptive statistics including numbers, means, standard error, and percentages were used to characterize the study population by hospital teaching status. Continuous variables were compared via least squares means, while proportional comparisons were conducted via Rao-Scott x² analysis. For comparing hospital charge and LOS between teaching hospital and non-teaching hospital, multiple statistical techniques were used. First, a simple univariate analysis in terms of least square means was used. Next, in order to reduce potential patient selection bias between teaching hospital and non-teaching hospital in this retrospective observational study, propensity scores¹⁶ were used to make adjustment. Propensity scores allow for analysis of observational data on a level comparable to randomized control trials^17-19, and their use has been widely accepted^20-23. The propensity score here is defined as the conditional probability (ranging from 0 to 1) of receiving treatment in teaching hospital based on observed co-variants, and estimated from the most popular multivariate logistic regression.

In this study, the co-variants are the patients, hospitals, and hospitalization variables. Specifically, patients variables include: 1) known comorbidities, including alcohol abuse, deficiency anemias, rheumatoid arthritis/ collagen, chronic blood loss anemia, chronic pulmonary disease, coagulopathy, depression, uncomplicated diabetes, diabetes with chronic complications, drug abuse, un-complicated hypertension, hypothyroidism, liver disease, fluid and electrolyte disorders, metastatic cancer, other neurological disorders, obesity, paralysis, psychoses, pulmonary circulation disorders, solid tumor without metastasis, valvular disease, and weight loss, and 2) others, such as age, gender, payer type²⁴ (public-Medicaid vs private-others), and median household income quartiles for patient’s ZIP Code. One variable, race, was excluded due to high missing rate (27.5%). Hospital variables include hospital bed size (small, medium, large), hospital region (northeast, midwest, south, west), and hospital location (rural, urban). Hospitalization variables include treatment (Transcervical-ICD 820.0, Pertrochanteric-ICD-820.2, Unspecified part of neck of femur-I CD-820.8), discharge quarter, number of diagnoses, number of procedures, and number of days from admission to procedure. According to a previous study, delay in operation directly relates to hospital charge and LOS²⁵.

After obtaining propensity scores, a common support test comparing propensity scores distributions between hospital types revealed considerable overlap indicating they were comparable. Then, from among three available propensity score analysis methods¹⁹ (regression, stratification, and matching), two of them, regression and stratification, were used to compare hospital charge and LOS between hospital types. We chose to use two separate statistical models because a positive result using two separate statistical methods holds greater validity than a single method alone.

In the first model, propensity scores were used as a continuous variable to make adjustment in comparing hospital charge and LOS between hospital types in the regression model. In the second model, patients from each cohort (teaching vs non-teaching) were matched into five equal strata based on their propensity scores. Propensity matching into quintiles alone has been shown to reduce bias by 90%¹⁹. Differences between hospital types for each stratum on hospital charge or LOS were calculated, then averaged across strata using stratum-specific weights, that is, the square of the standard error of the difference between means, and finally the overall effects of hospital teaching status on hospital charge and LOS were derived from the averaged difference divided by the variance of the estimated mean differences²⁶. Statistical analyses, after incorporating complex sample designs, were conducted using SAS (Cary, NC) version 9.2²⁷. The level of significance for all statistical tests was set at P<0.05.

Results

Characteristics of the 1,827 hospital discharges for hip fractured children and adolescents are presented in Table 1(A) and (B). Table 1(A) displays categorical variables, while Table 1(B) shows continuous variables. It can be seen from Table 1(A) that most patients were treated in teaching hospitals (76.2%), located at urban with large bed sizes. Patients were predominantly male (71.47%) and about two-thirds were white. Seventy-one percent of the sample reported a private payer as the primary source of insurance coverage, with 29% having a public payer. Family income level was nearly evenly distributed.

Table 1(A).

Characteristics of U.S. hospital discharges for hip fractured children and adolescents by hospital teaching status in 2006–for categorical variables

	Total (N=1827)		Teaching Hospital (n=1392)		Non-Teaching Hospital (n=435)		P value
		No.	%	No.	%	No.		%
Sex							0.2708
Male	1270	71.47	962	70.64	308	74.20
Female	507	28.52	400	29.36	107	25.80
Race/ethnicity							0.0003
White	822	61.6	588	57.93	234	73.27
Black	214	16	175	17.23	39	12.08
Other	299	22.4	252	24.84	47	14.65
Income quartile							0.8257
1	514	28.68	398	29.15	116	27.20
2	426	23.78	328	24.03	98	22.99
3	437	24.44	325	23.85	112	26.37
4	413	23.08	313	22.97	100	23.45
Payer type							0.3330
Private	1290	70.75	972	70.00	318	73.15
Public	422	29.59	417	30.00	*	26.85
Hospital bed size							<0.0001
Small	197	10.8	158	11.36	39	9.02
Medium	459	25.13	375	26.97	84	19.24
Large	1171	64.08	859	61.68	312	71.74
Hospital location Rural	121	6.65	20	1.41	101	23.24	<0.0001
Urban	1706	93.35	1372	98.53	334	76.76
Hospital region							<0.0001
Northeast	326	17.84	286	20.51	40	9.29
Midwest	375	20.51	286	20.51	89	20.46
South	685	34.49	514	36.95	171	39.25
West	442	24.16	307	22.02	135	31.01
Discharge quarter							0.8089
Jan.-Mar.	441	24.11	337	24.19	104	23.82
Apr.-Jun.	448	24.52	348	25.00	100	22.99
Jul.-Sep.	532	29.1	406	29.15	126	28.97
Oct.-Dec.	407	22.27	302	21.66	105	24.22
Treatment							0.7334
Pertrochanteric	846	46.31	637	45.77	209	48.07
Transcervical	737	40.3	564	40.48	173	39.72
Unspecificed	244	13.38	191	13.75	53	12.21
Neck

^*less than 10, not reported per data use agreement by AHRQ

Table 1(B).

Characteristics of U.S. hospital discharges for hip fractured children and adolescents by hospital teaching status in 2006–for continuous variables

	Teaching Hospital (n=1392)	Non-Teaching Hospital (n=435)	P value
	Mean (S.E.)	Mean (S.E.)
Age	12.88(0.15)	14.33(0.25)	<0.0001
No. of diagnoses	3.31(0.11)	2.86 (0.15)	0.0157
No. of procedures	2.03(0.07)	1.77(0.11)	0.0402
No. of days from admission to principal procedure	0.70(0.04)	0.46(0.04)	0.0001

The patients treated in teaching hospitals had a mean age 12.88 years, compared to the rest treated in non-teaching hospitals with a higher mean age 14.33 years. See Table 1(B). Also, the mean number of diagnoses, the mean number of procedures, and the mean waiting days from admission to principle procedure were larger at teaching hospitals, and those differences were significantly different.

Proportional comparisons in the distributions of the most frequently mentioned comorbidities (chronic pulmonary disease, deficiency anemias, fluid and electrolyte disorders, other neurological disorders, and obesity) revealed no difference by hospital teaching status. Seventy-three percent had no comorbidity, 20% had a single one, and only 0.1% had maximum six comorbidities. Treatment procedures proportions showed no differences.

Unadjusted hospital charges were significantly higher (17.36% more) in teaching vs non-teaching hospitals (p=0.0085). After propensity adjustment, however, these differences narrowed to 5.74% and were no longer statistically significant (p = 0.2940). From another perspective, the averaged across strata mean hospital charge difference between teaching and non-teaching hospitals was $5586, but not significant (95% CI [-$1486-$12657]). These results are detailed in Table 2. With regard to hospital LOS (Table 3), unadjusted rates were 17.03% longer at teaching hospitals (p=0.0198). After adjustment, however, this difference was only 6.22% and not significant (p=0.4220). Likewise, the averaged across strata mean LOS difference between teaching and non-teaching hospitals was 0.44 days, but this also was not significant (95% CI [-0.16, 1.03]).

Table 2.

Comparisons of hospital charge for hip fractured children and adolescents by hospital teaching status in 2006

	Teaching Hospital mean (S.E.), (US$)	Non-Teaching Hospitalmean (S.E.), (US$)	P value
Unadjusted	34662(1234)	29534(1490)	0.0085
Propensity score adjusted	34779(1303)	32891(1512)	0.2940

Table 3.

Comparisons of length of stay for hip fractured children and adolescents by hospital teaching status in 2006

	Teaching Hospital mean (S.E.), days	Non-Teaching Hospital mean (S.E.), days	P value
Unadjusted	4.26(0.17)	3.64(0.21)	0.0198
Propensity score adjusted	4.10(0.14)	3.89(0.23)	0.4220

Discussions

In this study, we have analyzed the impact hospital type had on total hospital charge and hospital length of stay after closed hip fracture treatment in 1827 pediatric patients using the 2006 KID data. Treatment of these injuries is costly, with annual US charges exceeding $60 million. After propensity score adjustment, we have shown that hospital type had minimal influence on hospital length of stay and hospital costs. We also observed some interesting demographic and geographic trends.

Several of these results merit further discussion. The present study showed clear gender disparity, namely males were more likely to sustain hip fracture injuries than females, and that children were more likely to be admitted to large, urban, teaching hospitals. Also, there were more patients from the southern region than any other region, and more fractures occurred in summer than any other seasons, likely reflecting climate variation. These results are consistent with previous studies which demonstrate seasonal and climate variations within pediatric trauma patterns²⁴. The results in Table 1(B) probably imply that teaching hospitals treated more serious patients (more diagnoses and procedures) and higher volume (longer waiting time).

Understanding the relationship between hospital teaching status and costs of care is particularly relevant in locales in which teaching and children’s hospitals compete with more efficient community hospitals. After using propensity scores to reduce potential selection bias here, we found that hospital charge was not significantly different between hospital types, and the same was true for LOS. These findings were consistent with findings seen in children with asthma¹³. This is particularly important given the increasingly competitive health care markets in which many teaching hospitals operate with functions of education, research, and care. Due to the limitations of the present database, it is unclear whether the observed differences in LOS and hospital charge are the results of enhancement of care or fulfillments of more functions by teaching hospitals.

There are several limitations within our study. Aside from the retrospective nature of our study, the KID is missing detailed, clinically important perioperative information such as blood loss and type of anesthetic used, which may have relationship to LOS and hospital charges. Since data capture ends at discharge, individuals who were hospitalized multiple times might have multiple records in the KID. The cost information provided by the KID is based on hospital charges, not actual costs; these are not always the same. Therefore, our estimation of total hospital charges may not reflect fully the financial impact on the patients and their families. In addition, not all hospitals in the United States were included in the KID.

Conclusions

This study analyzed characteristics of hip fractures that resulted in hospitalization in children by hospital teaching status, and suggested that teaching hospitals do not have higher charges or longer stay. Future research should be directed towards understanding the referral patterns after pediatric fractures, injuries appropriate for hospital transfer, and the economic impact of these patterns.