During the past decade, the number of children with asthma increased; however, the number of asthma hospitalizations for children decreased.
To assess the proportion of high-severity cases among children hospitalized with asthma and the association of high-severity asthma with patient and hospital characteristics.
The data set used was the Healthcare Cost and Utilization Project Nationwide Inpatient Sample. Records were selected of patients 18 years and younger who had the principal diagnosis of asthma. Records were analyzed of 29 077 patients at 746 hospitals in 1990 and 33 443 patients at 811 hospitals in 1995. Asthma severity was classified by All Patient Refined–Diagnosis-Related Groups. Cross-sectional logistic regression analysis was performed using survey data analysis software.
The most common diagnoses associated with high-severity asthma were respiratory distress and respiratory failure. The proportion of high-severity asthma cases did not change significantly between 1990 (4.2%) and 1995 (4.6%) (P = .08). Adolescents and boys were more likely to have high-severity asthma than children aged 5 to 12 years and girls. Western, southern, and north-central hospitals and urban teaching hospitals had a higher proportion of high-severity asthma cases than northeastern hospitals and nonteaching hospitals.
Between 1990 and 1995, the proportion of high-severity cases among children hospitalized with asthma did not change significantly. However, patient age, sex, region of the country, and hospital teaching status were associated with variations in the proportion of high-severity asthma cases.
DURING THE past decade in the United States, the number of children with self-reported asthma increased and the number of childhood deaths caused by asthma also increased. However, during the same period, the number of asthma hospitalizations for children decreased.1 Specifically, between 1990 and 1994, the number of children younger than 16 years with self-reported asthma increased by 26% to 4.1 million. During the same period, the number of children hospitalized with asthma decreased by 11% to 164 000.
In a similar study in Monroe County, New York, between 1991 and 1995, the annual hospitalization rate for children with asthma remained fairly constant. However, as a proportion of all childhood asthma hospitalizations, mild cases decreased and severe cases increased. Asthma severity was based on the worst recorded oxygen saturation during the first 24 hours of hospitalization.2 However, to our knowledge, no published study has used a large national sample of inpatient records to assess the severity of childhood asthma by patient and hospital characteristics.
The main objective of this study was to assess recent trends in asthma severity among children hospitalized in the United States. The study addressed 3 key questions. First, what was the proportion of high-severity cases among children hospitalized with asthma in 1990 and 1995? Second, what patient and hospital characteristics were associated with high-severity childhood asthma? Third, what was the impact of asthma severity on median total charges and median length of stay over time?
High-severity asthma is defined in this study as major or extreme complications such as respiratory distress or respiratory failure, respectively. The proportion of hospitalizations due to high-severity asthma is the number of hospitalizations for major or extreme asthma divided by the number of hospitalizations for all cases of asthma.
We hypothesized that the proportion of high-severity cases among children hospitalized with asthma increased between 1990 and 1995. The proportion of high-severity cases increases as the number of high-severity cases in the community increases, as the threshold for admitting low-severity cases increases, and as attending physicians document more comorbidities and complications in discharge records.
Nationwide inpatient sample (nis)
The data set used was the publicly available NIS for US hospital discharges occurring in 1990 and 1995. The NIS is part of the Healthcare Cost and Utilization Project sponsored by the Agency for Healthcare Research and Quality, Rockville, Md. The NIS contains uniform inpatient stay data collected from existing hospital discharge databases maintained by state agencies, hospital associations, and other private data organizations. It is designed to approximate a 20% sample of US nonfederal, short-term hospitals as defined by the American Hospital Association, and is stratified according to geographic region, ownership, location, teaching status, and bed size.
The NIS contains records for all stays in the sampled hospitals. Nationwide Inpatient Sample release 1 for 1990 was drawn from 11 states (Arizona, California, Colorado, Florida, Illinois, Iowa, Massachusetts, New Jersey, Pennsylvania, Washington, and Wisconsin) and contains 6.3 million inpatient records from 871 hospitals. Nationwide Inpatient Sample release 4 for 1995 was drawn from 19 states (the 11 previously mentioned plus Connecticut, Kansas, Maryland, Missouri, New York, Oregon, South Carolina, and Tennessee) and contains 6.7 million records from 938 hospitals.
The NIS categorizes hospitals by 4 regions: Northeast (Connecticut, Maine, New Jersey, New York, and Pennsylvania), North-central (Illinois, Iowa, Kansas, and Wisconsin), South (Florida, Maryland, Missouri, South Carolina, and Tennessee), and West (Arizona, California, Colorado, Oregon, and Washington). Hospital ownership categories are private nonprofit, public, and private investor owned. The hospital locations are urban or rural. A hospital is considered to be a teaching institution if it has either an American Medical Association–approved residency program or a membership in the Council of Teaching Hospitals. Rural hospitals are not split according to teaching status because rural teaching hospitals are rare. Hospital bed capacity is grouped into small, medium, and large categories. Bed size categories are specific to a hospital's location and teaching status. Detailed definitions of these variables are published elsewhere.3
Patient race or ethnicity categories include white, black, Hispanic, Asian or Pacific Islander, and Native American. Patients are classified into 5 primary payer groups: Medicaid, including fee-for-service and prepaid programs; commercial or a Blue Cross and Blue Shield preferred provider organization; private health maintenance organization or prepaid health plan (not Medicaid); self-pay; and other (Medicare, CHAMPUS, Title V, other government program, workman's compensation, and no charge). Patients are also grouped into 4 annual household income categories based on ZIP code of residence: $0 to $15 000, $15 001 to 30 000, $30 001 to 45 000, and greater than $45 000.
Classification of asthma severity
Severity-of-illness levels were assigned to each patient discharged using the All Patient Refined–Diagnosis Related Groups (APR-DRGs), version 15, software system for inpatient care. Diagnosis-related groups are a patient classification scheme that relates hospital case mix to costs. They are used by the Health Care Financing Administration for hospital payment for Medicare beneficiaries. The APR-DRGs expand DRGs to be more representative of pediatric patients and incorporate illness severity classes. The APR-DRGs were developed jointly by 3M Health Information Systems, Wallingford, Conn, and the National Association of Children's Hospitals and Related Institutions, Alexandria, Va. The classification system was clinically derived and extensively tested with historical data. Use of a National Association of Children's Hospitals and Related Institutions' database in research and development ensured sufficient case volume to test statistically rare but expensive conditions most commonly treated at tertiary hospitals.
The APR-DRGs are recognized as a state-of-the-art classification system for inpatient hospital care. About 1400 hospitals and other organizations, including 17 state health departments and data commissions, use the system.4 In the assessment of administrative data, the APR-DRG system is a reliable tool for case mix adjustment.5 However, the APR-DRG system has not been compared with other severity measurement systems for pediatric patients. These systems vary in their definitions of severity, pertinent patient populations, the role of diagnosis and surgery in quantifying risk, data requirements, system development, timing of reviews, and classification approach.6
The APR-DRG system divides the DRG into 4 severity-of-illness classes. The severity classes are based on groups of patients who were physiologically and functionally similar and who had a comparable pattern of resource intensity. Each secondary diagnosis in the discharge record is assigned to one of the 4 illness severity levels. The assignment of a patient discharge record to an illness severity class considers not only the severity level of the secondary diagnoses but also the interaction among secondary diagnoses, age, principal diagnosis, and the presence of certain operating room and non–operating room procedures.
In this study, low-severity asthma is defined as the APR-DRG classes of mild and moderate severity. High-severity asthma is defined as the APR-DRG classes of major and extreme severity. Typical secondary diagnoses by severity class are otitis media and upper respiratory tract infection in patients with mild asthma, pneumonia and atelectasis in patients with moderate asthma, acute respiratory distress in patients with major asthma, and respiratory failure in patients with extreme asthma.
Patient selection criteria
Children with asthma were selected from the 1990 and 1995 NIS by the following protocol. First, patient records with age 18 years and younger and principal diagnosis and/or first secondary discharge diagnosis as asthma were extracted using SAS statistical software (SAS Institute Inc, Cary, NC). Asthma was defined by the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM)7 as codes between 493.00 and 493.91. A total of 36 185 records in 1990 and 46 938 records in 1995 were extracted initially. Next, DRG codes and asthma severity classes were assigned to each record using the APR-DRG software. If the principal diagnosis was asthma by ICD-9-CM code or if the first secondary diagnosis was asthma by ICD-9-CM code and the DRG was a respiratory condition with a primary diagnosis attributable to asthma, then the record was included in the final data set.
Specifically, the following DRGs and primary diagnoses with the secondary diagnosis asthma were included:
Major respiratory procedure (DRG 120) and primary diagnosis atelectasis (ICD-9-CM code 518.0) or respiratory failure (ICD-9-CM code 518.81).
Minor respiratory procedure (DRG 121) and primary diagnosis spontaneous pneumothorax (ICD-9-CM code 512.8).
Respiratory system diagnosis with ventilator support longer than 96 hours (DRG 130) and primary diagnosis respiratory failure (ICD-9-CM code 518.81) or continuous mechanical ventilation (ICD-9-CM code 967.2).
Pulmonary edema and respiratory failure (DRG 133) and primary diagnosis respiratory failure (ICD-9-CM code 518.81) or respiratory arrest (ICD-9-CM code 799.1).
Pneumothorax and pleural effusion (DRG 143) and primary diagnosis pleural effusion (ICD-9-CM code 511.8 or 511.9) or spontaneous pneumothorax (ICD-9-CM code 512.0 or 512.8).
Respiratory system signs and symptoms (DRG 144) and primary diagnosis atelectasis (ICD-9-CM code 518.0), pulmonary insufficiency (ICD-9-CM code 518.82), or asphyxia (ICD-9-CM code 799.0).
In this manner, 29 077 records for 1990 and 33 443 records for 1995 were selected for inclusion in the final data set for analysis.
The Survey Data Analysis Software System, developed by the Research Triangle Institute, Research Triangle Park, NC, was used for data analysis. The system is specifically designed for analysis of data from stratified, clustered survey samples such as the NIS. It properly accounts for the sampling design in computing the SE of the estimates, thus greatly increasing the accuracy of the results.
Logistic regression analysis was used to assess the effect of explanatory variables on asthma severity. The severity was dichotomized into 2 groups, low vs high, and was studied as the dependent variable. Akaike information criteria were used in a stepwise procedure to select the best model.8 The final model selected using these criteria included patient age and sex and hospital region, location, teaching status, and bed size as covariates. Using the Akaike information criteria, none of the interaction terms were significant (P≤.05); therefore, they were excluded from the model. In addition to the variables described, year (1990 and 1995) was also included as an explanatory variable because the primary objective of the study was to assess the change in asthma severity across time.
The LOGISTIC procedure in the Survey Data Analysis Software System (SUDAAN) was used to perform the logistic regression analysis. The variables were estimated by solving the weighted score equations. The variances of the estimates were computed using a generalized estimating equations approach that accounted for the between-cluster variability and the complexity of the sampling design. This procedure provided consistent estimates for the variance of the variable estimates.
The SUDAAN was used to compute the variable estimates, their SEs, statistics for testing the significance of the explanatory variables, corresponding P values, odds ratio estimates, and 95% confidence intervals for the odds ratios.
Because patient race or ethnicity was not reported by some states, the race or ethnicity variable was excluded from the overall logistic regression because such systematically missing data might bias the results. However, the relative likelihood of high-severity cases by race or ethnicity and study years was assessed for states with high rates of reporting race or ethnicity.
The estimated number of children with asthma was similar in the 1990 and 1995 NIS based on projections from 20% stratified cluster samples (Table 1). In the 1990 and 1995 NIS, 746 and 811 hospitals, respectively, served children with asthma. In both years, the number of children hospitalized with asthma was highest in urban nonteaching and private nonprofit hospitals and was lowest in rural hospitals and for-profit hospitals.
Estimated Number of Children With Asthma in the National Sample by Hospital Characteristics in 1990 and 1995
During 1990 and 1995, the number of mild cases was highest and the number of extremely severe cases was lowest among children hospitalized with asthma (Figure 1). Between 1990 and 1995, the unadjusted proportion of mild severity cases among children hospitalized with asthma decreased slightly and the proportion of moderate, major, and extreme severity cases increased slightly in cross-sectional hospital surveys. However, overall, no statistically significant change in asthma severity among hospitalized children occurred across time (P = .08).
Estimated percentage of children hospitalized with asthma by severity class in 1990 and 1995.
To confirm this cross-sectional trend in childhood asthma severity, records of children with the principal diagnosis of asthma were extracted from the cohort of 373 identical hospitals included in the 1990 and the 1995 NIS. A total of 15 371 records in 1990 and 15 711 records in 1995 were analyzed. In this cohort of hospitals, the unadjusted proportion of mild childhood asthma cases decreased from 79.4% to 76.6%, while moderate severity cases increased from 16.9% to 19.0%. In the same hospital cohort, major severity cases increased from 3.3% to 3.9% and extreme severity cases increased from 0.4% to 0.5%.
In 1990 and 1995, only 50% and 53% of discharge records, respectively, reported secondary diagnoses; thus, half of the cases were classified as mild severity because no comorbidity or complication was documented.
The unadjusted, discharge-weighted proportion of high-severity cases was greatest for preschool-aged children in 1990 and for adolescents in 1995; for girls, Medicaid recipients, and lowest-income families in 1990 and 1995; and for self-pay children in 1995 only (Table 2). The proportion of high-severity cases was also greatest for Native Americans in 1990 and Asian or Pacific Islanders in 1995. Between 1990 and 1995, the unadjusted proportion of high-severity cases among children hospitalized with asthma substantially increased for adolescents, girls, self-pay children, lowest-income families, and Asian or Pacific Islanders and during spring months.
Estimated Number of Children Hospitalized With High-Severity Asthma in 1990 and 1995 by Patient Characteristics and Season*
The unadjusted, discharge-weighted proportion of high-severity cases among children hospitalized with asthma was greatest for large, western, and urban teaching hospitals in 1990 and 1995, public hospitals in 1990, and nonprofit institutions in 1995 (Table 3). From 1990 to 1995, the proportion of high-severity cases substantially increased at north-central, rural, private nonprofit, and large hospitals, and it substantially decreased in for-profit and public hospitals.
Estimated Number of Children Hospitalized With High-Severity Asthma in 1990 and 1995 by Hospital Characteristics*
Patient race or ethnicity was not associated with variation in severity among children hospitalized with asthma in 3 states (California, Massachusetts, and New Jersey) where high rates of patient race were reported in 1990 and 1995.
In a multiple logistic regression analysis adjusting for all factors in the model, a greater proportion of adolescents was more likely to have high-severity asthma than children aged 5 to 12 years (Table 4). A greater proportion of hospitalized boys was more likely to have high-severity asthma than girls. Hospitals in western, southern, and north-central regions were more likely to serve a greater proportion of children with high-severity asthma than hospitals in the Northeast. Urban teaching hospitals were more likely to provide care for a greater proportion of children with high-severity asthma than urban nonteaching hospitals. Race or ethnicity, family income based on ZIP code of residence, primary payer, hospital ownership, bed size, and season were not significantly (P>.05) associated with variations in severity among children hospitalized with asthma. After accounting for all of these variables, high-severity asthma cases were no greater in 1995 than in 1990.
Logistic Regression Analysis of Variables Associated With High Severity Among Children Hospitalized With Asthma
In 1990 and 1995, the median length of stay and the median total charges increased progressively with higher severity class (Table 5). From 1990 to 1995, the median length of stay decreased for children with low-severity asthma, but it did not change for children with high-severity asthma. Between 1990 and 1995, inflation-adjusted median total charges decreased slightly but not significantly for mild, moderate, and major severity cases and increased slightly but not significantly for extreme severity cases.
Median Length of Stay and Inflation-Adjusted Median Total Charges by Asthma Severity Class in 1990 and 1995*
In the cross-sectional and the cohort analyses of the NIS, the proportion of mild cases decreased slightly and the proportion of moderate, major, and extreme severity cases increased slightly among children hospitalized with asthma in 1990 and 1995; however, overall, no statistically significant change in asthma severity occurred across time. Between 1990 to 1992 and 1993 to 1994, the estimated average annual number of persons aged 0 to 4, 5 to 14, and 15 to 34 years with self-reported asthma increased by 47%, 18%, and 22%, respectively, based on the National Health Interview Survey. However, during the same periods, the estimated average annual number of asthma hospitalizations for persons aged 0 to 4, 5 to 14, and 15 to 34 years decreased by 13%, 8%, and 5%, respectively, based on the National Hospital Discharge Survey.1 Perhaps a higher proportion of low-severity asthma cases that did not require inpatient care occurred among the increased number of children with self-reported asthma.
The proportion of hospitalizations for high-severity asthma is affected by the number of admissions of complicated cases, the threshold for admitting low-severity cases, and the discharge coding by attending physicians. The quality of ambulatory care, including choice of preventive therapies and thresholds for admission, likely plays a key role in determining community hospitalization rates for childhood asthma.10
Discharge diagnoses probably are less sensitive indicators of childhood asthma severity than oxygen saturation. In 1990 and 1995, only half of childhood asthma discharge records reported secondary diagnoses. The addition of the secondary diagnosis atelectasis or asphyxia may increase the APR-DRG classification of asthma severity from mild to moderate.
In this study, the unadjusted proportion of high-severity asthma cases was greatest for hospitalized adolescents in 1995. Moreover, a greater proportion of adolescents was more likely to be hospitalized with high-severity asthma than other school-aged children. Between 1990 to 1992 and 1993 to 1994, the estimated average number of deaths with asthma as the underlying cause for children younger than 15 years increased by 15% from 148 to 170 based on the underlying cause of death data set. During this period among children aged 0 to 4 years, the asthma death rate decreased by 14% from 2.1 per million to 1.8 per million. However, among children aged 5 to 14 years, the asthma death rate increased by 23% from 3.0 per million to 3.7 per million.1
Hospitals in western, southern, and north-central regions had a higher probability of treating a greater proportion of high-severity asthma cases than hospitals in the Northeast. From 1990 to 1994, asthma prevalence rates were similar in all regions of the country. From 1988 to 1994, asthma hospitalization rates increased in the Northeast but decreased in the West and Midwest. During those 7 years, asthma death rates in the overall population were highest in the West, while asthma death rates among persons aged 5 to 34 years were highest in the Midwest and Northeast.1 Perhaps northeastern hospitals have a lower threshold for admitting low-severity childhood asthma cases than other regions of the country.
Urban teaching hospitals were more likely to serve a higher proportion of children admitted with high-severity asthma than urban nonteaching hospitals. Adjustment for age, sex, and geographic region revealed the expected likelihood of a greater proportion of high-severity asthma cases at teaching hospitals than at nonteaching institutions.
This study focused on the proportion of high-severity cases among children hospitalized with asthma. The NIS does not provide data about the population of children in communities that might be used to determine hospitalization rates. In fact, fewer than 10 states have conducted asthma prevalence surveys.11 National population-based studies of asthma diagnosis, severity, and management are needed to assess possible variations in hospitalization rates.
The cross-sectional study design limits the ability to draw causal inferences from the study data.12 Analyses, such as this one, that are based on administrative data sets can provide an inexpensive source of information with a large sample size and nationally representative data. At the same time, hospital claims records pose important limitations. This data set lacks important, complementary diagnostic and prognostic information available with concurrently collected clinical data. The assigned severity of illness may be influenced by and confounded with the competence of the hospital care. The severity ratings may reflect a combination of case mix and clinical management that could lead to unwarranted conclusions. These potential limitations must be cautiously acknowledged when interpreting the findings.
Study results may vary depending on which severity method is used for risk adjustment. No published outcome studies of children hospitalized with asthma have compared clinical measures with the APR-DRG classification system. Nevertheless, severity adjustment is essential before comparing patient outcomes across hospitals. Based on the consistent progression in median charges and median length of stay for childhood asthma calculated for the assigned illness severity levels (Table 5), the system was reasonably reliable. The APR-DRG system can be used to risk adjust capitation rates for hospitals serving a disproportionate share of children with high-severity asthma.
In conclusion, based on hospital discharge records, the proportion of high-severity cases among children hospitalized with asthma did not change between 1990 and 1995. However, patient age and sex, region of the country, and hospital teaching status were significantly associated with variations in the severity of asthma among hospitalized children. These study findings imply that reported secondary diagnoses in discharge records may not reliably document hypoxemia. They also imply that thresholds for asthma admission vary by region of the country. Between 1990 and 1995, the median length of stay decreased for children with low-severity asthma but remained the same for children with high-severity asthma. During the same period, inflation-adjusted median total charges did not change significantly for any severity class.
Additional research is needed to estimate state-level asthma prevalence and hospitalization rates. Such studies would provide more precise estimates of the burden of asthma among children and the frequency of inpatient care. Small area analyses of the quality and accessibility of preventive ambulatory asthma care, including thresholds for admission, are also needed.
Box Section Ref ID
Editor's Note: This study involving a national data set provides much information about children hospitalized in 1990 and 1995 for asthma. When the number of reported cases and deaths increases for a disease, we need to know much more about it.—Catherine D. DeAngelis, MD
Accepted for publication June 16, 1999.
This study was funded by grant R03-HS09564-01 from the Agency for Healthcare Research and Quality, Rockville, Md; and a grant from the Children's Hospital of Wisconsin Foundation. The All Patient Refined–Diagnosis Related Group software license was provided by 3M Health Information Systems, Wallingford, Conn.
Presented in part at the Ambulatory Pediatric Association Annual Meeting, San Francisco, Calif, May 2, 1999.
We thank Earnestine Willis, MD, MPH, and 2 anonymous reviewers for useful advice; and Janet Hatcher for creating the figure.
Corresponding author: John R. Meurer, MD, MM, Department of Pediatrics, Medical College of Wisconsin, 8701 Watertown Plank Rd, Milwaukee, WI 53226 (e-mail: firstname.lastname@example.org).
Centers for Disease Control and Prevention, Surveillance for asthma: United States, 1960-1995. MMWR Morb Mortal Wkly Rep.
1998;471- 27Google Scholar
KJ Increase in admission threshold explains stable asthma hospitalization rates. Pediatrics.
1999;104454- 462Google ScholarCrossref
J New version of APR-DRGs released. Child Hospitals Today.
1998;620- 21Google Scholar
J The use of administrative data as the first step in the continuous quality improvement process. Am J Med Qual.
1996;11(suppl)S35- S38Google Scholar
LI Risk Adjustment for Measuring Health Care Outcomes. Ann Arbor, Mich Health Administration Press1994;
Not Available, International Classification of Diseases, Ninth Revision, Clinical Modification. 5th ed. Salt Lake City, Utah Medicode Publications1998;
H Information theory and an extension of the maximum likelihood principle. Petrov
Feds. Proceedings of the Second International Symposium on Information Theory.
Budapest, Hungary Akademiai Kiado1973;267- 281Google Scholar
US Bureau of Labor Statistics, Consumer Price Index Detailed Report. Washington, DC US Bureau of Labor Statistics1997;
et al. Does quality of care affect rates of hospitalization for childhood asthma? Pediatrics.
1996;9818- 23Google Scholar
Centers for Disease Control and Prevention, Forecasted state-specific estimates of self-reported asthma prevalence: United States, 1998. MMWR Morb Mortal Wkly Rep.
1999;471022- 1025Google Scholar
DT Quasi-Experimentation: Design and Analysis Issues for Field Settings. Boston, Mass Houghton Mifflin1979;295- 340