Objective. Hospital care for children with viral lower respiratory illness (VLRI) is highly variable, and its relationship to severity and impact on outcome is unclear. Using the Pediatric Comprehensive Severity Index, we analyzed the correlation of institutional practice variation with severity and resource utilization in 10 children's medical centers.
Methods. Demographics, clinical information, laboratory results, interventions, and outcomes were extracted from the charts of consecutive infants with VLRI from 10 children's medical centers. Pediatric Component of the Comprehensive Severity Index scoring was performed at admission and at maximum during hospitalization. The correlation of patient variables, interventions, and resource utilization at the patient level was compared with their correlation at the aggregate institutional level.
Results. Of 601 patients, 1 died, 6 were discharged to home health care, 4 were discharged to rehabilitative care, and 2 were discharged to chronic nursing care. Individual patient admission severity score correlated positively with patient hospital costs (r = 0.48), but institutional average patient severity was negatively correlated with average institutional costs (r = −0.26). Maximal severity score correlated well with costs (r = 0.66) and length of stay (LOS;r = 0.64) at the patient level but poorly at the institutional level (r = 0.07 costs; r= 0.40 LOS). The institutional intensity of therapy was negatively correlated with admission severity (r = −0.03) but strongly correlated with costs (r = 0.84) and LOS (r = 0.83).
Conclusions. Institutional differences in care practices for children with VLRI were not explained by differences in patient severity and did not affect the children's recovery but correlated significantly with hospital costs and LOS.
- respiratory syncytial virus
- lower respiratory illness
- practice variation
- pediatric intensive care
- intravascular catheters
- Comprehensive Severity Index
- severity scoring
One percent to 2% of all children in the United States are hospitalized for viral lower respiratory illness (VLRI) during their childhood.1,,2 Variation in the treatment of these children has been well described and reflects a lack of consensus regarding optimal therapy.3–5 Ascertainment of optimal care is difficult because our therapies are supportive, not curative, and most children do well irrespective of differences in therapy. Consequently, there is a propensity to persist in care practices that may offer little or marginal benefit. Determination of best care practices requires consideration of the impact of treatment not only on morbidity and mortality but also on resource utilization.
We evaluated the care of children who were hospitalized with VLRI across 10 children's medical centers. Using the Pediatric Comprehensive Severity Index (PCSI),6–8 we examined the relative contributions of severity versus practice variation on differences in pediatric intensive care unit (PICU) admission rates, intubation, hospital costs, and hospital length of stay (LOS) at these institutions.
The Human Investigation Committees of each of the 10 participating centers approved this study, which was funded by a grant from the Agency for Health Care Policy and Research and directed by International Severity Information Systems, Inc. (ISIS; Salt Lake City, UT).
As part of a larger study of 16 506 pediatric hospital admissions, the hospital records of all children who were younger than 12 months and had the principal diagnoses of bronchiolitis (International Classification of Disease—Ninth Revision [ICD-9] code 466.1) and respiratory syncytial virus (RSV) pneumonia (ICD-9 code 480.1) and were admitted to 10 children's medical centers from April 1, 1995, to September 30, 1996, were reviewed. These 2 ICD-9 codes were combined because we believed that these 2 diagnoses exist on a continuum, they constitute the bulk of pediatric VLRI, and the same severity matrix in the PCSI is used for both diagnoses. We chose children who were younger than 12 months because they are at the greatest risk of severe illness with VLRI. The 10 children's medical centers all were tertiary care hospitals with accredited pediatric residency programs, full-time pediatric faculty, and PICUs staffed by full-time pediatric intensivists.
Comorbidities specifically examined included heart disease, a history of preterm birth, a history of hospitalization for wheezing in the previous 6 months, and a history of hospitalizations for bronchiolitis. Children with bronchopulmonary dysplasia (ICD-9 code 770.7) were excluded.
Severity scoring using the PCSI was done at admission (Admission Comprehensive Severity Index [ACSIC]) and at maximum (Maximum Comprehensive Severity Index [MCSIC]) for the entire hospital course.
All data collection was retrospective. Patient demographic data that included patient name, birth date, gender, admission and discharge dates, PICU admission and discharge dates, discharge disposition, Diagnostic Related Groups code, and International Classification of Diseases—Ninth Revision—Clinical Modification(ICD-9-CM) diagnostic and procedure codes were submitted directly on electronic media to ISIS, Inc, from hospital databases at each site. Fully allocated total costs for each patient, excluding physician's fees, were used, as defined at each site. These were downloaded directly from patient files. Financial representatives from each institution met via telephone conference calls to standardize the types of data sent. Cost outliers (<$4/d and >$13 000/d) and transferred patients were excluded. One site did not record these data in a similar format, and their costs were excluded from the analysis. Information on clinical findings, laboratory results, interventions, and outcomes was collected by trained chart abstractors from the medical records. Every patient's record was reviewed completely from admission to discharge.
All data were entered into a database using a software program developed for this purpose by ISIS, Inc, and subsequently transferred directly into a SAS database (SAS, Inc, Cary, NC) for analysis.
To ensure comparability among chart abstractors, the data collector at each institution first underwent a 3-day training session with instructors from ISIS. For reliability, each chart abstractor sent photocopies of 4 charts along with corresponding clinical summary reports. Photocopied charts then were abstracted independently by a member of the training team responsible for reliability, and the results were compared. Nine of the 10 institutions achieved a 95% or better agreement rate after the first reliability test. After a review session with the data collector from the 10th institution, a second reliability test resulted in reliability scores at or above the desired 95% agreement rate. Reliability testing was repeated when data collection was at least 50% complete. Each data collector performed well, achieving at least a 95% agreement rate.
The PCSI is a modification of the Comprehensive Severity Index (CSI) for adults. CSI was developed by a consensus panel of 150 specialist physicians at Johns Hopkins University Medical School.9–15 In CSI, severity criteria for each diagnosis were selected by the consensus panel on the basis of objective clinical findings, including elements of history, physical examination, and laboratory findings, and rated from 1 to 4 (1 = normal or mild severity, 2 = moderate, 3 = severe, 4 = life-threatening). These then were assembled into diagnosis-specific (ICD-9-CM) severity matrices. Final severity scoring considered all of the patient's diagnoses to obtain an overall patient severity level and was scaled both categorically (from 1 to 4) and continuously (on a scale not subject to any preset maximum). Criteria that were common to more than 1 matrix were scored only once, using the matrix that gave the highest number of points. For both discrete and continuous measures, higher scores indicated higher severity.
CSI was designed to be diagnosis specific and defined severity a priori rather than on statistical regression models constructed to explain the variation in a defined outcome (eg, mortality). Because scoring is (relatively) independent of treatment, it does not require “recalibration” as treatments and/or outcomes change.
The PCSI was adapted from the adult CSI using similar methodology. Pediatric physician specialists from the 10 participating institutions modified the 838 severity matrices from the adult CSI to produce >1400 age- and diagnosis-specific matrices. Pediatric-specific algorithms for weighting principal and secondary diagnoses also were developed. Final computation of the overall PCSI severity score involves the following:
Scoring of each ICD-9-CM diagnosis encoded
Weighting of principal and secondary diagnoses. Secondary diagnoses that are associated directly with either the principal or with other secondary diagnoses are not considered to avoid multiple coding of a single disease process.
Points from severity criteria shared by more than 1 diagnosis are counted only once in computation of the continuous severity score.
Menu-driven microcomputer software that incorporated the matrix structure, algorithms, and logic of the PCSI and integrated this into the existing CSI software system was developed. The specific matrix used for “lower respiratory illness” in this study included criteria such as blood gases, white blood cell counts, radiology findings, pulmonary findings on clinical examination, temperature, and respiratory rate.
Descriptive statistics (mean, standard deviation, percent) are used to describe patient characteristics. Patient, treatment, and outcome variables for children with and without various comorbidities are compared using χ2 and 2-sample ttests. Correlation analyses are used to associate patient variables and severity of illness with costs, LOS, PICU admission, and intubation. Comorbidities, significant clinical and laboratory findings, and severity of illness are combined in regression analyses to determine the concurrent effects on measures of resource utilization (costs, LOS, PICU admission, and intubation).
Correlation and regression analyses were performed with 2 different units of analysis: 1) at the patient level with the patient as the unit of analysis and 2) at the institutional level. The aggregate institutional level analysis is subject to the problem of ecological inference, but we performed it to demonstrate in a succinct way the extensive variation in practice across these 10 sites.
A total of 853 consecutive infants with the ICD-9diagnostic code for bronchiolitis (466.1) or RSV pneumonia (480.1) as principal diagnosis were identified. Excluded were 244 children who were older than 12 months, 4 with bronchopulmonary dysplasia and 4 with incomplete data. Thus, 601 (70%) children met study criteria, 82 with the principal diagnosis of RSV pneumonia (480.1) and 519 with bronchiolitis (466.1). Demographic, comorbidity, and admission findings for the 601 children are shown in Table 1.
There was 1 death. Six children (1%) were discharged to short-term home health care, 4 (0.7%) were discharged to rehabilitative care, and 2 (0.3%) were discharged to chronic nursing care. The average hospital LOS was 4.5 ± 4.2 days, and average hospital costs were $6789 ± 9131.
Correlation of Patient Variables and PCSI Scores With LOS and Costs at the Patient Level
The patient variables that significantly correlated with LOS or costs included a history of prematurity, cardiac disease, rales on admission physical examination, atelectasis/infiltrate on admission chest radiograph, and the diagnosis of RSV pneumonia (ICD-9code 480.1). Regression analysis showed that the correlation of the aggregate of these variables with both LOS and costs wasr = 0.50. Admission PCSI (ACSIC) correlated as well with LOS (r = 0.49) and costs (r = 0.48) as the aggregate of all of the patient variables (r= 0.50 for both). Maximum PCSI (MCSIC) had the strongest correlation with LOS and costs (r = 0.66 for costs andr = 0.64 for LOS). Consequently, although these patient variables were distributed unevenly across institutions (data not shown), either PCSI score (ACSIC or MCSIC) can serve as a reasonable proxy for the patient variables that significantly correlated with LOS or costs and thus are used in the subsequent analyses.
Correlation of Institutional Average PCSI Scores With Institutional Resource Utilization
In contrast to their performance at the patient level, institutional average PCSI scores correlated poorly with institutional average resource utilization (Table 2). Institutional average ACSIC scores correlated negatively with institutional frequency of PICU admission, frequency of intubation, and average costs. Institutional average MCSIC scores correlated negatively with frequency of PICU admission and had poor correlation with frequency of intubation and average costs relative to their correlation at the patient level.
Correlation of Institutional Care Practices With Severity and Resource Utilization
Because differences in average patient severity for each institution did not account for institutional differences in resource utilization, we examined the correlation of institutional care practices with severity. The interventions that demonstrated significant institutional variability are summarized in Table 3. To quantify institutional differences in the intensity of therapy, we ranked the institutions from 1 to 10 with respect to the frequency or duration of use of each of these interventions (with 1 the lowest, 10 the highest, and average ranks used for ties) and added the ranks (Table 4). The resulting rank sums are semiquantitative estimates of the relative intensity of therapy at each institution. The institutional rank sums of interventions were compared with the institutional average admission severity score, institutional average costs, and institutional average LOS. Institutional rank sums of interventions had a negative correlation with institutional average admission severity score but, in contrast, correlated strongly with institutional average costs and institutional average LOS (Table 5).
In a multi-institutional Canadian study, Wang et al3noted significant variation in the frequency of interventions performed for children with VLRI. Although they were unable to adjust for patient variables, they posited that such variation could not be explained solely by differences in disease severity and suggested that “practice preferences” among hospitals must play a role. Our data delineate similar practice variation in American hospitals. In addition, our data demonstrate that intensity of therapy bore little relationship to severity of illness but was a primary determinant of resource utilization, as Wang et al3 had suggested previously.
Such variation is not surprising. Nearly all of the therapies used in VLRI are supportive with unclear indications and indeterminate efficacy. Our study suggests that many are used indiscriminately. Antibiotics were used in 64% of patients despite the accepted viral causation of this disease and convincing data demonstrating that bacterial superinfection is rare.16,,17 Similarly, despite a demonstrated lack of efficacy of corticosteroids in VLRI,18–21 1 institution used them in 61% of patients. The use of inhaled β agonists in 92% of our patients is more difficult to address because the literature is equivocal.22–34 There are no data, however, on the benefit of continuous nebulized bronchodilators, a therapy used in 6% of these patients. Use of other therapies, including ribavirin, chest physiotherapy, furosemide, and invasive monitoring, was similarly idiosyncratic. Greater institutional use of each of the above interventions was associated with higher costs and some with longer LOS (data not shown).
The variable use of intensive care and intensive care interventions merits particular comment, if only because these are so costly. Although admission severity score (ACSIC) correlated with PICU admission at the patient level (r = 0.39), it was actually negatively correlated at the institutional level (r = −0.29). The rationale for PICU admission cannot be discerned from our data, although the fact that most PICU patients (82%) were not intubated suggests that factors other than need for ventilatory support were important. The institutional variation in the frequency of intubation (0%–26%) and the discordance between frequency of intubation and severity across institutions suggest that the threshold for intubation differed at different sites. Use of intravascular catheters was similarly institution dependent. The influence of physician practice preferences on intensive care interventions in lower respiratory illness and the impact on costs and complications have been reported previously.4
Patients with similar severity of illness received very different care at different institutions. The negative correlation of institutional average admission severity with the institutional rank sum of interventions (r = −0.03) suggests that disease severity on admission was not a strong determinant of treatment at the institutional level. The strong correlation of rank sum of intervention with costs and LOS, however, is evidence that greater intensity of care had a direct effect on resource utilization. There is no evidence that greater intensity of care affected morbidity or mortality in view of the almost uniformly good patient outcomes.
Our study demonstrates that institutional “practice preferences” are a major determinant of care for children who are hospitalized with VLRI. These practice preferences seem to have little impact on recovery but have a significant impact on resource utilization. This suggests the need for a more rational and cost-effective approach to the care of these patients.
The treatment of children with VLRI differs widely across children's medical centers. Much of this variation seems to represent institutional or individual physician practice preferences that bear little relationship to severity. Given the self-limited nature of VLRI, as well as the costs and risks of interventions, a more conservative approach to treatment should be considered.
This work was funded by a grant for the Agency for Health Care Policy and Research, Contract No. 290-95-0042.
Collaborating Members of the Rating of Illness Severity in Kids (RISK) Study Group: Douglas F. Willson, MD, University of Virginia Children's Medical Center, Charlottesville, VA; Robert Pettignano, MD, Egleston Children's Hospital, Atlanta, GA; Adalberto Torres, MD, Arkansas Children's Hospital, Little Rock, AR; Ann Thompson, MD, Children's Hospital of Pittsburgh, Pittsburgh, PA; J. Michael Dean, MD, Primary Children's Medical Center, Salt Lake City, UT; Britt Nelson, MD, Cook Children's Medical Center, Fort Worth, TX; Stephen Johnson, MD, Kaiser Permanente Hospital, Los Angeles, CA; Gilbert Goldman, MD, Loyola University Medical Center, Chicago, IL; Robert Gomez, MD, Children's Hospital of the King's Daughters, Norfolk, VA; and David Bergman, MD, Lucille S. Salter Children's Hospital, Palo Alto, CA.
- Received October 24, 2000.
- Accepted February 5, 2001.
Reprint requests to (D.F.W.) Division of Pediatric Critical Care, University of Virginia Children's Medical Center, Box 386, University of Virginia Health Sciences Center, Charlottesville, VA 22908. E-mail:
- VLRI =
- viral lower respiratory illness •
- PCSI =
- Pediatric Comprehensive Severity Index •
- PICU =
- pediatric intensive care unit •
- LOS =
- length of stay •
- ICD-9 =
- International Classification of Diseases—Ninth Revision •
- RSV =
- respiratory syncytial virus •
- ISIS =
- International Severity Information Systems •
- CSI =
- Comprehensive Severity Index •
- ACSIC =
- Admission Comprehensive Severity Index •
- MCSIC =
- Maximum Comprehensive Severity Index •
- ICD-9-CM =
- International Classification of Diseases—Ninth Revision—Clinical Modification
- Wang EEL,
- Law BJ,
- Boucher FD,
- et al.
- Torres A,
- Bergman D,
- Dean JM,
- et al.
- Torres A,
- Bergman D,
- Dean JM,
- et al.
- Horn SD,
- Horn RA
- Andersen BFP,
- Hornsleth EBA,
- Jensen A,
- et al.
- Wang EEL,
- Milner R,
- Allen U,
- Maj H
- Ho L,
- Collis G,
- Landau LI,
- Le Souf PN
- Kristjansson S,
- Lodrup Carlson KC,
- Wennergren G,
- et al.
- Henry RL,
- Milner AD,
- Stokes GM
- Dobson JV,
- Stephens-Groff SM,
- McMahon SR,
- et al.
- Flores G,
- Horwitz RI
- Copyright © 2001 American Academy of Pediatrics