To the Editor.—
In a recent article by Chang,1 hospitalization records extracted from the Nationwide Inpatient Sample (NIS) were used to describe the epidemiology of Kawasaki syndrome (KS) during 1988–1997 in the United States. Chang reported that the incidence of KS increased over the 10-year study period. We question the accuracy of these findings, because the methods used for analyzing the NIS data were inappropriate.
The NIS is an annual stratified probability sample of hospitals from participating statewide data organizations designed to approximate a 20% sample of US community hospitals. It is produced as part of the Healthcare Cost and Utilization Project (HCUP), sponsored by the Agency for Healthcare Research and Quality in partnership with the states.2 Because the NIS is a complex stratified sample of hospitals, appropriately scaled discharge weights are provided with the database to obtain national estimates of the number of hospitalizations for a given year.2,3 However, Chang did not apply these discharge weights to obtain estimates that take into account the design of the database.1 The use of discharge weights to obtain national estimates is particularly important because the number of states, hospitals, and hospital discharges increased in the NIS during 1988–1997.3
We analyzed the NIS database by using appropriate weighting and analytic methods to estimate annual KS hospitalizations and rates for children <5 years of age for 1988 through 1997 (Table 1). The estimated number of annual hospitalizations ranged from 1997 in 1988 to 3440 in 1991. The annual estimated KS hospitalization rate fluctuated, but did not show an increase during the study period. Because of the limited number of states participating in the NIS during 1988 through 1992, particularly in 1988 with only 87 participating states, estimates of rare conditions in special populations (eg, children) in the earlier years of NIS data may be less reliable than those generated in later years.3 Furthermore, the 1997 KS hospitalization rate was likely affected by the inclusion, for the first time, of children discharges from Hawaii, the state with the highest incidence of KS in the United States.4
The unit of analysis for the NIS database is hospitalizations rather than patients.2,3 Multiple hospitalizations for the same patient have been reported to account for about 10% of KS hospitalizations.5 Chang’s article states that patients with KS in the NIS were identified by the unique sequence number and only the first hospitalization was included in the analysis.1 However, the unique sequence number is a yearly “dummy count variable” and does not enable identification of multiple hospitalizations for a given patient.2,3 The NIS database has no unique patient identifier.
The NIS is a valuable tool for describing national hospitalizations when appropriate analysis and weighting techniques are used.2,3 Appropriate analysis of the NIS data does not support Chang’s report that the incidence of KS in the United States increased over the 10-year study period. Our conclusion is based on analytical methods recommended by HCUP. Failure to use such methods in Chang’s study led to inaccurate results and misinterpretations of the data.
- ↵Chang RR. Hospitalization for Kawasaki disease among children in the United States, 1988–1997. Pediatrics.2002;109(6) . Available at: http://www.pediatrics.org/cgi/content/full/109/e87
- ↵Healthcare and Cost Utilization Project (HCUP). 1997 Nationwide Inpatient Sample (NIS) Technical Documentation. Rockville, MD; Agency for Healthcare Research and Quality, 2000
- Copyright © 2003 by the American Academy of Pediatrics
The comments from Holman et al correctly point out the methodological errors in my study.1 The incidence rates shown in Figure 2 of the article were calculated by extrapolating the number of hospitalizations without using the weighting techniques recommended by the HCUP.2 In addition, the mention of the “unique sequence number” in the paper was a misstatement. In the NIS database, each unique sequence number identifies a hospitalization, not a patient. Therefore, the procedure of using the sequence numbers eliminated duplicating records, not multiple hospitalizations of the same patients. It is possible that multiple hospitalizations may affect the accuracy of the estimates of Kawasaki syndrome (KS) incidence.
The results of KS incidence calculation reported by Holman et al (Table 1) serve as corrections for the miscalculated KS incidences in my paper. I appreciate the valuable comments from Dr Holman and his colleagues and regret the errors made in my paper.
- ↵Chang RR. Hospitalizations for Kawasaki disease among children in the United States, 1988–1997. Pediatrics2002;109(6) . Available at: http://www.pediatrics.org/cgi/content/full/109/6/e87
- ↵Healthcare and Cost Utilization Project (HCUP). 1997 Nationwide Inpatient Sample (NIS) Technical Documentation. Rockville, MD: Agency for Healthcare Research and Quality: 2000