CURRENT LITERATURE AND CLINICAL ISSUES
Association of rug shampooing and Kawasaki disease The cause of Kawasaki disease remains unclear.I, 2 In 1982, Patriarca et al. 3 reported the results of an epidemiologic study done in Denver, Col0. During interviews the parents of the first two patients volunteered the information that they had shampooed their rugs during the 30 days before their children became ill. This led the investigators to include this factor in the case-control study. The results indicated that the application of rug shampoo during the month before the onset of the disease was associated with the occurrence of Kawasaki disease. Since then, several contradictory case-control studies have been published.47 The purpose of this report is to evaluate the published casecontrol studies that have investigated the association between Kawasaki disease and rug shampooing for adherence to accepted standards for case-control studies. 8 These standards include (1) the definition of Kawasaki disease case status, (2) the appropriate selection of control subjects, (3) the accurate measurement of exposure to rug shampoo, (4) the evaluation of potential bias that may yield an "unfair" comparison between case subjects and control subjects, and (5) the analysis and interpretation of data. THE STUDIES The case-control study follows a paradigm that proceeds from effect (disease) to cause (exposure). In this design, case subjects are compared with control subjects with respect to an exposure that may be relevant to the disease under study. 8 The case-control method is an appropriate choice to investigate the etiologic hypothesis that use of rug shampoo in a household is associated with the occurrence of Kawasaki disease because it provides the investigator with an indirect method for determining relative risk and allows for efficient sampling of diseased children with a relatively uncommon illness. Five detailed studies of the association between rug shampoo and Kawasaki disease have been published (Table). Two of the studies demonstrated a significant associReprint requests: Stephen R. Daniels, MD, PhD, Divisionof Cardiology, Children's Hospital Medical Center, Elland and Bethesda Avenues, Cincinnati, OH 45229.
9/34/26565
ation between rug shampooing and the occurrence of Kawasaki disease3, 7; the other three found no association.4-6 In addition, Rauch 9 reported the results of seven studies conducted by the Centers for Disease Control that have not been published or were published only in abstract form. In two, parents of patients with Kawasaki disease did report a significantly higher proportion of rug shampooing than did parents of control subjects; in five of these studies. there was no significant association. However. there is insufficient detail in the report by Rauch to permit analysis of the methods of these studies. DEFINITION
OF CASES
The definition of case status was essentially the same for each of the studies: each employed the Centers for Disease Control criteria. 10 There is currently no definitive diagnostic test for Kawasaki disease, so clinical criteria are necessary to determine whether a child has the disease. Cases may be included in the study at the time of diagnosis (incident cases) or determined from a review of medical records (retrospective cases). All the studies except that of Fatiea et al. 7 used incident cases. Fatica et al. gathered retrospective cases from the 6 years before the study. Three of the studies distinguished between cases that occurred as part of an outbreak and others that occurred sporadically.3, 4, 7 It is not clear whether this distinction was useful, because it appears unlikely that there would be different causes for the two presentations. SELECTION
OF CONTROL SUBJECTS
All five studies used a matching procedure based on age, race. and gender. Four sources of control subjects have been used (Table): (1) neighborhood control subjects---children who lived nearby, usually selected by the parents of a case subject; (2) pediatric control subjects--children from the same pediatric practice as the case subjects; (3) clinic control subjects children from a hospital-based pediatric clinic; and (4) random control subjects children found by means of a telephone survey derived from an 11% random sample of birth certificates matched for birth date, gender, and county of residence. None of the studies employed a hospitalized control group.
485
486
Daniels and Speeker
The Journal of Pediatrics March 1991
T a b l e . Studies of rug shampoo and Kawasaki disease Control subjects
Case subjects
Study
Proportion with exposure to rug n shampooing
Season
n
Proportion with exposure to rug shampooing
Patriarca et al. 3 23 (1982)
0.48
April-June
86
0.10
Lin et al. 4 (1985)
36
0.14
Jan.-May
72
0.14
Rogers et al. 5 (1985) Klein et al. 6 (1986)
11
0.64
Oct.-Jan.
22
0.68
15
0.40
Nov.-March 30
0.43
Fatica et al. 7 (1989)
63
0.24
63
0.03
Source of control subjects
Matching Criteria
Neighborhood (n = 30) Age, race, Random (birth certificate) gender (n = 56) Pediatric practice Age, race, (n = 36) gender Neighborhood--selected by parent (n = 36) Pediatric practice Age, race, gender Neighborhood--selected Age, gender by parent (n = 22) Pediatric practice (n = 8) Clinic Age, race, gender
Results RR = 4.6
No association
No association No association
RR = 9.4
RR, Relative risk.
Persons with similar socioeconomic status tend to have similar exposures and may have similar house-cleaning practices, including shampooing of rugs. Although none of the studies investigated the socioeconomic status of the study subjects, the method of control subject selection used in the studies (except that of Fatica et al., 7 who used a clinic population, and Patriarca et al., 3 who used a random control group in part of their analysis) probably resulted in similar socioeconomic status for case subjects and control subjects. This raises the issue of whether the studies' matching procedure was too extensive, making the case subjects and control subjects too similar. The result may be a selection bias against finding a difference in exposure between case subjects and control subjects. The problem of overmatching can be avoided by using multiple control groups from different sources to evaluate potential bias. Two of the studies used multiple control groups that consisted of neighborhood and pediatric practice control subjects.4, 6 However, children from the same neighborhood may attend the same pediatric practice and have similar socioeconomic status. It is therefore likely that the neighborhood and pediatric control groups were similar to each other, as well as similar to the case subjects, with respect to local environmental exposures. MEASUREMENT
OF EXPOSURE
Most of the studies used a telephone interview with the parents of case subjects and control subjects to determine the presence or absence of exposure to rug shampooing.
Klein et al. 6 used both a telephone interview and a personal interview; both were conducted according to a predetermined protocol that did not seem to differ between case and control subjects. Fatica et al. 7 did not present their method for determining exposure in detail, but the method was similar for case and control subjects. Differences in the approach to the parents of case and control subjects, or in the way questions are posed, can be a source of information bias, which refers to the systematic misclassification of subjects with respect to disease or exposure status. For example, if the interviewer is aware of the research hypothesis and a standard protocol is not used, information regarding the exposure may be more actively sought for the case subjects than for the control subjects. The reports do not mention whether the interviewers were unaware of the research hypothesis or the case-control status of the subjects. One type of information bias is referred to as recall bias. It occurs when there are systematic differences in memory of the exposure between case and control subjects. An important factor that may influence memory of the exposure is the timing of the interviews with respect to the application of rug shampoo. This time interval varied among the studies. In the study of Patriarca et al., 3 the case subjects were interviewed within 13 days of the onset of Kawasaki disease. In the study of Fatica et al., 7 this interval varied from 6 months to 4 years or longer. It is more important if there are systematic differences in the time interval between case and control subjects. In most of the studies, the parents of case subjects were asked about exposure to rug shampoo
Volume 1t 8 Number 3
during the 30 or 60 days before the onset of their child's illness. Control parents were asked about exposure during the 30 or 60 days before the interview. Accurate recall can be a problem if there is a long interval between the onset of Kawasaki disease and the interview, as in the study of Fatica et al. 7 This discrepancy may be particularly important for a disease that has a seasonal occurrence and an exposure that may be seasonal. The incidence of Kawasaki disease has a peak occurrence in the spring and fall. It is possible that rug cleaning may also be more common in the spring and fall. In the study of Fatica et al., 7 the control subjects were asked about rug shampooing during the 30 days before the interview, but the interview could occur at any time during the year. Thus there may have been a bias against rug shampoo exposure in the control group in this study, the investigators being more likely to be asking about an exposure during the spring or fall for the case subjects but were contacting control subjects throughout the year. Recall bias may also result when the parents of children who have been sick are more likely to search their memory for exposures than the parents of control subjects, who have not been ill or hospitalized, iv If the potential exposure has been publicized in either the lay press or the medical literature, it is also possible that the case families have read about the exposure or have been asked about it by physicians caring for the child; therefore they may be more likely to recall the exposure than control families would. Rauch 9 suggested that for those patients who had a single known time of exposure, the interval between exposure and disease clustered around 13 to 30 days. He argued that if this clustering is nonrandom, then this finding supports a real association between rug shampooing and Kawasaki disease and suggests a lack of recall bias. If recall bias were a problem, one might expect parents of case patients to recall more recent events during the month before the onset of disease. However, this does not shed light on the memory of exposure in control subjects, whose parents may have forgotten about rug cleaning. It may be helpful for the interview to include specific information that better defines the exposure, such as the usual frequency of rug shampooing or the type of cleaner used, to determine whether there is a difference between the case and control subjects. Rogers et al. 5 included this comparison in their study and extended their comparison to other activities such as wetting a carpet and cleaning draperies or upholstered furniture) Fatica et al. 7 reported on the frequency of rug shampooing practices in the families of case patients but did not compare this factor with that for the control subjects. Other investigators also studied the type of carpeting and specific rug cleaning activity.7 It also may be useful to search for exposure to rug shampoo out-
R u g shampooing and Kawasaki disease
487
side the home, particularly if the children spend a substantial portion of their time in day care. Rogers et al. s collected information from the person who cared for a case or control subject outside the home when the child had spent at least 3 hours once each week in day care) It may also be important to measure the effect of rug shampoo on the local environment to gain insight into possible pathophysiologic mechanisms. Klein et al. 6 included collection of house dust mites but found no differences between case and control subjects. A N A L Y S I S OF DATA AND I N T E R P R E T A T I O N OF R E S U L T S All the studies used an appropriate data analysis strategy for matched case-control studies. However, only Klein et al. 6 used multiple logistic regression techniques that may be useful for investigating potential confounders or interactions or both. "Confounding" refers to the effect of an extraneous variable that partially or wholly accounts for the observed association between the study exposure and disease. Thus the apparent association (or lack of association) between rug shampoo and Kawasaki disease found in a study may actually be due to another (confounding) variable. To be a confounder, an extraneous variable must satisfy both of two conditions: (1) it is a "risk factor" for the disease, and (2) it is associated with, but is not a consequence of, the exposure (rug shampooing).8 If a variable is suspected of being a confounder, it can be accounted for in the design of the study by using a matching procedure or it can be investigated in the data analysis.8 Most of the reported studies matched subjects according to age, race, and gender. Additional potential confounders would include season of disease occurrence and socioeconomic status; these known risk factors for Kawasaki disease may be associated with rug shampooing and could account for an observed relationship. Other variables may be associated with rug shampoo use, but their status as a risk factor for Kawasaki disease is not completely known. These would include the presence of pets in the home, the number of children in the home, and the geographic area of residence. For studies in which no association is detected, it is necessary to consider the power of that study to detect an association. The determinants of power include the sample size and the baseline frequency of rug shampooing in the control group. The study of Lin et al. 4 reported a power of 0.79, the study of Rogers et al. 5 had a power of 0.18, and the study of Klein et al. 6 had a power of 0.41 to detect a relative risk as strong as 4.6, which was the odds ratio reported in the study of Patriarca et a l ) As can be seen in the Table,
488
Daniels and Specker
the proportion of control subjects exposed to rug shampoo varied widely from one study to another. The question of whether rug shampoo application is associated with the occurrence of Kawasaki disease remains unanswered, in part because of the methodologic limitations of some of the published studies. Also missing is a clear biologic understanding of the proposed exposure-disease relationship. The possibilities include a chemical exposure, an insect vector (e.g., dust mites), or a viral infection. The association between the disease and cleaning activities does not appear to be limited to a single type of detergent; this would make a chemical exposure less likely. At least two studies have found no association between the presence or number of dust mites and disease. 9 The potential connection between rug shampooing and viral infection is unclear. The acceptance of rug shampooing as a causal factor in the development of Kawasaki disease will require investigators to meet a number of operational criteria: a consistently documented strong association in a variety of study settings, a clear temporal relationship between rug shampooing and the subsequent development of Kawasaki disease, demonstration of the biologic plausibility of the relationship, and coherence with the current understanding of the pathogenesis of the disease. A dose-response relationship would further support a causal relationship.13 In the meantime, there is inadequate evidence on which to base any sound clinical or public health recommendations. Future research on this question should (1) focus on incident cases and justify division of cases into epidemic and sporadic groups, (2) use multiple control groups, including a group hospitalized with other acute illnesses similar in severity to Kawasaki disease, (3) include a sufficiently large sample size to ensure adequate power to detect an association if it is present, (4) obtain a valid, detailed assessment of exposure to rug shampooing, (5) minimize recall bias by ensuring that the time of obtaining information on the exposure is similar for case and control subjects, (6) reduce potential interviewer bias by using standardized questionnaires and interviewers who are unaware of either the
The Journal of Pediatrics March 199l research hypothesis or the case or control status of the subject, and (7) include an appropriate assessment of potential confounding.
Stephen R. Daniels, MD, PhD* Bonny Specker, PhD Divisions of Cardiology and Neonatology Department of Pediatrics University of Cincinnati College of Medicine Children's Hospital Medical Center Cincinnati, OH 45229 REFERENCES
1. Kawasaki T, Kasaki F, Okawa S, et ah A new infantile acute febrile mucocutaneous lymph node syndrome (MLNS) prevailing in Japan. Pediatrics 1974;54:271-6. 2. Bell DM. Kawasaki syndrome: still a mystery after 20 years [Editorial]. JAMA 1985;254:801. 3. Patriarca PA, Rogers MF, Morens DM, et al. Kawasaki syndrome: association with the application of rug shampoo. Lancet 1982;2:578-80. 4. Lin FC, Bailowitz A, Koslowe P, Israel E, Kaslow RA. Kawasaki syndrome: a case-control study during an outbreak in Maryland. Am J Dis Child 1985;139:277-9. 5. Rogers MF, Kochel RL, Hurwitz ES. Kawasaki syndrome: is rug shampoo important? Am J Dis Child 1985;139:777-9. 6. Klein BS, Rogers MF, Patrican LA, et al. Kawasaki syndrome: a controlled study of an outbreak in Wisconsin. Am J Epidemiol 1986;124:306-16. 7. Fatica NS' Ichida F' Engle MA' Lesser ML' Rug shamp~176 and Kawasaki disease. Pediatrics 1989;84:231-4. 8. Schlesselman JJ. Case-control studies: design, conduct, analysis. New York: Oxford University Press, 1982. 9. Rauch AM. Kawasaki syndrome:reviewof new epidemiologic and laboratory developments.Pediatr Infect Dis J 1987;101621. 10. Melish ME. Kawasaki syndrome (mucocutaneouslymph node syndrome). Pediatr Rev 1980;1:107-14. 11. Sackett DL. Bias in analytic research. J Chronic Dis 1979; 32:51-68. 12. Hill AB. Principles of medical statistics. 9th ed. New York: Oxford University Press, 1971:309-23. *Recipient of an Established Investigator Award from the American Heart Association.
Association of rug shampooing and Kawasaki disease The cause of Kawasaki disease remains unclear.I, 2 In 1982, Patriarca et al. 3 reported the results of an epidemiologic study done in Denver, Col0. During interviews the parents of the first two patients volunteered the information that they had shampooed their rugs during the 30 days before their children became ill. This led the investigators to include this factor in the case-control study. The results indicated that the application of rug shampoo during the month before the onset of the disease was associated with the occurrence of Kawasaki disease. Since then, several contradictory case-control studies have been published.47 The purpose of this report is to evaluate the published casecontrol studies that have investigated the association between Kawasaki disease and rug shampooing for adherence to accepted standards for case-control studies. 8 These standards include (1) the definition of Kawasaki disease case status, (2) the appropriate selection of control subjects, (3) the accurate measurement of exposure to rug shampoo, (4) the evaluation of potential bias that may yield an "unfair" comparison between case subjects and control subjects, and (5) the analysis and interpretation of data. THE STUDIES The case-control study follows a paradigm that proceeds from effect (disease) to cause (exposure). In this design, case subjects are compared with control subjects with respect to an exposure that may be relevant to the disease under study. 8 The case-control method is an appropriate choice to investigate the etiologic hypothesis that use of rug shampoo in a household is associated with the occurrence of Kawasaki disease because it provides the investigator with an indirect method for determining relative risk and allows for efficient sampling of diseased children with a relatively uncommon illness. Five detailed studies of the association between rug shampoo and Kawasaki disease have been published (Table). Two of the studies demonstrated a significant associReprint requests: Stephen R. Daniels, MD, PhD, Divisionof Cardiology, Children's Hospital Medical Center, Elland and Bethesda Avenues, Cincinnati, OH 45229.
9/34/26565
ation between rug shampooing and the occurrence of Kawasaki disease3, 7; the other three found no association.4-6 In addition, Rauch 9 reported the results of seven studies conducted by the Centers for Disease Control that have not been published or were published only in abstract form. In two, parents of patients with Kawasaki disease did report a significantly higher proportion of rug shampooing than did parents of control subjects; in five of these studies. there was no significant association. However. there is insufficient detail in the report by Rauch to permit analysis of the methods of these studies. DEFINITION
OF CASES
The definition of case status was essentially the same for each of the studies: each employed the Centers for Disease Control criteria. 10 There is currently no definitive diagnostic test for Kawasaki disease, so clinical criteria are necessary to determine whether a child has the disease. Cases may be included in the study at the time of diagnosis (incident cases) or determined from a review of medical records (retrospective cases). All the studies except that of Fatiea et al. 7 used incident cases. Fatica et al. gathered retrospective cases from the 6 years before the study. Three of the studies distinguished between cases that occurred as part of an outbreak and others that occurred sporadically.3, 4, 7 It is not clear whether this distinction was useful, because it appears unlikely that there would be different causes for the two presentations. SELECTION
OF CONTROL SUBJECTS
All five studies used a matching procedure based on age, race. and gender. Four sources of control subjects have been used (Table): (1) neighborhood control subjects---children who lived nearby, usually selected by the parents of a case subject; (2) pediatric control subjects--children from the same pediatric practice as the case subjects; (3) clinic control subjects children from a hospital-based pediatric clinic; and (4) random control subjects children found by means of a telephone survey derived from an 11% random sample of birth certificates matched for birth date, gender, and county of residence. None of the studies employed a hospitalized control group.
485
486
Daniels and Speeker
The Journal of Pediatrics March 1991
T a b l e . Studies of rug shampoo and Kawasaki disease Control subjects
Case subjects
Study
Proportion with exposure to rug n shampooing
Season
n
Proportion with exposure to rug shampooing
Patriarca et al. 3 23 (1982)
0.48
April-June
86
0.10
Lin et al. 4 (1985)
36
0.14
Jan.-May
72
0.14
Rogers et al. 5 (1985) Klein et al. 6 (1986)
11
0.64
Oct.-Jan.
22
0.68
15
0.40
Nov.-March 30
0.43
Fatica et al. 7 (1989)
63
0.24
63
0.03
Source of control subjects
Matching Criteria
Neighborhood (n = 30) Age, race, Random (birth certificate) gender (n = 56) Pediatric practice Age, race, (n = 36) gender Neighborhood--selected by parent (n = 36) Pediatric practice Age, race, gender Neighborhood--selected Age, gender by parent (n = 22) Pediatric practice (n = 8) Clinic Age, race, gender
Results RR = 4.6
No association
No association No association
RR = 9.4
RR, Relative risk.
Persons with similar socioeconomic status tend to have similar exposures and may have similar house-cleaning practices, including shampooing of rugs. Although none of the studies investigated the socioeconomic status of the study subjects, the method of control subject selection used in the studies (except that of Fatica et al., 7 who used a clinic population, and Patriarca et al., 3 who used a random control group in part of their analysis) probably resulted in similar socioeconomic status for case subjects and control subjects. This raises the issue of whether the studies' matching procedure was too extensive, making the case subjects and control subjects too similar. The result may be a selection bias against finding a difference in exposure between case subjects and control subjects. The problem of overmatching can be avoided by using multiple control groups from different sources to evaluate potential bias. Two of the studies used multiple control groups that consisted of neighborhood and pediatric practice control subjects.4, 6 However, children from the same neighborhood may attend the same pediatric practice and have similar socioeconomic status. It is therefore likely that the neighborhood and pediatric control groups were similar to each other, as well as similar to the case subjects, with respect to local environmental exposures. MEASUREMENT
OF EXPOSURE
Most of the studies used a telephone interview with the parents of case subjects and control subjects to determine the presence or absence of exposure to rug shampooing.
Klein et al. 6 used both a telephone interview and a personal interview; both were conducted according to a predetermined protocol that did not seem to differ between case and control subjects. Fatica et al. 7 did not present their method for determining exposure in detail, but the method was similar for case and control subjects. Differences in the approach to the parents of case and control subjects, or in the way questions are posed, can be a source of information bias, which refers to the systematic misclassification of subjects with respect to disease or exposure status. For example, if the interviewer is aware of the research hypothesis and a standard protocol is not used, information regarding the exposure may be more actively sought for the case subjects than for the control subjects. The reports do not mention whether the interviewers were unaware of the research hypothesis or the case-control status of the subjects. One type of information bias is referred to as recall bias. It occurs when there are systematic differences in memory of the exposure between case and control subjects. An important factor that may influence memory of the exposure is the timing of the interviews with respect to the application of rug shampoo. This time interval varied among the studies. In the study of Patriarca et al., 3 the case subjects were interviewed within 13 days of the onset of Kawasaki disease. In the study of Fatica et al., 7 this interval varied from 6 months to 4 years or longer. It is more important if there are systematic differences in the time interval between case and control subjects. In most of the studies, the parents of case subjects were asked about exposure to rug shampoo
Volume 1t 8 Number 3
during the 30 or 60 days before the onset of their child's illness. Control parents were asked about exposure during the 30 or 60 days before the interview. Accurate recall can be a problem if there is a long interval between the onset of Kawasaki disease and the interview, as in the study of Fatica et al. 7 This discrepancy may be particularly important for a disease that has a seasonal occurrence and an exposure that may be seasonal. The incidence of Kawasaki disease has a peak occurrence in the spring and fall. It is possible that rug cleaning may also be more common in the spring and fall. In the study of Fatica et al., 7 the control subjects were asked about rug shampooing during the 30 days before the interview, but the interview could occur at any time during the year. Thus there may have been a bias against rug shampoo exposure in the control group in this study, the investigators being more likely to be asking about an exposure during the spring or fall for the case subjects but were contacting control subjects throughout the year. Recall bias may also result when the parents of children who have been sick are more likely to search their memory for exposures than the parents of control subjects, who have not been ill or hospitalized, iv If the potential exposure has been publicized in either the lay press or the medical literature, it is also possible that the case families have read about the exposure or have been asked about it by physicians caring for the child; therefore they may be more likely to recall the exposure than control families would. Rauch 9 suggested that for those patients who had a single known time of exposure, the interval between exposure and disease clustered around 13 to 30 days. He argued that if this clustering is nonrandom, then this finding supports a real association between rug shampooing and Kawasaki disease and suggests a lack of recall bias. If recall bias were a problem, one might expect parents of case patients to recall more recent events during the month before the onset of disease. However, this does not shed light on the memory of exposure in control subjects, whose parents may have forgotten about rug cleaning. It may be helpful for the interview to include specific information that better defines the exposure, such as the usual frequency of rug shampooing or the type of cleaner used, to determine whether there is a difference between the case and control subjects. Rogers et al. 5 included this comparison in their study and extended their comparison to other activities such as wetting a carpet and cleaning draperies or upholstered furniture) Fatica et al. 7 reported on the frequency of rug shampooing practices in the families of case patients but did not compare this factor with that for the control subjects. Other investigators also studied the type of carpeting and specific rug cleaning activity.7 It also may be useful to search for exposure to rug shampoo out-
R u g shampooing and Kawasaki disease
487
side the home, particularly if the children spend a substantial portion of their time in day care. Rogers et al. s collected information from the person who cared for a case or control subject outside the home when the child had spent at least 3 hours once each week in day care) It may also be important to measure the effect of rug shampoo on the local environment to gain insight into possible pathophysiologic mechanisms. Klein et al. 6 included collection of house dust mites but found no differences between case and control subjects. A N A L Y S I S OF DATA AND I N T E R P R E T A T I O N OF R E S U L T S All the studies used an appropriate data analysis strategy for matched case-control studies. However, only Klein et al. 6 used multiple logistic regression techniques that may be useful for investigating potential confounders or interactions or both. "Confounding" refers to the effect of an extraneous variable that partially or wholly accounts for the observed association between the study exposure and disease. Thus the apparent association (or lack of association) between rug shampoo and Kawasaki disease found in a study may actually be due to another (confounding) variable. To be a confounder, an extraneous variable must satisfy both of two conditions: (1) it is a "risk factor" for the disease, and (2) it is associated with, but is not a consequence of, the exposure (rug shampooing).8 If a variable is suspected of being a confounder, it can be accounted for in the design of the study by using a matching procedure or it can be investigated in the data analysis.8 Most of the reported studies matched subjects according to age, race, and gender. Additional potential confounders would include season of disease occurrence and socioeconomic status; these known risk factors for Kawasaki disease may be associated with rug shampooing and could account for an observed relationship. Other variables may be associated with rug shampoo use, but their status as a risk factor for Kawasaki disease is not completely known. These would include the presence of pets in the home, the number of children in the home, and the geographic area of residence. For studies in which no association is detected, it is necessary to consider the power of that study to detect an association. The determinants of power include the sample size and the baseline frequency of rug shampooing in the control group. The study of Lin et al. 4 reported a power of 0.79, the study of Rogers et al. 5 had a power of 0.18, and the study of Klein et al. 6 had a power of 0.41 to detect a relative risk as strong as 4.6, which was the odds ratio reported in the study of Patriarca et a l ) As can be seen in the Table,
488
Daniels and Specker
the proportion of control subjects exposed to rug shampoo varied widely from one study to another. The question of whether rug shampoo application is associated with the occurrence of Kawasaki disease remains unanswered, in part because of the methodologic limitations of some of the published studies. Also missing is a clear biologic understanding of the proposed exposure-disease relationship. The possibilities include a chemical exposure, an insect vector (e.g., dust mites), or a viral infection. The association between the disease and cleaning activities does not appear to be limited to a single type of detergent; this would make a chemical exposure less likely. At least two studies have found no association between the presence or number of dust mites and disease. 9 The potential connection between rug shampooing and viral infection is unclear. The acceptance of rug shampooing as a causal factor in the development of Kawasaki disease will require investigators to meet a number of operational criteria: a consistently documented strong association in a variety of study settings, a clear temporal relationship between rug shampooing and the subsequent development of Kawasaki disease, demonstration of the biologic plausibility of the relationship, and coherence with the current understanding of the pathogenesis of the disease. A dose-response relationship would further support a causal relationship.13 In the meantime, there is inadequate evidence on which to base any sound clinical or public health recommendations. Future research on this question should (1) focus on incident cases and justify division of cases into epidemic and sporadic groups, (2) use multiple control groups, including a group hospitalized with other acute illnesses similar in severity to Kawasaki disease, (3) include a sufficiently large sample size to ensure adequate power to detect an association if it is present, (4) obtain a valid, detailed assessment of exposure to rug shampooing, (5) minimize recall bias by ensuring that the time of obtaining information on the exposure is similar for case and control subjects, (6) reduce potential interviewer bias by using standardized questionnaires and interviewers who are unaware of either the
The Journal of Pediatrics March 199l research hypothesis or the case or control status of the subject, and (7) include an appropriate assessment of potential confounding.
Stephen R. Daniels, MD, PhD* Bonny Specker, PhD Divisions of Cardiology and Neonatology Department of Pediatrics University of Cincinnati College of Medicine Children's Hospital Medical Center Cincinnati, OH 45229 REFERENCES
1. Kawasaki T, Kasaki F, Okawa S, et ah A new infantile acute febrile mucocutaneous lymph node syndrome (MLNS) prevailing in Japan. Pediatrics 1974;54:271-6. 2. Bell DM. Kawasaki syndrome: still a mystery after 20 years [Editorial]. JAMA 1985;254:801. 3. Patriarca PA, Rogers MF, Morens DM, et al. Kawasaki syndrome: association with the application of rug shampoo. Lancet 1982;2:578-80. 4. Lin FC, Bailowitz A, Koslowe P, Israel E, Kaslow RA. Kawasaki syndrome: a case-control study during an outbreak in Maryland. Am J Dis Child 1985;139:277-9. 5. Rogers MF, Kochel RL, Hurwitz ES. Kawasaki syndrome: is rug shampoo important? Am J Dis Child 1985;139:777-9. 6. Klein BS, Rogers MF, Patrican LA, et al. Kawasaki syndrome: a controlled study of an outbreak in Wisconsin. Am J Epidemiol 1986;124:306-16. 7. Fatica NS' Ichida F' Engle MA' Lesser ML' Rug shamp~176 and Kawasaki disease. Pediatrics 1989;84:231-4. 8. Schlesselman JJ. Case-control studies: design, conduct, analysis. New York: Oxford University Press, 1982. 9. Rauch AM. Kawasaki syndrome:reviewof new epidemiologic and laboratory developments.Pediatr Infect Dis J 1987;101621. 10. Melish ME. Kawasaki syndrome (mucocutaneouslymph node syndrome). Pediatr Rev 1980;1:107-14. 11. Sackett DL. Bias in analytic research. J Chronic Dis 1979; 32:51-68. 12. Hill AB. Principles of medical statistics. 9th ed. New York: Oxford University Press, 1971:309-23. *Recipient of an Established Investigator Award from the American Heart Association.
