| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Guest Access | Sign In via User Name/Password
|
|
|
|||||||
Guest Access | Sign In via User Name/Password |
|||||||||
U.S. Department of Health and Human Services Kansas City, MO
Correspondence to: Kazim Sheikh, MD, U.S. Department of Health and Human Services, Centers for Medicare & Medicaid Services, 601 E. 12th St, Kansas City, MO 64106; e-mail: ksheikh{at}cms.hhs.gov
To the Editor:
In a study of concordance between the clinical diagnosis and autopsy diagnosis, Dr. Tai and his colleagues1 concluded that knowledge of the correct (clinical) diagnosis would have altered therapy in 44% of the discordant cases and perhaps would have prevented the deaths of the patients.
In this study of patients admitted to a medical intensive care unit (MICU), autopsies were performed on only 91 of the 401 patients (22.3%) who had died in the MICU. Consequently, the clinical diagnoses could be compared with the autopsy diagnoses in this selected subsection of the fatalities. The authors compared the deceased patients who were subjected to autopsy with those who were not with respect to their age, sex, race, APACHE (acute physiology and chronic health evaluation) III score, and the length of stay in MICU. The two groups were similar except for age—the autopsy group was significantly younger. However, none of these factors were determined to predict a correct diagnosis, and the factors that were associated with a correct diagnosis were not known to the authors or they were not included in the analysis. Consequently, the authors could not investigate possible selection bias. The selection bias2 affects the results of a study in the same way as the nonresponse bias does in a survey or in a follow-up observational study, where information on the outcome among the nonrespondents is not known. Bias is strongly suspected if the respondents are not representative of the study population. In other words, if nonrespondents differ from the respondents with respect to the risk factors for the study outcome, the results based on only the respondents would most probably be biased, particularly when the response rate is very low.3 4
In the discussion of Tai et al,1 the authors compared the level of discordance (19.8%) in their study (with autopsy diagnosis for 22.3% of 401 patients) to the level of discordance (30%) in a similar, larger study5 (with autopsy diagnosis for 96% of 400 patients). But, they failed to declare that the difference in the levels of discordance in the two studies was statistically insignificant (p = 0.135), most probably due to the small numbers in the former study. If the autopsy rate in their study had also been 96%, the authors might also have found greater discordance.
The authors considered an association between the discordance in diagnosis and the conditions and procedures that lead to admission into the MICU, but the numbers were too small for any meaningful results. This study should be regarded as inconclusive because of a probable selection bias that could not be investigated and because of the small number of autopsy diagnoses.
Disclaimer: The views expressed in this letter do not represent the views and policies of the Centers for Medicare & Medicaid Services or the United States.
References
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
