Published online October 2, 2006
PEDIATRICS
Vol. 118
No. 4
October 2006, pp.
1798
(doi:10.1542/peds.2006-1775)
Predictors of Fatality in Hemolytic Uremic Syndrome Lack Numbers
Michael Hawkes, MD, BEng, FRCPC
Abdulrahman Al-Odayani, MD
Division of Infectious Diseases
Hospital for Sick Children
Toronto, Ontario, Canada M5G 1X8
To the Editor.—
As an exercise in evidence-based medicine, fellows in the clinical training program at our institution dissect a current article at a monthly journal-club meeting. We recently reviewed the case-control study on predictors of death in hemolytic uremic syndrome by Oakes et al.1 We have the following concerns regarding a conceptual error in the statistical analysis.
Oakes et al described the causes of death in children who presented with hemolytic uremic syndrome over a 30-year period in Utah. Using information from a registry of 358 patients, they used a case-control design to determine the effect of prognostic variables, both clinical (bloody diarrhea, oliguria, anuria, fever, seizures, coma, edema, vomiting, dehydration, and lethargy) and laboratory parameters (hematocrit, white blood cell count), on the outcome of interest: death. They used both univariate and multivariate analyses to identify key prognostic variables. They presented the results of the univariate analysis as a "relative risk"; for example, fatal cases were associated with a relative risk of elevated hematocrit of 7.58 (P = .00045). Given the study design, this analysis is backward: the risk of the prognostic factor (elevated hematocrit) is given as a function of the outcome (death)! A more conventional approach for a case-control study, which provides more intuitive and useful information, would be to use an odds ratio, which approximates the relative risk of the outcome (death) given the presence or absence of the prognostic variable (elevated hematocrit). Using the data presented, the odds ratio of death for a patient with an elevated hematocrit is 16.8 (P = .00002, Fisher's exact test), meaning that the odds of dying are 
17-fold higher in a patient with an elevated hematocrit. We view this as a fundamental, conceptual error, highlighting the importance of a clear distinction between outcome variables and prognostic variables and the utility of the odds ratio in approaching the analysis of a case-control study on prognosis. The error provided an excellent platform for discussion at our educational journal club!
Second, the authors used conditional logistic regression to determine which variables were independently associated with death. Only an elevated white blood cell count and an elevated hematocrit were independent predictors, yet the authors' conclusions include oligoanuria and dehydration as important prognostic variables. We feel that the analysis supports only the elevated white blood cell count and the elevated hematocrit as significant prognostic variables.
We also noted incidental discrepancies between the text and Table 3 and request clarification from the authors: P value for edema (text: P = .035; table: P = 0.35); percent lethargy among nonfatal cases (text: 49.1%; table: 66.1%); P value for lethargy (text: P = .016; table: P = 0.3); and seizures listed in the table as significant (P < .05) variable but not in text.
REFERENCES
- Oakes RS, Siegler RL, McReynolds MA, Pysher T, Pavia AT. Predictors of fatality in postdiarrheal hemolytic uremic syndrome.
Pediatrics. 2006;117
:1656
–1662[Abstract/Free Full Text]
- Laupacis A, Wells G, Richardson WS, Tugwell P. Users' guides to the medical literature: V. How to use an article about prognosis. Evidence-Based Medicine Working Group.
JAMA. 1994;272
:234
–237[CrossRef][ISI][Medline]
PEDIATRICS (ISSN 1098-4275). ©2006 by the American Academy of Pediatrics
