PEDIATRICS Vol. 119 No. 1 January 2007, pp. 219a-220 (doi:10.1542/peds.2006-2659)
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
LETTER TO THE EDITOR |
Predictive Model for Lyme Meningitis: A Reply
Robert A. Avery, DO, MA11 Division of Neurology
Children's Hospital of Philadelphia
Philadelphia, PA 19104
Gary Frank, MD, MS2
2 Department of Pediatrics
Children's Healthcare of Atlanta
Atlanta, GA 30342
Stephen C. Eppes, MD3
3 Division of Infectious Diseases
A.I. duPont Hospital for Children
Wilmington, DE 19899
To the Editor.—
We appreciate Porwancher's interest1 in our article, which described the Lyme meningitis (LM)–prediction model.2 Despite his methodologic concerns, a prospective validation study is still warranted because the distinction between aseptic and LM remains difficult and can result in unnecessary antibiotic therapy.
Patients with erythema migrans (EM) were included in our study because there are no studies demonstrating that those with or without this pathognomonic finding have a different clinical course in LM. To exclude patients with EM is not only illogical, but it would also create a selection bias. Although we rely on the visual diagnosis of EM, additional support of a diagnosis using our prediction model may be helpful, because some clinicians may not feel comfortable committing a child to long-term parental antibiotics on the basis of EM alone.
The concern over misclassification of patients without EM using laboratory criteria only applies to approximately one third of our study population. Over half of the patients without EM had negative cerebrospinal fluid (CSF) viral study results (culture or polymerase chain reaction) or the presence of cranial neuritis, which made the chance of misclassification unlikely. Our experience with commercially available polymerase chain reaction for Borrelia burgdorferi DNA in CSF3 and intrathecal antibody tests for B burgdorferi have been unhelpful and are no longer routinely ordered in our hospitals. The concern that previous infection with B burgdorferi would have caused misclassification of aseptic meningitis cases as current Lyme cases is more theoretical than practical. When there are clinically compatible physical and CSF findings associated with positive serum antibody testing, the overwhelming likelihood is that there is a causal association; this is especially true in children in whom (a) there is a smaller lifetime opportunity to have had subclinical infection with B burgdorferi and (b) there are fewer disease states that could cause aseptic CSF pleocytosis, compared with adults.
Porwancher argues that we used incorporation bias in our model because we included predictors that are part of the Centers for Disease Control and Prevention case definition. However, lymphocytic meningitis and headache are certainly not found exclusively in LM and occur regularly in aseptic meningitis, thereby making the distinction between them difficult. In attempting to discriminate between 2 diseases, shared symptoms are commonly incorporated into predictive models.4,5
It is unlikely that a significant portion of our subjects had pretreated bacterial meningitis and were misclassified as having aseptic meningitis, thereby artificially decreasing the average number of CSF mononuclear cells. Nearly one half of the patients classified as having aseptic meningitis had a positive viral CSF culture or CSF polymerase chain reaction result for enterovirus DNA. None of the patients with negative viral studies were treated with a full course of intravenous antibiotics for possible bacterial meningitis by their attending physicians.
Despite eliminating 10 subjects from the aseptic-meningitis group for the second logistic regression analysis, both analyses reached an identical level of significance when compared with a constant-only model and revealed similar levels for the Hosmer-Lemeshow test and Nagelkerke R2 effect size.
A previously unpublished letter to the editor identified an accidentally omitted minus sign from the beginning of the exponent in the denominator of the predicted probability formula. However, the data presented in our study were calculated correctly.
We agree with Porwancher that clinicians need a more accurate method of diagnosing Lyme disease. Despite our previously acknowledged study limitations, a prospective validation study of our prediction model is still warranted and, we hope, would provide evidence-based decision-making that would serve to decrease unnecessary diagnostic testing and antibiotic use.
REFERENCES
1. Porwancher R. Predictive model for Lyme meningitis [letter].
Pediatrics. 2006;118
:438
–439
2. Avery RA, Frank G, Glutting JJ, Eppes SC. Prediction of Lyme meningitis in children from a Lyme disease–endemic region: a logistic-regression model using history, physical, and laboratory findings. Pediatrics. 2006;117(1) . Available at: www.pediatrics.org/cgi/content/full/117/1/e1
3. Avery RA, Frank G, Eppes SC. Diagnostic utility of Borrelia burgdorferi cerebrospinal fluid polymerase chain reaction in children with Lyme meningitis. Pediatr Infect Dis J. 2005;24 :705 –708[CrossRef][Web of Science][Medline]
4. Friedman MJ, Attia MW. Clinical predictors of influenza in children.
Arch Pediatr Adolesc Med. 2004;158
:391
–394
5. Nigrovic LE, Kuppermann N, Malley R. Development and validation of a multivariable predictive model to distinguish bacterial from aseptic meningitis in children in the post–Haemophilus influenzae era.
Pediatrics. 2002;110
:712
–719
PEDIATRICS (ISSN 1098-4275). ©2007 by the American Academy of Pediatrics
CiteULike 
Connotea 
Del.icio.us 
Digg 
Facebook 
Reddit 
Technorati 
Twitter What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
