Neurology -- Correspondence for Miller et al., 58 (4) 542-548

Correspondence published:

Reply to Letter to the Editor: Steven Miller, Anthony James Barkovich (2 May 2002)

Seizure-associated brain injury in term newborns with perinatal asphyxia: Craig Campbell, George Wells and Pierre Jacob (2 May 2002)

Reply to Letter to the Editor 2 May 2002

Steven Miller
University of California at San Francisco,
Anthony James Barkovich
Send Correspondence to journal:
Re: Reply to Letter to the Editor

smiller{at}itsa.ucsf.edu Steven Miller, et al.

As Campbell et al. point out, we developed a score to estimate the severity of seizures in the neonatal period. [1] Unfortunately there is no single concrete measure of neonatal seizure "severity" published and accepted in the literature. The aim of this study was quite clearly not to discuss the "design, application and analytical issues" related to the score. Campbell et al. suggests that because of this, this simple clinical score of seizure severity is not valid. We do not feel that this is correct. Our understanding of validity is "how well the measurements represents the phenomenon of interest", that is the severity of seizures in the newborn. [2] Neonatal neurologists, expert in the diagnosis, management and study of neonatal seizures, concluded that a score incorporating seizure frequency and onset, neonatal EEG abnormalities, and the number of anticonvulsant medications used had "face validity", a subjective assessment that the score is reasonable. The literature and our experience caring for newborns with seizures, indicate that in the context of perinatal depression there is no evidence for a relationship between "seizure type" and brain injury or neurologic outcome. Furthermore, those caring for newborns with seizures would agree that the duration of an individual seizure is difficult, if not impossible, to estimate. Campbell et al. questions why newborns without seizures were scored as zero. At our institution, depressed newborns without seizures are not routinely studied with EEG or sedated with anti-convulsants. On practical and conceptual levels it is unclear what meaningful information this would add to the newborns without seizures. In the absence of a gold standard, no clinical score is perfect. We chose those variables for the score that incorporated the most important aspects of the phenomenon under study while minimizing the potential overlap in information provided by each component of the score.
The seizure score we developed has the advantage of using information that is readily available in the charts of newborns with seizures. The seizure score is also reliable. A child neurologist without knowledge of the initial seizure score reviewed ten randomly selected charts from this cohort. There was perfect agreement between both independent evaluators (Kappa=1.0, p<0.001).
Campbell et al. also raises concerns regarding model building. Our logistic regression models meet the statistical assumptions required for this analysis. Furthermore, we limited our statistical testing to our a priori hypothesis that the seizure score is associated with the relative concentrations of lactate and N-acetylaspartate.
The most convincing and substantive measure of validity is criterion validity, the "degree to which a measurement correlates with an external criterion of the phenomenon under study". [2] We feel that brain injury is the external criterion of most importance in neonatal seizures. Thus, if the goal of our paper had in fact been to validate our seizure score as a measure of seizure severity, we would have succeeded in demonstrating criterion validity. The severity of seizures as measured by our seizure score, in term newborns with perinatal depression, was associated with brain injury as measured by proton MR spectroscopy.
References:
1) Miller SP, Weiss BS, Barnwell A et al. Seizure-associated brain injury in term newborns with perinatal asphyxia. Neurology 2002;58:542- 548.
2) Hulley SB, Martin JN, Cummings SR. Planning the Measurements: Precision and Accuracy. In: Hulley SB, Cummings SR, Browner WS, Grady D, Hearst N, Newman TB. Designing Clinical Research. Second Edition. Philadelphia: Lippincott Williams and Wilkins, 2001.

Seizure-associated brain injury in term newborns with perinatal asphyxia 2 May 2002

Craig Campbell
Children's Hospital of Eastern Ontario Ottawa Canada,
George Wells and Pierre Jacob
Send Correspondence to journal:
Re: Seizure-associated brain injury in term newborns with perinatal asphyxia

ccampbell{at}cheo.on.ca Craig Campbell, et al.

The recent article by Miller et al. [1] explores the important area of neonatal asphyxia and seizures using 1H-MRS markers of neuronal metabolism and integrity. The authors indeed found changes in basal nuclei lactate/choline ratios and intervascular boundary zone lactate/choline and NAA/choline ratios suggesting that neonatal seizures alter brain metabolism and neuronal integrity.
The author's have created and applied as a their main predictor variable for seizure severity a "seizure score" consisting of four scores relating to seizure frequency, seizure onset, EEG characteristics and anticonvulsant therapy. There is no reference to prior use of the score and the design, application and analytical issues related to the development of the score that should be of concern to the reader are not addressed in the paper.
Such scores may be a source of powerful information but choosing the factors that contribute to the content and construct validity of this type of score can be difficult. [2] In the development of this score it is not clear why items such as seizure type, seizure duration, and a more detailed frequency score were included. In addition, it is unclear on what basis a difference in seizure onset of 24 hours has been given a 10% difference in seizure "severity". Individual subscales of a score must individually contribute to the overall score without substantial overlap in information. [3] In this situation the factors impacting on severity have questionable independence and may not lend new information to the overall score. For example, neonates having status epilepticus may be more likely to receive multiple different medications than someone who had only one brief seizure. The ultimate result is that bigger score differences are forced between patients then would be the case with more independent measures. The results of the formal procedures to assess criterion validity should be reported.
Practically, the score seems to have been assigned retrospectively despite the prospective nature of the study possibly creating information bias. The authors do not describe the inter-rater reliability of the chart review nor the manner in which they handled missing information. The score appears to have been applied only to the group of children who had clinical seizures which is evident in table 3 where the mean and standard deviation of the seizure score in the control group are both zero. This is of considerable concern as there are measures in the score (EEG and anticonvulsant use) that could be applicable to children not having clinical seizures. Again, by not applying the seizure score to all children the result is to create a greater difference between the groups. In addition, it is not clear if the whole group or only the 33 children with seizures were included in the linear regression model. If all the children were included, than given that the controls all had a seizure score of zero, the model may be simply finding the significance based on the presence or absence of clinical seizures rather than finding a true linear relation between the variables. Other issues in model building such as whether the individual components of the seizure severity score were examined in a univariate fashion or independently in the multivariate model would be of interest to explore the independent contribution of the factors.
In order to better understand the population the reader would be helped by a description of the seizure score parameters in the children with seizures as well as clinical details on the whole sample. For example, were there children on midazolam or paralytic agents in either group? These details may impact on the detection of clinical seizures, as well as seizure score measures such as frequency, EEG and anticonvulsant use.
The task of developing a seizure severity score and carrying out this study is clearly valuable and holds the potential benefit of understanding the relations between neonatal seizures and subsequent metabolic and functional neurological outcome. In this study, however, the reader is not made aware of the methodologic issues related to the seizure score that was used. The concerns about the measurement of seizure severity should temper the conclusions drawn in this study and should generate an increased interest toward validly and reliably developing measures of seizure severity in newborns.
References:
1. Miller SP et al. Seizure-associated brain injury in term newborns with perinatal asphyxia. Neurology 2002; 58: 542-548.
2. McDowell I and Newell C. Chapter 2: The theoretical and technical foundations of health measurement. In: Measuring Health A Guide to Rating Scales and Questionnaires (2nd Edition). New York: Oxford University Press, 1996.
3. Streiner DL and Norman GR. Chapter 5: Selecting the items. In: Health Measurement Scales A Practical Guide to Their Development and Use (2nd Edition). New York: Oxford University Press, 1995.