Statistics from Altmetric.com
Editor,—While I applaud Tatman et al’s attempt to test James’s modification of the Glasgow coma scale for use in intubated children,1 I have some concerns about their study. First, as some of the observations are on the same child, and the attending nurse may have had some knowledge of the child’s condition before the assessment, the observations are not independent and thus may not occur by chance (the basis of Cohen’s κ2). Second, Cohen’s κ and weighted κ (derived from Cohen’s κ) are sensitive to an unequal distribution of the marginal totals,3 which are present in this study. Third, the verbal and grimace scales were not performed on the same group of patients, further complicating the comparison. These factors, together with the lack of significance between the grimace and verbal scores, make it difficult to conclude that the grimace scale is more reliable than the verbal score.
I agree with the authors that the grimace and verbal scales may measure different cortical functions, but would add that age may influence the clinical significance of these scales. In the development of coma scales, I think classification of the response should be based on components that predict differences in outcome or detect physiologically important changes—for example, seizures—rather than fitting the number of categories in other scales. As the authors point out, reporting individual components is much more informative than summated scores. Furthermore, increasing the number of categories is likely to worsen the interobserver reliability.4 Finally, there should be some agreement about eliciting the response to a painful stimulus. Tatman et al appear to have used only nailbed pressure, while many clinicians in the UK and elsewhere apply supraorbital or sternal pressure. Thus, before we accept the grimace scale, it needs to undergo further interobserver studies, its sensitivity in detecting clinically important events and the relation between its components and outcome needs to be assessed.
Drs Whitehouse and Powell comment:
We thank Dr Newton for making several pertinent points regarding the validity of our study. He highlights the problems of translating an ideal experimental design into everyday clinical practice.
Although some children were observed more than once, the observations were not necessarily interdependent. The coma state rather than the child was the unit of study. To reduce intrapatient variation, a different nurse attended on each occasion and more than 24 hours elapsed between readings. Nowadays much research is undertaken alongside routine clinical care, and as procedures such as obtaining parental consent can take a great deal of time, using multiple readings per child is an attractive compromise.
While κ is theoretically sensitive to unequal distribution of the marginal totals (and to the number of categories), in clinical practice an even distribution of the coma states would rarely occur. Our distribution of coma states (shown in tables 3 and 4) may be similar to that of other paediatric intensive therapy units. We emphasise that the numerical values for κ should be broadly interpreted.
Verbal and grimace scores had to be carried out on different groups of patients because verbal scores cannot be measured in intubated patients.
A strength of our study is that it evaluates the child’s Glasgow coma scale when used by routinely trained bedside nurses. These are the people relied on in clinical practice to perform, record, and act on changes in the child’s coma state. We believe that, for nurses, nailbed pressure is easier to perform in a standard, reliable way than supraorbital pressure.
We look forward to further publications on the validation of this child’s Glasgow coma scale, and our grimace scale in the future.