| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Ann Thorac Surg 1997;64:1225-1226
© 1997 The Society of Thoracic Surgeons
Oxford Heart Centre, John Radcliffe Hospital, Oxford, OX3 9DU, United Kingdom
To the Editor:
McKhann and associates drew two main conclusions in their recent article [1]: (1) neuropsychological changes vary according to the specific cognitive domains, and (2) some of the domains (eg, visual construction and executive function) show "late cognitive changes" when assessed at 1 year. The validity of these conclusions, however, depends entirely on the definition used to operationalize cognitive decline. The conclusions of McKhann and associates are based on a definition of cognitive impairment of just 0.5 standard deviation (SD) deterioration (after standardization to Z scores). By employing such a restrictive definition, they leave little room for intrasubject variation, and in the absence of a relevant control group, there is a very real possibility that their conclusions are based, at least in part, on results that might be expected by chance alone.
McKhann and associates justify their use of the restrictive 0.5 SD cut-off by suggesting that a larger deterioration in performance is required to classify a subject as impaired, because improvement would be expected to occur through practice. The rationale underlying this cut-off, however, is flawed for several reasons. First, the process of standardizing raw scores to Z scores should cancel the effects of practice. Applying the cut-off could therefore result in a subject being classified as "impaired" without having declined by more than 0.5 SD in raw scores. Second, there is evidence of considerable fluctuation in practice effects in normal subjects on psychometric tests [2].
Two brief examples serve to demonstrate how small variations in raw scores may result in a subject being potentially misclassified as impaired when using this definition. Supposing a patient scores 40 (total score) on the Rey Auditory Verbal Learning Test preoperatively, then in the current study (where the overall group mean is 40.68 with an SD of 8.76) the converted Z score is -0.08. If this same patient produces a raw score of 37 at 12 months, his or her Z score is now -0.59. Despite the raw score decreasing by just 3 points (that is, less than half the preoperative SD), the 0.5 SD cut-off classifies the subject as impaired. Similarly, a patient scoring 8 on the Delayed Recall Trial of the Rey Auditory Verbal Learning Test (group mean = 7.51; SD = 3.07) will have a Z score of +0.16. At 1 year, this patient (using the 0.5 SD cut-off) would be described as "impaired" if his or her score decreased by just 1 point to 7 (Z score equivalent = -0.38). These two examples clearly show that small changes in raw scores at 12 months, given such a cut-off, are sufficient to classify subjects as impaired. Although such performance drops conform to the 0.5 SD definition, it is not clear whether one should interpret them as clinically significant, particularly as alternate test forms were employed.
In their attempt to control for the effects of retesting, McKhann and associates ignore the possibility that the observed small changes in raw scores may represent natural fluctuations in normal performance. To avoid false positives, all definitions of impairment should be based on performance of relevant control groups.
References
Johns Hopkins Hospital, 600 N Wolfe St, Baltimore, MD 21287
To the Editor:
We appreciate the interest of our colleagues from the Oxford Heart Centre in our article. Regarding the choice of a criterion for decline, we agree that the choice of 0.5 standard deviation (SD) change from baseline score may appear unduly "restrictive" and perhaps not necessarily reflect clinical significance. However, in attempting to identify factors associated with cognitive decline after coronary artery bypass grafting, we considered any decline of potential scientific interest. The conversion of raw scores to Z scores does not cancel the effects of practice, because the Z scores, as explained in the statistical methodology section, were based only on preoperative distribution of scores.
We agree that there may be some fluctuations in practice effects in normal subjects when tested repeatedly on neuropsychological tests. These effects are somewhat test specific and also influenced by factors like age and educational levels. The study referenced by Browne and colleagues is unfortunately not very applicable to coronary artery bypass grafting populations, because it was based on only 10 subjects less than 50 years of age tested over 2- to 4-week intervals.
The remainder of the comments by Browne and colleagues appear to be based on a misunderstanding of our methodologic approach of using only the preoperative scores for Z score calculations. Thus, a raw score of 37 at 12 months corresponds to a Z score of [40.68 - 37]/8.76 = -0.42 (not -0.59). The choice of using only the preoperative scores for the basis of the Z score calculations was motivated by the fact that both the means and SDs would fluctuate somewhat over time. It is also a somewhat more conservative approach, because the preoperative standard deviations in general tended to be somewhat larger than those for 1-month and 1-year follow-ups. It is thus possible to calculate the amount of decline in raw scores required to meet the 0.5 SD criterion for decline for each test by dividing the preoperative SD by 2. For the Rey Auditory Verbal Learning test total score it would be 8.76/2 = 4.38. Therefore, a patient with a preoperative score of 40 would need to decline to a score of 35 to meet the 0.5 SD criteria.
The use of the 0.5 SD criterion for decline was not intended as a definition of "impairment," but simply as a threshold for cognitive change that might be of potential scientific (as opposed to clinical) significance in the search for possible correlates of adverse cognitive outcomes after coronary artery bypass grafting.
We completely agree, however, that methods for determining decline should ideally rely on data from a group of appropriate nonsurgical controls. These should be of similar medical and demographic characteristics as the patients in the surgical group exposed to the same tests and testing intervals. We have designed such a study and will begin enrollment of patients this summer.
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| ANN THORAC SURG | ASIAN CARDIOVASC THORAC ANN | EUR J CARDIOTHORAC SURG |
| J THORAC CARDIOVASC SURG | ICVTS | ALL CTSNet JOURNALS |
