Ann Thorac Surg 2005;80:917
© 2005 The Society of Thoracic Surgeons
Original article: Cardiovascular
Invited commentary
David C. Bellinger, PhD, MS
Department of Neurology, Harvard Medical School, Children’s Hospital Boston, Farley Basement Box 127, 300 Longwood Ave, Boston, MA 02115
(Email: david.bellinger{at}childrens.harvard.edu).
The demonstration that grouping neuropsychological test scores, either empirically or theoretically, into putative "domains" can be misleading is extremely important. Domain scores are used both to increase statistical power (tests grouped together are assumed to reflect a latent construct such as "memory") and reduce a type I error rate otherwise inflated by multiple comparisons. However, Lewis and colleagues’ [1] findings imply that the assumption of stable latent constructs is questionable and that basing analyses on invalid domain scores might introduce outcome misclassification, increasing the likelihood of type II error (bias towards the null). Interpreting individual test scores rather than domain scores may be a safer approach.
Nevertheless, analyses based on test clusters can be informative. How test scores cluster might be sample-specific, varying with stable characteristics (eg, socioeconomic status) or transient characteristics (eg, fatigue, intoxication). As noted, although tests are labeled as assessments of "attention" or "language," good performance on any test requires that one do several things reasonably well. In the absence of other data, the reason for poor performance is uncertain. Therefore, differences between pre-cardiac and post-cardiac surgery factor structures may shed light on the nature of the impact of surgery on cognitive function. In this study, the factor structures for CABG patients were identical at baseline and at 3 months postoperatively. However, at postoperative day 7, tests clustered differently, and in ways that did not clearly map onto the expected theoretical "domains." Indeed, at least one endpoint from each of the tests administered loaded onto a single mega-factor, "complex attention/manual dexterity." Perhaps because of fatigue or lack of stamina, patients’ performance on all tests on this occasion was reduced by limited ability to attend to instructions, sustain attention, and monitor performance. Tests that would not cluster under usual circumstances might do so when energy and stamina are low. That patients’ performance was generally slower and more variable at the 7 day postoperative assessment than at the others is consistent with this hypothesis. The reason for the instability in the factor structure of controls’ test scores is puzzling. Perhaps the tests whose factor assignments shifted between assessments are those with the lowest reliability coefficients. Not all tests are equal, psychometrically.
Certainly this study suggests that interpreting patterns in test scores in terms of neuropsychological "domains" is simplistic, but let’s not throw out the baby with the bathwater.
 |
References
|
|---|
- Lewis MS, Maruff PT, Silbert BS. Examination of the use of cognitive domains in postoperative cognitive dsyfunction after coronary artery bypass graft surgery Ann Thorac Surg 2005;80:910-917.[Abstract/Free Full Text]
Top
References