| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Ann Thorac Surg 2007;83:374-376
© 2007 The Society of Thoracic Surgeons
a Department of Neurology, The Johns Hopkins University School of Medicine
b Department of Biostatistics, The Johns Hopkins Bloomberg School of Public Health
* Address correspondence to Dr Selnes, Division of Cognitive Neuroscience, Johns Hopkins Medical Institutions, Read Hall–East 2, Baltimore, MD 21205-1910 (Email: oselnes{at}jhmi.edu).
An important clinical as well as research use of neuropsychological testing is to evaluate changes in cognitive performance over time. Prospective follow-up studies allow identification of changes in the progression of a disease, recovery, and possible effects of pharmacological interventions. In the context of cardiac surgery, preoperative baseline testing followed by one or more postoperative assessments has become an established method for evaluating the possible impact of the use of cardiopulmonary bypass on cognitive outcomes.
However, it is well known that the reported incidence of cognitive decline after coronary artery bypass grafting (CABG) has varied quite significantly from one study to another. It has been widely assumed that this variability has resulted from differences in patient populations, follow-up time intervals, and the choice of neuropsychological measures. More recently it has become apparent that even more important factors contributing to variability in the estimates of rates of postoperative CABG cognitive decline are the choice of statistical criteria for defining decline and the inclusion of control groups. One of the most noteworthy examples of the importance of these factors is the Octopus Study in which the reported incidence of postoperative cognitive decline changed dramatically from 31% to 8% after a different criterion for decline and inclusion of control subjects was applied [1].
In this issue of The Annals of Thoracic Surgery, Keith and colleagues [2] draw attention to another potential methodological shortcoming in the assessment of cognitive changes after CABG. They note that the preoperative baseline evaluation may be unreliable because of situational and patient-related variables. This, together with practice effects from repeated exposure to the same neuropsychological tests, can lead to an underestimate of the degree of cognitive change associated with the use of cardiopulmonary bypass. They argue that the use of a control group can not entirely account for the effects of repeated administration of the neuropsychological tests because of an interaction between the cardiac surgery and practice effects.
Therefore they recommend dispensing with the preoperative baseline assessment and instead rely on comparison of postoperative cross-sectional assessment of cardiac surgical patients with well-matched controls to evaluate possible adverse cognitive effects of the surgery. Although we agree that serial assessment of change in performance on cognitive tests is not without its methodological challenges, the proposal to replace this method with a single, cross-sectional postoperative evaluation of cognitive performance will introduce far more serious obstacles to an unbiased assessment of cognition after CABG. The preoperative variability in cognitive performance in candidates for CABG can be attributed in part to demographic variables, such as age and education, but in addition there is variability due to cerebrovascular disease. The majority of CABG patients have some degree of coexisting cerebrovascular disease [3], and the impact of this on their preoperative cognitive performance can only be established through cognitive testing; it can not be estimated based on factors such as duration and severity of hypertension, hypercholesterolemia, and diabetes.
Keith and colleagues [2] identify two specific factors that may result in inadvertent lower preoperative performance of candidates for CABG: (1) elevated levels of anxiety, and (2) distractive settings for those patients who are tested in their hospital room. They suggest that elevated preoperative anxiety levels generally interfere with neuropsychological performance. Although this seems intuitively to be a reasonable assumption, the empirical support is weak at best. In non-CABG populations, a curvilinear relationship between anxiety and neuropsychological test performance has been reported [4]. Mild to moderate levels of anxiety actually can produce mild improvements in performance, whereas severe anxiety is associated with some decline in performance. A recent study that examined anxiety and neuropsychological test performance before surgery concluded that low levels of preoperative depression and anxiety states do not affect neuropsychological functioning among CABG candidates [5]. Similar conclusions were reached by Andrew and colleagues [1], who reported a very low incidence of severe preoperative anxiety, and no relationship with preoperative neurocognitive performance [6].
Even if mild to moderate levels of anxiety did indeed result in lower cognitive performance, the assumption that this problem could be solved by only postoperative testing may not be correct. There are data to suggest that the incidence of anxiety may be higher after surgery than before, whereas the incidence of depression and stress remains relatively unchanged preoperatively to postoperatively [1]. In addition, degree of postoperative anxiety was a significant predictor of worse (postoperative) performance on measures of psychomotor speed and memory. If a subset of patients has elevated anxiety levels even after surgery, a single, cross-sectional postoperative assessment of cognition may therefore be even more difficult to interpret.
The second obstacle to a reliable preoperative baseline noted by Keith and colleagues [2] is that in some studies a subgroup of patients are tested in their hospital room shortly before surgery. Since most patients are tested in a nonhospital setting after surgery, their performance may thus improve simply as a result of better testing circumstances.
However there may be alternate explanations for the lower cognitive performance of patients who are preoperatively tested in a hospital setting. As noted by Keith and colleagues [2], abnormal neuropsychological test performance in candidates for CABG even before surgery has been "convincingly demonstrated." In several of these studies the preoperative testing was done 1 to 2 days before surgery and not in a hospital setting [7–9]. Therefore, lower than expected preoperative cognitive test performance can not be accounted for by a distractive test setting alone. There are a number of reasons why some patients are tested preoperatively in the hospital, including the status of their coronary artery disease. There is evidence that cardiovascular disease, even at subclinical levels, can influence performance on neuropsychological tests [10]. Therefore it is possible that the lower cognitive performance of patients who are tested in the hospital before surgery may reflect effects of longstanding vascular disease of the brain.
If a subset of CABG patients does have abnormal neuropsychological test performance before surgery, and several studies suggest that this may be true independent of testing circumstances, this presents an additional problem for the cross-sectional postoperative approach. In the absence of baseline testing, any postoperative impairment would be attributed to the effects of surgery. Several years ago it was observed that a substantial number of patients were depressed after surgery. It was formerly believed that CABG was the cause of postoperative depression until preoperative studies documented that most patients with postoperative depression were actually depressed even before surgery.
As a method for classifying the presence or absence of clinically relevant postoperative cognitive impairment, Keith and colleagues [2] propose that postoperative cognitive test scores of individual patients can be compared with those in existing normative databases. The availability of large normative data sets for many neuropsychological tests has improved substantially in the past 2 decades, and in theory therefore, this may be an option, although it would depend on the research question being addressed. In addition, the success of this approach would also depend on having age-stratified and education-stratified normative data sets for all tests included in the assessment battery collected from the same population of normal controls. Currently, normative data for each of the tests included in a battery are typically derived from populations that may differ in terms of education, age, gender, intelligence, and cultural background. Another limitation of current normative data sets is that very few provide base rates of "impaired" neuropsychological performance among their normal controls. Studies that have reported such data have found that 20% of their subjects had at least two scores in the impaired range on separate tests [11]. Finally, to avoid attributing differences in postoperative cross-sectional scores to factors other than surgery, the control data set should be collected from individuals with demographic and medical risk factors similar to those of the study patients.
Keith and colleagues [2] note that a postoperative cross-sectional assessment may work particularly well in randomized clinical trials, because "any differences between the groups will necessarily indicate the effect of intervention." Although in theory this is true, there are examples in the literature that even in the best randomized clinical trials there may be unanticipated preoperative differences in performance on some cognitive tests. For example, Nathan and colleagues [12] reported statistically significantly higher scores on measures of attention in their control group before surgery.
From a statistical perspective the proposal to rely on cross-sectional comparisons of the post-CABG response, ignoring baseline values, is a high-risk strategy because it is likely to increase both bias and variance of the estimated CABG effect. By comparing the change from baseline, we eliminate any differences between the CABG and control groups that persist with time. These differences are substantial in our own studies. It is a violation of Occam’s razor, which favors simpler theories, to worry about a two-way interaction, but not about one of the main effects. Keith and colleagues [2] postulated a two-way interaction of surgery with the learning effect, but use a method that is potentially seriously biased by the main effect of surgical group, one of the components of their interaction. Ignoring a main effect to account for its interaction favors more complex, not simpler explanations. We need to focus on main effects before considering possible interactions among them.
The proposed cross-sectional analyses will also be more variable (ie, it will use the available data less efficiently) than the comparison of change from baseline. In our studies and those of others of cognitive outcomes after CABG, the typical correlation among repeated measures within 3 to 12 months of neuropsychological tests is between 0.65 and 0.85 [13, 14]. When the repeated measures correlation is >0.5, the comparison of change is more precise than the cross-sectional comparison at one time. When the correlation is 0.75, the middle of our range, it is twice as precise. In other words, ignoring the baseline data has the effect of discarding half of the available information, even if there are no baseline differences between the two groups.
Although the search for the cause of cognitive changes after CABG has not yet identified a single, causative factor, there is nonetheless considerable progress in the field. More recent studies of cognitive outcomes have included control groups as well as baseline cognitive assessments [7, 13, 15, 16]. Older studies lacking such controls were forced to rely on arbitrary statistical criteria for determining decline and, as a result, overestimated the incidence of postoperative cognitive decline. Current studies suggest that postoperative cognitive changes in most cases are relatively mild, and generally they are reversible by 3 months after surgery. The cause of these transient cognitive changes is likely to be multi-factorial [17], and the focus of contemporary studies is no longer just on procedure-related factors, but also on patient-related variables [18].
We believe that the proposal by Keith and colleagues [2] to dispense with the preoperative baseline testing would be a step in the wrong direction for CABG cognitive outcomes research. There does not seem to be substantive empirical support for their arguments that the preoperative cognitive test performance of patients having CABG is biased because of situational or psychological factors. Therefore we conclude that the assumptions underlying consensus statement 3, stating that the measurement of "individual change in performance from baseline to a time after operation is essential," remain valid [19], and that longitudinal change in performance should still be considered the method of choice for studying cognitive outcomes after cardiac surgery.
 |
Acknowledgments |
|---|
 Top Acknowledgments References |
|---|
|
References |
|---|
|
TopAcknowledgmentsReferences |
|---|
Related Article
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 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 |
