The reliability of neuropsychologic changes after cardiac surgery and a deeper problem posed by practice effects

Abstract

To the editor:Two articles published in the February 2001 issue of the Journal of Cardiothoracic and Vascular Anesthesia discussed methods for quantifying postoperative changes in the cognitive performances of cardiac surgery patients.1Slade P Sanchez P Townes B Aldea GS The use of neurocognitive tests in evaluating the outcome of cardiac surgery: Some methodologic considerations.Cardiothorac Vasc Anesth. 2001; 15: 4-8Abstract Full Text Full Text PDF PubMed Scopus (18) Google Scholar, 2Andrew MJ Baker R Bennetts J et al.A comparison of neuropsychologic deficits after extracardiac and intracardiac surgery.Cardiothorac Vasc Anesth. 2001; 15: 9-14Abstract Full Text Full Text PDF PubMed Scopus (30) Google Scholar The purpose of this commentary is to pose several questions regarding a data analysis method recommended by Slade et al1Slade P Sanchez P Townes B Aldea GS The use of neurocognitive tests in evaluating the outcome of cardiac surgery: Some methodologic considerations.Cardiothorac Vasc Anesth. 2001; 15: 4-8Abstract Full Text Full Text PDF PubMed Scopus (18) Google Scholar and used by Andrew et al,2Andrew MJ Baker R Bennetts J et al.A comparison of neuropsychologic deficits after extracardiac and intracardiac surgery.Cardiothorac Vasc Anesth. 2001; 15: 9-14Abstract Full Text Full Text PDF PubMed Scopus (30) Google Scholar the reliable change interval (RCI). I am in favor of abandoning the practice of treating impairment as a categorical variable and recommend that researchers adopt traditional parametric statistical techniques that would permit them to determine whether nonsurgery control groups reliably outperform cardiac surgery groups postoperatively on neuropsychologic tests. A second issue concerns problems associated with the influences of practice effects produced by multiple administrations of neuropsychologic tests. Although Slade et al1Slade P Sanchez P Townes B Aldea GS The use of neurocognitive tests in evaluating the outcome of cardiac surgery: Some methodologic considerations.Cardiothorac Vasc Anesth. 2001; 15: 4-8Abstract Full Text Full Text PDF PubMed Scopus (18) Google Scholar and Andrew et al2Andrew MJ Baker R Bennetts J et al.A comparison of neuropsychologic deficits after extracardiac and intracardiac surgery.Cardiothorac Vasc Anesth. 2001; 15: 9-14Abstract Full Text Full Text PDF PubMed Scopus (30) Google Scholar discussed problems associated with practice effects, neither article addressed the most disturbing issue posed by practice effects.At face value, the task of determining whether cardiac surgery causes changes in cognitive performance appears to be a relatively straightforward matter; one would simply administer neuropsychologic assessments before and after surgery, and differences in those scores (ie, change scores) would reflect an effect of surgery. As Slade et al1Slade P Sanchez P Townes B Aldea GS The use of neurocognitive tests in evaluating the outcome of cardiac surgery: Some methodologic considerations.Cardiothorac Vasc Anesth. 2001; 15: 4-8Abstract Full Text Full Text PDF PubMed Scopus (18) Google Scholar and Andrew et al2Andrew MJ Baker R Bennetts J et al.A comparison of neuropsychologic deficits after extracardiac and intracardiac surgery.Cardiothorac Vasc Anesth. 2001; 15: 9-14Abstract Full Text Full Text PDF PubMed Scopus (30) Google Scholar noted, however, this is not the case,3Kneebone AC Andrew MJ Baker RA Knight JL Neuropsychologic changes after coronary artery bypass grafting: Use of reliable change indices.Ann Thorac Surg. 1998; 65: 1320-1325Abstract Full Text Full Text PDF PubMed Scopus (82) Google Scholar and rigorous scrutiny of research methodologies and data analysis techniques is needed to overcome some of the subtle, but formidable obstacles involved in studying the neuropsychologic consequences of cardiac surgeries. The authors of these articles appropriately called attention to the fact that the quantitative methods that have been widely adopted for analyzing effects of cardiac surgery on cognitive and psychomotor performance are not ideal because they (1) fail to adequately account for the influence of practice effects on postoperative performances on neuropsychologic tests and (2) rely on an arbitrary threshold for defining preoperative and postoperative performance changes as impairment, improvement, or no change.Practice effects are performance changes (usually in the direction of improvement) that result from participants' repeated (ie, preoperative and postoperative or baseline and follow-up) exposures to the neuropsychologic instruments. Because surgery-induced changes are superimposed on practice-related performance improvements, detecting neuropsychologic changes caused by cardiac surgery is more complicated than it initially appears. As Slade et al1Slade P Sanchez P Townes B Aldea GS The use of neurocognitive tests in evaluating the outcome of cardiac surgery: Some methodologic considerations.Cardiothorac Vasc Anesth. 2001; 15: 4-8Abstract Full Text Full Text PDF PubMed Scopus (18) Google Scholar and Andrew et al2Andrew MJ Baker R Bennetts J et al.A comparison of neuropsychologic deficits after extracardiac and intracardiac surgery.Cardiothorac Vasc Anesth. 2001; 15: 9-14Abstract Full Text Full Text PDF PubMed Scopus (30) Google Scholar correctly noted, the best way to control for practice effects is to include a noncardiac surgery control group that completes the neuropsychologic test battery at the same test-retest intervals used for the surgery group. When a nonsurgery control group is included, several statistical methods are available that can separate practice and group (ie, surgery v control) effects on performance. Some of the statistical options that can be used to handle the problem of practice effects are subsequently discussed.Defining reliable changeTo evaluate the reliability of neuropsychologic performance changes associated with cardiac surgery, one could use traditional parametric inferential statistical methods, such as analysis of variance (ANOVA), to determine whether the distribution of change scores (baseline to follow-up) of cardiac surgery patients on neuropsychologic tests reliably differed from that obtained from a control group that did not undergo cardiac surgery. Instead of using traditional parametric statistics to assess the reliability of between-group (ie, cardiac surgery v control) differences, however, many studies in this area have presented categorical data showing the number of cardiac surgery patients who were postoperatively impaired, unimpaired, or improved. Slade et al1Slade P Sanchez P Townes B Aldea GS The use of neurocognitive tests in evaluating the outcome of cardiac surgery: Some methodologic considerations.Cardiothorac Vasc Anesth. 2001; 15: 4-8Abstract Full Text Full Text PDF PubMed Scopus (18) Google Scholar and Andrew et al2Andrew MJ Baker R Bennetts J et al.A comparison of neuropsychologic deficits after extracardiac and intracardiac surgery.Cardiothorac Vasc Anesth. 2001; 15: 9-14Abstract Full Text Full Text PDF PubMed Scopus (30) Google Scholar were critical of the method used for defining these categories that is reported most frequently in the literature, the standard deviation (SD) method. The SD method categorizes postoperative change using this formula: (postoperative score—preoperative score)/SD. The SD is derived from the preoperative scores of a nonsurgical control group. When the SD method is used, a surgery patient is classified as impaired if his or her transformed score on a given instrument was ≤ −1.0. Performance would be classified as improved if the score was ≥1.0. Performance would be classified as unchanged if the score was (anywhere) between −1.0 and 1.0.There are several reasons why one might be critical of the SD method. One concern that some,3Kneebone AC Andrew MJ Baker RA Knight JL Neuropsychologic changes after coronary artery bypass grafting: Use of reliable change indices.Ann Thorac Surg. 1998; 65: 1320-1325Abstract Full Text Full Text PDF PubMed Scopus (82) Google Scholar including Slade et al1Slade P Sanchez P Townes B Aldea GS The use of neurocognitive tests in evaluating the outcome of cardiac surgery: Some methodologic considerations.Cardiothorac Vasc Anesth. 2001; 15: 4-8Abstract Full Text Full Text PDF PubMed Scopus (18) Google Scholar and Andrews et al,2Andrew MJ Baker R Bennetts J et al.A comparison of neuropsychologic deficits after extracardiac and intracardiac surgery.Cardiothorac Vasc Anesth. 2001; 15: 9-14Abstract Full Text Full Text PDF PubMed Scopus (30) Google Scholar have expressed about this method is that a 1 SD change in performance would occur by chance alone with about a 16% likelihood (p = 0.16) when no other factors (eg, surgery or practice effects) influenced performance. This criterion is more liberal than the conventional α level traditionally used to reject the null hypothesis (0.05).The classification method preferred by Slade et al1Slade P Sanchez P Townes B Aldea GS The use of neurocognitive tests in evaluating the outcome of cardiac surgery: Some methodologic considerations.Cardiothorac Vasc Anesth. 2001; 15: 4-8Abstract Full Text Full Text PDF PubMed Scopus (18) Google Scholar and Andrew et al,2Andrew MJ Baker R Bennetts J et al.A comparison of neuropsychologic deficits after extracardiac and intracardiac surgery.Cardiothorac Vasc Anesth. 2001; 15: 9-14Abstract Full Text Full Text PDF PubMed Scopus (30) Google Scholar the RCI, was first introduced by Jacobson and Truax4Jacobson NS Truax P Clinical significance: A statistical approach to defining meaningful change in psychotherapy research.J Consult Clin Psychol. 1991; 59: 12-19Crossref PubMed Scopus (7142) Google Scholar to attempt to quantify psychotherapy-induced performance changes in individual patients. The RCI differs from the SD method in 3 respects. First, the RCI uses the 90% confidence interval as the criterion for defining reliable change. A change score (ie, postoperative—preoperative) that exceeded 1.64 × SD would be located in the extreme 10% of the sample distribution (5% on each tail of the distribution). A criterion that requires a change score to fall outside of the 90% confidence interval of the control group's mean to be classified as an impairment or improvement (if positive) could be considered reliable given the conventional belief that changes that are expected by chance alone are only 1 in 20 times as reliable (the choice of α = 0.05 is itself an arbitrary convention). Second, the RCI method includes a parameter that reflects an assessment instrument's test-retest reliability (ie, random score changes across test administrations). This parameter is the standard error (SE) of the measurement computed from the control group's raw data. Third, the magnitude of the practice effect is determined based on the control group's raw data and removed mathematically. The formula used to compute RCI is: RCI = (SEdiff)* (± 1.64) + practice effect, where and where (SD1) is the SD of the baseline score, and (rxx) is the test-retest reliability coefficient. The practice effect is the mean of the control group's difference scores between the first test session and the follow-up test session.If the goal is to determine whether an individual patient's change score on an instrument is greater than would be expected by chance (or other uncontrolled factors, such as practice effects), the RCI method would have several advantages over the SD method. At issue is whether the group that underwent cardiac surgery had postoperative test scores that reliably differed from scores produced by the control group, taking into account factors such as practice effects. To determine this, one must compute an inferential statistic on the group data; otherwise, one has only an incidence summary that does not quantify the overall reliability of group differences on neuropsychologic performance. Because the primary goal of much research on this topic is to discover how to improve postoperative neuropsychologic outcome, it seems that the question of whether the effects are reliable at the group level should be an important focus of the quantitative analyses. At this point, one has several choices.One choice is to use one of the data transformations described earlier (ie, RCI or SD) to create a discrete variable (impaired, not impaired, improved) and determine if the distribution of surgery patients across these categories differs reliably from the control group's (eg, X = square). This approach is weakened by 2 considerations. First, information is lost whenever a continuous variable is transformed to a discrete variable. Second, nonparametric statistics tend to overestimate the probability of a type 1 error and sacrifice statistical power.5Welkowitz J Ewen RB Cohen J Introductory Statistics for the Behavioral Sciences. Harcourt Brace Jovanovich, Orlando, FL1976Google Scholar A second choice is to perform a one-way between-groups ANOVA on the transformed postoperative scores (expressed as a continuous variable) to determine whether the groups reliably differed. If one chose to use a one-way, between-groups ANOVA as the inferential statistic, the RCI data transformation would be preferable for the reasons discussed earlier (ie, it factors in the influence of practice effects and test-retest reliability on postoperative scores). Perhaps the most elegant approach to the problem is to use a mixed-factor ANOVA that includes treatment (control v cardiac surgery) as the between-group factor and session (1 v 2) as a repeated-measures factor, using either the raw data or a data transformation that yields a continuous variable. Differences in magnitude of change between the baseline and follow-up sessions that are a function of group (surgery v control) are reflected in the F ratio of the session (baseline v follow-up) × group interaction.Although there is nothing inherently wrong with determining where individual participant's data points are located within the sampling distribution (as computing the RCI or a z-score does), there is no clear justification for defining region of the distribution as the region that equals impairment. In this respect, the RCI method is no less arbitrary than the SD method that it was intended to replace. The RCI method does incorporate a definition of reliability that is in line with the conventional α level (0.05) applied when statistical methods are used for null hypothesis testing with group data. Neuropsychologic performance is a continuous variable, however, and impairments are no doubt manifest in different degrees across individuals in the surgery sample. When a threshold for defining impairment (or improvement) is set at some criterion level and cases that do not reach that threshold are functionally excluded from the overall analysis (regardless of how far they fall from the threshold), information is lost. The problem is not specific to the RCI method because no mathematical formula could permit one to determine where the threshold that separates impairment from nonimpairment should be placed. Ultimately, more valuable information is obtained when impairment is treated as a continuous variable, and one applies conventional parametric inferential statistical methods (ie, mixed-factor ANOVA), reports effect size, and quantifies the probability of the type 1 error when one uses statistics to reject the null hypothesis.Statistical reliability versus clinical significanceAs Slade et al1Slade P Sanchez P Townes B Aldea GS The use of neurocognitive tests in evaluating the outcome of cardiac surgery: Some methodologic considerations.Cardiothorac Vasc Anesth. 2001; 15: 4-8Abstract Full Text Full Text PDF PubMed Scopus (18) Google Scholar noted, Jacobson and Truax4Jacobson NS Truax P Clinical significance: A statistical approach to defining meaningful change in psychotherapy research.J Consult Clin Psychol. 1991; 59: 12-19Crossref PubMed Scopus (7142) Google Scholar introduced the RCI method to define meaningful clinical change in the psychotherapy setting, the implication being that clinical significance is solely a question of effect size. Meaningful clinical change is not strictly a quantitative issue, however. Suppose one computed RCI scores for 2 different tests that measured performance in 2 different neuropsychologic domains, memory and psychomotor speed, and on both measures the postoperative scores placed the subjects outside of the 90% confidence interval Would it be reasonable to claim that the clinical meaningfulness of the changes on both tests would be equivalent? Of course not, because memory and psychomotor speed impairments may not affect the individual's functioning in everyday life to the same degree. Small changes in one domain may be functionally more important than larger changes in the other. It would not be reasonable to defend the view that if the RCI on the memory test placed the patient's memory score around the 80% confidence interval and the psychomotor speed score outside the 90% confidence interval those 2 changes were not equally meaningful. The data gathered in studies on the neuropsychologic effects of cardiac surgery do not permit one to determine how large the change must be on a given instrument for that change to be considered clinically significant. Determining clinical significance (or meaningfulness) is separate from the issue of statistical reliability, and it requires a detailed analysis of the relationship between scores on the neuropsychologic instrument in question and performance on tasks encountered in everyday life. Few would argue that changes in visual contrast sensitivity could be usefully compared, in terms of their clinical meaningfulness, with changes in auditory pitch sensitivity on statistical grounds alone; it is easy to see that one is comparing apples with oranges in such a case. The fact is overlooked that the same is true when performances on different neuropsychologic tests are compared.Practice effects: A different dimension of the problemIt is widely acknowledged that neuropsychologic tests are highly susceptible to practice effects, and neuropsychologic impairment in cardiac surgery patients may be superimposed on the practice-related improvements. As Benedict6Benedict RHB Cognitive function after open-heart surgery: Are postoperative neuropsychological deficits caused by cardiopulmonary bypass?.Neuropsychol Rev. 1994; 4: 223-255Crossref PubMed Scopus (52) Google Scholar and others7Murkin JM Newman SP Stump DA Blumenthal JA Statement of consensus on assessment of neurobehavioural outcomes after cardiac surgery.Ann Thorac Surg. 1995; 59: 1289-1295Abstract Full Text PDF PubMed Scopus (528) Google Scholar have pointed out, postoperative neuropsychologic impairment in cardiac surgery patients typically has been manifest in terms of smaller than expected practice effects, relative to the magnitudes of practice effects observed in nonsurgery control patients. That is, postoperative performances of cardiac surgery patients did not decline per se; their scores just did not improve as much at the follow-up session as the control subjects' did. One possible explanation for the reduced size of the practice effect in cardiac surgery patients is that the surgery itself produced some new neuropsychologic impairment that disrupted a process involved in performing the task. Another possible explanation is that cardiac surgery patients simply did not learn as much as the nonsurgery patients about the tests during the preoperative test session. By this account, the entire postoperative deficit may reflect a preoperative learning impairment, and cardiac surgery itself may be irrelevant to the reduced size of the practice effect observed in cardiac surgery patients. It is well established that illness and stress can impair learning and memory.8Smith A Thomas M Whitney H After-effects of the common cold on mood and performance.Ergonomics. 2000; 43: 1349Google Scholar, 9Sapolsky RM Glucocorticoids, stress, and their adverse neurological effects: Relevance to aging.Exp Gerontol. 1999; 34: 721-732Crossref PubMed Scopus (386) Google Scholar In the situation of cardiac surgery patients, one would expect the combined effects of illness and stress on learning and memory to be substantial Studies that report impairments in cardiac surgery patients that are manifest only as smaller than expected practice effects cannot conclude that surgery, rather than a preoperative learning deficit, caused any postoperative differences between surgery and control patients unless the expected practice effect is based on data obtained from control subjects who are well matched with the cardiac surgery patients in terms of the nature and the severity of cardiovascular disease and the stress of impending major surgery. This aspect of the practice effect problem is particularly disturbing not only because it means that the causes of postoperative deficits reported in many previous studies are ambiguous, but also because it will not be an easy problem to address methodologically in future studies.These concerns expressed in this commentary are in no way intended to be critical of the outstanding contributions by Slade et al1Slade P Sanchez P Townes B Aldea GS The use of neurocognitive tests in evaluating the outcome of cardiac surgery: Some methodologic considerations.Cardiothorac Vasc Anesth. 2001; 15: 4-8Abstract Full Text Full Text PDF PubMed Scopus (18) Google Scholar and Andrew et al2Andrew MJ Baker R Bennetts J et al.A comparison of neuropsychologic deficits after extracardiac and intracardiac surgery.Cardiothorac Vasc Anesth. 2001; 15: 9-14Abstract Full Text Full Text PDF PubMed Scopus (30) Google Scholar These authors are to be commended for their ongoing efforts to improve the methods used to investigate cardiac surgery effects on neuropsychologic performance.Department of Psychology University of North Carolina at Wilmington Wilmington, NCdoi:10.1053/jcan.2001.26559 To the editor:Two articles published in the February 2001 issue of the Journal of Cardiothoracic and Vascular Anesthesia discussed methods for quantifying postoperative changes in the cognitive performances of cardiac surgery patients.1Slade P Sanchez P Townes B Aldea GS The use of neurocognitive tests in evaluating the outcome of cardiac surgery: Some methodologic considerations.Cardiothorac Vasc Anesth. 2001; 15: 4-8Abstract Full Text Full Text PDF PubMed Scopus (18) Google Scholar, 2Andrew MJ Baker R Bennetts J et al.A comparison of neuropsychologic deficits after extracardiac and intracardiac surgery.Cardiothorac Vasc Anesth. 2001; 15: 9-14Abstract Full Text Full Text PDF PubMed Scopus (30) Google Scholar The purpose of this commentary is to pose several questions regarding a data analysis method recommended by Slade et al1Slade P Sanchez P Townes B Aldea GS The use of neurocognitive tests in evaluating the outcome of cardiac surgery: Some methodologic considerations.Cardiothorac Vasc Anesth. 2001; 15: 4-8Abstract Full Text Full Text PDF PubMed Scopus (18) Google Scholar and used by Andrew et al,2Andrew MJ Baker R Bennetts J et al.A comparison of neuropsychologic deficits after extracardiac and intracardiac surgery.Cardiothorac Vasc Anesth. 2001; 15: 9-14Abstract Full Text Full Text PDF PubMed Scopus (30) Google Scholar the reliable change interval (RCI). I am in favor of abandoning the practice of treating impairment as a categorical variable and recommend that researchers adopt traditional parametric statistical techniques that would permit them to determine whether nonsurgery control groups reliably outperform cardiac surgery groups postoperatively on neuropsychologic tests. A second issue concerns problems associated with the influences of practice effects produced by multiple administrations of neuropsychologic tests. Although Slade et al1Slade P Sanchez P Townes B Aldea GS The use of neurocognitive tests in evaluating the outcome of cardiac surgery: Some methodologic considerations.Cardiothorac Vasc Anesth. 2001; 15: 4-8Abstract Full Text Full Text PDF PubMed Scopus (18) Google Scholar and Andrew et al2Andrew MJ Baker R Bennetts J et al.A comparison of neuropsychologic deficits after extracardiac and intracardiac surgery.Cardiothorac Vasc Anesth. 2001; 15: 9-14Abstract Full Text Full Text PDF PubMed Scopus (30) Google Scholar discussed problems associated with practice effects, neither article addressed the most disturbing issue posed by practice effects.At face value, the task of determining whether cardiac surgery causes changes in cognitive performance appears to be a relatively straightforward matter; one would simply administer neuropsychologic assessments before and after surgery, and differences in those scores (ie, change scores) would reflect an effect of surgery. As Slade et al1Slade P Sanchez P Townes B Aldea GS The use of neurocognitive tests in evaluating the outcome of cardiac surgery: Some methodologic considerations.Cardiothorac Vasc Anesth. 2001; 15: 4-8Abstract Full Text Full Text PDF PubMed Scopus (18) Google Scholar and Andrew et al2Andrew MJ Baker R Bennetts J et al.A comparison of neuropsychologic deficits after extracardiac and intracardiac surgery.Cardiothorac Vasc Anesth. 2001; 15: 9-14Abstract Full Text Full Text PDF PubMed Scopus (30) Google Scholar noted, however, this is not the case,3Kneebone AC Andrew MJ Baker RA Knight JL Neuropsychologic changes after coronary artery bypass grafting: Use of reliable change indices.Ann Thorac Surg. 1998; 65: 1320-1325Abstract Full Text Full Text PDF PubMed Scopus (82) Google Scholar and rigorous scrutiny of research methodologies and data analysis techniques is needed to overcome some of the subtle, but formidable obstacles involved in studying the neuropsychologic consequences of cardiac surgeries. The authors of these articles appropriately called attention to the fact that the quantitative methods that have been widely adopted for analyzing effects of cardiac surgery on cognitive and psychomotor performance are not ideal because they (1) fail to adequately account for the influence of practice effects on postoperative performances on neuropsychologic tests and (2) rely on an arbitrary threshold for defining preoperative and postoperative performance changes as impairment, improvement, or no change.Practice effects are performance changes (usually in the direction of improvement) that result from participants' repeated (ie, preoperative and postoperative or baseline and follow-up) exposures to the neuropsychologic instruments. Because surgery-induced changes are superimposed on practice-related performance improvements, detecting neuropsychologic changes caused by cardiac surgery is more complicated than it initially appears. As Slade et al1Slade P Sanchez P Townes B Aldea GS The use of neurocognitive tests in evaluating the outcome of cardiac surgery: Some methodologic considerations.Cardiothorac Vasc Anesth. 2001; 15: 4-8Abstract Full Text Full Text PDF PubMed Scopus (18) Google Scholar and Andrew et al2Andrew MJ Baker R Bennetts J et al.A comparison of neuropsychologic deficits after extracardiac and intracardiac surgery.Cardiothorac Vasc Anesth. 2001; 15: 9-14Abstract Full Text Full Text PDF PubMed Scopus (30) Google Scholar correctly noted, the best way to control for practice effects is to include a noncardiac surgery control group that completes the neuropsychologic test battery at the same test-retest intervals used for the surgery group. When a nonsurgery control group is included, several statistical methods are available that can separate practice and group (ie, surgery v control) effects on performance. Some of the statistical options that can be used to handle the problem of practice effects are subsequently discussed.Defining reliable changeTo evaluate the reliability of neuropsychologic performance changes associated with cardiac surgery, one could use traditional parametric inferential statistical methods, such as analysis of variance (ANOVA), to determine whether the distribution of change scores (baseline to follow-up) of cardiac surgery patients on neuropsychologic tests reliably differed from that obtained from a control group that did not undergo cardiac surgery. Instead of using traditional parametric statistics to assess the reliability of between-group (ie, cardiac surgery v control) differences, however, many studies in this area have presented categorical data showing the number of cardiac surgery patients who were postoperatively impaired, unimpaired, or improved. Slade et al1Slade P Sanchez P Townes B Aldea GS The use of neurocognitive tests in evaluating the outcome of cardiac surgery: Some methodologic considerations.Cardiothorac Vasc Anesth. 2001; 15: 4-8Abstract Full Text Full Text PDF PubMed Scopus (18) Google Scholar and Andrew et al2Andrew MJ Baker R Bennetts J et al.A comparison of neuropsychologic deficits after extracardiac and intracardiac surgery.Cardiothorac Vasc Anesth. 2001; 15: 9-14Abstract Full Text Full Text PDF PubMed Scopus (30) Google Scholar were critical of the method used for defining these categories that is reported most frequently in the literature, the standard deviation (SD) method. The SD method categorizes postoperative change using this formula: (postoperative score—preoperative score)/SD. The SD is derived from the preoperative scores of a nonsurgical control group. When the SD method is used, a surgery patient is classified as impaired if his or her transformed score on a given instrument was ≤ −1.0. Performance would be classified as improved if the score was ≥1.0. Performance would be classified as unchanged if the score was (anywhere) between −1.0 and 1.0.There are several reasons why one might be critical of the SD method. One concern that some,3Kneebone AC Andrew MJ Baker RA Knight JL Neuropsychologic changes after coronary artery bypass gra