Are we using our brains? Diagnosis of postoperative cognitive dysfunction

In 1951, Dennis and colleagues performed the first human cardiopulmonary bypass operation using an oxygenator; unfortunately, the patient died shortly afterwards [1]. Even with the early closed type of valvular surgery (without opening the rib cage), the mortality figures were high. It was Gibbon who performed the first successful operation of this kind; however, only one of his first four patients was still alive 1 year postoperatively [2]. With further experience, the problem was largely addressed and by 1968, Favaloro had carried out a total of 55 more intricate cardiac graft operations with only two deaths [3]. Today, in our Cardiothoracic Centre at the University Hospital of Wales, mortality following cardiac graft surgery is less than 1% [4]. Subtle postoperative changes in brain function have been identified as an important side-effect of this type of surgery. Interestingly, the first report of these changes was published within a few years of these pioneering operations [5]. Progress was made with respect to this mild brain impairment: as early as 1957, researchers began to employ batteries of psychometric tasks to aid diagnosis of cognitive deterioration [6]. Subtle brain injury following cardiac surgery has since been the subject of much research. The incidence of cognitive deterioration has been examined at various time points, from 18 h postoperatively [7], to discharge [8–10], to up to 5 years postoperatively [11–13]. Associated risk factors [14–16] have also been explored, with major advances being made in non-cardiac surgical patients [14, 15]. The term postoperative cognitive dysfunction (POCD), used to describe subtle brain impairment, was adopted from International Study of Postoperative Cognitive Dysfunction (ISPOCD), one of the large multi-national studies [14]. Many studies have since explored associations between POCD and biochemical markers [17–19], all driven with the aim of developing strategies to predict diagnosis and hence reduce the incidence. The question is, how far have we really progressed? Currently, we can only make a formal diagnosis of POCD during research as the tools are too cumbersome for routine clinical practice. This is because both pre-operative and postoperative testing are required and the battery of psychometric tests used to make the diagnosis usually includes 5–17 different tasks [8, 20], taking at least 25 min to complete. A wide variation in the incidence of POCD is quoted [21, 22]. This could be dependent upon the tests employed as different groups of researchers use different batteries [21–23]. Furthermore, POCD also occurs after non-cardiac surgery [14, 15], although perhaps the incidence is not as high as with cardiac surgery [24]. However, this may again be a reflection of the battery of psychometric tasks chosen and the statistical test employed [24]. Finally, the cause of POCD is multifactorial, including, but not limited to: the inflammatory response; microemboli; hypoperfusion; genetic susceptibility; and the use of cardiopulmonary bypass [24–26]. It has been suggested that off-pump cardiac surgery may reduce the incidence compared to on-pump cardiac surgery, although the results are not yet conclusive [10, 13]. Evidence suggests that increasing age is a risk factor [14]. More important and alarming is the recent link demonstrated between POCD and mortality [15]. With our increasingly aged population and this particular association, the problem of POCD must not be overlooked. Certain key issues require answers. How can we reduce the incidence of POCD? If it occurs, can we reverse or reduce its impact? These questions, however, can only be addressed if POCD is diagnosed correctly in the first place. An often-used definition of POCD is deterioration in cognition associated with surgery [24]. However, this definition is vague; quantification of symptoms such as a loss of attention, memory and learning impairment, deterioration in perceptual and verbal abilities, and abstract thinking is difficult, and combining such impairments even more so. Correct definition and diagnosis are very basic problems. Studies from the 1960s and 1970s highlight the frustration of investigators with diagnosis [27, 28]. Severe injury such as stroke was easy to diagnose. From the very beginning, subtle brain injury was diagnosed by the adoption of a psychological approach, which did not necessarily provide reliable results. Researchers wanted to devise a standard way of diagnosing the condition. The first important document, paving the way for diagnosis of subtle brain injury following cardiac surgery, was the publication of a consensus statement in 1995 [29]. Experts identified and approved 14 key points that they anticipated researchers would take heed of in further research. Some of these points also had recommendations, including a list of five core tasks for psychometric testing. The consensus statement does not say whether these tasks are all that are required in a battery and sheds no light at all upon the statistical method to use in combination with the battery to diagnose POCD. However, very few researchers have actually employed all five tasks [10, 17]. Another consensus recommendation is the use of a control group to help test for practice effects, although no mention is made of what comprises a control group. The consensus statement is outdated, vague and subject to individual interpretation. Perhaps researchers are not to be blamed for not embracing it. There are a number of issues to raise. First, although there are many statistical methods that can be used to define POCD, such as the standard deviation rule, 20% rule and the reliable change index, at present there is no consensus as to the ‘best test’. All have advantages and disadvantages: the two former statistical methods are much simpler to apply and do not require a control group; whereas the more complicated reliable change index rule, usually employed with a control group, does consider practice effects. However, the make-up of a control group is always a problem for researchers. Authors sometimes claim one method to be better than another, suggesting higher scores in terms of sensitivity and specificity [21]. A highly sensitive test is good at identifying true positives, e.g. patients who have brain impairment following surgery who are identified as having POCD. Highly specific tests would be good at identifying patients who do not have true brain impairment and are not classified as having POCD. To judge how good a test is, information is required about both. However, at present it is not possible to identify when POCD has actually occurred, so we cannot calculate sensitivities and specificities. Clearly, it is going to be difficult to explore variables affecting POCD: it is difficult to relate uncertainties. Second, is it time to revisit the consensus statement? Would it not be better if all researchers used the same diagnostic approach? Currently, there is no gold standard and researchers apply different batteries of tasks and different statistical tests; one review listed 14 different approaches [30]. Studies from 2007 highlight this problem, with batteries consisting of between 9 and 19 tasks being used in the different studies, and with the 20% rule, one standard deviation rule and Z-score approaches (with raw Z-scores or calculated percentage incidences using the reliable change index rule) to diagnose POCD [11, 13, 17, 26, 31, 32]. Finally, is a basic redefinition required? Would it be so very incorrect to start at the point when the patient complains of being positively aware of a cognitive impairment? Interestingly, in one early study following coronary artery bypass surgery, a battery of tests was employed, but in addition, patients were asked whether they thought their cognition was unchanged [9]. They found that 79% had deteriorated in at least one task and 49% demonstrated deterioration in two or more tasks using the standard deviation rule. Of those with an impairment, 52% did not notice any cognitive disability in hospital, 38% had noticed a deterioration and 10% considered themselves disabled by their impaired cognition. Is this useful? Did the remaining 21% of the population notice any cognitive impairment? Deiner and Silverstein say that self reports of cognition status and results from psychometric testing do not always correlate [33]. However, severe self-reported problems do happen, contradicting normal results from cognitive testing [34]. Research tools are not necessarily of use in helping to diagnose and support individual patients. We would like to suggest a new research direction, otherwise the debate will continue over many more decades. We must progress; time is precious and our older population continues to increase in size. Perhaps we should start at the end of the story. If patients say that their cognition is impaired, then we should take that seriously and develop new tests appropriate to the problems that they themselves have identified. This would represent a new body of individualised research. Armed with more specific information, we may be able not only to revisit the consensus statement, but also to focus valuable medical resources on patients particularly at risk [15]. No external funding and no competing interests declared.