- Top of page
Summary: Background: While some patients experience a decline in memory function following an anterior temporal lobe resection, there is considerable individual variation in the extent, nature, and direction of postoperative memory change. Patients with surgically remediable temporal lobe epilepsy differ in etiology, the extent and type of underlying pathology, and on demographic and epilepsy-related variables, all of which may have an impact on their pre- and postoperative neuropsychological functioning. This study examined the relationship between these variables and postoperative memory decline.
Methods: Logistic regression was used to examine the effects of age, laterality of surgery, age of onset of epilepsy, underlying pathology and preoperative level of memory function on postoperative verbal learning in 288 patients who had undergone an anterior temporal lobe resection. One hundred twenty-five patients underwent a right temporal lobe resection (RTL), 163 patients underwent a left temporal lobe resection (LTL).
Results: In the group as a whole, 25% of the patients demonstrated a significant postoperative deterioration in verbal learning. Postoperative deterioration in verbal learning was significantly associated with higher levels of preoperative function in both the RTL and LTL groups. Older age at the time of the operation and a lower verbal IQ were additional significant predictors for the RTL group. The presence of cortical dysgenesis was a significant predictor of postoperative decline in the LTL group. The logistic regression models accurately identified 3/4 of those who experienced a postoperative decline in memory, using a cutoff of 0.25 or above to identify high risk.
Conclusions: Our analyses suggest that the majority of patients with a high risk of significant postoperative memory decline can be reliably identified preoperatively. These models are valuable tools helping patients make an informed decision regarding surgery.
While many patients experience a decline in memory function following an anterior temporal lobe resection, there is considerable individual variation in the extent, nature, and direction of postoperative memory change (Baxendale and Thompson, 2005; Hermann et al., 1995). Patients with surgically remediable temporal lobe epilepsy differ in etiology, the extent and type of underlying pathology, and on a number of demographic and epilepsy-related variables all of which may have an impact on their pre- and postoperative neuropsychological functioning. Preoperative memory ability and later age of seizure onset and at surgery have been suggested as important predictors of postoperative memory decline (Hermann et al., 1995; Chelune et al., 1991; Hermann et al., 1992). Further, quantitative and qualitative MRI data mean that many pathologies can now be identified and quantified preoperatively (Kuzniecky et al., 1997; Watson et al., 1997).
A number of studies have attempted to predict postoperative change in memory using the data available preoperatively. The majority has examined the relationship between variables derived from a single source and postoperative memory change. Data from baseline neuropsychological tests (Chelune et al., 1991; Davies et al., 1998; Jokeit et al., 1997) and the intracarotid amobarbital procedure (IAP) (Bell et al., 2000; Chiaravalloti and Glosser, 2001; Kneebone et al., 1995; Loring et al., 1990) have been found to predict some of the variance in postoperative memory function. Similarly, preoperative measures of hippocampal integrity including direct recordings from hippocampal neurons (Cameron et al., 2001), 1H MRS (Hanoglu et al., 2004) and hippocampal volumes and T2 relaxation times derived from preoperative MRI studies have been shown to predict some aspects of postoperative memory function (Baxendale et al., 1998; Wendel et al., 2001). More recently, a number of functional imaging paradigms have shown some promise in the prediction of both postoperative decline in both visual (Janszky et al., 2005) and verbal (Richardson et al., 2004) memory skills. Collectively, the results from these studies suggest that a number of preoperative markers of the structural and functional integrity of the ipsilateral and contralateral hippocampi predict postoperative memory decline, consistent with the model of functional adequacy/functional reserve proposed by Chelune (1995).
The definition of memory decline is an important consideration in these studies. Reliable change indices (RCIs) are statistical calculations that allow the determination of meaningful change in scores on specific tasks by taking account of practice effects and the measurement error inherent in test–retest situations (Jacobson and Truax, 1991). In essence, RCIs determine whether a patient's performance on a particular test has significantly changed relative to their baseline and may be more accurate than classifications based on the standard deviation of the test alone. Notwithstanding the ongoing debate regarding the best RCI methodology (Temkin, 2004) RCIs are increasingly used in the study of cognitive change following temporal lobe epilepsy surgery (Engman et al., 2004; Hermann et al., 1996; McDonald et al., 2004; Stroup et al., 2003).
Although a number of studies have employed a multivariate approach to examine the relationship between measures of structure and function in preoperative patients (Baxendale et al., 1998; Hendriks et al., 2004; Lencz et al., 1992), fewer studies to date have employed a multivariate approach to predict postoperative decline. Stroup et al. (2003) developed a multivariate risk factor model to predict postoperative verbal memory decline in individual patients, using the side of the proposed resection, MRI findings other than exclusively unilateral mesial temporal sclerosis, preoperative measures of verbal memory and IAP performance as their independent variables. Their logistic regression analyses demonstrated that all five risk factors were significantly and independently associated with memory outcome, side of surgery having the strongest association, and preoperative immediate verbal memory the weakest.
While indices from the IAP may predict postoperative memory function, this effect may be dependent on the type of stimuli used in the IAP paradigm. Kirsch et al. (2005) found that IAP asymmetry scores obtained from a paradigm using mixed stimuli did not predict postoperative verbal memory function. In addition, the IAP is not always available for all patients preoperatively. As clinical confidence in structural and functional imaging paradigms has increased, a number of centres across Europe and Australia no longer routinely perform the IAP on all preoperative patients (Duncan et al., 2005; Helmstaedter, 2005).
Predictive models of postoperative memory change clearly have great potential in the preoperative counselling of patients. The aim of this study was to develop a multivariate model using data from widely available, noninvasive preoperative evaluations to predict verbal memory decline in patients undergoing temporal lobe epilepsy surgery.
- Top of page
Our findings suggest that the majority of patients who are at high risk of postoperative memory decline can be reliably identified preoperatively, using noninvasive indices of structure and function. The results suggest that the level of preoperative function is an important risk factor for a postoperative decline in verbal learning for all temporal lobe surgery patients. A low verbal IQ and higher age at the time of surgery are additional risk factors for RTL patients. The presence of cortical dysgenesis is a significant additional risk factor for LTL patients.
Postoperative decline in verbal learning was associated with higher levels of preoperative verbal learning in both the RTL and LTL groups. This finding adds to a convergent body of literature linking high preoperative function with a greater likelihood of postoperative decrease, in both RTL and LTL patients and across a variety of memory indices (Chelune et al., 1991; Hermann et al., 1995; Powell et al., 1985). Hermann et al. (1995) found that adequacy of preoperative memory performance was a nonspecific predictor, associated with a decrease in postoperative memory performance for both left and right TL patients and for multiple types of memory indices. These findings cannot be simply explained as an artifact of regression toward the mean. Our definition of change, based on RCIs is statistically robust and designed to overcome the difficulties associated with both regression toward the mean and practice effects in repeat assessments. In clinical terms it would seem that those who have the most, have the most to lose following epilepsy surgery.
Consistent with the majority of previous studies, we found that an older age at time of surgery was a significant risk factor for postoperative decline, particularly for RTL patients. Hermann et al. (1995) found that later age of epilepsy onset and older chronological age were significant and selective predictors of episodic memory decrease in left TL patients. Similarly, Helmstaedter and Elger (1996) found a significant correlation between older age at the time of surgery and postoperative decline on the German auditory verbal learning test. It is also interesting to find that a low verbal IQ is associated with increased risk of verbal memory decline in RTL patients. It seems probable that older age and a lower verbal IQ may be associated with more limited compensatory reserves following surgery.
In our study, the presence of cortical dysgenesis was associated with postoperative decline, particularly in LTL patients. The presence of cortical dysgenesis is a strong prognostic indicator of poor postoperative seizure control (Sisodiya et al., 1997). In addition it seems that it is associated with more widespread postoperative cognitive decline than the other pathologies that commonly underlie medically intractable epilepsy (Baxendale et al., 1999). It is not clear whether this is due to continued postoperative seizures, or the structural disturbance that results from the surgery. Improved seizure control in postoperative patients has been associated with improved cognitive function in a number of studies (Helmstaedter et al., 1992; Helmstaedter and Elger, 1996; Novelly et al., 1984). Since our aim was to provide a clinically useful model to predict postoperative memory decline in preoperative patients, this could not be considered in our analyses. However, it is not surprising to note that some of the factors associated with a poor cognitive outcome identified in the current study, have also been associated with a poor outcome in terms of seizure control (Baxendale et al., 1999).
There is some debate within the literature regarding the merits of prediction models based on the combined data from RTL and LTL groups, compared with separate models for each group. In their original study, Stroup et al. (2003) used laterality of surgery as a predictor variable. The group later reported no significant differences in their data when the RTL and LTL groups were analyzed separately (Langfitt and Stroup, 2004). However, our analysis produced different solutions for the RTL and LTL groups. While a high level of preoperative verbal learning was risk factor for both groups, Verbal IQ and the age at the time of surgery were additional significant risk factors for the RTL group, while the presence of cortical dysgenesis was significant for the LTL group. The use of separate regression equations for RTL and LTL groups is certainly advantageous in the identification of specific risk factors for each group and provides a valuable insight into the different mechanisms that may underlie the decline in each group. However, the regression solution based on the whole group, which examines the risk associated with the side of surgery together with the other factors, produced the most accurate model in terms of both sensitivity and specificity.
In addition to identifying these risk factors, our logistic regression models can also be used to predict the unique probabilities of memory decline on each task for individual patients. Towards this aim and to maximize the clinical application of these findings we now use an automated program to calculate the risk of postmemory decline for prospective surgical candidates. In this program, a series of screens allows the patient's preoperative details to be entered and their individual risk of decline is then calculated. It would not be appropriate for the majority of other centres to use our exact regression equations to predict decline in their own surgical populations. The equations are specific to the AMIPB, which is not widely used outside the United Kingdom. Although the AMIPB is similar in many key respects to the RAVLT and the CVLT, these equations will only hold true for the AMIPB test scores and the definitions of significant decline that are inherently part of the test. However, the similarities between clinical list learning tasks and the memory abilities they tap are such that the variables identified as significant in our equations should hold true and our methods can certainly be applied universally.
While it is certainly a clinical bonus to have such a tool at one's disposal, it is also worth noting that the majority of the patients in this study did not demonstrate a significant postoperative decline in verbal learning. The levels of cognitive stability recorded in our study are consistent with a number of other reports (Hermann, 1988; Rausch, 1991; Selwa et al., 1994; Hermann et al., 1995), but deserves special emphasis since the risks of temporal lobe surgery to memory function are routinely highlighted in every neuropsychological study of TLE patients. However, until recently the individual risks have received little attention.
We did not address the magnitude of deterioration in memory function in this study. Patients were dichotomized on statistical criteria, albeit criteria that were designed to maximize the clinical relevance and statistical significance of postoperative change and minimize the powerful effects of regression toward the mean. However the clinical significance of a drop of 15 percentile points will obviously be very different to a drop of 30 percentile points or more. Similarly the “real world” effect of deterioration in memory function will to some extent be dependent on the “starting point” of the drop. A drop of 15 percentile points, moving someone from a superior/high average level to an average level of function may not necessarily mean that they would not be able to carry out their everyday activities, although they may of course notice the decline. However, a similar magnitude of decline from a low average to an impaired level may result in the loss of key abilities. In the same way that the patients who deteriorate on the memory tests postoperatively are a heterogeneous group, caution should be used in the grouping of patients who do not demonstrate postoperative deterioration. Patients with gross impairments preoperatively may not measurably deteriorate postoperatively, but are not easily equated with individuals who have maintained an average or high level of function on a task across the surgical procedure.
Whether our definition of deterioration is clinically meaningful for a given patient will depend on the patient's capacity for cognitive compensation, the nature and extent of environmental demands on these abilities, and to some extent his or her psychosocial resources and support systems. Although certain neuropsychological measures may be more relevant to everyday memory failures than others, there is generally a poor correlation between objective measures of memory function and patient's complaints (Baxendale and Thompson, 2005). McGlone (1994) suggests that many postoperative patients may be under the mistaken impression that their memory function improves after surgery, influenced mainly by their positive surgical outcome. Thus while the regression analyses created in this study may provide a rational basis for the counselling of preoperative temporal lobectomy patients about the relative risks of postoperative memory decline on objective measures, they should be used with caution in the wider discussion of postoperative memory performance.
Since we no longer routinely conduct an IAP on our patients prior to surgery, we were not able to include data on language dominance in our analyses. Although we excluded all confirmed patients with atypical language representation, some cases will have been included in the series. Given the large numbers involved in the study, this small known confound in the generation of the equations is unlikely to have had a very significant impact on the results. However, atypical language representation will be a very significant factor in the subsequent clinical application of these regression equations in predicting postoperative memory decline for individual patients. It is likely that the equations generated will not be valid in the prediction of memory decline in patients with atypical language representation. Further work is underway to integrate data from our current clinical fMRI language paradigms to address this important issue.
A postoperative follow-up period of 1 year was used in this study. In the longer term, if patients maintain their seizure freedom, they may eventually stop taking their antiepileptic medication. Durwen et al., (1989) found that verbal memory performance improved significantly under reduced medication in LTLE patients. Thus the memory deficits recorded 1 year after the surgery may not be static and a number of factors may lead to an improvement or further decline in function over time.
Notwithstanding these caveats, this model is currently serving as a clinically useful and practical guide for counselling prospective epilepsy patients of the risks of temporal lobe surgery. Further work is underway to enhance the predictive power of the regression model using quantitative measures of structural integrity (Hippocampal volumes and T2 relaxation times) and measures derived from fMRI paradigms, and additional measures of neuropsychological function.