Worsening of quality of life after epilepsy surgery

Participants.

Participants were the subset of 348 patients who underwent temporal lobe resection while enrolled in the Multicenter Study of Epilepsy Surgery (MSES) and who had complete data sufficient for these analyses. The MSES is an ongoing, prospective, observational study of the long-term outcomes of epilepsy surgery at seven centers in the Northeast and Central United States. Baseline characteristics and initial outcomes of the cohort are described elsewhere.^4,16 Preoperatively, all patients had had monthly consciousness-impairing seizures for at least the past 2 years, despite trials of two or more antiseizure medications, and were evaluated for resective surgery, using a common protocol, including intracarotid amobarbital testing to determine hemispheric speech dominance.¹⁷ Each institution's human subjects' protection board approved all study procedures.

Procedures.

Prior to surgery, patients completed the Quality of Life in Epilepsy Inventory (QOLIE)-89, an epilepsy-targeted survey of HRQOL that includes the Rand Short Form 36 (SF-36) as a generic core and a 53-item epilepsy-targeted supplement.¹⁸ Item responses are summarized into 17 subscales, 4 superordinate domains (Physical Health, Mental Health, Cognitive, and Epilepsy-Targeted), and an overall score (QOLIE-89 Total), based on factor analyses.^18,19 Baseline neuropsychological testing included the California Verbal Learning Test (CVLT), a 16-item, multi-trial list learning test commonly used to assess verbal memory before and after temporal lobectomy.^20,21 Follow-up procedures included quarterly telephone calls to obtain information from patient diaries about seizure occurrence, annual administration of the QOLIE-89, and re-administration of the neuropsychological tests at 2 and 5 years following surgery.

Seizure control.

Long-term seizure control and HRQOL can vary, as up to a quarter of patients remit or relapse during extended postoperative follow-up and HRQOL is related to recent seizure control.⁴ Different systems for classifying seizure control in the previous year reflect differences in HRQOL to different degrees,²² but there has been no corresponding analysis of methods for classifying long-term outcome. Preliminary analyses of the 2- and 5-year QOLIE-89 observations separately revealed nonsignificant trends supporting our primary hypothesis. We therefore decided to perform a joint analysis of the 2- and 5-year data in order to increase power, recognizing that seizure control and HRQOL remain stable in the majority of patients, but that late changes in seizure control are reflected in changes in HRQOL.⁴

In order to have large enough groups for meaningful analysis, we classified long-term seizure control as good, intermediate, and poor, based on patients' combined remission status at 2 and 5 years. Remission was defined as absence of consciousness-impairing seizures for at least the previous year (equivalent to Engel Class 1a and 1b). Good outcome patients were those in remission at both the 2-year and at the 5-year QOLIE-89. The vast majority (>95%) of these patients had been in full remission since discharge from surgery. Poor outcome patients were those who were in remission at neither the 2-year nor at the 5-year QOLIE-89. Approximately half of these patients (52%) had never entered remission during the entire 5 years of follow-up. The rest had entered remission at some point, but had relapsed at the time of the 2- and the 5-year QOLIE-89. Intermediate outcome patients were those in remission at the 2-year or at the 5-year QOLIE-89, but not at both. To determine whether QOL differed among seizure-free patients with or without persisting auras, we repeated our primary analyses after dividing the Good outcome group into those in 1 year remission from all seizures at 2 and 5 years and those in 1 year remission from all conscious-impairing seizures but having auras at 2 or 5 years.

Memory decline.

We classified patients as having had memory decline (MD+) or not (MD−) depending on whether they showed a postoperative decline of 1 SD or more from presurgery to the 2-year evaluation on the Long Delay Free Recall trial of the CVLT. We focused on verbal memory because verbal memory problems are more commonly reported than nonverbal memory problems by patients before and after temporal resection.^6,21,23,24 A 1 SD decline is unlikely to be explained by statistical artifacts of the test (e.g., test-retest variability, practice effects), as it corresponds roughly to the lower limit of the reliable change index for this test.²⁵ The clinical significance of this level of decline will vary across patients depending on different factors (e.g., demands of their job on verbal memory). Patients with a similar level of decline are more likely to report subjective memory decline than patients with lesser degrees of decline.^7,8 To assess the potential influence of comorbid depression on memory, we examined correlations between postoperative memory decline and scores on the Beck Depression Inventory (BDI) at baseline and 2 and 5 years following surgery.

In patients who were missing memory data from the 2-year evaluation, we used data from the 5-year evaluation to determine whether memory had declined after surgery. Although late memory change has been described, the largest memory changes generally are noted at the first postoperative evaluation.^15,26 Preliminary analyses in this cohort (unpublished data) showed that significant group declines in verbal memory occurred only between the presurgical and the 2-year evaluations and not between the 2- and 5-year evaluations, consistent with results in a different cohort.²⁷

Statistical analyses.

The primary analysis was a repeated measures analysis of variance on the total QOLIE-89 score, using a 3 (Seizure Outcome) × 2 (Memory Decline) factorial design with Time (Baseline, 2 years, and 5 years) as the repeated measures factor (SAS PROC Mixed).²⁸ Planned contrasts tested the difference in HRQOL change from baseline to 5 years among patients with and without memory decline (i.e., the Memory Decline × Time interaction) in each seizure outcome group. These contrasts tested the hypothesis that the impact of memory decline on long-term, self-reported HRQOL differs among patients with different patterns of postoperative seizure control. A second set of planned contrasts tested the significance of change from baseline to 5 years separately in the six groups defined by seizure control and memory decline status.

Secondary analyses using the same design examined changes in QOLIE-89 domain and subscale scores, in order to further describe how HRQOL changes in association with memory decline and persisting seizures. Analyses of demographic differences among different groups were conducted using t test, χ², and Fisher exact test, as appropriate.

Imputation of missing QOLIE-89 data.

Longitudinal follow-up studies often have data missing at critical time points that limit the subjects available for analysis and reduce power to detect effects in small subgroups. Preliminary analyses indicated trends consistent with our hypotheses, so we sought to improve power and generalizability by increasing the sample size through imputation of missing QOLIE-89 surveys. Missing baseline scores were imputed as the score predicted from the most statistically significant and parsimonious in a series of exploratory multiple regression analyses of the entire MSES sample in which different demographic variables were used to predict the baseline QOLIE-89 values.

We imputed missing 2- or 5-year QOLIE-89 data using QOLIE-89 data from comparable postoperative periods.²⁹ For those missing the 2-year survey, we used scores from the survey from the nearest period with the same seizure remission status. For example, if the subject was in remission at 1 year and 2 years, but had relapsed prior to the third year QOLIE-89, the survey data from year 1 were used to impute the 2-year score. If remission status was the same across all three periods, the average of the 1-year and 3-year score was used. Data for the 2-year survey were not imputed if remission status at the 1-year and 3-year surveys differed from status at the time the 2-year survey should have been administered or if all three surveys were missing. For patients missing the 5-year survey, the last observation was carried forward, provided remission status was the same for the two periods and any intervening ones. Otherwise, the 5-year survey was not imputed.

To test the validity of this strategy, we compared actual vs predicted values in a subsample of 173 MSES patients who had completed QOLIE-89s in three consecutive follow-up periods and who were clinically stable throughout the entire postoperative period (i.e., were always or never in remission since surgery).

To test for bias due to non-random missing data, we repeated the primary analysis using only those cases with non-missing surveys at all time points (i.e., excluding cases with imputed data). We also used individual growth curve analysis to include additional patients with missing QOLIE-89 values that could not be imputed.³⁰ Finally, analyses were repeated using only patients who had undergone anterior temporal lobectomy (ATL).

Subjects.

Information on seizure control sufficient for our analyses was available in 247/348 (71%) temporal resection patients. Forty-three of the 101 with missing seizure control data had left the study by 2 years (6 due to death, 22 due to withdrawal or protocol noncompliance, 13 due to loss to follow-up, and 2 for unrecorded reasons). The remaining 58 had either left the study by 5 years (6 due to death, 19 due to withdrawal or protocol noncompliance, 10 due to loss to follow-up) or had not yet come to the 5-year follow-up (n = 23).

Data for defining verbal memory change were available for 147/247 (60%). Of the 147, 32 cases had a missing QOLIE-89 observation. Missing observations were successfully imputed in 23 (6 baseline, 17 follow-up), yielding a final analysis sample of 138. Most of the 138 patients in the analysis had undergone modified ATL (n = 131). Six lesionectomies (4 nondominant, 2 dominant) and 1 nondominant, neocortical resection accounted for the other surgical types. These cases were distributed across the seizure and memory outcome groups in proportions similar to the ATL patients. Demographic characteristics of the sample are shown in table 1. The proportions of patients in the seizure and memory outcome groups are shown in table 2. There was no difference in the proportion of patients with memory decline across the three seizure outcome groups (p = 0.66).

Patients with insufficient seizure or QOL follow-up data were more likely to be younger, male, less well educated, and to have a shorter duration of epilepsy (all p < 0.05). There were no differences between those with and without sufficient seizure and HRQOL data in race, IQ, employment status, or other demographic characteristics. Those who did not have sufficient cognitive testing data were younger (34 vs 37 years) and more likely to be male (56% vs 39%), nonwhite (25% vs 15%), and have less than a high school education (21% vs 9%), compared to those who did have sufficient data (all p < 0.05). There were no differences between those with and without sufficient cognitive testing data in side of surgery, presence of hippocampal sclerosis, or level of presurgical memory function (all p > 0.10). Classification of memory decline was based on 2-year results in 104 cases (75%) and 5-year results in 34.

Validity of the imputation strategy.

The final regression equation predicted the baseline QOLIE-89 total score (p < 0.0001, r-square = 0.25), using gender, number of comorbid conditions, seizure frequency, seizure severity, and employment. Similar results were obtained for predicting the baseline domain and subscale scores. Among the 173 stable MSES patients who completed three consecutive postoperative QOLIE-89s, the middle QOLIE-89 Total score and the average of the preceding and following observations were highly correlated (r = 0.85, p < 0.0001) and there was no difference between the group means (55.72 vs 55.47, paired t = 0.63, p < 0.50). The absolute difference between the middle observation and the average of the preceding and following observations was less than the minimally important clinical change (MCIC)³¹ in 90% of the sample. This suggests that any bias and error introduced by our imputation strategy was not clinically significant in the majority of cases.

Primary analyses: Total QOLIE-89 scores.

Results of the primary analysis of total QOLIE-89 scores are represented in the figure. There were main effects of Seizure Outcome (p < 0.0001) and Time (p < 0.0001), with interactions for Seizure Outcome × Time (p < 0.0001), Seizure Outcome × Memory Decline (p < 0.01), and Seizure Outcome × Memory Decline × Time (p < 0.05).

• Download figure
• Open in new tab
• Download powerpoint

Figure Change in health-related quality of life

Change in health-related quality of life among temporal resection patients with Good, Intermediate, and Poor seizure outcome, with (MD+) and without (MD−) postoperative memory decline.

Planned contrasts of the Time × Memory Decline interaction within each of the three seizure outcome groups showed that HRQOL improved in the Good and Intermediate groups, regardless of memory decline (p [Time × Memory Decline] > 0.20, both groups). Contrasts of within-groups change from baseline to 5 years indicated improvement in the Good and Intermediate groups, with and without memory decline (all p < 0.01). Despite the appearance of the figure, the two Intermediate groups did not differ (p [Time × Memory Decline] = 0.75).

As hypothesized, change scores differed between the groups with poor seizure outcome (Poor-MD+ and Poor-MD−) (p [Time × Memory Decline] < 0.02). Poor MD− patients did not change (p = 0.25), but HRQOL declined in Poor MD+ patients (p < 0.03) (figure). Degree of memory decline did not correlate with BDI scores at any point, nor with BDI change from baseline (all r < 0.10, p > 0.20).

When the 32 patients with imputed QOLIE-89 data were excluded (n = 115), the overall pattern of change among groups did not change. Main effects of Time and Seizure Outcome remained (p < 0.001), but the two- and three-way interactions showed only trends (p = 0.07, p = 0.12, respectively). The pattern of results and significance levels did not change when cases with non-imputable data were included in the growth curve analysis (n = 147), nor when the seven non-ATL patients were excluded from the analysis (n = 131). This suggests that there were no detectable biasing effects introduced by imputing missing data, nor by the inclusion of patients with different types of surgery.

Reclassifying Good outcome subjects as with (n = 16) or without (n = 61) persisting auras did not substantially alter the results. Main effects and the Seizure Control × Memory decline interaction remained, but the Seizure Outcome × Memory Decline × Time interaction showed only a trend (p = 0.13). Those with and without auras in the Good seizure outcome group showed similar QOL improvements, irrespective of memory decline (p > 0.30). There continued to be a difference in change between those with vs without memory decline in the Poor seizure outcome group (p = 0.04).

Secondary analyses: Domain and subscale scores.

The same general pattern of main effects and two-way interactions was observed for the four domain scores. Across all domains, mean scores for the Poor MD+ group tended to change in a direction opposite that of changes in the other groups, although there was a Seizure Outcome × Memory Decline × Time interaction only for the Cognitive (p = 0.03) and Epilepsy Targeted domains (p = 0.04). The Poor MD+ group declined on the Cognitive domain score (p = 0.02), compared to other groups who either improved or did not change. The Poor MD+ group did not change (p = 0.18) on the Epilepsy Targeted domain, while all other groups improved (all p < 0.03).

There was a Seizure Outcome × Memory Decline × Time interaction (p < 0.05) on subscales measuring role limitations due to physical problems, role limitations due to emotional problems, work/driving/social function, and memory. Change scores differed between the Poor MD+ and Poor MD− groups on role limitations due to emotional problems and work/driving/social function (p [Time × Memory Decline] < 0.05) and tended to do so for memory (p = 0.10). Only the Poor MD+ group declined on these three subscales (all p < 0.05).

Characteristics of the Poor MD+ group.

Compared to the rest of the sample (n = 127), the Poor MD+ patients (n = 11) were more likely to have had a dominant hemisphere resection (73% vs 38%, p = 0.05), lower baseline IQ (85 vs 93, p = 0.04), later age at epilepsy onset (23.6 vs 14.1, p = 0.008), and shorter duration of epilepsy (17.3 vs 25.3, p = 0.05). They tended to be less likely to have hippocampal atrophy on MRI (45% vs 72%, p = 0.08). The mean memory decline (i.e., post-pre change score) in the Poor MD+ group did not differ from the mean memory decline in the Good MD+ and Intermediate MD+ groups (p = 0.27), indicating that the HRQOL declines in the Poor MD+ group were not due to a greater degree of memory decline.