Address correspondence and reprint requests to Dr. M. Jones-Gotman at Montreal Neurological Institute, 3801 University Street, Montreal, Quebec H3A 2B4, Canada. E-mail: email@example.com
Summary: Purpose: We investigated possible cognitive effects of topiramate (TPM) in polypharmacy on patients with intractable epilepsy.
Methods: Study 1 evaluated 22 consecutively admitted patients whose antiepileptic drugs (AEDs) on admission to the Montreal Neurological Hospital included TPM. Performance on neuropsychological tests administered on and subsequently off TPM was analyzed. Four patients also were tested before taking TPM, allowing comparisons off, then on, and then off the drug again. Measures included intellectual function, verbal and nonverbal memory, language, word and design fluency, somatosensory sensitivity, and motor skills. In Study 2, 16 patients at the Minnesota Epilepsy Group were tested first off, then on TPM with nine cognitive tasks that measured concentration, verbal fluency, language, and psychomotor speed.
Results: In Study 1, significant (p ≤ 0.01) improvements were observed off TPM on 13 measures including verbal and nonverbal fluency and certain verbal and perceptual tasks. Notably, verbal learning and memory were unaffected; a limited effect was observed on nonverbal memory. Patients tested 3 times scored better in both tests off TPM compared with on this drug. In Study 2, declines on TPM were observed on all measures, significantly (p ≤ 0.05) for tests of fluency, sustained concentration, and visual motor processing speed.
Conclusions: TPM was associated with declines in fluency, attention/concentration, processing speed, language skills, and perception; working memory but not retention was affected. As the two studies used an opposite order of testing on versus off TPM, our results clearly show a performance decrement while patients are taking TPM, without respect to which condition is tested first.
It is known that antiepileptic drugs (AEDs) can cause adverse effects on cognitive function (1,2). Topiramate (TPM) has been used with considerable success for seizure control (3–8). It is well tolerated and is considered one of the most effective antiepileptic agents in comparison with other AEDs (1). It has been associated, however, with side effects including ataxia, confusion, dizziness, weight loss (2), and (rarely) also with acute psychotic symptoms (9). In a study of new AEDs, TPM had the highest overall incidence of adverse effects, even at low doses (1). According to subjective reports from our patients, we note that some do not experience any altered cognitive abilities when taking TPM. However, neuropsychological tests have shown significant impairments even in patients who do not report cognitive changes (10). Some of these effects have been described as “abnormal thinking,” mental slowing, and word-finding difficulties (11). Another study reported that patients described themselves as having slow thoughts, decreased cognition, difficulty calculating, dulled thinking, and blunted mental reactions (12). Some patients are aware of such side effects but prefer to continue to take the drug, possibly because of the improved seizure control (13). Adverse cognitive effects and behavioral problems also have been seen in children (10,14,15).
Studies that elaborate on which cognitive effects are specifically altered and the extent to which patients have been affected have begun to appear more recently. Burton and Harden (1996) assessed attention span in 10 adult patients with epilepsy who were receiving a range of TPM dosages over an average of 3 months. To assess attention, a digit-span test was performed weekly for the duration of TPM administration. Four of nine subjects showed a correlation between a higher dosage of TPM and poorer forward digit-span scores (16).
Several studies have tested a wider range of cognitive abilities. In one study of 37 patients taking TPM, cognitive impairments in concentration, psychomotor slowing, memory deficits, and dysphasia were seen in some patients (17). A preliminary report from the Montreal Neurological Hospital (MNH) indicated lower scores when taking TPM compared with off the drug on several tasks including attention, language comprehension, fluency, arithmetic, naming, and mood (18).
Martin et al. (19) compared the cognitive effects of TPM with two other of the newer AEDs, gabapentin (GBP) and lamotrigine (LTG), in 17 healthy young adult volunteers. Tests of attention, psychomotor speed, language, memory, and mood were examined. The group of subjects taking TPM showed declines in attention and letter and category word fluency when tested 3 h after large initial doses. These neurocognitive effects on TPM remained apparent at 2 and 4 weeks. The subjects taking GBP and LTG, however, did not show any significant declines on neuropsychological tests when tested 3 h after large initial doses and after 2 and 4 weeks. Although the study revealed adverse cognitive effects for TPM in comparison with other AEDs, the authors acknowledged that the dose-titration schedule for TPM and LTG was more rapid than that normally used in the clinical setting. Moreover, their results addressed the side effects in the first month of TPM use only.
Aldenkamp et al. (20) compared the cognitive effects of TPM with valproate (VPA) when used as adjunct therapy to carbamazepine (CBZ) in patients with localization-related epilepsy. Short-term verbal memory, as measured by the Recognition of Words Test, showed reduced scores with both TPM and VPA, although the effect of TPM was greater when baseline and titration scores were compared. Performance on another verbal memory test, the Rey Auditory Verbal Learning Test (RAVLT), gave conflicting results within their study and was thus inconclusive. Other tests of motor speed, alertness/reaction speed, information-processing speed, and memory did not show statistically significant differences between TPM and VPA. The authors concluded that the difference in cognitive effects between TPM and VPA as adjunctive treatment was small and that gradual introduction of TPM could reduce the extent of cognitive impairment.
Thompson et al. (21) reported improvements in verbal fluency, verbal learning, and digit span when TPM was withdrawn or reduced. They also found a decline on TPM in verbal intelligence, which is considered to be relatively insensitive to AED effects. No significant changes in nonverbal learning, recall, or perceptual analysis were observed, however. The authors concluded that the cognitive changes could not be due merely to an acute effect or to rapid drug introduction because the subjects had been taking TPM for at least 3 months before testing.
Although some of the cognitive effects of TPM have been described, the specific effects of this drug and the extent to which it alters function are yet uncertain. Furthermore, among the existing studies, one investigated patients because of cognitive complaints, causing bias in the sample (21); another study had a fast dose-titration schedule for TPM, and it examined effects only in the first month of TPM use (19).
The present investigation comprises two studies in which epilepsy patients were tested on versus off TPM. The first was a prospective study of consecutively admitted patients who were taking TPM when admitted to the MNH for intensive monitoring; they were tested on a large number of neuropsychological measures on admission and were subsequently retested after treatment with TPM had been discontinued. In the second study, patients at the Minnesota Epilepsy Group (MEG) underwent neuropsychological assessment before initiation of TPM therapy and were later retested after reaching a therapeutic dose of the drug.
Study 1: Patients at the MNH tested first on, then off, topiramate
Twenty-two patients with intractable epilepsy and who were taking TPM on admission to the MNH participated in Study 1. Four of these patients also had been tested in the past, before TPM administration. Table 1 provides descriptive information about the patients. All patients were being evaluated as possible surgical candidates. Among them, 11 of 22 had nonfocal seizure onset, whereas 11 had focal seizure loci in various sites. The patients who were tested 3 times (before, during, and off TPM) (see Results) all had focal seizure onset. All of them received TPM as a part of polypharmacy. The other drugs were CBZ (12 subjects), LTG (four), VPA (four), clobazam (CLB; eight), phenytoin (PHT; three), primidone (PRM; one), and phenobarbital (PB; one). Four of the 22 also had either one (two patients) or two (two patients) other drugs discontinued, among which were LTG (two), CBZ (two), CLB (one), and VPA (one). Because we tested consecutive patients on a TPM regimen, we made no exclusions based on IQ. Only one of the 22 patients had a Full-Scale IQ <70 (IQ was 59, tested while taking TPM). Table 2 lists the number of AEDs that the patients were taking during neuropsychological testing, and gives information about mean dosages and the length of time that they had been taking TPM before participating in Study 1. As it is the usual MNH procedure to reduce AEDs during investigation of potential surgical candidates, the only alteration in the usual investigation for participants in this study was that they underwent additional neuropsychological evaluation after discontinuation of TPM was complete. All subjects gave informed consent according to the declaration of Helsinki (BMJ 1991;203:1194).
Twenty-two tasks were administered, some of which consisted of more than one measure. Table 3 shows the tasks used and the measures analyzed within them. They break down into the following categories.
Table 3. Study 1: paired samples test
Test 1: on TPM
Test 2: off TPM
WAIS-R subtest means are scaled scores; all others are mean raw scores.
Significant (p ≤ 0.01)
Digit Span (WAIS-R)
Chicago Word Fluency
Jones–Gotman Design Fluency
Boston Naming Test
WMS Visual Reproduction Test: delayed recall
Chapman–Cook Speed of Reading Test
Block Design (WAIS-R)
Complex Figure Memory Test: copy
Abstract Design List Learning: recall on trial 1
Digit Symbol (WAIS-R)
Complex Figure Memory Test: delayed
WMS Visual Reproduction Test: immediate recall
Picture Completion (WAIS-R)
Two-point Discrimination Test: left hand
Two-point Discrimination Test: right hand
WMS Paired-Associate learning: delayed recall
Abstract Word List Learning: recall on trial 1
Pinch Strength: left hand
WMS Paired-Associate Learning: learning score
Abstract Word List Learning: recall on trial 4
RAVLT: false positive errors (recognition)
Pinch Strength: right hand
Abstract Design List Learning: delayed recall
Abstract Design List Learning: recall on trial 4
Simple Tapping: left hand
RAVLT: Recall after 30-minute delay interval
RAVLT: Recall after an interference trial
RAVLT: Recall on trial 1
RAVLT: Recognition list 2
Sequential Tapping: right hand
Abstract Word List Learning: delayed recall
RAVLT: Recall on trial 5
Sequential Tapping: left hand
Simple Tapping: right hand
RAVLT: Recognition list 1
WMS Logical Memory: immediate recall
WMS Logical Memory: delayed recall
Six subtests of the Wechsler Adult Intelligence Scale–Revised (WAIS-R) (22) were administered. These were Digit Span, Arithmetic, Block Design, Digit Symbol, Similarities, and Picture Completion. A Full Scale IQ rating also was calculated. For one patient who was 14 years old, the Wechsler Intelligence Scale for Children–Revised (WISC-R) was used (23).
Four verbal memory tests were used: the Paired-Associate Learning and Logical Memory subtests from the Wechsler Memory Scale (WMS) (24), the Abstract Word List Learning Test (25), and the Rey Auditory Verbal Learning Test, or RAVLT (26). The largest number of measures for a single test was on the RAVLT, for which we analyzed seven measures (listed in Table 3). The nonverbal, visuoperceptual memory tasks were the Complex Figure memory test (27–29), the Abstract Design List Learning Test (25), and WMS Visual Reproduction test (24).
Three language tasks were used: the Boston Naming Test (30), the Token Test (31,32) for language comprehension, and the Chapman-Cook Speed of Reading test.
The two fluency tasks used were the Chicago Word Fluency (verbal) (33) and Jones-Gotman Design Fluency (figural, nonverbal) (34–36) tests.
Sensory and motor
A two-point discrimination test was used to determine somatosensory sensitivity on the hands (37), and motor functions were sampled with tests of pinch strength (38), simple tapping, sequential tapping, and bimanual sequential tapping (39).
To test for possible significant differences due to TPM, we used paired t tests to compare patients' neuropsychological test scores on and off the drug. To determine whether low-functioning and high-functioning patients' cognitive abilities might be affected differently by TPM, we also calculated a Pearson correlation coefficient between the on-to-off difference score of each measure and Full Scale IQ. It was not possible to administer all tests to all of the patients; therefore the sample size is not the same across all measures.
Change in scores ON versus OFF topiramate
Because multiple t tests were performed, we accepted as significant only results with a p value ≤0.01. Among the 22 tasks examined, several consisted of more than one measure (Table 3). Thirteen measures showed a significant improvement off TPM compared with on (Fig. 1), and no result went in the other direction. Thus four of the six subtests from the WAIS-R (Digit Span, Arithmetic, Block Design, and Digit Symbol) showed significant improvements off TPM compared with on. Both verbal and nonverbal fluency and all three of the language tests showed significant improvements off TPM. There were no significant effects on verbal learning and memory, and robust effects were seen on only two measures of the nonverbal memory tasks. These were the delayed recall on the WMS Visual Reproduction task and the recall on the first learning trial of the Abstract Design List Learning task. The somatosensory measure was not significant, and the only significant motor change was on bimanual sequential tapping.
In addition to the 13 measures showing significant improvements off TPM, nine cognitive measures showed changes at a p ≤ 0.05 but p > 0.01; we defined this result as not significant, but the reader is referred to Table 3, which shows the p values for every measure analyzed. To be noted are the 19 tasks for which there was clearly no difference between the on and off TPM conditions: these consisted primarily of memory tasks and motor measures.
Because the range of possible scores among the different tests varies widely, we used a percentage change score for graphic presentation of the data, so that the results can be compared visually more easily. The percentage changes on tasks for which the differences were significant at p ≤ 0.01 or less are depicted in Fig. 1.
Effect of overall level of cognitive function
To explore the possibility that TPM may have a different effect depending on overall cognitive level, we calculated a Pearson Correlation Coefficient between difference scores and Full Scale IQ. There was a significant correlation between Full Scale IQ and the arithmetic subtest of the WAIS-R (r = 0.655, p = 0.002), but no other measure correlated with IQ. The IQ rating was obtained in the first testing session, while patients were taking TPM.
Patients tested before, during and off topiramate (question of practice effects)
Four patients were tested 3 times, as they had had a prior investigation, before starting to take TPM. Thus for those four patients, it was possible to compare performance while taking the drug with two “off TPM” conditions. This allowed us to assess whether patients' improvements in the “off TPM” conditions were attributable only to practice effects. However, results are available for a reduced number of tests because there were only five measures on which all four patients were tested 3 times. These measures were the Token Test, Chicago Word Fluency, and three subtests from the WAIS-R (scaled scores): Digit Span, Arithmetic, and Digit Symbol. The patients scored better off TPM on each of the five measures during both of the off-TPM testing sessions. Figure 2 shows the percentage loss on these five measures when the patients initiated TPM use (comparing the first “off TPM” condition with “on TPM”), and the percentage improvement when they discontinued TPM (comparing the “on TPM” with the second “off TPM” condition).
Study 2: Patients at MEG tested first off, then on, topiramate
The subjects in this study were 16 patients with intractable epilepsy, who had undergone clinical and cognitive assessment before treatment with TPM. Patient report of cognitive complaints was not a consideration in patient selection. However, the incidence of cognitive complaints while taking TPM was estimated by retrospective chart review. This was based on patient responses to standard questions posed by the nurse or physician about medication side effects during routine office visits. Follow-up testing was conducted after a therapeutic dose of TPM had been attained. Two of the 16 patients were evaluated while still inpatients and were therefore tested after a more rapid titration of TPM. The mean dose of TPM for all 16 patients was 359 mg (SD = 207) with a dose range of 100 to 700 mg (Table 2). Demographic information is presented in Table 1. Most patients were young adults who had completed the equivalent of a high school degree, with cognitive abilities in the average to low-average range at baseline, with the exception of one who had a Full-Scale IQ of 55. Eight patients had diffuse or bilateral seizure onset, whereas eight had focal findings on EEG (four left, four right). Six patients had undergone surgery before participation in this study. Informed consent was obtained from all patients.
Description of neuropsychological tests
Cognitive assessment on TPM was conducted with measures previously established as sensitive to medication side effects. The battery included tests of sustained concentration, verbal fluency, visual motor processing, and psychomotor speed. Table 4 shows the measures and their corresponding constructs.
Table 4. Study 2: Cognitive battery
Test 1: Before TPM
Test 2: With TPM
Name of test
Means are raw scores.
CPT, Continuous Performance Task; COWAT, Controlled Oral Word Association Test.
Digit symbol (WAIS-R, WAIS-III)
Visual motor processing speed
Digit span (WAIS-R, WAIS-III)
Serial Digit Learning
VIGIL CPT: Omission errors
Grooved Pegboard (Dominant Hand)
VIGIL CPT: Reaction Time
The following four measures of attention and concentration were included: Digit Span and Digit Symbol subtests of the WAIS-R and WAIS-III (22,40), Serial Digit Learning (a supraspan digit learning task) (41), and a computerized continuous performance test of sustained attention, the VIGIL (42). The Digit Symbol task also reflects speed and accuracy of visual motor processing.
Verbal fluency was assessed with the Controlled Oral Word Association Test (“C, F, L”; 43) and Animal Naming (44).
The Grooved Pegboard task (45) and the reaction time score from the VIGIL (42) were used to assess psychomotor speed.
Paired t tests were conducted to compare group test scores off and on TPM. Not all tests were administered to all patients. Sample size for each individual measure is indicated in Table 4. To assess individual change associated with TPM, clinically meaningful decline on each measure was defined as negative change equaling at least one standard deviation with one exception. For Serial Digit Learning, significant decline was defined as a decrease in percentile rank greater than one quartile.
On TPM therapy, mean scores on all nine cognitive measures declined relative to baseline (Fig. 3). For each measure, we accepted as significant declines with a p ≤ 0.05. With this criterion, a statistically significant decline in performance on TPM was noted on three measures: Digit Span, Digit Symbol, and COWAT (Table 4).
Individually, 15 (94%) of 16 patients displayed clinically meaningful decline on at least one measure based on objective criteria described earlier. The results of one of the two patients who had been subjected to more rapid titration were consistent with those of the group as a whole, showing losses on retest on one or more measures. The only patient without a significant decline was the other one tested after more rapid TPM titration. According to medical record review, only 10 (63%) of 16 patients who demonstrated decline on testing complained of any cognitive difficulty when questioned. Subjective report of the patient appears to be an unreliable indicator of cognitive decline in this sample.
The results of these studies show that TPM use is associated with decline in several aspects of cognition. Although the findings do not suggest a global decrease in cognitive function, several specific areas appear to be severely affected. Both verbal and nonverbal fluency scores improved by ≥70% after TPM was discontinued in Study 1, and verbal fluency scores decreased significantly on TPM in Study 2 (in which nonverbal fluency was not tested). This suggests that the drug greatly limits divergent thinking. This is in agreement with previous findings of decreased word fluency on TPM (19,21), but extends those findings to include nonverbal divergent thinking.
Digit Symbol, Block Design, and the copy of a complex figure were affected by TPM therapy. These tasks all require visuoperceptual skills, and thus visual perception may contribute to the deficits noted on those tasks. The significant changes in Digit Span and Digit Symbol also can be considered a reflection of impaired attention, consistent with other reports (16,19). However, discrepant findings were reported by Aldenkamp et al. (20), with respect both to attention and to visual perception. They reported results from two timed tasks that were considered attention tasks; however, one also required decision making, and the other, visual perception. Contrary to our results, they did not find significant differences between patients taking VPA versus taking TPM (in both cases also taking CBZ) on those tasks (20).
Our finding that arithmetic is affected is consistent with other reports (13,18), and reflects a reduced ability to hold and manipulate information in working memory. Digit Span, another measure of working memory, also showed significant improvement off TPM (Study 1), and a decline on TPM compared with baseline (Study 2).
All three of the language tests administered in Study 1 and one of the verbal fluency measures in Study 2 showed significant change associated with TPM, which is surprising in the sense that only the fluency tasks require speed. Thus the deficit goes beyond a simple slowing of verbal abilities and is more than just an attention deficit. The poorer naming and language comprehension on TPM suggests that the drug leads to “dulled thinking,” as reported by Jones (12) and Crawford (11). Although this interpretation is consistent with the subjective reports of some patients who are aware of their impaired cognition, Study 2 determined that patient report alone might be unreliable as an indicator of cognitive change.
Interestingly, learning and memory are affected very little by TPM. The only memory measures that showed a meaningful change (Study 1) were delayed recall of the WMS drawings and the first recall trial on the Abstract Design List Learning task (there were no effects on subsequent trials of the nonverbal learning task). Thus in both cases in which there were effects, patients were required to recall designs after a single exposure. The poor performance on TPM on these measures most likely reflects the same perceptual deficit as noted on the Digit Symbol and Block Design tasks and the copy of a complex figure. We conclude from this that TPM does affect visual perception, and that it combines with “dulled thinking” to produce an impairment in nonverbal memory when further exposure to the material to be remembered is not possible.
There was no difference between the on and off TPM conditions on any of the 14 measures of verbal learning and memory in Study 1 or on the one learning measure included in Study 2. This is compelling evidence that TPM affects cognition in an immediate sense, but that information can be processed and retained, given time and practice. Our findings are consistent with those of Aldenkamp et al. (20), who also did not find significant losses with the RAVLT. Thompson et al. (21), however, reported impaired verbal learning on TPM. The discrepancy among these results remains for further investigation.
It is interesting that the only motor task on which we found a loss on TPM was bimanual sequential tapping. This is the one motor measure that requires the most cognitive processing, as patients must tap in a different specific sequential order with the two hands simultaneously. The task involves attention, perception, and the capacity to monitor and coordinate out-of-phase movements (39).
On the basis of informal observation, we hypothesized that patients with lower intellectual capacity may show less an effect on TPM than higher functioning ones, owing to overall poor function, or a “floor effect.” If true, this could mean that TPM would be better tolerated cognitively by such patients. We examined this notion with the patients and data of Study 1, for which there were a large number of measures to analyze. We performed a correlation analysis that compared Full Scale IQ, as an estimate of overall function, with the change scores for each of the other measures. As the full IQ was tested only in the first session, while patients were taking TPM, the IQ scores used for this analysis are surely underestimated to some extent. However, there was still a wide range of IQ ratings (59 to 112), even if the absolute numbers were shifted downward. Among the measures included in the correlation analysis were IQ subtests that were given both on and off TPM. One could expect them to correlate with the overall IQ measure, and the only significant correlation was between IQ and the Arithmetic subtest. Thus not even the other IQ subtests correlated with “overall intellectual ability,” clearly repudiating our hypothesis. This means that even for very low functioning individuals, TPM effects on cognition must be taken into consideration.
Although the time interval between patients being tested on and off TPM in Study 1 was ≥12 days, it is possible that the improved scores off TPM were due at least in some part to a practice effect. Study 2, however, refutes the idea that the improved scores off TPM were due only to practice effects, because those results show declines when the patients were tested on TPM, even though it was a second testing. Three measures were statistically significant: Digit Span, Digit Symbol, and COWAT. However, negative change was noted on all measures at the second testing (on TPM). Moreover, the results of Study 1 for those patients tested before, during, and off TPM clearly suggest that regardless of what practice effects may be occurring in multiple testing sessions, TPM depresses cognitive function, and its discontinuation alleviates it.
Additional issues have been debated regarding neuropsychological results in TPM studies. Some authors have contended that a slow titration of the drug minimizes cognitive effects, and that the cognitive deficits should resolve over time as the drug is better tolerated (20). We could not control for titration in Study 1 because our sample was of patients admitted consecutively for intensive monitoring and who were being treated with TPM on admission. However, when tested, they had been taking the drug for an average of 16 months (range, 1.5–48 months). Moreover, the fact that they were serially admitted potential surgical candidates limits the likelihood of selection bias, because all patients were taken off TPM regardless of whether they had cognitive complaints. Two patients in Study 2 could have been subject to rapid titration effects, but one had results similar to those of the rest of the patients in this study, whereas the other evidenced no significant decline. The remaining patients had an average duration of TPM treatment of ≥6 months before second testing.
There were some limitations in our study. First, TPM was administered in polytherapy together with other AEDs, which were different for each subject and consisted of an average of 1.6 other drugs in Study 1 and 2.3 drugs in Study 2. Plasma concentrations of TPM may be affected by other AEDS either positively or negatively (2), which might potentiate cognitive effects: the effects from monotherapy may impart a different profile. It also would be instructive to compare cognitive change related to TPM with those related to a different AED with the same methods as used here, and such a study is planned. Second, the average TPM dosage varied across the subjects. Factors such as age, sex, duration, or type of epilepsy may affect the action of TPM. To some extent, we have tried to account for these factors by comparing patients with themselves on and off the drug. Four of the 22 patients in Study 1 had either one or two additional drugs discontinued at the time of the second testing. The drugs removed, however, differed among patients, and it is highly unlikely that this factor could make a systematic difference in the results. Third, we were unable to analyze effects as a function of side or site of cerebral abnormality because too few of the patients had clearly enough defined seizure foci. Finally, because the study was carried out in the context of clinical evaluation, the neuropsychological testers were in most cases aware of the drug condition.
Both studies provide strong evidence that TPM is associated with a variety of adverse effects on cognition. Objective neuropsychological testing is necessary to document type and extent of cognitive effects, and patient report cannot be considered a reliable indicator of such effects. Moreover, this study reveals that TPM effects must be taken into account for individuals at all IQ levels.
We therefore recommend that when TPM therapy is prescribed for a patient, a brief cognitive screen that includes digit span, fluencies, and arithmetic should be conducted before TPM administration, and alternate equivalent versions that test those skills should be given after therapeutic levels of the drug are attained. The cognitive results should then be weighed with the improvement in seizure control to allow an informed appraisal of the overall value of the treatment.
Acknowledgment: We thank the patients who graciously agreed to undergo additional neuropsychological testing for this study, Rhonda Amsel for statistical advice, and Zia Lakdawalla. This work was supported by a Summer Research Studentship in Medicine from Rx & D Health Research Foundation to Suzee E. Lee, and by grant MT144991 from the Canadian Institutes of Health Research to M. Jones-Gotman.