Chapter  77:  Management of Treatment-Resistant Epilepsy, Volumes 1 and 2

Question 1: What are the definitions of treatment-resistant epilepsy used in the literature?

• Treatment resistance is infrequently defined in the literature. Less than one third of the surveyed publications reported any definition of this term.

• When treatment resistance was defined, definitions typically included the number of AEDs a patient tried before being considered treatment-resistant. Some definitions also included seizure frequency, duration of illness, and whether AEDs were administered at maximum tolerable doses.

• Drug trials tended to require fewer failures of AED treatment compared to surgical trials. This is because a very thorough assessment of drug regimens is usually attempted before surgery is considered. Assessments are usually less thorough when giving a patient another AED.

• Despite the fact that reports of clinical trials and review articles regularly use terms such as “intractable,” “refractory,” or “treatment-resistant” to describe patients for whom one or more treatments have failed, no consensus exists as to precisely what these terms mean.

Question 2: Which methods of rediagnosing or reevaluating treatment-resistant epilepsy lead to, or can be expected to lead to improved patient outcomes?

We partitioned this question into four subquestions. The first two subquestions addressed differential diagnosis of epileptic seizures from nonepileptic seizures. The remaining two subquestions addressed the differential diagnosis of different seizure types. Whether we addressed some questions depended on the findings for previous questions.

Question 3: Is there evidence that patients with treatment-resistant epilepsy are not optimized at their current level of treatment?

• Not all patients with treatment-resistant epilepsy receive optimized AED treatment.

• The percentage of patients with treatment-resistant epilepsy who are not receiving optimized therapy is difficult to estimate. This is because of a lack of relevant, large, population-based studies. Further, many studies of AEDs do not report whether patients comply with their AED regimens.

Question 4: Which drug treatment strategy, (A) sequential monotherapy, (B) polytherapy, or (C) optimized current therapy leads to improved outcomes for patients with treatment-resistant epilepsy, and (D) What are the relative improvements obtained with each strategy?

Based on the recommendation of the partners, for the purposes of this question, sequential monotherapy is defined as changing a patient's drug regimen from one or many AEDs to a single, different AED. Polytherapy is defined as changing a patient's drug regimen from one or many AEDs to a different multiple-AED regimen. In this report, all polytherapy trials were trials of a single add-on AED. Optimized current therapy was defined as changing the dose and/or the frequency of administration. Based on the recommendation of the partners, we also included the removal of one or more drugs within this definition.

Question 5: Which methods of nondrug treatment for epilepsy after initial treatment failure lead to improved outcomes for patients with treatment-resistant epilepsy?

Question 6: Which social, psychological or psychiatric services for treatment-resistant epilepsy lead to, or can be expected to lead to improved patient outcomes?

• There were too few acceptable studies addressing this question to permit analysis.

Question 7: What characteristics of treatment-resistant epilepsy interfere with ability to obtain and maintain employment, or attend and perform well in school?

• There were too few acceptable studies addressing this question to permit analysis.

Question 8: What is the mortality rate of patients with treatment-resistant epilepsy?

• Persons with treatment-resistant epilepsy are approximately 2 to 10 times more likely to die compared to people in the general population. This excess mortality in persons with treatment-resistant epilepsy is largest among younger individuals.

• Sudden unexpected death appears to be a major cause of death among patients with treatment-resistant epilepsy, representing 6 percent to 55 percent of the total deaths in studies that reported relevant data.

• Drowning rates are higher among treatment-resistant patients with epilepsy compared to the general population. Higher quality evidence is needed to determine the precise magnitude of the difference in drowning rates.

• There is insufficient evidence to determine whether accident-related mortality, or mortality due to pneumonia, aspiration, suicide or cancer is higher among persons with epilepsy compared to the general population.

Question 9: Is there a correlation between the number and/or type of seizure and sudden death?

• Generalized tonic-clonic seizures appear to increase the risk of sudden death.

• The relationship between overall seizure frequency and sudden death is uncertain.

Studies of diagnostics

The lack of an accepted gold standard for the differential diagnosis of epileptic seizures from nonepileptic seizures makes evaluating the utility of any given diagnostic problematic. This is because of the difficulty in verifying that the diagnostic decisions that result from the use of the test are correct. Given this lack of an acceptable gold standard, attempting to determine whether the use of a diagnostic improves patient outcomes may offer a fruitful avenue for future research. Such an approach requires determining whether the use of the diagnostic of interest ultimately leads to improved patient outcomes and, as a consequence, requires a prospective, randomized controlled trial.

Because a diagnosis of epilepsy is not made based on the findings of a single diagnostic technology, studies are needed to evaluate the effectiveness of different clinical algorithms that utilize data collected from combinations of diagnostic technologies. Again, this approach would require a prospective, randomized controlled trial.

Studies of treatment

In the literature on drug strategies, an important direction for future research involves direct comparisons between the drug strategies for treatment-resistant epilepsy. None of the studies included in our assessment of drug strategies made direct comparisons between sequential monotherapy and polytherapy. Ideally, a trial would randomize patients to different drug strategies, and compare seizure frequency outcomes as well as adverse effects of treatment.

Another area for future research on drugs concerns the adverse effects patients experience from their pretrial drug regimens and changes in these adverse effects on the new treatment regime. Changes in the frequency and severity of the adverse effects associated with each drug treatment strategy need to be evaluated, because patients and clinicians seek to reduce adverse effects as well as seizure frequency.

Prospective studies of surgical interventions are needed. This approach would allow seizure and nonseizure-related outcome measures to be recorded at multiple followup periods (1 year, 2 year, 5 year, etc.) rather than the single mean or median followup reported in most retrospective studies. Better reporting of patient characteristics is also needed and, if possible, individual patient characteristics should be reported when study sizes are small (less than 20 patients). Studies reporting standardized quality of life measures, validated for patients with epilepsy, would help in determining the effect of surgery on this important nonseizure-related outcome. Studies reporting other types of nonseizure-related outcome measures, such as employment, education, and cognitive function data, are also needed.

Higher quality controlled trials are particularly lacking for the nonmedical treatments such as education and training in skills that may help prevent seizures or enable patients to better adapt to seizures. This area constitutes another important direction for future research.

Studies of patient characteristics related to employment and school

Reporting of employment and schooling status among patients with treatment-resistant epilepsy is particularly lacking in both the medical and nonmedical treatment literature. The ideal study design to address this question would be a prospective cohort study using multiple regression techniques to evaluate the potential correlation between specific patient characteristics and the ability to work or attend school both before and after treatment. This is an area in particular need of future research and higher quality studies.

Studies of mortality

The present literature has a number of large (mostly retrospective) studies that have calculated standardized mortality rates (SMRs) for overall mortality, but few studies have calculated separate SMRs for specific causes of death or subgroups of specific ages. To generate meaningful data, cohort studies must enroll sufficient numbers of patients and follow the patients for sufficient periods. The most useful study of mortality among patients with treatment-resistant epilepsy would be a large prospective study that followed patients for several years. In addition to calculating an SMR for overall mortality, the study would calculate SMRs for specific causes of death, especially those that could be related to epilepsy (such as accidents, drowning, and motor vehicle accidents).

Large prospective studies where all suspected sudden unexpected death in epilepsy (SUDEP) cases receive an autopsy are needed. An autopsy is particularly important because it provides the best evidence that the death did not have an explainable cause. This would increase the accuracy of estimates of SUDEP rates for different age subgroups of patients with treatment-resistant epilepsy.

More prospective case-control studies using multiple regression analysis would be useful to address the potential relationship between SUDEP and seizure type or frequency. Future studies would ideally include a hundred patients or more to ensure that there is adequate statistical power to detect correlations. Multiple regression analysis is needed to reduce the effect of possible confounding variables and increase the likelihood that an observed statistically significant correlation represents an actual causal relationship.

Neurobiology

An epileptic seizure can be defined clinically as an intermittent, stereotyped, disturbance of consciousness, behavior, emotion, motor function, or sensation that results from abnormal cortical neuronal discharge and recur without provocation.⁴ The discharge may result in an almost instantaneous loss of consciousness, alteration of perception or impairment of psychic function, convulsive movements, disturbance of sensation, or some combination of these events.¹ The onset of the abnormal neuronal discharge may be widespread and bilateral or it may be localized. In the latter instance, the abnormal discharge arises from an assemblage of excitable neurons, called a focus, in any part of the cerebral cortex that may or may not be associated with a visible lesion.^a Cortical excitation may then spread to the adjacent cortex and to the contralateral cortex through interhemispheric pathways as well as to subcortical areas such as the basal ganglion, thalamus, and brainstem. Clinical manifestations of a seizure occur when the excitation reaches these areas. In rare instances, death may occur due to sustained cessation of respiration, derangement of cardiac action, or some unknown cause.

Etiology and Pathology of Epilepsy

The focal cortical lesions responsible for the abnormal discharges associated with epileptic seizures may arise from a variety of causes. Epilepsies in which no pathological lesion can be found are referred to as primary or idiopathic epilepsies and include certain generalized tonic-clonic and absence seizure conditions.¹ The underlying cause of these seizure types is probably genetic. Secondary or symptomatic epilepsies are associated with a discernable lesion. Secondary epilepsies include simple partial seizures and complex partial seizures. The lesions may be zones of neuronal loss and scarring (sclerosis or gliosis), vascular malformations, tumors, or cortical dysplasia. Many patients with temporal lobe epilepsy have a condition called mesial temporal sclerosis (MTS) that is characterized by a loss of volume and scarring in the hippocampus and adjacent gyri on one or both sides. Posttraumatic epilepsy may occur after head trauma, brain surgery, and various infections that are responsible for creating focal lesions that result in epilepsy.

Signs, Symptoms, and Characteristics of Epilepsy and Treatment-Resistant Epilepsy

Epilepsy seizures may be classified according to etiology, site of origin, clinical form, frequency, or electrophysiologic characteristics.¹ Classification schemes have repeatedly changed, but the most commonly used scheme is the one adopted by the Commission on Classification and Terminology of the International League Against Epilepsy.⁵ This classification is based primarily on the clinical form of the seizure and its electroencephalographic (EEG) features. Seizures are divided into partial seizures (a focal or localized onset can be discerned) and generalized seizures (bilateral origin and diffuse cerebral cortical involvement from the onset). Partial seizures that develop into generalized seizures are referred to as secondarily generalized seizures. Partial seizures (also called focal seizures) are further classified as simple when consciousness is maintained, and complex if consciousness is altered or lost. Simple partial seizures can be motor, sensory, autonomic, or psychic. Simple partial seizures are also called auras and may be a precursor to a complex seizure or may constitute the entire seizure. Generalized seizures may be convulsive or nonconvulsive. The common convulsive type is the tonic-clonic seizure and the common nonconvulsive type is the absence seizure that is characterized by a brief lapse of consciousness.

Patients with treatment-resistant epilepsy may experience one or more of the various types of epileptic seizures depending on the etiology of their condition. The following is a brief description of the clinical characteristics of each type of epileptic seizure that a patient with treatment-resistant epilepsy may experience. Adams and Victor's Principles of Neurology was used a guide in preparing these descriptions.¹

Generalized tonic-clonic seizures are sometimes preceded by subjective phenomena (prodromes) that may take the form of psychological changes or myoclonic jerks of the trunk or limbs. Most often, the seizures occur without warning and the patient loses consciousness and falls to the ground. The seizure starts with an initial flexion of the trunk, opening of the mouth and eyelids, and upward deviation of the eyes. The tonic phase follows with protracted extension of the back and neck and then the arms and legs. Breathing is also impaired. The tonic phase lasts for 10 to 20 seconds. The clonic phase follows the tonic phase with a mild generalized tremor that represents relaxation of the tonic contractions. The rate of contractions gradually lessens over 30 seconds. Breathing is still impaired until the end of the clonic phase. Finally, all movements end. Upon regaining consciousness, the patient usually experiences a period of disorientation and/or fatigue.

Absence seizures are very brief, always associated with impaired consciousness, and may or may not have accompanying abnormal motor activity. Some are so short as to resemble daydreaming. The seizure comes without warning and consists of a sudden interruption of consciousness. These patients usually do not fall. After 2 to 10 seconds the patients become conscious again and resume their preseizure activity. Absence seizures are the most common form of epileptic seizure of childhood and rarely begin before 4 years of age or after puberty. Absence seizures tend to occur frequently, with as many as several hundred occurring in a single day. This type of seizure may be the only type seen in childhood and the attacks diminish in frequency with age and eventually disappear. However, absence seizures may be replaced in some instances by generalized tonic-clonic seizures.

Partial seizures are the product of focal abnormalities in some part of the cerebral cortex. Such lesions are often associated with focal EEG abnormalities. Simple partial seizures most often derive from a focal area in the sensory and motor cortex, while complex partial seizures most often derive from the temporal lobe of one side. Somatosensory seizures usually have a focus in the precentral or postcentral convolution. The sensation is described as numbness or tingling that usually starts in the lips, fingers, or toes and spreads to adjacent parts of the body. Visceral seizures (vague feelings in the thorax and abdomen) are some of the most frequent simple partial seizures. Other less common types of simple partial seizures include visual seizures (sensation of darkness and flashes of light), auditory hallucinations (buzzing or roaring in the ears), vertiginous sensations, and olfactory hallucinations (abnormal odors).

Loss of consciousness distinguishes complex partial seizures from simple partial seizures. However, unlike generalized tonic-clonic seizures, the patient suffers a period of altered behavior and consciousness with no later recollection, instead of a complete loss of control of thought and action. Any type of complex partial seizure may develop into other forms of secondary generalized seizures. Complex partial seizures are not exclusive to any age group but show an increased incidence in adolescence and adult years.

The Lennox-Gastaut Syndrome is a neurological syndrome characterized by frequent seizures of various types. These patients have intellectual impairment and serious neurologic disease. The syndrome is usually diagnosed between 2 and 6 years of age.

Temporal lobe epilepsy, which is typically manifested by complex partial seizures described above, commonly starts before 10 years of age. By adolescence or early adulthood, this condition often becomes treatment-resistant.⁶ Recurrent seizures may cause damage in the hippocampus resulting in a progressive loss in hippocampal volume as long as the seizures persist.⁷

Arroyos, Brodie, Avanzini, et al.³ suggest that treatment-resistant epilepsy may be a distinct condition within epilepsy characterized by progressive neuronal, cognitive, and psychosocial deterioration. These authors point out that several markers have been advanced as indicators of treatment-resistant epilepsy. These include age at onset younger than 1 year, type of epilepsy (partial epilepsies or catastrophic epilepsies of childhood), failure of the first AED, use of more than two drugs, duration of treatment without achieving control, and specific pathologies. However, these markers may not be especially sensitive or specific for treatment-resistant epilepsy.³

Epidemiology of Epilepsy

Epilepsy is among the most common serious neurologic condition, with a prevalence rate 10 times higher than multiple sclerosis and 100 times higher than motor neuron disease.⁸ An International League Against Epilepsy (ILAE) Commission Report from 1997 gives the prevalence of active epilepsy as 40 to100 in 10,000.⁹ The ILAE prevalence and incidence rates are based on a review of selected studies, mostly from developed countries, by Sander and Shorvon .¹⁰ Both rates varied widely among the studies included in this review. Their review found prevalence rates that ranged from 1.5 to 31 per 1,000 and incidence rates that varied from 11 to 134 per 100,000. The authors believe that the highest prevalence and incidence rates are more accurate because studies with these findings used more intensive and sophisticated case ascertainment methods. They further state that the low rates found in some studies were probably due to deficiencies in patient reporting (physical manifestations are transient and not observed by a clinician, patients are unaware of or deny their condition) and in diagnosis of epilepsy (syncope and psychogenic seizures are misdiagnosed as epilepsy, diagnostic criteria are unspecified or loosely defined). Sander and Shorvon¹⁰ also believe that retrospective review of patient medical records in most case ascertainment studies also leads to an underestimate of prevalence and incidence.

Burden of Illness

In addition to the immediate effects of seizures, patients with epilepsy may also have to contend with the following burdens:^{4, 11, 12}

• interference with normal activities

• increased possibility of accidental injury and even death (addressed in Key Question 8 in this report)

• impaired physical, psychological, and social functioning

• economic loss

• diminished quality of life

• psychiatric disorders, in particular, anxiety and depression¹³

Electronic Database Searches

To obtain information for this report, we systematically searched 23 electronic databases. These were:

• Center for International Rehabilitation Research Information and Exchange (CIRRIE) (searched November 30, 2001)

• Cochrane Database of Systematic Reviews (through 2001, Issue 4)

• Cochrane Registry of Clinical Trials (through 2001, Issue 4)

• Cochrane Review Methodology Database (through 2001, Issue 4)

• Cumulative Index to Nursing and Allied Health (CINAHL) (1988 through January 11, 2002)

• Database of Reviews of Effectiveness (Cochrane Library) (through 2001, Issue 4)

• ECRI Health Devices Alerts (1977 through January 2002)

• ECRI Health Devices Sourcebase (through January 2002)

• ECRI Healthcare Standards (1975 through January 2002)

• ECRI International Health Technology Assessment (IHTA) (1990 through January 2002)

• ECRI Library Catalog (through January 2002)

• ECRI TARGET (through January 2002)

• Embase (Excerpta Medica) (1975 through January 2002)

• ERIC (Educational Resources Information Center) (searched January 8, 2002)

• Health and Psychosocial Instruments (HAPI) (through April 27, 2001)

• LocatorPlus (through January 2002)

• NDA Pipeline (searched November 1, 2001)

• PsycINFO (1975 through January 31, 2002)

• PubMed® (MEDLINE®, PreMEDLINE®, HealthSTAR) (1975 through January 2002)

• Rehabdata (searched April 24, 2001)

• U.K. National Health Service (NHS) Economic Evaluation Database (EED) (through January 2002)

• U.S. Centers for Medicare & Medicaid Services (CMS) (formerly HCFA) (through January 2002)

• U.S. National Guidelines Clearinghouse^TM (NGC) (through January 2002)

Search Strategies

We employed different searches for different sections of the report, including different searches for different questions. The strategies for these different searches, given in PubMed® /MEDLINE® syntax, are provided in Appendix A.

Other Sources

In addition to the above searches, we also reviewed the bibliographies and reference lists of all studies included in this Evidence Report, and searched Current Contents—Clinical Medicine® on a weekly basis.

Meta-Analyses

We performed meta-analyses of data from RCTs and uncontrolled trials. We performed meta-analyses of RCTs to estimate an average effect of treatment. We performed meta-analytic threshold analyses of uncontrolled studies to determine whether an intervention was plausibly effective. All meta-analyses, except meta-regressions, were performed with software programs developed by ECRI. This software has been extensively validated using published examples and hand calculations. Meta-regressions were performed using SPSS (version 10.1) Statistical Software (Copyright © SPSS Inc., 1989-2000).

Studies of Interventions

Our evaluation of the quality of interventional studies employed three tools. The first was an evidence hierarchy we used to determine which studies to retrieve, include and, in certain questions, to determine whether there was greater potential for bias in some included studies than in others. The second tool we used was a checklist for evaluating each study's internal validity. Finally, we considered the difficulties inherent in certain outcome measurements.

Studies of Diagnostics

The biases that can affect studies of diagnostics are different from those that can affect studies of interventions. The checklist we employ for determining whether a diagnostic study was potentially affected by a bias incorporated items suggested by Lijmer, Mol, Heisterkamp, et al.,²⁴ Irwig, Tosteson, Gastsonis, et al.,²⁵ Gann,²⁶ Begg and Greenes,²⁷ and Ransohoff and Feinstein.²⁸ These biases are:

Studies of Mortality

Two questions in this Evidence Report, Questions #8 and #9, concern epilepsy-related mortality rates. Studies examining mortality rates may be biased by factors that are different from those that bias the results of other kinds of studies. Therefore, we examined studies of mortality for the following potential biases:

External Validity

We evaluated each study's external validity (generalizability) according to patient characteristics appropriate to each question. These characteristics are provided as we address each question. We did not evaluate external validity when evidence-based conclusions could not be reached.

Evidence Tables vs. Tables

The results of our analyses of internal and external validity and of our meta-analyses are presented in two types of tables. Evidence Tables contain detailed information on each of the studies used in an assessment and the results of meta-analyses of these studies. The Evidence Tables tend to be large and are therefore contained in a separate volume of this report. They are organized according to the Key Questions addressed in this report. Other tables appear in the Results chapter following the discussion of each intervention being assessed. These tables are intended to provide a brief listing and description of the studies and the outcomes reported in these studies. Tables addressing the internal validity of studies used in an assessment are presented in Appendix B.

Figures

Figures are also presented in the Results chapter after the discussion of each intervention assessed in this report. Figures are designed to present summary information in the form of a forest plot (array of study effect sizes usually with a summary estimate), graphs of threshold analyses and meta-regressions, or other appropriate graphical presentations.

Internal Review

Throughout the preparation of this report, the five analysts and the Project Manager held numerous meetings to determine the strategy and methods of analysis. The Project Manager then individually reviewed each completed section of the report, and suggested changes. Upon completion of these changes, the individual sections were assembled into an initial draft report that was again reviewed by the Project Manager. Subsequent to changes made in response to this draft, it was distributed to the five analysts in the project team for review. Suggested changes were reviewed by the Project Manager and discussions were held among the project team to determine which suggestions would be incorporated. Upon incorporation of the appropriate changes, the draft report was sent for external review.

External Review

To select peer-reviewers for the draft Evidence Report, ECRI prepared a list of 27 potential reviewers. This list was submitted to AHRQ, which approved all reviewers. Letters inviting these individuals to review the draft report were then mailed. Twenty-five individuals responded to these letters, 19 agreed to review the draft Evidence Report, and nine individuals returned reviews.

Upon receipt of reviews, ECRI revised the draft report accordingly. ECRI also prepared a document describing the disposition of all substantive reviewer comments and supplied this document to AHRQ for review and approval.

Question specific inclusion criteria

As noted in the Methodology section, the general inclusion criteria listed in that section do not apply to this question. Rather, we included:

1. Any clinical study that was evaluated in Questions 2 to 6, that enrolled at least 50 patients, and that was published in 1996 or later. All such studies meeting the initial inclusion criteria for each question were included, regardless of whether they were later excluded from the analysis of that question. We abstracted definitions from clinical studies in an effort to obtain a broad sample of definitions. We did not include articles retrieved for Key Questions 7 – 9, because the nature of these studies made them less likely to include definitions of treatment-resistant epilepsy.

2. A random sample of 100 review articles on treatment-resistant epilepsy published between 1996 and 2001, inclusive. We chose this random sample by using a random number generator to assign a random number to each of the 298 review articles identified in our searches for this Evidence Report. We chose to use a random sample rather than a comprehensive dataset out of consideration of the time and budget for this project.

3. Any evidence-based clinical practice guidelines identified during our searches. We termed a guideline as “evidence-based” if it was included in the National Guidelines Clearinghouse (NGC).^c

Evidence base

Table 3. Definitions of treatment resistance

Table 3. Definitions of treatment resistance

Source	Number of Publications Selected for Question	Number of Publications Reporting Definitions	Percentage of Publications Reporting Definitions
Question 2	6	3	50%
Question 3	10	4	40%
Question 4	26	7	27%
Question 5	39	9	23%
Question 6	1	1	100%
Research Articles Total	82	24	29%
Treatment Guidelines	3	1	33%
Review Articles	100	21	21%
Grand Total	185	46	25%

Design and conduct of included studies

This question addresses definitions, not an intervention or diagnostic. As such, an evaluation of the quality of the literature is not relevant.

Definitions in Included Articles

Of the 82 clinical studies that met our inclusion criteria for this question, only 24 (29 percent) reported an explicit definition of “intractable”, “refractory”, “treatment-resistant,” or any similar term. The remainder merely stated that the patients they enrolled had treatment-resistant epilepsy (or some equivalent term) without defining that term. Of the 24 articles reporting a definition, five definitions did not include any specific information (e.g. “incompletely controlled by existing therapy”).³⁰ One study defined treatment resistance in terms of seizure frequency with no mention of treatment.³¹

Of the remaining 19 studies, 15 reported the number of AEDs patients tried before being considered treatment-resistant. Two studies required at least one AED, four required at least two, and four required three. Five were nonspecific (e.g. “multiple”). Six of the studies named the AEDs that they required patients to have tried before being considered treatment-resistant.

Definitions in Clinical Practice Guidelines

Definitions in Review Articles

Of the 100 review articles surveyed, 79 articles defined treatment-resistant epilepsy as the presence of uncontrolled seizures or a similar term. Two of the remaining definitions were not specific.^{33, 34} Of the remaining 19 reviews, eight reported the number of AEDs patients tried before being declared treatment-resistant. Three of the eight reviews required at least two AEDs and three required three. Two of the eight reviews were not specific (e.g. “several”). Only one of the reviews named the AEDs that they required patients to have tried before being considered treatment-resistant.

Twelve reviews mentioned intolerable side effects or ineffectiveness at maximum tolerated dose as a reason to consider drug treatment unsuccessful. Four reviews mentioned frequent seizures as part of their definition, but none of these quantified what was meant by “frequent.” Rather, their effect on the ability of the patient to lead a normal life was considered the proper criterion in three of these four reviews.

Definitions Implied by Inclusion Criteria and Patient Characteristics in Clinical Studies

Because definitions were infrequently reported, we examined the inclusion/exclusion criteria and the characteristics of patients in clinical studies to determine the characteristics of patients deemed to have treatment-resistant epilepsy. These characteristics may imply a definition. However, the requirements of a trial are not necessarily the same as the requirements of a patient seeking treatment. A patient experiencing one seizure a year may be considered treatment-resistant but is unlikely to be included in a clinical trial. Thus, patient inclusion criteria may be biased toward enrolling more severely ill patients.

Of the 82 clinical studies included, eight specifically examined pediatric patients, two focused on Lennox-Gastaut syndrome, while two examined mesial temporal sclerosis (MTS), and one examined non-MTS focal lesions.

Figure 2. Minimum number of AEDs: different patient (more...)

   Figure 2. Minimum number of AEDs: different patient types

*“>1” indicates that the study used a nonspecific term such as “several”

Figure 3. Minimum number of AEDs: different treatments (more...)

   Figure 3. Minimum number of AEDs: different treatments

*“>1” indicates that the study used a nonspecific term such as “several”

Figures 2

3

In those studies that reported a minimum number of AEDs, the majority required at least one AED. This proportion did not differ dramatically from the proportion in studies of pediatric patients or studies in which there was no special patient group.

All FDA drug studies and most non-FDA drug studies reported that a minimum number of AEDs must have been tried without success before a patient was considered treatment-resistant. In both cases, the minimum number was nearly always one. Most studies of surgery (80 percent) did not report a minimum number of AEDs that had been tried. However, when a number was reported, it was always greater than one. This difference between drug and surgical trials probably reflects differences in trial qualifications rather than differences in definitions of treatment resistance.

Figure 4. Minimum baseline seizure frequency: different (more...)

   Figure 4. Minimum baseline seizure frequency: different patient types

Figure 4

Figure 5. Minimum baseline seizure frequency: different (more...)

   Figure 5. Minimum baseline seizure frequency: different treatments

Figure 5

Do all patients diagnosed with epilepsy that is deemed to be treatment-resistant truly have epilepsy?

To address Question 2A, we looked for studies that attempted to estimate the prevalence of patients with nonepileptic seizures among populations of patients with a diagnosis of treatment-resistant epilepsy. These nonepileptic seizures may have been the sole seizure type experienced by a patient (in which case the patient was misdiagnosed), or they may have occurred in addition to true epileptic seizures (in which case the patient was correctly diagnosed with epilepsy but the additional diagnosis describing the nonepileptic seizures was missed). In the former case, patients would not be expected to respond satisfactorily to treatment with AEDs. In the latter case, the epileptic seizures may be well controlled by AEDs, and the seizures experienced by the patient are nonepileptic in nature. In either case, such patients would, unless given a new diagnosis, remain incorrectly labeled as exclusively having treatment-resistant epilepsy.

Evidence base

Table 4. Evidence base for determining if patients (more...)

Table 4. Evidence base for determining if patients diagnosed with treatment-resistant epilepsy actually have epilepsy

Reference	Study Design	Country in Which Study Performed	Size (N)	Multicenter	Number of Centers
Zaidi (2000)38	Cross-sectional case series	United Kingdom	74	Yes	2a
Holmes (1998)39	Cross-sectional case series	United States	379	No	1
Henry (1997)37	Cross-sectional case series	United States	145	No	1
Arnold (1996)40	Cross-sectional case series	United States	45	No	1
Slater (1995)41	Cross-sectional case series	United States	101	No	1

a

Patients enrolled at two centers but diagnostic reassessment was performed at a single study center

Five articles met both the general inclusion and the question-specific inclusion criterion presented above. These five articles are listed in Table 4. Details of these studies are presented in Evidence Tables 6 through 9.

All five articles in the evidence base were cross-sectional, case series. In these studies, a series of patients (total N = 744) were given diagnostic reassessment in order to determine the prevalence of patients with nonepileptic seizures among specific subgroups of patients, all of who were considered, prior to reassessment, to have treatment-resistant epilepsy.

Four of the five articles included in Table 4 described studies that were carried out at a single center. The remaining article described a study in which patients were recruited at two different centers. However, all patients in this latter study had their diagnosis reassessed at a single study center by a single diagnostic team.

Synthesis of study results

Figure 6. Prevalence of nonepileptic seizures

   Figure 6. Prevalence of nonepileptic seizures

Prevalence of nonepileptic seizures among patients diagnosed with treatment-resistant epilepsy

NES: Non-epileptic seizure

ES: Epileptic seizure

The prevalence of nonepileptic seizures among the patients in each of the five studies used to address Question 2A are presented in Evidence Table 9 and are summarized in Figure 6. This figure demonstrates that between 8.3 percent and 37.6 percent of patients believed to have treatment-resistant epilepsy turned out to either not have epilepsy or to suffer from a combination of both epileptic and nonepileptic seizures. With the exception of the study of Henry and Drury,³⁷ the majority of these patients suffered from nonepileptic seizures alone. Only a small proportion of patients had a combination of both epileptic and nonepileptic seizures (Range: 0 percent to 1.0 percent).

The patients examined in the study of Henry and Drury³⁷ were undergoing presurgical evaluation. Such patients were probably assessed more often and/or more completely by clinicians who specialize in epilepsy compared to most other patients with a diagnosis of treatment-resistant epilepsy. Consequently, that no patients in their sample suffered from nonepileptic seizures alone is not a surprise. However, that 8.3 percent of the patients in this study experienced a combination of epileptic and nonepileptic seizures is surprising. Followup of these patients revealed that the changes in patient management that resulted from the reassessment led to a complete cessation of seizures in three patients (25 percent). Whether these three patients would have been identified prior to surgery had the study not been performed is unknown. However, they may possibly have undergone unnecessary surgery.

To determine an overall estimate of the prevalence of patients with nonepileptic seizures among patients who, prior to re-evaluation, were considered to have treatment-resistant epilepsy, we performed a meta-analysis. This meta-analysis did not include prevalence data abstracted from the study of Henry and Drury³⁷ for reasons explained above. The results of this homogenous (Q = 0.28; p = 0.96439) fixed-effects meta-analysis are presented in Evidence Table 10.

This meta-analysis shows that the proportion of patients who were misdiagnosed as having treatment-resistant epilepsy was substantial (35 percent; CI: 29 percent to 41 percent). Therefore, a problem of misdiagnosis clearly exists in clinical practice. However, these findings do not accurately represent the proportion of misdiagnosed patients in the overall population of patients with treatment-resistant epilepsy because none of the studies included in the meta-analysis were population-based. Also, the patients included in the four studies that we did meta-analyze represent a subpopulation of patients referred for specialist evaluation of their seizures.

Which diagnostic modalities are useful in differentiating seizure types commonly mistaken for epilepsy from true epileptic seizures?

Based on the results of Question 2A, we addressed Question 2B by evaluating the evidence on the diagnostic technologies most commonly used to differentiate epileptic seizures from nonepileptic seizures. As stated above, in clinical practice, the differential diagnosis of epileptic seizures is usually based on information from many sources (medical history, etc.). The clinical diagnosis is seldom based on one diagnostic technology alone. Ultimately then, to answer this question, diagnostic performance data from each analysis of each individual diagnostic technology must be combined into a single decision model which better describes the true clinical picture. As will be seen, a paucity of available evidence precluded the construction of such a model. Thus, we were limited to an analysis of the clinical utility of individual “stand alone” diagnostic technologies.

Question specific inclusion criteria

In addition to employing the general inclusion criteria, we included articles if they met the following criteria:

1. The study must have evaluated the effectiveness of a diagnostic technology used for the differential diagnosis of epileptic seizures from nonepileptic seizures.

2. The patients enrolled in the study were not restricted to only those with treatment-resistant epilepsy. Because the intent of Question 2B is to determine the utility of those diagnostic technologies that have been used to differentiate epileptic seizures from nonepileptic seizures, addressing this question requires a study that enrolls both kinds of patients.

3. The study must have either reported diagnostic test performance characteristics (e.g., sensitivity and specificity) or presented data in a format that allows calculation of test performance characteristics based on a comparison with some “reference” standard.^d Alternatively, the study must have included followup data that allow conclusions about the effects of using a diagnostic on patient outcomes.

Number of articles addressing each diagnostic

Table 5. Articles addressing each diagnostic

Table 5. Articles addressing each diagnostic

Diagnostic	Number of Articles
Blood Prolactin Levels	8
Minnesota Multiphasic Personality Inventory	6
Video-EEG	6
Ambulatory-EEG	5
Provocation techniques	4
Routine EEG	4
Creatinine kinase levels	3
Tilt table	2
Auditory evoked potentials	1
Hypnotic recall	1
Magnetic Resonance Imaging	1
Single Photon Emission Computed Tomography	1
Tongue biting	1
Computed Tomography	0

Forty-three articles met the inclusion criteria for Question 2B. The numbers of articles that address each of the diagnostics meeting the inclusion criteria are presented in Table 5. A full list of articles and the diagnostics that they addressed are presented in Evidence Table 11.

The most common type of excluded article reported on a case-series study in which a group of patients was diagnosed with a given modality, and this diagnosis was then used to influence medical management. However, none of these studies reported whether these management changes led to improvements in patient outcomes. Although this study design is seen by some as being a legitimate design for the assessment of a diagnostic,⁴² this assumes that the diagnostic test was accurate. Requiring the assumption that a diagnostic be accurate in order to assess the accuracy of that same diagnostic is circular reasoning. It also assumes perfect sensitivity and specificity of the test, which is not possible. This sort of study design is particularly common in the literature on EEG technologies (i.e. routine EEG, ambulatory-EEG, and video-EEG).

Of the fourteen diagnostics considered, only four (blood prolactin levels, the MMPI, video-EEG, and ambulatory-EEG) were addressed by five or more studies. As per the general inclusion criteria specified in the Methodology section, data about diagnostics not addressed by at least five studies are not considered further in this report. Consequently, we do not include further information about provocation techniques, routine EEG, creatinine kinase levels, tilt tables, auditory evoked potentials, hypnotic recall, MRI, SPECT, tongue biting, or CT in this report.

Blood Prolactin Level Monitoring

Interest in the use of blood prolactin levels as a diagnostic tool for the differentiation of epileptic seizures from nonepileptic seizures began in 1976 when Ohman, Walinder, Balldin et al.⁴³ reported that blood prolactin levels rose following epileptic seizures induced by electroconvulsive therapy. Blood levels of prolactin were found to peak approximately 30 to 40 minutes after the seizure occurred and then decline to normal preseizure levels. These findings were confirmed by Trimble⁴⁴ who demonstrated that blood prolactin levels increased following spontaneous epileptic seizures and noted that blood prolactin levels did not rise following a nonepileptic, psychogenic seizure (also known as a pseudoseizure, hysterical seizure, or psychological seizure). Since then, a number of reports have been published that have assessed the relationship between blood or plasma prolactin levels in patients with epilepsy and a number of different nonepileptic seizure types.^44–57 In the following section of the report, we assess the evidence related to the value of measuring blood prolactin levels in differentiating epileptic seizures from nonepileptic seizures.

Synthesis of study results

The assessment of study quality presented above indicates that, given the present evidence, definitive conclusions cannot be drawn about whether blood prolactin level measurements have a useful role in differentiating epileptic seizures from nonepileptic seizures. Acknowledging this, we have instead evaluated the available data with the aim of determining the plausibility of blood prolactin measurements having a role in differentiating epileptic seizures from nonepileptic seizures.

Table 6. Differential diagnoses of seizures

Table 6. Differential diagnoses of seizures

	Epileptic Seizure Type					Nonepileptic Seizure Type
Reference	Mixed ES	GTCS	CPS	SPS		Mixed NES	PsyS	SynS
Lusic (1999)53					From
Anzola (1993)45					From
Zelnik (1991)56					From
Mishra (1990)57					From
Wroe (1989)50					From
Lusic (1999)53					From

CPS Complex partial seizure

ES Epileptic seizure

FS Febrile seizure

GTCS Generalized tonic-clonic seizure

NES Nonepileptic seizure

PsyS Psychogenic seizure

SPS Simple partial seizure

SynS Syncopal seizure

Not all of the studies in the present evidence base evaluated the ability of blood prolactin level measurements to differentiate epileptic seizures from the same type of nonepileptic seizure. The specific differential diagnoses assessed by each of the studies included in this section of the report are presented in Table 6.

These studies primarily assessed the ability of blood prolactin level measurements to differentiate several epileptic seizure types (mixed seizures, generalized tonic-clonic seizures, complex partial seizures, and simple partial seizures) from two paroxysmal seizure disorders that are often misdiagnosed as epileptic. These two nonepileptic seizure types were syncopal seizures and psychogenic seizures. Thus, the findings of this assessment are not applicable to the differentiation of epileptic seizures from any other nonepileptic seizure type.

Differentiating epileptic seizures from syncopal seizures

This section summarizes the findings of the studies that reported on the diagnostic utility of blood prolactin level measurement in differentiating epileptic seizures from syncopal seizures. Three of the five studies (Anzola,⁴⁵ Lusic, Pintaric, Hozo, et al.,⁵³ and Zelnik, Kahana, Rafael, et al.⁵⁶) presented data on this differentiation. Of these, two (Lusic, Pintaric, Hozo, et al.⁵³ and Anzola⁴⁵) presented dichotomous diagnostic performance data that allowed sensitivity, specificity, PPV and NPV of the test at a predetermined threshold to be directly determined. These terms are defined in Evidence Table 19. Data from these studies are presented in Evidence Table 20.

Typically, diagnostic performance data is captured by Receiver Operator Characteristic (ROC) curves that describe the trade off between sensitivity and specificity. If enough studies report appropriate data, the data can be meta-analytically combined into a single summary ROC (SROC) curve. Because the present data set consisted of only three studies, we have not attempted such a meta-analysis. As mentioned earlier, Zelnik, Kahana, Rafael, et al.⁵⁶ did not present their diagnostic performance data in a typical 2 by 2 format. Instead, they summarized their data in the form of mean (and standard deviation) blood prolactin levels.

Because the present data could not be meta-analyzed, and because ROC curves synthesized from the available continuous data sets may be misleading, we have instead summarized these data as the effect size, Hedges' d. Although this effect size cannot be used to describe the diagnostic performance of blood prolactin measurements, it does allow diagnostic data to be compared and contrasted among studies that reported this data in different formats.

Figure 7. Blood prolactin: discrimination between epileptic (more...)

   Figure 7. Blood prolactin: discrimination between epileptic and syncopal seizures

The data from the three studies reporting the diagnostic utility of blood prolactin level measurement in differentiating epileptic seizures from syncopal seizures is summarized in Figure 7. The effect sizes with confidence intervals that overlap zero indicate that the diagnostic test did not discriminate epileptic seizures from syncopal seizures any better than chance. Thus, data from Lusic, Pintaric, Hozo, et al.⁵³ did not show that blood prolactin level measurements were useful in differentiating epileptic seizures from nonepileptic seizures. The studies by Anzola⁴⁵ and Zelnik, Kahana, Rafael, et al.,⁵⁶ however, both found that the test did discriminate these two seizure types from one another statistically significantly better than chance.

Figure 8. Differences in threshold when evaluating (more...)

   Figure 8. Differences in threshold when evaluating test performance in studies of blood prolactin measurement

Exploration suggests that Lusic, Pintaric, Hozo, et al.⁵³ did not find that the test was significantly better than chance because the performance characteristics data were collected at a threshold that was not optimal for the test. This is illustrated in Table 5, which shows three point estimates plotted in ROC space. These point estimates, along with their confidence intervals, came from the dichotomous data presented by Lusic, Pintaric, Hozo, et al.⁵³ and Anzola⁴⁵ (Evidence Table 20). All three point estimates can conceivably originate from a single underlying ROC curve (Figure 8).^e However, because Lusic, Pintaric, Hozo, et al.⁵³ chose to use a lower threshold compared to any of the thresholds used by Anzola⁴⁵, the point estimate falls nearer the chance line. Thus, given the available data, blood prolactin measurements may plausibly provide information that aids in differentiating epileptic seizures from syncopal seizures. Further data are required, however, before stating that this test performs well enough to be used in actual clinical practice.

Differentiating epileptic seizures from psychogenic seizures

Two of the five included studies attempted to use blood prolactin levels to differentiate epileptic seizures from psychogenic seizures. These data are presented in Evidence Table 20 and 21. Wroe, Henlet, John et al.⁵⁰ presented dichotomous diagnostic performance data that allowed the sensitivity, specificity, PPV and NPV of the test at a predetermined threshold to be directly determined. Mishra, Gahlaut, and Kumar⁵⁷ presented summary statistics (means, standard deviations, etc.) that describe the distributions of blood prolactin levels in the two diagnostic groups.

Figure 9. Blood prolactin: discrimination between epileptic (more...)

   Figure 9. Blood prolactin: discrimination between epileptic and psychogenic seizures

We summarized the available data from these two studies in terms of Hedges' d. This summary, shown in Figure 9, suggests that blood prolactin measures can plausibly provide information that aids in differentiating epileptic seizures from psychogenic seizures.

Figure 10. Blood prolactin: discrimination between (more...)

   Figure 10. Blood prolactin: discrimination between different epileptic seizure types and psychogenic seizures

Data abstracted from Mishra (1990), GTCS: Generalized tonic-clonic seizures, CPS: Complex partial seizures, SPS: Simple partial seizures

In addition to the data presented in Figure 9, Mishra, Gahlaut, and Kumar⁵⁷ also presented data about the effectiveness of blood prolactin level measurement in differentiating three types of epileptic seizures (generalized tonic-clonic seizures; complex partial seizures; and simple partial seizures) from psychogenic seizures. These data, presented in Figure 10, suggest that while blood prolactin level measurements may be of some use in differentiating generalized tonic-clonic seizures and complex partial seizures from psychogenic seizures, the test appears to have little or no value in differentiating patients with simple partial seizures from those with psychogenic seizures. The data from Mishra, Gahlaut, and Kumar⁵⁷ show that blood levels of prolactin do not increase following simple partial seizures. Thus, blood prolactin levels will probably have little value in differentiating simple partial epileptic seizures from syncopal seizures as well.

Minnesota Multiphasic Personality Inventory

The Minnesota Multiphasic Personality Inventory-2 (MMPI-2) and its predecessor (MMPI) instruments are among the most widely used and widely researched tests of adult psychopathology. These instruments provide a broad psychological profile across a number of domains. Some investigators believe that the psychological profile of a patient with psychogenic seizures (either alone or in combination with epileptic seizures) may be different from that of patients with epileptic seizures alone.

Video-EEG

Video-EEG monitoring is used in clinical practice to verify the seizure type, to localize the area of seizure onset if surgery is being considered, and to verify the diagnosis of epilepsy if the diagnosis is in doubt. Video-EEG monitoring consists of the simultaneous recording of EEG brain wave activity combined with time synchronized video recording of the patient. This diagnostic procedure is performed on an in-patient basis, and requires highly specialized equipment and dedicated space. Patients are monitored for extended periods in order to capture typical seizure events on video and simultaneously capture EEG activity during that event. In some centers, patients' medications are withdrawn in order to increase the chance of recording a seizure.

Ambulatory-EEG

Ambulatory-EEG monitoring is the recording of EEG brain wave activity remotely from the hospital environment. Ambulatory-EEG, like video-EEG, has the advantage over routine EEG of allowing EEG traces to be recorded continuously over long periods. This increases the chance of recording an ictal event. Unlike video-EEG, no video record of seizure events is available, and patients or their caregivers must accurately record the occurrence of a typical seizure event in order to temporally compare the occurrence of a seizure with the EEG trace.

Early clinical investigations documented the ability of ambulatory-EEG to record identifiable focal and generalized epileptiform activity. In 1983, a cassette tape ambulatory-EEG system was introduced. This system had continuous 8-channel recording capability, real-time identification, gain adjustment, and filter adjustments. Since then, improvements in computer technology have led to the development of instruments that can perform portable continuous recording of more than 16 channels with sampling rates of over 200 Hz.

Comments on EEG Technologies

That the evidence base for the three EEG technologies (routine ictal and interictal EEG, ambulatory-EEG and video-EEG) did not reach the required minimum of five acceptable studies is surprising. Even with the expansion of the inclusion criteria so that we included articles published between 1980 and 1985 and articles describing retrospective studies, we did not reach five studies. This may be because all three of these EEG technologies are commonly used as aids in the diagnosis of epilepsy, and video-EEG is considered by many to be the “gold standard” for the differentiation of epileptic seizures from nonepileptic seizures.⁵⁸ Some reviews of the use of video-EEG in the differential diagnosis of epileptic seizures from nonepileptic seizures include studies in which provocation was used in an attempt to induce a seizure that was then captured by video-EEG, citing such studies as evidence of the effectiveness of video-EEG in combination with provocation. Without exception, however, these studies used video-EEG as a “reference standard” against which the effectiveness of provocation was measured. Thus, such studies cannot be considered as part of the evidence base for the diagnostic utility of video-EEG and instead form the evidence base for seizure provocation techniques.⁵⁹

For example, a number of studies of patients with implanted electrodes have demonstrated that epileptic seizures originating in the medial or orbital surface of the frontal lobe, the parietal lobe, or the temporal lobe, often occur in the absence of a measurable EEG abnormality when the EEG is performed using scalp electrodes.^62–65 These types of seizures may arguably be relatively rare. However, given that the appearance of a nonepileptic seizure is often very similar to epileptic seizures originating in the medial or orbital surface of the frontal lobe, the parietal lobe, or the temporal lobe,⁶⁶ these are the very patients who are the most likely to be misdiagnosed as having epileptic seizures. Thus, some false-negative decisions must be assumed to occur when video-EEG is used.

The fact that evidence-based conclusions were not drawn in the present report regarding the ability of vEEG to differentiate epileptic seizures from nonepileptic seizures should not be interpreted as evidence that this technology is not effective or useful. Indeed, vEEG may very well have an important role in diagnostic algorithms that are designed to make such a differential diagnosis. Until more high quality studies become available, however, the diagnostic performance characteristics of vEEG and its place in such diagnostic algorithms cannot be determined.

Is seizure type in some patients with treatment-resistant epilepsy misdiagnosed in some patients?

There are two purposes to the present question. First, to establish whether there is evidence in the peer-reviewed literature to indicate that some patients believed to suffer from a specific seizure type actually suffer from a different seizure type (either alone or in combination with the originally diagnosed seizure type) and would, therefore, not be expected to respond satisfactorily to their current treatment regimen. The second purpose is to quantify, if relevant, the prevalence of these patients among the population of patients thought to suffer from a particular seizure type.

Which diagnostic modalities are useful in differentiating between different seizure types?

Because Question 2C cannot be answered in an evidence-based fashion, Question 2D could not be addressed in an evidence-based fashion.

Question specific inclusion criteria

In addition to the general inclusion criteria (see Methodology section), we used the following criteria to determine whether a study was included:

1. The study must have reported information indicating that some patients in the study did not meet at least one of the criteria for optimization described above. Thus, we are seeking only evidence of nonoptimization, not a percentage of patients who are optimized.

2. The study must have been published in 1975 or later. This ensured the use of evidence on standard AEDs as well as evidence on newer agents.

3. All drugs in the study must have been cleared for marketing in the United States by the Food and Drug Administration. If the study included some patients on non-FDA approved drugs, it was required to report data separately from patients on FDA-approved drugs, and we abstracted results only from the latter group of patients. This criterion was determined by the Expert Panel and the Technical Experts.

Excluded studies

We did not exclude any studies for reasons of quality.

Evidence base

Design and conduct of included studies

Synthesis of study results

Question specific inclusion criteria

Although we divided this question into four subsections (one for each treatment strategy, and one for comparisons between strategies), we employed the same inclusion criteria for studies of each strategy. Thus, in addition to the general inclusion criteria described in the Methodology section, we included trials for this question if they met all of the following criteria:

1. The trial must have been published in 1975 or later. For school- and work-related outcomes, the trial must have been published in 1985 or later. Including trials since 1975, as well as more recent trials, facilitated incorporation of data from standard AEDs that may no longer be the focus of clinical research.

2. Before the trial, patients must have received unsuccessful treatment with at least one of the following drugs: carbamazepine, ethosuximide, phenobarbital, phenytoin, primidone, or valproate. The Technical Experts determined that these six drugs are standard AEDs.

3. All drugs received by patients during the trial must be cleared for marketing in the United States by the U.S. Food and Drug Administration (FDA). If the trial included some patients on non-FDA-approved drugs, it was required to report data separately from patients on FDA-approved drugs, and we abstracted results only from the latter group of patients. We included trials employing off-label usage of drugs for the treatment of epilepsy. Confining the question to only FDA-approved drugs was in accordance with the wishes of the Expert Panel and Technical Experts.

4. If a trial reported that some patients had been noncompliant, then results must have been reported separately for patients who were compliant. Noncompliant patients may not be treatment-resistant, and their seizure rates may drop during a trial. Whether the improvement in noncompliant patients was due to better compliance or to the beneficial effect of the trial drug cannot be determined. Therefore, the outcomes of noncompliant patients were not included.

5. If there were five or more placebo-controlled randomized trials on a specific drug treatment strategy, then other trials with other designs (e.g., trials that used a low AED dose as a control) were not considered. We adopted this criterion because results of placebo-controlled randomized trials are more easily interpreted compared to results of trials that employed other control groups.

6. Trial must be a Phase II or III efficacy trial. Earlier trials (Phase I) were not primarily intended to reduce seizures, and later trials (Phase IV) involved drugs whose effectiveness had already been documented by other trials.

7. Trials that used a crossover design must have reported results for the first period (i.e., before the crossover), or must have reported that seizure frequencies returned to baseline at the end of the washout period. In a crossover trial, the use of a drug at the start of a trial may have potentially influenced the effectiveness of a different drug used later in the trial. If seizure frequency returned to baseline at the end of the washout period, then the evidence suggests that the first drug is no longer active, and data for the second drug are interpretable. However, if a return to baseline was not reported, then we only abstracted data for the first period.

To include the maximum number of potentially relevant studies, we did not require studies to report patients' seizure frequencies at baseline. However, baseline seizure frequencies do provide a measure of the severity of patients' initial conditions, thus aiding in the interpretation of study findings. For example, suppose a treatment eliminated all seizures. Such an outcome would be more impressive if patients' baseline seizure frequencies were 20 per month than if frequencies were only five per month. Because the baseline frequency helps place the study results in proper context, ideally all studies would report this frequency.

Sequential Monotherapy

Sequential monotherapy involves administering a single AED not yet received by any of the patients. Patients can receive any number of AEDs prior to the initiation of the new drug. However, all prior AEDs must be withdrawn from patients' drug regimens in order to investigate the effect of the novel monotherapy drug. In this section, we describe the evidence base for sequential monotherapy, assess the quality of these trials with respect to both internal and external validity, and analyze the trials' results for all relevant outcomes.

Design and conduct of included studies

Table 7. Drugs and doses in studies of sequential monotherapy (more...)

Table 7. Drugs and doses in studies of sequential monotherapy

	Felbamate	Gabapentin		Lamotrigine	Oxcarbazepine	Primidone	Tiagabine	Topiramate	Valproate
Reference	3600 mg/day	2400 mg/day	3600 mg/day	500 mg/day	2400 mg/day	750 mg/day	36 mg/day	1000 mg/day	150 μg/ml
Sachdeo (2001)68
Beydoun (2000)86
Kanner (2000)87
Schachter (1999)79
Gilliam (1998)76
Bergey (1997)78
Beydoun (1997)85
Beydoun (1997)83
Sachdeo (1997)84
Devinsky (1995)75
Schachter (1995)88
Theodore (1995)89
Faught (1993)77
Totals	3	1	1	1	3	1	1	1	1

mg/day Maximum dose in milligrams per day

μg/ml Maximum dose in micrograms per milliliter

Relevant design aspects of the 13 included studies appear in Evidence Tables 27 through 30. To assess the effect of sequential monotherapy, the ideal study would have randomly assigned patients to receive either a new drug as monotherapy, or to receive the same drug regimens used before the trial. None of the 13 studies employed this design. Twelve studies were randomized and controlled, but patients in the control groups did not receive their prestudy drug regimens. Instead, all prestudy drugs were withdrawn. In three studies, patients in the control groups received a placebo alone, and in the other nine studies, patients in the control groups received a low dose of a drug. Because these control groups do not address whether sequential monotherapy causes an improvement over patient's prestudy drug regimens, we did not abstract data from the control groups. Instead, we abstracted data from only the high-dose active-drug group in each of the 12 controlled studies. The 13^th study did not have a control group, thus we abstracted data from the single group in that study. Among the 13 studies, eight drugs were given as monotherapy: felbamate (three studies), oxcarbazepine (three studies), gabapentin (two studies), lamotrigine (one study), primidone (one study), tiagabine (one study), topiramate (one study), and valproate (one study) (Table 7).

Internal validity

In evaluating internal validity, we determined whether the results were potentially biased by the threats to validity that are discussed in the Methodology section. Although other questions in this report consider the potential for attrition bias, we do not consider it here because attrition was a study outcome. As discussed earlier, the control groups of these studies are not relevant to the question. Consequently, for the purpose of this report, the studies can be viewed as case series and susceptible to several threats to internal validity (see Appendix B). All were potentially affected by both regression bias and extraneous event bias. Further, most studies were potentially affected by sample specification bias (12/13 studies) and measurement bias (10/11 studies that reported the method of seizure measurement).

External validity

In our appraisal of the external validity of studies of sequential monotherapy, we considered aspects of patient enrollment as well as the actual characteristics of patients in the studies. All patient characteristics appear in Evidence Tables 31 through 34. All 13 studies enrolled patients because of seizure type (partial seizures), thus the results of these studies are not applicable to the treatment of generalized seizures. Three studies enrolled adults only,^75–77 and the remaining 10 studies enrolled both children and adults. The mean age of patients in the studies ranged from 33.4 to 37 years. The proportion of patients who were female ranged from 0.43 to 0.63 and was greater than 0.50 in nine of the 11 studies that reported this characteristic. Median seizure frequency ranged from 5.5 to 13.4 seizures per month, and mean seizure frequency ranged from 6.3 to 70.7 seizures per month. The proportion of patients receiving two or more prior AEDs ranged from 0 to 1. The proportion was less than 0.5 in nine of the 11 studies that reported this patient characteristic. As a whole, then, the results of these studies apply primarily to adults with treatment-resistant epilepsy who experience partial seizures.

Synthesis of study results

Table 8. Outcomes in studies of sequential monotherapy (more...)

Table 8. Outcomes in studies of sequential monotherapy

Reference	Seizure Frequency	Adverse Effects	Quality of Life	Mood	Cognitive Function	Ability to Return to Work	Ability to Return to School	Ability to Hold a Driver's License	Mortality
Sachdeo (2001)68
Beydoun (2000)86
Kanner (2000)87
Schachter (1999)79
Gilliam (1998)76
Bergey (1997)78
Beydoun (1997)85
Beydoun (1997)83
Sachdeo (1997)84
Devinsky (1995)75
Schachter (1995)88
Theodore (1995)89
Faught (1993)77
Totals	13	13	2	3	2	0	0	0	5

In this section, we assess the results separately for each of the following outcomes: seizure frequency, adverse effects, quality of life, mood, cognitive function, functional status/ability, ability to return to work, ability to return to school, ability to hold a driver's license, and mortality. We included freedom from seizures as a seizure frequency outcome. The outcomes reported by each study are listed in Table 8. All outcomes from all of the studies appear in Evidence Tables 35 through 38. Seizure frequency and adverse effects were each reported by all 13 studies.

In cases where meta-analysis was feasible, we used random effects models. We employed these models because the included studies investigated different drugs. Therefore, these studies cannot be viewed as having been sampled from a population of studies with a fixed mean. Random effects models employ statistical methods that are most applicable when studies use different variations of a treatment.

Also, these studies reported data on an intent-to-treat basis. This was particularly important because they employed a priori exit criteria (such as doubling of monthly seizure frequency) to limit harms to patients. If any patient met an exit criterion, investigators removed the patient from the study and reinstituted the patient's prior AED regimen. Consequently, the analyses described below included all patients who were randomized to receive high-dose monotherapy.

Seizure frequency

Table 9. Seizure frequency outcomes in studies of sequential (more...)

Table 9. Seizure frequency outcomes in studies of sequential monotherapy

Reference	Mean Absolute Seizure Frequency	Median Absolute Seizure Frequency	Median Absolute Difference From Baseline	Median Percent Difference From Baseline	Number of Patients Seizure-free	Number of Patients With ≥75 Percent Reduction	Number of Patients With ≥50 Percent Reduction	Number of Patients With Any Reduction	Mean Rank of Seizure Frequency	Number of Patients With Doubling of Monthly Seizure Frequency	Number of Patients With Doubling of Highest Consecutive Two-Day Seizure Frequency	Median Time to Exit	Mean Time on Monotherapy	Risk Ratio of Time to Exit
Sachdeo (2001)68
Beydoun (2000)86
Kanner (2000)87
Schachter (1999)79
Gilliam (1998)76
Bergey (1997)78
Beydoun (1997)85
Beydoun (1997)83
Sachdeo (1997)84
Devinsky (1995)75
Schachter (1995)88
Theodore (1995)89
Faught (1993)77
Totals	1	2	1	1	6	2	4	1	1	5	5	4	4	2

Details of the seizure frequency results are presented in Evidence Table 35. The studies we included for this question reported 14 different measures of seizure frequency (Table 9). Only three seizure frequency outcomes were reported by five or more studies: the percentage of patients who were seizure-free during the study, the percentage of patients whose monthly seizure frequency doubled during the study (vs. baseline), and the percentage of patients whose highest two-day seizure frequency doubled during the study (vs. baseline). We emphasize that, because seizure frequency changes over time, a study's length of followup influences seizure frequency measurements. For example, a given patient is more likely to be seizure-free during a short-term study than a long-term study. Therefore, for each outcome, we considered the length of followup in the studies that reported the outcome.

Figure 11. Threshold analysis: sequential monotherapy (more...)

   Figure 11. Threshold analysis: sequential monotherapy and seizure freedom

Percentage of patients who were seizure-free during the study. This outcome was reported by six studies and ranged from 9 percent to 28 percent. Because of the lack of relevant control groups, we performed a threshold analysis (see the Methodology section for a discussion of this approach). In this analysis, we compared the results obtained in patients who received sequential monotherapy to those of a synthetic control group in which we varied the percentage of seizure-free patients. The percentage at which the difference between the monotherapy and “control” group became statistically nonsignificant is the threshold. The results of this analysis appear in Figure 11. Each summary estimate in the figure is based on Cohen's h. The summary estimate calculated at the 0 percent point on the graph (no patients in a synthetic control group were seizure-free) was 0.81 (CI: 0.64 to 0.98, p <0.000001) and corresponded to 16 percent (CI: 10 percent to 22 percent)ⁱ of patients experiencing no seizures during sequential monotherapy. The summary estimate became nonsignificant (no statistically significant difference between monotherapy and control patients in the number of patients becoming seizure-free) when the proportion of patients in the synthetic control group reached 10 percent.

Figure 12. Threshold analysis: monotherapy and seizure (more...)

   Figure 12. Threshold analysis: monotherapy and seizure freedom (long-term studies)

We next performed a second threshold analysis of these data to test the effect of followup period on seizure-free status. Two of the six studies that reported seizure freedom employed short followup times (8 days in Bergey, Morris, Rosenfeld, et al.,⁷⁸ and 10 days in Schacter, Vasquez, Fisher, et al.).⁷⁹ In the other four studies, patients were followed for at least 16 weeks. The percentage of patients who were seizure-free was greater than 25 percent in both short-term studies, but was less than 14 percent in all of the long-term studies. Therefore, our second threshold analysis included only the four studies with longer followup times (Figure 12). The summary estimate calculated at the 0 percent point on the graph (no patients in a synthetic control group were seizure-free) was 0.67 (CI: 0.47 to 0.87, p <0.000001) and corresponded to 11 percent (CI: 5 percent to 18 percent) of patients experiencing no seizures during sequential monotherapy. The summary estimate became nonsignificant (no statistically significant difference between monotherapy and control patients in the number of patients becoming seizure-free) when the proportion of patients in the synthetic control group reached 6 percent.

In summary, approximately 11 percent of patients are seizure-free during long-term studies of sequential monotherapy. However, given the designs of these studies, whether the new drug actually caused any of the patients to become seizure-free during the study is not clear. Further, seizure frequencies change from month to month, and some patients with treatment-resistant epilepsy may experience periods without seizures, and some patients may have been misdiagnosed (Question 2). These latter patients may be more likely to become seizure-free. Even if some patients become seizure-free as a result of sequential monotherapy, the majority of patients (approximately 89 percent) continue to have seizures despite receiving a new drug as monotherapy. A firm conclusion about whether sequential monotherapy produces any new seizure-free patients would require the use of a relevant control group (i.e., continuation of prior drug regimens).

Figure 13. Threshold analysis: monotherapy and doubling (more...)

   Figure 13. Threshold analysis: monotherapy and doubling of monthly seizure frequency

Percentage of patients whose monthly seizure frequency doubled. Freedom from seizures measures the percentage of patients who experienced maximum benefit. In contrast, seizure doubling indexes the percentage of patients who experienced significant harm. This outcome was reported by five studies and ranged from 9 percent to 29 percent. All five studies followed patients for at least 16 weeks, thus the concern about study duration does not apply to seizure doubling. Due to the lack of relevant control groups, we performed a threshold analysis of this outcome (Figure 13). Each summary estimate in the figure is based on Cohen's h. The summary estimate calculated at the 0 percent point on the graph (no patients in a synthetic control group had a doubling of monthly seizure frequency) was 0.82 (CI: 0.64 to 0.99, p <0.000001) and corresponded to 16 percent (CI: 10 percent to 23 percent) of patients experiencing a doubling of monthly seizure frequency. The summary estimate became nonsignificant (no statistically significant difference between monotherapy and control patients in the number of patients experiencing a doubling of monthly seizure frequency) when the proportion of patients in the synthetic control group reached 10 percent.

Sequential monotherapy cannot be directly considered the cause of the doubling in seizure frequency because of the lack of a true control group. However, three factors do suggest a causal relation. First, at the beginning of the sequential monotherapy studies, all prestudy drugs were removed from patients' regimens and replaced with a new AED. Presumably, the original AEDs were already reducing seizure frequency. The removal of these drugs, therefore, may have caused seizures to increase. Second, a doubling of monthly seizure frequency was set by investigators as an a priori exit criterion. A doubling of seizure frequency resulted in immediate removal from the study, and all prestudy drugs were reinstituted. This suggests that investigators believed that the doubling of monthly seizure frequency was being caused by sequential monotherapy. Third, given the possibility that patients enter drug trials when they are relatively sick, they would not be expected to become even worse. Instead, based on regression-to-the-mean, reductions in seizure frequency would be expected. Each of these factors suggest that sequential monotherapy caused dramatic increases in seizures in some patients. A definitive conclusion about this possibility would require randomization of patients to either sequential monotherapy or a continuation of the prestudy drug regimen.

Figure 14. Threshold analysis: monotherapy and doubling (more...)

   Figure 14. Threshold analysis: monotherapy and doubling of two-day seizure frequency

Percentage of patients whose highest two-day seizure frequency doubled. This outcome also measures the percentage of patients who experienced significant harm during studies of sequential monotherapy. It was reported by five studies, and ranged from 4 percent to 23 percent. All five studies followed patients for at least 16 weeks. Due to the lack of relevant control groups, we performed a threshold analysis of this outcome (Figure 14). Each summary estimate in the figure is based on Cohen's h. The summary estimate calculated at the 0 percent point on the graph (no patients in a synthetic control group had a doubling of two-day seizure frequency) was 0.76 (CI: 0.56 to 0.96, p <0.000001) and corresponded to 14 percent (CI: 8 percent to 21 percent) of patients experiencing a doubling of two-day seizure frequency. The summary estimate became nonsignificant (no statistically significant difference between monotherapy and control patients in the number of patients experiencing a doubling of two-day seizure frequency) when the proportion of patients in the synthetic control group reached 8 percent.

As stated previously, whether sequential monotherapy was the cause of doubling of two-day seizure frequency cannot be determined without a true control group. However, the same three factors discussed above apply to this outcome as well. Based on these factors, sequential monotherapy in some patients appears to have caused doubling in two-day seizure frequency. Combining the estimates of the two seizure increase outcomes that result in exiting a trial, 30 percent of patients in studies of sequential monotherapy experience a doubling in either monthly seizure frequency or two-day seizure frequency. Therefore, some patients may be experiencing large seizure increases as a direct result of sequential monotherapy. To provide a definite answer, randomizing patients to receive either sequential monotherapy or a continuation of the prestudy drug regimen would be necessary.

Adverse effects

In clinical practice, a physician prescribing an AED for a patient with treatment-resistant epilepsy must consider not only the possible reduction of seizure frequency, but also the possible adverse effects of the new drug. Before entering studies, patients with treatment-resistant epilepsy were already experiencing adverse effects from their prestudy antiepileptic drug regimens. None of the studies reported these patients' prestudy adverse effects, and none reported whether the adverse effects observed during the study were more or less severe compared to patients' prestudy adverse effects. This latter outcome would have been informative because patients (and physicians) seek to reduce adverse effects as well as seizure frequency.

Table 10. Overview of adverse effects of sequential (more...)

Table 10. Overview of adverse effects of sequential monotherapy

Reference	Drug	Dose in mg/day	Percent of Patients Who Experienced Any Adverse Effect	Name of Most Commonly Experienced Adverse Effect	Percent of Patients Who Experienced This Adverse Effect
Sachdeo (2001)68	Oxcarbazepine	2400	NR	Headache	11% (5/45)
Beydoun (2000)86	Oxcarbazepine	2400	NR	Dizziness	46% (19/41)
Kanner (2000)87	Primidone	750	53% (16/30)	Irritability	37% (11/30)
Schachter (1999)79	Oxcarbazepine	2400	91% (46/51)	Nervous system	45% (23/51)
Gilliam (1998)76	Lamotrigine	500	75% (57/76)	Dizziness	20% (15/76)
Bergey (1997)78	Gabapentin	3600	73% (29/40)	Ataxia	20% (8/40)
Beydoun (1997)85	Valproate	150 μG/mL	NR	Tremor	64% (61/96)
Beydoun (1997)83	Gabapentin	2400	88% (80/91)	Dizziness	25% (23/91)
Sachdeo (1997)84	Topiramate	1000	NR	Paresthesia	58% (14/24)
Devinsky (1995)75	Felbamate	3600	NR	NR	NR
Schachter (1995)88	Tiagabine	36	95% (91/96)	Dizziness	35% (34/96)
Theodore (1995)89	Felbamate	3600	NR	NR	NR
Faught (1993)77	Felbamate	3600	NR	Headache	34% (19/56)

mg/day Milligrams per day

NR Not reported

μG/ml Micrograms per milliliter

All 13 included studies of sequential monotherapy reported adverse effects of the new drug treatment. The overall percentage of patients who experienced any side effects was reported by six studies and ranged from 53 percent to 95 percent (Table 10). Dizziness was the most common adverse effect in four studies, and headache was the most common adverse effect in two studies. All details of the adverse effects in the 13 studies appear in Evidence Table 36.

Figure 15. Threshold analysis: monotherapy and trial (more...)

   Figure 15. Threshold analysis: monotherapy and trial exits due to adverse effects

Percentage of patients who exited trials due to adverse effects. To summarize the available data on adverse effects, we focused on whether the adverse effects in a given patient were severe enough to warrant discontinuation of the new drug (i.e., trial exit). This outcome is a marker of treatment failure. All 13 included trials reported the percentage of patients who exited trials due to adverse effects, and it ranged from 0% to 29%. As with seizure frequency, due to the lack of relevant control groups we performed a threshold analysis (Figure 15). Each summary estimate in the figure is based on Cohen's h. The summary estimate calculated at the 0 percent point on the graph (no patients in a synthetic control group exited due to adverse effects) was 0.47 (CI: 0.24 to 0.71, p <0.000073) and corresponded to 5 percent (CI: 1 percent to 12 percent) of patients exiting trials due to adverse effects. The summary estimate became nonsignificant (no statistically significant difference between monotherapy and control patients in the number of patients exiting trials due to adverse effects) when the proportion of patients in the synthetic control group reached 2 percent.

Quality of life

Only two studies of sequential monotherapy reported quality of life outcomes (Evidence Table 37).^{80, 81} Evidence Table 39 lists the scales and subscales used to measure quality of life in these studies. Due to the small number of studies, we did not perform meta-analyses of the results. There were no statistically significant changes in quality of life in either of the two studies. The lack of statistical significance may have been due to insufficient power. An estimate of the power of pre- vs. posttests would require knowledge of the correlation between baseline and outcome measurements. However, the authors did not report these correlations and therefore the power of this study to detect statistically significant quality of life changes could not be determined. Many of the subscales showed a nonsignificant improvement over baseline. However, these subscales are not independent (i.e., improvement on one subscale is likely to result in improvement in another subscale). Therefore, firm evidence-based conclusion about the influence of sequential monotherapy on quality of life cannot be based on these data.

Mood

Three studies of sequential monotherapy reported outcomes related to mood (Evidence Table 37).^80–82 Evidence Table 40 lists the scales and subscales used to measure mood in these studies. Each of the three studies investigated a different drug for sequential monotherapy. Two of the three studies used the same set of scales (Dodrill, Arnett, Hayes, et al.⁸¹ and Dodrill, Arnett, Shu, et al.⁸⁰). As with quality of life, the small number of studies precluded any meta-analysis. None of the subscales in the study by Dodrill, Arnett, Hayes, et al.⁸¹ showed statistically significant changes in mood. In the study by Dodrill, Arnett, Shu, et al.,⁸⁰ one of eight subscales (the Vigor-Activity subscale) exhibited a statistically significant decrement from baseline in mood. As discussed in the quality of life section, insufficient power may have prevented these studies from detecting changes in mood. However, incomplete reporting in the published literature prevents investigating this possibility.

Ketter, Malow, Flamini, et al.⁸² reported that mood and psychiatric symptom scores changed after 2 weeks of sequential monotherapy in patients given felbamate. After the removal of all AEDs, patients' mood scores significantly worsened relative to baseline for seven of the 11 subscales. At both week 1 and week 2 of felbamate monotherapy, the decrements persisted. Thus, the initiation of felbamate monotherapy did not return patients' mood scores to baseline. With longer followup, patients' mood scores may potentially have returned to baseline or even improved over baseline. However, these findings suggest that the first phase of sequential monotherapy (i.e., the drug reduction phase) may cause significant worsening of mood and psychiatric symptom scores. However, because there was only one study reporting such changes, firm evidence-based conclusions cannot be drawn about the general effect of sequential monotherapy on mood.

Cognitive function

Only two studies of sequential monotherapy reported cognitive function (Evidence Table 38).^{80, 81} The two studies used the same subscales for measuring cognitive function (Evidence Table 41). Due to the small number of studies that reported the effect of sequential monotherapy on cognitive function, we did not perform meta-analyses of these data. In the study described by Dodrill, Arnett, Hayes, et al.⁸¹, none of the 19 cognitive function subscales were significantly different from baseline. Of the 19 subscales in the study described by Dodrill, Arnett, Shu, et al.,⁸⁰ four showed a statistically significant improvement from baseline. These results may have been caused by a practice effect and not by tiagabine (see the discussion of instrumentation bias in the Methodology section). The power of these studies to detect changes in cognitive function could not be calculated because the authors did not report the correlations between baseline and outcome measurements.

Functional status/ability

No studies of sequential monotherapy reported this outcome.

Ability to return to work

No studies of sequential monotherapy reported this outcome.

Ability to return to school

No studies of sequential monotherapy reported this outcome.

Ability to hold a driver's license

No studies of sequential monotherapy reported this outcome.

Mortality

Five of the 13 studies of sequential monotherapy (38 percent) reported whether any patients died during the study. Three of the five studies reported that no patients died,^{78, 83, 84} and the other two studies each reported one death.^{68, 85} The authors did not attribute either death to the treatment. The mortality rates in these five studies ranged from 0 percent to 2 percent (Evidence Table 42).

Polytherapy

Polytherapy is defined as the administration of a multiple-drug regimen in which at least one of the drugs is novel to each patient. As with sequential monotherapy, patients received any number of drugs prior to the initiation of a new drug. Most polytherapy interventions involve the addition of a single novel drug to patients' regimens (referred to as “add-on” treatment). In this section, we describe the evidence base for polytherapy, assess the quality of these trials with respect to both internal and external validity, and analyze the trials' results for all relevant outcomes.

Design and conduct of included studies

Table 11. Drugs and doses in trials of polytherapy

Table 11. Drugs and doses in trials of polytherapy

Reference	Drug	Trial Dose(s)a	Total Number of Trials That Used This Drug/Dose Combination
Faught (2001)94	Zonisamide	400	1
Ben-Menachem (2000)95	Levetiracetam	3000	2
Betts (2000)96	Levetiracetam	2000, 4000	1, 1
Cereghino (2000)97	Levetiracetam	1000, 3000	1, 2
Glauser (2000)98	Oxcarbazepine	Tailored to weight	1
Appleton (1999)99	Gabapentin	Tailored to weight	1
Biton (1999)100	Topiramate	Tailored to weight	3
Duchowny (1999)30	Lamotrigine	Tailored to weight	2
Elterman (1999)101	Topiramate	Tailored to weight	3
Korean Topiramate Study Group (1999)102	Topiramate	600	4
Sachdeo (1999)103	Topiramate	Tailored to weight	3
Uthman (1998)104	Tiagabine	16, 32, 56	1, 2, 1
Sachdeo (1997)105	Tiagabine	32b, 32c	2, 1
Ben-Menachem (1996)90	Topiramate	800	2
Chadwick (1996)106	Gabapentin	1200	3
Faught (1996)91	Topiramate	200, 400, 600	1, 2, 4
Privitera (1996)107	Topiramate	600, 800, 1000	4, 2, 1
Sharief (1996)108	Topiramate	400	2
Tassinari (1996)109	Topiramate	600	4
Willmore (1996)110	Valproate	Tailored to weight	1
Anhut (1994)111	Gabapentin	900, 1200	2, 3
Messenheimer (1994)112	Lamotrigine	400	1
Bourgeois (1993)113	Felbamate	3600	1
Felbamate Study Group (1993)114	Felbamate	Tailored to weight	1
Matsuo (1993)115	Lamotrigine	300, 500	1, 1
McLean (1993)116	Topiramate	600, 1200, 1800	4, 1, 1
Schmidt (1993)117	Zonisamide	Tailored to weight	1
Sivenius (1991)118	Gabapentin	900	2
UK Gabapentin Study Group (1990)119	Gabapentin	1200	3
Jawad (1989)70	Lamotrigine	Tailored to weight	2

a

Maximum dose in milligrams per day

b

Based on 16 milligrams twice per day

c

Based on 8 milligrams four times a day

Aspects of the trial designs appear in Evidence Tables 43 through 46, and the patient characteristics appear in Evidence Tables 47 through 53. All 30 trials were randomized, placebo-controlled, add-on trials. In these trials, patients continued to take their pretrial drug regimens, and either a placebo or a new drug was added to those regimens. Nine add-on drugs were investigated in these trials: topiramate (9 trials), gabapentin (6 trials), lamotrigine (4 trials), levetiracetam (3 trials), tiagabine (2 trials), zonisamide (2 trials), felbamate (2 trials), oxcarbazepine (1 trial), and valproate (1 trial). No single dose of a given drug was used in all trials of that drug (Table 11). Of the 29 drug doses, 20 (69 percent) were employed by only one trial, and no drug dose was employed by more than four trials. Ten trials (33 percent) individualized the dose to each patient based on weight. These observations highlight the wide variation among trials' implementations of the polytherapy strategy.

Internal validity

For each trial of polytherapy, we determined whether the results were potentially biased by the factors noted in the Methodology section. Other questions in this report consider the potential for attrition bias, but for polytherapy, we did not consider it because attrition was a study outcome. All 30 trials of polytherapy were randomized and placebo-controlled. Thus, they were free from many potential threats to internal validity (see Appendix B). We meta-analytically tested selection bias with respect to several patient characteristics, and trials were free from potential selection bias in all but two cases (Evidence Table 54 through 58; also see discussion in Appendix B). All of the trials were free from five potential biases (sampling, regression, investigator, patient, and extraneous event). However, all of the trials had potential measurement bias. In addition, 90 percent of the trials had sample specification bias.

External validity

In our appraisal of the external validity of trials of polytherapy, we considered aspects of patient enrollment as well as the actual characteristics of patients in the trials (Evidence Tables 43 to 53). Twenty-seven trials (90 percent) enrolled patients because of seizure type: 25 for partial seizures, and two for generalized seizures. Two trials enrolled only patients with Lennox-Gastaut syndrome, and one trial included patients with any seizure type or syndrome. Six trials enrolled children only, 23 trials enrolled adults only, and one trial enrolled both children and adults. In the six trials of children, the mean age ranged from 7.9 to 13.0, and in the 23 trials of adults, the mean age ranged from 29.4 to 38.0. The proportion of patients who were female ranged from 0.14 to 0.69, and was less than 0.50 in 21 of the 28 trials that reported this characteristic. Median seizure frequency ranged from 1 to 80 seizures per month, and mean seizure frequency ranged from 7.3 to 68.7 seizures per month. The proportion of patients who had received two or more prior AEDs ranged from 0 to 0.81. This proportion was greater than 0.5 in 15 of the 18 trials that reported this patient characteristic. As a whole, then, the characteristics of the patients in these studies are not particularly unusual.

Synthesis of study results

Outcomes in trials of polytherapy

Outcomes in trials of polytherapy

Reference	Seizure Frequency	Adverse Effects	Quality of Life	Mood	Cognitive Function	Ability to Return to Work	Ability to Return to School	Ability to Hold a Driver's License	Mortality
Faught (2001)94
Ben-Menachem (2000)95
Betts (2000)96
Cereghino (2000)97
Glauser (2000)98
Appleton (1999)99
Biton (1999)100
Duchowny (1999)30
Elterman (1999)101
KTSG (1999)102
Sachdeo (1999)103
Uthman (1998)104
Sachdeo (1997)105
Ben-Menachem (1996)90
Chadwick (1996)106
Faught (1996)91
Privitera (1996)107
Sharief (1996)108
Tassinari (1996)109
Willmore (1996)110
Anhut (1994)111
Messenheimer (1994)112
Bourgeois (1993)113
FSG (1993)114
Matsuo (1993)115
McLean (1993)116
Schmidt (1993)117
Sivenius (1991)118
UKGSG (1990)119
Jawad (1989)70
Totals	30	30	2	1	1	0	0	0	9

In this section, we assess the results separately for each of the relevant outcomes (Table 12). All reported outcomes appear in Evidence Tables 59 through 62. Seizure frequency and adverse effects were each reported by all 30 trials, whereas the other outcomes were not commonly reported.

In cases where meta-analyses were conducted, we used random-effects models because, as shown in Table 12, the trials employed a variety of drugs and doses. The trials are therefore not derived from a population of trials with a fixed mean. Our meta-analytic syntheses of trial results yield approximate estimates of the typical effect of adding a new AED to patients' prior AED regimens. However, these estimates have limited generalizability because the effect of a new AED may depend on the other AEDs in patients' regimens. Each trial employed a control group of patients who received an add-on placebo, but the prior regimens were different among different trials (and among patients in a single trial). Thus, the 30 trials did not administer the exact same “control” treatment. Because the treatments and controls differ across trials, the summary effect sizes from random-effects meta-analyses can only be used as approximate estimates of the effect of adding a new AED and may be best suited for use as starting points in future research. The actual effect on seizure frequency or adverse effects in any single patient is likely to depend on the specific drug to be added as well as characteristics of the AEDs already in use.

We performed all meta-analyses on an intent-to-treat basis. This means that we included all randomized patients in our analyses, not solely the patients who completed the trials. If a patient exited early from a trial and the authors did not report the relevant outcome for that patient, we assumed that seizure frequency did not decrease for that patient. This is a reasonable assumption because all patients who respond to a drug would likely be reported as responders.

Seizure frequency

Seizure frequency outcomes in trials of polytherapy (more...)

Seizure frequency outcomes in trials of polytherapy

	Absolute Monthly Seizure Frequency		Absolute Difference From Baseline		Absolute Percent Difference From Baseline		Number of Patients With
Reference	Mean	Median	Mean	Median	Mean	Median	Seizure-free	≥75 Percent Reduction	≥50 Percent Reduction	≥25 Percent Reduction	Any Reduction	Any Increase	≥50 Percent Increase	≥75 Percent Increase	Mean Response Ratioa	Mean Adjustment Response Ratioa
Faught (2001)94
Ben-Menachem (2000)95
Betts (2000)96
Cereghino (2000)97
Glauser (2000)98
Appleton (1999)99
Biton (1999)100
Duchowny (1999)30
Elterman (1999)101
KTSG (1999)102
Sachdeo (1999)103
Uthman (1998)104
Sachdeo (1997)105
Ben-Menachem (1996)90
Chadwick (1996)106
Faught (1996)91
Privitera (1996)107
Sharief (1996)108
Tassinari (1996)109
Willmore (1996)110
Anhut (1994)111
Messenheimer (1994)112
Bourgeois (1993)113
FSG (1993)114
Matsuo (1993)115
McLean (1993)116
Schmidt (1993)117
Sivenius (1991)118
UKGSG (1990)119
Jawad (1989)70
Totalsb	2	11	1	3	2	24	18	17	27	8	5	6	4	2	2	4

a

The response ratio is the ratio (T-B)/(T+B) where T is the number of seizures a month during treatment and B is the number of seizures a month during baseline. Some authors adjusted the response ratio in order to account for differences between centers in multi-center trials (using ANOVA).

The included trials reported 20 different measures of seizure frequency (Table 13). Seven measures were reported by five or more trials. One was a measure of absolute seizure frequency (median percentage reduction), and the remaining six were dichotomous measures. We did not analyze two of the dichotomous measures (75 percent or more reduction and 25 percent or more reduction) because they provided data that was effectively captured by other dichotomous measures (seizure-freedom, 50 percent or more reduction, and any reduction). The use of multiple seizure types and multiple study intervals necessitated that we adopt two selection rules for abstracting data from an included study.^j First, if a study reported the same seizure frequency measure for more than one seizure type, we selected the most general type for inclusion in any meta-analyses.^k Second, if a study reported the same seizure frequency measure for different study intervals, we selected the longest interval for inclusion in any meta-analyses^l These selection rules permitted us to focus our analyses on the most general and widely reported seizure frequency measures.

In considering meta-analyses of the seizure frequency outcomes, nine of the 30 trials each contained three or more groups of patients. From each of these nine trials, therefore, multiple effect sizes can be computed (e.g., dose 1 vs. placebo, and dose 2 vs. placebo). Multiple effect sizes within a single trial are statistically dependent. Ideally, we would analyze these data using general linear models that account for this dependence. This, however, was precluded by the relative paucity of data. Therefore, to avoid this dependence, for each meta-analysis we selected only one drug dose from each trial. Thus, the effect size we computed for each trial was based on the difference between outcomes in one add-on drug group and the add-on placebo group. In some meta-analyses (“high-dose”), we selected the highest-dose group in each trial, whereas in other meta-analyses (“low-dose”), we selected the lowest-dose group in each trial. Trials with only one add-on drug group appear in both the high-dose and low-dose meta-analyses. Consequently, the high-dose meta-analysis was not independent of the low-dose meta-analysis.

A comparison between the results of a high-dose meta-analysis with those of the corresponding low-dose meta-analysis can be viewed as a form of sensitivity analysis. Larger effect sizes may be expected a priori in the high-dose meta-analysis (i.e., larger effects with higher doses). Performing both analyses permits us to estimate the robustness of the results. Although this approach allows us to estimate the effect of high- and low-dose polytherapy, it has the disadvantage that each meta-analysis uses only a subset of the available data. Consequently, some information is lost in our analysis.

Median percentage reduction. Twenty-four of the 30 included trials reported the median percentage reduction in seizures. However, none of these trials reported the dispersion about these medians (e.g., variances, standard deviations, interquartile ranges). Therefore, effect sizes could not be calculated and a meta-analysis was not conducted with these data.

Figure 16. Median percentage reduction in seizures (more...)

   Figure 16. Median percentage reduction in seizures after polytherapy

Note: In this plot, positive numbers represent reductions in seizures, whereas negative numbers represent increases in seizures.

Because the median percentage of seizure reduction was a commonly reported seizure frequency outcome, we plotted a summary of the published findings. This plot (Figure 16) depicts the 24 statistical comparisons to placebo that were reported.^m Twenty-two of these 24 comparisons were statistically significant in favor of the add-on drug. The remaining two comparisons also favored the add-on drug, but the differences were not statistically significant. The range of medians was -18 percent to 13 percent for the groups that received add-on placebo, and 13 percent to 51 percent for the add-on drug groups (as a convention in the epilepsy literature, negative numbers represent percentage increases from baseline, and positive numbers represent percentage decreases from baseline).

The data in Figure 16 provide evidence for a placebo effect: 14 of the 17 placebo groups (82 percent) had a median percentage reduction that was greater than zero (i.e., a beneficial effect indicated by the rightward shift on the x-axis). This percentage is significantly larger than 50 percent (two-tailed sign test, p = 0.013). The size of this placebo effect cannot be estimated because the trials did not report dispersion statistics for median percentage reduction. The observed medians, however, do indicate that patients with treatment-resistant epilepsy have fewer seizures when a placebo is added to their drug regimens. This placebo effect does not influence our investigations of polytherapy because all trials were placebo-controlled and therefore all effect sizes involved comparisons to placebo groups. However, the placebo effect does underscore the need for placebo controls in treatment trials involving patients with epilepsy, because if a treated group improves, part of that improvement may be due to the initiation of any medical intervention rather than to the intervention itself.

Seizure-freedom. Eighteen of the 30 included trials reported the percentage of patients who became seizure-free. Two of these trials, however, only reported seizure-freedom for a severe type of partial seizure (secondarily generalized seizures) that was experienced by only a subset of patients before the trials.^{90, 91} They did not report freedom from a seizure type that all patients had experienced before the trial. Thus, we analyzed the 16 trials of polytherapy that did report the latter kind of seizure-freedom results.

Figure 17. Forest plot: polytherapy and seizure-freedom (more...)

   Figure 17. Forest plot: polytherapy and seizure-freedom (high-dose)

Figure 18. Forest plot: polytherapy and seizure-freedom (more...)

   Figure 18. Forest plot: polytherapy and seizure-freedom (low-dose)

We first performed the high-dose meta-analysis. The percentage of patients who were seizure-free ranged from 0 percent to 9 percent in the high-dose groups and from 0 percent to 2 percent in the add-on placebo groups. The effect sizes are plotted in Figure 17, and the details of the random-effects meta-analysis appear in Evidence Table 63. The random-effects summary statistic (Cohen's h) was 0.29 (CI: 0.20 to 0.37). Patients who received a high-dose of add-on drug were statistically significantly more likely to become seizure-free compared to patients who received add-on placebo. The estimated summary percentages were 5 percent for the high-dose groups (CI: 3 percent to 7 percent) and 1 percent for the placebo groups (CI: 0 percent to 1.4 percent) as calculated from the back-transformed Cohen's h. Similar results were observed in the low-dose meta-analysis for seizure-freedom (Figure 18 and Evidence Table 64). The summary Cohen's h (0.28, CI: 0.20 to 0.36) was only slightly lower than the high-dose meta-analysis. The estimated summary percentage was 5 percent (CI: 3 percent to 7 percent) in the low-dose groups.

We performed four sensitivity analyses separately for the high-dose and low-dose meta-analyses. The sensitivity analyses involved recalculating the meta-analysis after separately removing the trial with the largest effect size, the smallest effect size, the largest sample size, and the smallest sample size. None of the four sensitivity analyses overturned our findings (Evidence Table 65 and Table 66).

In summary, the evidence suggests that adding a drug to patient's regimens increases the likelihood of becoming seizure-free. This finding occurred in both the high-dose and low-dose groups, and multiple sensitivity analyses did not overturn the results. However, seizure-freedom was analyzable for only 16 of the 30 trials of polytherapy.

Figure 19. Forest plot: polytherapy and 50 percent (more...)

   Figure 19. Forest plot: polytherapy and 50 percent seizure reduction (high-dose)

Figure 20. Forest plot: polytherapy and 50 percent (more...)

   Figure 20. Forest plot: polytherapy and 50 percent seizure reduction (low-dose)

50 percent reduction. Twenty-seven trials reported the percentage of patients who experienced 50 percent or more reduction in seizures. As with seizure-freedom, we performed both a high-dose meta-analysis and a low-dose meta-analysis. The range was 13 percent to 50 percent in the high-dose groups and 0 percent to 25 percent in the placebo groups. A plot of the effect sizes appears in Figure 19, and the statistical details of the meta-analysis are in Evidence Table 67. The random effects summary statistic (Cohen's h) was 0.52 (CI: 0.43 to 0.62). Patients who received a high-dose of add-on drug were significantly more likely to experience 50 percent reduction compared to patients who received add-on placebo. The estimated summary percentages were 35 percent for the high-dose groups (CI: 31 percent to 38 percent) and 13 percent for the placebo groups (CI: 10 percent to 15 percent). We observed similar results with the low-dose meta-analysis (Figure 20 and Evidence Table 68). The random-effects summary Cohen's h was 0.45 (CI: 0.35 to 0.55), and the estimated summary percentage for the low-dose groups was 31 percent (CI: 27 percent to 36 percent).

As described previously, we performed four sensitivity analyses for both the high-dose and low-dose meta-analyses. None of these analyses overturned our findings (Evidence Tables 69 and 70).

These studies suggest that when a drug is added to patients' drug regimens, approximately one-third of patients will experience a 50 percent or more reduction in seizures. As mentioned above, however, the generalizability of this finding may be limited.

FIgure 21. Forest plot: polytherapy and any seizure (more...)

   FIgure 21. Forest plot: polytherapy and any seizure reduction (high-dose)

Figure 22. Forest plot: polytherapy and any seizure (more...)

   Figure 22. Forest plot: polytherapy and any seizure reduction (low-dose)

Any reduction. Five trials reported the percentage of patients who experienced any reduction in seizures. As with other measures, we performed both a high-dose meta-analysis and a low-dose meta-analysis. The range was 61 percent to 80 percent in the high-dose groups and 41 percent to 72 percent in the placebo groups. A plot of the effect sizes appears in Figure 21, and the statistical details of the meta-analysis are in Evidence Table 71. The random effects summary statistic (Cohen's h) was 0.37 (CI: 0.19 to 0.55). Patients who received a high-dose of add-on drug were significantly more likely to experience a reduction compared to patients who received add-on placebo. The estimated summary percentages were 70 percent for the high-dose groups (CI: 61 percent to 77 percent) and 52 percent for the placebo groups (CI: 44 percent to 61 percent). We obtained similar results with the low-dose meta-analysis (Figure 22 and Evidence Table 72). The random-effects summary Cohen's h was 0.31 (CI: 0.15 to 0.47), and the estimated summary percentage for the low-dose groups was 67 percent (CI: 59 percent to 74 percent).

We performed the four sensitivity analyses for both the high-dose and low-dose meta-analyses. None of these overturned our findings (Evidence Table 73 and 74).

These studies suggest that when certain AEDs are added to patients' drug regimens, approximately two-thirds of patients will experience some reduction in seizures. This analysis, like the previous one, may have limited generalizability.

Figure 23. Forest plot: polytherapy and any seizure (more...)

   Figure 23. Forest plot: polytherapy and any seizure increase (high-dose)

FIgure 24. Forest plot: polytherapy and any seizure (more...)

   FIgure 24. Forest plot: polytherapy and any seizure increase (low-dose)

Any increase. Six trials reported the percentage of patients who experienced any increase in seizures. One of these trials, however, reported this outcome for a specific seizure type that was experienced by only a subset of patients before the trial. Thus, we analyzed seizure increase data from the other five trials. As with other measures, we performed both a high-dose meta-analysis and a low-dose meta-analysis. The range was 16 percent to 38 percent in the high-dose groups and 28 percent to 44 percent in the placebo groups. A plot of the effect sizes appears in Figure 23, and the statistical details of the meta-analysis are in Evidence Table 75. The random effects summary statistic (Cohen's h) was 0.38 (CI: 0.23 to 0.53). Patients who received a high-dose of add-on drug were significantly less likely to experience an increase compared to patients who received add-on placebo. The estimated summary percentages were 21 percent for the high-dose groups (CI: 15 percent to 28 percent) and 39 percent for the placebo groups (CI: 32 percent to 46 percent). We observed similar results with the low-dose meta-analysis (Figure 24 and Evidence Table 76). The random-effects summary Cohen's h was 0.39 (CI: 0.22 to 0.57), and the estimated summary percentage for the low-dose groups was 20 percent (CI: 15 percent to 27 percent).

We performed the four sensitivity analyses for both the high-dose and low-dose meta-analyses. None of these analyses overturned our findings (Evidence Table 77 and 78).

These data suggest that when certain AEDs are added to patients' drug regimens, approximately 20 percent of patients will experience an increase in seizures. This analysis, like the previous ones, may have limited generalizability.

Adverse effects

Table 14. Overview of adverse effects of polytherapy (more...)

Table 14. Overview of adverse effects of polytherapy

Reference	Drug and Dose (mg/day)	Percent of Patients Who Experienced Any Adverse Event	Name of Most Commonly Experienced Adverse Event	Percent of Patients Who Experienced This Adverse Event
Faught (2001)94	Zonisamide 400	NR	Somnolence	15% (18/118)
Ben-Menachem (2000)95	Levetiracetam 3000	55% (100/181)	Asthenia	13.8% (25/181)
Betts (2000)96	Levetiracetam 2000	83% (35/42)	Asthenia	31% (13/42)
Betts (2000)96	Levetiracetam 4000	84% (32/38)	Somnolence	45% (17/38)
Cereghino (2000)97	Levetiracetam 1000	89% (87/98)	Infection	28% (27/98)
Cereghino (2000)97	Levetiracetam 3000	89% (90/101)	Infection	27% (27/101)
Glauser (2000)98	Oxcarbazepine 1800	91% (125/138)	Vomiting	36% (50/138)
Appleton (1999)99	Gabapentin 1800	NR	Viral infection	11% (13/119)
Biton (1999)100	Topiramate 400	NR	Upper respiratory tract infection	41% (16/39)
Duchowny (1999)30	Lamotrigine 750	94% (92/98)	Somnolence	24% (24/98)
Elterman (1999)101	Topiramate 400	NR	Upper respiratory tract infection	41% (17/41)
KTSG (1999)102	Topiramate 600	81% (74/91)	Anorexia	21% (19/91)
Sachdeo (1999)103	Topiramate 600	NR	Somnolence	42% (20/48)
Uthman (1998)104	Tiagabine 16	NR	Nervous system	69% (42/61)
Uthman (1998)104	Tiagabine 32	NR	Nervous system	70% (62/88)
Uthman (1998)104	Tiagabine 56	NR	Nervous system	77% (44/57)
Sachdeo (1997)105	Tiagabine 32	NR	Nervousness	10.5% (11/105)
Ben-Menachem (1996)90	Topiramate 800	NR	Fatigue	79% (22/28)
Chadwick (1996)106	Gabapentin 1200	67% (39/58)	Somnolence	12% (7/58)
Faught (1996)91	Topiramate 200	NR	Dizziness	36% (16/45)
Faught (1996)91	Topiramate 400	NR	Dizziness	33% (15/45)
Faught (1996)91	Topiramate 600	NR	Dizziness	35% (16/46)
Privitera (1996)107	Topiramate 600	NR	Fatigue	38% (18/48)
Privitera (1996)107	Topiramate 800	NR	Abnormal thinking	44% (21/48)
Privitera (1996)107	Topiramate 1000	NR	Dizziness	38% (18/47)
Sharief (1996)108	Topiramate 400	NR	Somnolence	35% (8/23)
Tassinari (1996)109	Topiramate 600	NR	Headache	27% (8/30)
Willmore (1996)110	Valproate 90 mg/kg	NR	Nausea	48% (37/77)
Anhut (1994)111	Gabapentin 900	63% (33/52)	Somnolence	22% (24/111)
Anhut (1994)111	Gabapentin 1200	68% (76/111)	Somnolence	13% (7/52)
Messenheimer (1994)112	Lamotrigine 400	NR	Rash	7% (3/44)
Bourgeois (1993)113	Felbamate 3600	NR	Headache	40% (12/30)
FSG (1993)114	Felbamate 3600	NR	Anorexia	49% (18/37)
Matsuo (1993)115	Lamotrigine 300	NR	Headache	32% (23/71)
Matsuo (1993)115	Lamotrigine 500	NR	Dizziness	54% (39/72)
McLean (1993)116	Gabapentin 600	88% (89/101)	Dizziness	25% (13/53)
McLean (1993)116	Gabapentin 1200	91% (49/54)	Somnolence	36% (36/101)
McLean (1993)116	Gabapentin 1800	87% (46/53)	Somnolence	20% (11/54)
Schmidt (1993)117	Zonisamide 20 mg/kg	59% (42/71)	Fatigue	23% (16/71)
Sivenius (1991)118	Gabapentin 900	NR	Drowsiness	25% (4/16)
UKGSG (1990)119	Gabapentin 1200	62% (38/61)	Somnolence	14.8% (9/61)
Jawad (1989)70	Lamotrigine 400	NR	NR	NR

mg/day Milligrams per day

NR Not reported

mg/kg Milligrams per kilogram

All 30 included studies of polytherapy reported adverse effects of the new drug treatment. The overall percentage of patients who experienced any side effects was reported by 16 studies, and ranged from 55 percent to 94 percent (Table 14). Somnolence was the most common adverse effect in nine studies, and dizziness was the most common adverse effect in four studies. All details of the adverse effects in the 30 studies appear in Evidence Table 60.

To summarize the available data on adverse effects, we focused on whether the adverse effects in a given patient were severe enough to warrant discontinuation of the new drug (i.e., trial exit). In trials of polytherapy, an add-on drug may be more likely or less likely to be discontinued due to adverse effects compared to add-on placebo.

Figure 25. Forest plot: polytherapy and trial exits (more...)

   Figure 25. Forest plot: polytherapy and trial exits due to adverse effects (high-dose)

Figure 26. Forest plot: polytherapy and trial exits (more...)

   Figure 26. Forest plot: polytherapy and trial exits due to adverse effects (low-dose)

Percentage of patients exiting trials due to adverse effects. All 30 trials of polytherapy reported this outcome. We meta-analyzed these data using the same methods that we used to analyze seizure frequency. The effect sizes are plotted in Figure 25, and the details of the high-dose meta-analysis appear in Evidence Table 79. The random-effects summary Cohen's h was significantly negative (-0.18, CI: -0.26 to -0.11). Thus, patients in the high-dose groups were significantly more likely to exit trials due to adverse effects compared to patients in placebo groups. The estimated summary percentages were 8 percent for the high-dose groups (CI: 6 percent to 10 percent) and 4 percent for the placebo groups (CI: 2 percent to 5 percent). Similar results were observed for the low-dose meta-analysis (Figure 26 and Evidence Table 80). The random-effects summary statistic was -0.16 (CI: -0.23 to -0.08), and the estimated summary percentage for the low-dose groups was 7 percent (CI: 5 percent to 9 percent).

We performed the same sensitivity analyses and they did not overturn any of our findings (Evidence Table 81 and 82).

Thus, adding a certain AED to a patient's drug regimen is more likely to cause adverse effects resulting in trial exit compared to adding a placebo. This finding persisted through multiple sensitivity analyses.

Figure 27. Tradeoff between seizure frequency and adverse (more...)

   Figure 27. Tradeoff between seizure frequency and adverse effects

Tradeoff between seizure frequency and adverse effects. We next evaluated the tradeoff between seizure frequency and adverse effects in trials of polytherapy. In the section on seizure frequency, we concluded that adding a drug to a patient's regimen is more likely to reduce seizures compared to adding a placebo. However, in the section on adverse effects, we concluded that adding a drug is also more likely to cause adverse effects resulting in trial exit. To illustrate the tradeoff, we constructed a scatterplot in which the horizontal axis represented the effect size for 50 percent seizure reduction and the vertical axis represented the effect size for exiting the trial due to adverse effects (Figure 27). We inverted the vertical axis so that the ideal drug would fall in the upper right quadrant of the plot (corresponding to fewer seizures and fewer adverse effects). Forty groups of patients who received an add-on drug are included in the plot (corresponding to the 27 trials that reported both 50 percent seizure reduction and adverse effect attrition). Thirty-one of 40 patient groups (78 percent) were in the lower right quadrant (fewer seizures and more adverse effects), and seven groups (18 percent) were in the upper right quadrant. This plot demonstrates the tradeoff between seizure frequency and adverse effects. However, reductions in both seizure frequencies and side effects also seem to occur.

Quality of life

Only two of the included trials of polytherapy reported quality of life (Evidence Table 61).^{92, 93} The two trials used different scales to measure quality of life (Evidence Table 83). Due to the small number of trials, we did not perform meta-analyses of the results. Instead, we created plots indicating the trials' results for all reported subscales. In the trial by Cramer, Arrigo, Van Hammee, et al.,⁹² four of the nine subscales of quality of life showed a statistically significant advantage of levetiracetam over placebo. Each of the other five subscales showed a nonsignificant advantage of levetiracetam. A statistical power analysis of this trial was not possible due to the lack of reporting of measures of dispersion. The results of this trial suggest that polytherapy with levetiracetam improves some aspects of quality of life. In the trial by Dodrill, Arnett, Sommerville, et al.,⁹³ no statistically significant effect was found on any of the 10 subscales of quality of life. A statistical power analysis of this trial could not be conducted because these were pre-post comparisons and the authors did not report the correlations between baseline and outcome measurements. Because only two trials reported quality of life outcomes after polytherapy, evidence-based conclusions could not be made about the influence of polytherapy on quality of life.

Mood

One trial of polytherapy (add-on tiagabine) reported mood outcomes (Evidence Table 61).⁹³ The trial used eight subscales to measure mood (Evidence Table 84). None of the eight subscales showed a statistically significant improvement in mood after add-on tiagabine. A statistical power analysis of this trial was not possible because the authors did not report the correlations between baseline and outcome measurements. Because this is only one trial, drawing any evidence-based conclusions about whether polytherapy affects mood is not possible.

Cognitive function

Only one of the included trials of polytherapy reported cognitive function (Evidence Table 62).⁹³ The subscales for measuring cognitive function in this trial appear in Evidence Table 85. Because there was only one trial, we created a plot indicating its results for all reported subscales. Of the 19 subscales of cognitive function, only one (the Benton Visual Retention test, Form F) demonstrated a statistically significant effect. Patients in the placebo group improved from baseline more compared to patients who received tiagabine. A power analysis of this trial could not be conducted because the authors did not report the correlations between baseline and outcome measurements. Because only one trial addressed this issue, evidence-based conclusions cannot be made about the influence of polytherapy on cognitive function.

Functional status/ability

No trials of polytherapy reported this outcome.

Ability to return to work

No trials of polytherapy reported this outcome.

Ability to return to school

No trials of polytherapy reported this outcome.

Ability to hold a driver's license

No trials of polytherapy reported this outcome.

Mortality

Nine of the 30 trials of polytherapy (30 percent) reported whether any patients died during the trial. The mortality results of these trials are listed in Evidence Table 86. The mortality rates ranged from 0 percent to 2 percent. Five of the nine trials reported that no patients died, three trials each reported one death, and one trial reported two deaths. None of the authors attributed the deaths to the add-on drugs.

Optimization of Current Drug Therapy

The previous two parts of the present question addressed strategies related to the use of new AEDs or new combinations of AEDs in patients with treatment-resistant epilepsy. In this section, we assess strategies designed to optimize the effectiveness of a patient's current drug regimen. Strategies designed to optimize current drug therapy seek to improve patient outcomes by either; (1) reducing seizure frequency without increasing the incidence (or intensity) of the side effects associated with AED treatment or, (2) by reducing the side effects of AED without increasing seizure frequency, seizure severity, or the onset of a new seizure type. Ideally, a drug regimen would both decrease seizure frequency and reduce side effects. However, as shown above, this rarely occurs in patients with treatment-resistant epilepsy, and a trade-off exists between the intensity of drug treatment and the incidence and severity of associated side effects. Thus, in order for an optimization strategy to be of value, it must either lead to reductions in seizure frequency or reductions in side effects (and/or improvements in quality of life, cognitive function, and mood) while not leading to increases in the other.

Published literature describes three different methods for optimizing drug therapy in patients with treatment-resistant epilepsy: (1) increasing the dose of the current drug (or drugs) to maximum tolerable levels, (2) modifying the frequency of dosing, and (3) reducing the total number of drugs. In this subquestion, we evaluate the literature pertaining to all three of these strategies.

Synthesis of study results

The assessment of study quality presented above indicates that, given the present evidence base, definitive conclusions cannot be drawn about whether implementation of the drug reduction strategy is effective in improving outcome in patients with treatment-resistant epilepsy. Acknowledging this, we have instead evaluated the available data with the aim of determining whether the implementation of this strategy may plausibly be effective in improving outcome among patients with treatment-resistant epilepsy.

Table 15. Outcomes in studies of drug reduction

Table 15. Outcomes in studies of drug reduction

	Reported Seizure Outcomes					Reported Nonseizure Outcomes
Reference	Difference in Absolute Seizure Frequency	Percent Change in Seizure Frequency	Percent Patients Seizure-free	Percent of Patients With >50 Percent Reduction in Seizure Rate	Percent of Patients With an Increase in Seizure Frequency	Mood	Cognitive Function	Adverse Events	Mortality
Controlled trials performed outside of the United States
May (1992)121	a
Duncan (1990)122						b	c	d
Thompson (1982)126
Case series performed in the United States
Mirza (1993)120
Case series performed outside of the United States
Specht (1989)123	e	e	e	e	e
Callaghan (1984)124				?
Schmidt (1983)125	e	e	e	e	e
Number of articles addressing outcome	4f	2	4	4	3	2	3	5	7

a

May, Bulmahn, Wolhlhuter et al.¹²¹ reported that no statistically significant between groups differences in seizure frequency were seen but did not present any data

b

Mood data abstracted from Kendrick, Duncan, and Trimble¹³⁶

c

Cognitive function data abstracted from Duncan, Shorvon, and Trimble¹³⁷

d

Adverse events data abstracted from Duncan, Shorvon, and Trimble¹³⁸

e

Data calculated by ECRI from individual patient data

f

Does not include May, Bulmahn, Wolhlhuter et al.¹²¹ (see footnote a above).

Not all of the outcomes listed by the Technical Expert Panel (see question-specific inclusion criteria above) were reported on in all of the articles in the present evidence base. The reported outcome measures and the articles that contained data pertaining to these outcome measures are presented in Table 15.

Seizure frequency outcomes

As stated previously, the goal of the drug reduction strategy is to remove a drug (or drugs), thereby reducing the occurrence of (or the risk for) adverse effects associated with the use of AEDs. This goal must be accomplished without increasing seizure frequency to unacceptably high levels. Although reductions in seizure frequency are desirable and may indeed occur, they are not, in this instance, to be expected. Consequently, studies needed only to demonstrate that implementation of the drug reduction strategy resulted in other benefits such as reductions in adverse events, increases in cognitive function, increases in quality of life, reduced cost, etc. This means that trials that evaluate changes in seizure frequency that result from drug reduction strategy must also demonstrate, through hypothesis testing, that clinically meaningful increases in seizure frequency did not occur.

In such trials, which are akin to studies of therapeutic equivalence,^p classical hypothesis testing (with the usual null hypothesis that there is no difference between the interventions) is inappropriate.^127–131 This is because the desired result of a bioequivalence study would be to prove the null hypothesis by showing that no increases in seizure frequency occurred in the treatment group when compared to the comparison group.^q An alternative hypothesis allows meaningful statistical analyses to be performed. In this instance, the alternative hypothesis is that seizure frequency increases in the treatment group will be less than a prespecified level, δ, above the seizure frequencies seen in the comparison group (H_A: X_DRS-X_c<δ; where H_A = alternative hypothesis). Thus, to demonstrate that implementation of a drug reduction strategy does not lead to increases in seizure frequency, any difference in seizure frequency between the treatment group and the comparison group (along with its 95 percent confidence intervals) must fall entirely below δ. Confidence intervals that extend above δ indicate that the alternative hypothesis has not been refuted and implementation of the strategy may lead to increases in seizure frequency.

Table 16. Possible decisions based on hypothesis test (more...)

Table 16. Possible decisions based on hypothesis test

	Testing H0: XC = XDRS		Testing H0': XDRS≥Xc + δ
True Difference XDRS - Xc	Fail to Reject (good for DRS)	Reject (bad for DRS)	Reject (good for DRS)	Fail to Reject (bad for DRS)
XDRS - Xc = 0 (good for DRS)	Correct Decision	Type I error	Correct Decision	Type II error
XDRS - Xc = δ (bad for DRS)	Type II error	Correct Decision	Type I error	Correct Decision

Adapted from Blackwelder¹²⁸

DRS Drug reduction study

H₀ Null hypothesis (standard)

H₀' Null hypothesis (therapeutic equivalence studies)

X_c Mean seizure frequency in control group

X_DRS Mean seizure frequency in drug reduction therapy group

δ Predefined difference in mean seizure frequency above which use of drug reduction study is unacceptable

As is the case with conventional hypothesis testing, a study should be designed with adequate power to avoid the possibility of making Type II statistical errors. As shown in Table 16, when performing hypothesis testing using the alternative hypotheses, the “standard” rules of a Type I error and a Type II error become reversed. Thus, a Type I error is made if the difference X_DRS-X_c is less than δ when, in fact, the difference is greater than or equal to δ, and a Type II error is made when the difference is greater than or equal to δ when it is actually less than δ.

Given the information above, the seizure outcomes of importance in this evaluation are those that assess increases in seizure frequency. Outcomes that assess improvements in seizure frequency (proportion of patients seizure-free, proportion of patients achieving a greater than 50 percent decrease in seizure frequency), though interesting, are of secondary importance. As a result, we have focused this section of the report on three seizure frequency outcomes (absolute seizure frequency, percentage change in seizure frequency, and proportion of patients with an increase in seizure frequency). Data pertaining to the remaining seizure frequency outcomes are summarized in Evidence Table 99 but are not discussed further.

Absolute seizure frequency.Two of the three controlled trials included for this subquestion presented data on (mean or median) absolute seizure frequency. Two studies are too few to allow a quantitative analysis to be performed. As a result, we present the findings of our semi-quantitative analysis of the available data. These data are presented in Evidence Table 99.

As discussed above, to demonstrate that seizure frequency does not increase in patients using a drug reduction strategy, the strategy must be shown not to cause clinically important increases in seizure frequency (X_DRS- X_c ≤ δ). This requires the authors to explicitly state what they consider a meaningful increase in seizure frequency (δ). Based on this seizure frequency, they should then state the size of the study (power) necessary to overturn the null hypothesis that seizure frequency in patients who received drug reduction will increase above this predefined seizure frequency.^r Neither of the two controlled trials that reported this outcome stated what they considered to be a clinically important difference in seizure frequency, nor did they perform a power analysis.^s

All of the statistical analyses presented in these two articles tested the traditional null hypothesis that no between-groups differences in seizure frequency exist. Thus, their analyses essentially attempted to prove the null hypothesis that there was no change in seizure frequency. As discussed above, this is inappropriate.

Because the investigators did not determine the power of their study and because their statistical analyses were not appropriate for the clinical question of interest, the seizure frequency analyses in the articles are of limited value. However, summary data from these studies may still be used to provide some useful information. This can be accomplished by calculating the mean difference in seizure frequency (and its CI) between the drug-reduction group and the control group for each study. The upper CI of this difference can then be used to determine the maximum magnitude of increase in seizure frequency that will not lead to the alternative hypothesis (H_A: X_DRS - X_c ≤ δ) being accepted over the null hypothesis (H₀: X_DRS - X_c >δ). Such an approach, however, requires that the study report seizure frequency data in such a way that a difference can be calculated.

Duncan, Shorvon, and Trimble¹²² summarized their seizure frequency data in terms of mean seizure frequencies along with its standard deviation. The range cannot be used to calculate a valid standardized between-groups difference. No other measures of dispersion were reported. As a result, the seizure frequency data presented by Duncan, Shorvon, and Trimble¹²²cannot be used to determine whether implementation of the drug reduction strategy leads to clinically important increases in seizure frequency.

Unlike Duncan, Shorvon, and Trimble,¹²² Thompson and Trimble¹²⁶ presented mean seizure frequency along with its dispersion (expressed in terms of standard deviations). However, the analysis described above still cannot be performed, because the technique is sensitive to pretreatment differences in seizure frequency. Although no statistically significant between-groups differences in seizure frequency data were detected at pretreatment, a between-groups differences in seizure frequency at baseline did exist, and these differences were large enough to lead to biased posttreatment effect size estimates. For example, the mean pretreatment frequency for partial seizures in the drug reduction group was 21.1 (SD: 34.6) seizures per week compared to 6.8 (SD: 9.7) per week in the control group. Thus, patients in the drug reduction arm were experiencing more than three times the number of seizures per week compared to the patients in the control arm at study onset. Consequently, the study is biased against finding that the implementation of the drug reduction strategy will lead to increases in seizure frequency.

To summarize, the data from the currently available controlled trials could not be used to draw evidence-based conclusions about whether or not implementation of the drug reduction strategy leads to increases in seizure frequency.

Although none of the four included case series reported on this outcome, two studies did present individual patient data that allowed us to summarize the seizure frequency data both pre and post implementation of the drug reduction strategy. These data are presented in Evidence Table 99. They do not suggest that seizure frequencies increase following implementation of the drug reduction strategy. However, because these data originate from two uncontrolled studies and, because seizure frequency in patients with treatment-resistant epilepsy commonly demonstrates regression to the mean (see Methodology section),^{23, 132} this observation does not provide convincing evidence to support the contention that implementation of the drug reduction strategy does not lead to increases in seizure frequency.

As will be seen in the following sections, other seizure frequency-based outcomes suggest that drug reduction strategy may lead to large increases in seizure frequency in some patients.

Mean or median percentage change in seizure frequency. None of the three included controlled trials presented data on the percentage change in seizure frequency following implementation of the drug reduction strategy. Thus, conclusions about this outcome can only be based on case series data.

Two of the four case-series studies presented individual patient data that allowed us to assess this outcome. These data show that the median percentage change in seizure frequency from baseline was -0.09 percent^t (Range: -100 percent to 412 percent) in the study by Specht, Boenigk, Wolf, et al.¹²³ and -0.12 percent (Range: -100 percent to 2,678 percent) in the study by Schmidt.¹²⁵ In both studies, more than 50 percent of the patients experienced a reduction in seizure frequency following implementation of the drug reduction strategy, and just under 50 percent of the patients experienced an increase in seizure frequency from baseline (43 percent of patients in the study by Specht, Boenigk, Wolf, et al.¹²³ and 47 percent of patients in the study of Schmidt¹²⁵). The proportion of patients who experience an increase in seizure frequency from baseline is addressed in more detail in the following section of the report.

Thus, these data suggest that a high proportion of patients (close to 50 percent) may experience increases in seizure frequency following the implementation of the drug reduction strategy. The data also suggest that some patients may experience decreases in seizure frequency. Given that regression to the mean is known to influence seizure frequency data,^23,132 some of these observed reductions in seizure frequency were probably a manifestation of this bias. The only other possible explanation is that the withdrawn drug was somehow causing seizures.

Proportion of patients with an increase in seizure frequency. None of the three included controlled trials presented data on the proportion of patients with an increase in seizure frequency following implementation of the drug reduction strategy. Thus, conclusions on this outcome can only be based on data from case series.

One of the four case series presented data on increases in seizure frequency. Callaghan, O'Dwyer, and Keating¹²⁴ reported that three of the 35 patients (9 percent) included in their study demonstrated an increase in seizure frequency. This information, however, is of limited value because the authors did not define what they meant by “worse.” Consequently, the magnitude of the reported increase in seizure frequency in these three patients cannot be determined, and no conclusions can be drawn as to whether these increases were clinically important.

Figure 28. Increase in seizure frequency and drug reduction (more...)

   Figure 28. Increase in seizure frequency and drug reduction strategies

Percentage presented in parentheses is the actual proportion of patients with seizure frequencies greater than the percent increase in seizure frequency shown on the X-axis. The diamond and error bars represent the effect size and 95% CI.

Two other articles presented individual patient data that allowed us to calculate the proportion of patients with an increase in seizure frequency (Schmidt¹²⁵ and Specht, Boenigk, Wolf, et al.¹²³). Because the magnitude of a clinically important increase in seizure frequency remains ambiguous, we believed that arbitrarily reporting the proportion of patients above any single frequency was inappropriate. Instead, we calculated the proportion of patients that demonstrated increases in seizure frequency above a series of percentage increases from baseline (thresholds). These data, which are presented in Evidence Table 99, are summarized in Figure 28.

This figure shows that a statistically significant proportion of patients in both case series exhibited large (>100 percent) increases in seizure frequency when compared to baseline (27.8 percent of patients in the study by Schmidt¹²⁵ and 8.6 percent in the study by Specht, Boenigk, Wolf, et al.¹²³). In neither study did these patients have unusually low seizure frequency rates at the onset of the study, suggesting that implementation of the drug reduction strategy will result in increased seizure frequency in a significant proportion of patients.

Mood

Two of the three controlled trials presented data on changes in mood following the implementation of a drug reduction strategy. Two studies are too few to allow a quantitative analysis of the available data to be performed. As a result, we present the findings of our semi-quantitative analysis of the available data. These data are presented in Evidence Table 99.

Both Thompson and Trimble¹²⁶ and Duncan, Shorvon, and Trimble¹²² presented mood data collected using two validated self-administered psychometric instruments. These instruments were the Middlesex Hospital Questionnaire (MHQ) and the Mood Adjective Checklist (MACL).

The MHQ is a self-administered questionnaire that measures six domains and provides a composite score. This instrument is commonly used as an aid in the diagnosis of clinical depression. The six domains that are assessed include: Free-floating anxiety (F-FA), phobic anxiety (PHO), obsessive-compulsive (OBS), somatic anxiety (SOM), depressive traits (DEP), and hysteric (HYS) traits. Although Duncan, Shorvon, and Trimble¹²² presented data for all six domains, Thompson and Trimble¹²⁶ only reported on two (F-FA and DEP). Neither Duncan, Shorvon, and Trimble¹²² nor Thompson and Trimble¹²⁶ found a statistically significant between-groups difference in any of the domains measured using the MHQ following completion of drug reduction. Nor were any trends in the data detected that would indicate that mood either improved or deteriorated following drug reduction.

The MACL is a standardized scale commonly used to detect alterations in mood across five domains. These domains provide measures of anxiety, fatigue, hostility, vigor, and depression, along with a composite score. Although both studies measured mood alterations using this instrument, only Thompson and Trimble¹²⁶ presented relevant data in their article. Again, as was the case with reporting of the data obtained using the MHQ, Thompson and Trimble¹²⁶ did not report data for all of the measured domains (in this case data for the domain “hostility” was not reported) and no explanation was provided as to why this was the case. Analysis of data abstracted from Thompson and Trimble¹²⁶ did not find statistically significant between-group differences in any of the domains measured using the MACL. Nor were any trends in the data detected that would indicate that mood either improved or deteriorated following drug reduction. This finding was corroborated by Duncan, Shorvon, and Trimble¹²² who reported that, “There were no statistically significant differences between the four groups^u on the anxiety, depression, fatigue, vigor, or hostility subscales of the Mood Adjective Checklist.”¹²²

Cognitive function

All three of the controlled trials included for the present subquestion presented data on changes in cognitive function following implementation of a drug reduction strategy when compared to a control group comprised of patients who were maintained on their current polytherapy drug regimen. Three studies are too few to allow a quantitative analysis to be performed. As a result, we present the findings of our semi-quantitative analysis of the available data. These data are presented in Evidence Table 99.

All three studies measured cognitive function using a series of standardized clinical tests. These tests included tests of concentration and attention, memory, and tests of psychomotor performance.

Figure 29. Drug reduction strategies and tests of concentration/attention (more...)

   Figure 29. Drug reduction strategies and tests of concentration/attention

Tests of concentration/attention. All three controlled trials measured concentration/attention before and after the implementation of the drug reduction strategy. These data are summarized in Figure 29. May, Bulmahn, Wohlhuter et al.¹²¹ used the d2 test and the modified version of the Frankfurt Concentration Test for Children (FCTC). Duncan, Shorvon, and Trimble¹²² used the Letter Cancellation Task (LCT), and Thompson and Trimble¹²⁶ used the Stroop test (ST) and a test of visual scanning speed (VSS).

Data from only one of the three studies, Duncan, Shorvon, and Trimble,¹²² suggested that concentration improvement was statistically significant among patients who had undergone drug-reduction when compared to patients in the control group. The only statistically significant posttreatment benefit was seen in patients who were removed from sodium valproate (t = 4.245; p = 0.000108^v) followed by patients who were removed from phenytoin (t = 1.965; p = 0.056). Assessment of the pretreatment LCT data, however, suggested the presence of selection bias, with patients who were removed from sodium valproate having statistically significantly higher baseline LCT scores compared to those in the control group (t = 3.404; p = 0.00140). Thus, the posttreatment between-groups difference was essentially the same as the pretreatment difference.

No such bias was found to have affected the LCT scores on removal of phenytoin and these data suggest that removal of phenytoin may lead to an improvement in concentration/attention in some patients. However, interpretation of the importance of a mean improvement of 18 points is difficult because the authors did not indicate if such a between-groups difference was clinically important.

Figure 30. Drug reduction strategies and the Frankfurt (more...)

   Figure 30. Drug reduction strategies and the Frankfurt Concentration Test for Children

Pre- and posttreatment Frankfurt Concentration Test for Children data from May (1992)¹²¹

May, Bulmahn, Wohlhuter et al.¹²¹ argued that their FCTC data showed a statistically significant between-groups difference in patients in the drug reduction arm (all of whom had phenytoin removed). Figure 30 shows graphically their reported pre-and posttreatment FCTC data. The data, as presented in the article, can lead to different conclusions. Changes in FCTC score seen from baseline between the two arms of their study were compared instead of the posttreatment data alone. Because FCTC scores improved in the reduction group and declined in the control group, the comparison found a significant between-groups difference. As shown in Figure 30, the changes in FCTC could reasonably be argued to be due to regression to the mean rather than an effect of treatment.

Figure 31. Drug reduction strategies and tests of memory (more...)

   Figure 31. Drug reduction strategies and tests of memory

Abbreviations:

LGT: Lern- und Gedachtnis Test

DSF: Digit scan forwards

DSB: Digit scan backwards

PHT: Phenytoin vs. Control

CBZ: Carbamezapine vs. Control

VPA: Valprioc acid vs. Control

IR: Immediate recall

DR: Delayed recall

R: Recognition

Memory. All three controlled trials measured memory before and after the implementation of the drug reduction strategy. May, Bulmahn, Wohlhuter et al.¹²¹ measured memory using a digit span and an immediate recall of pictures, and a delayed-recall task at the end of the test session that were taken from the Lern- und Gedachtnis-test (LGT-3). Duncan, Shorvon, and Trimble¹²² measured memory using a digit span task derived from the Wechsler Adult Intelligence Scale. Thompson and Trimble¹²⁶ used an immediate-recall and delayed-recall of pictures task that they developed and validated themselves.^133–135 Data on the effects of drug reduction on memory collected in these studies are summarized in Figure 31.

Figure 32. Drug reduction strategies and digital scanning (more...)

   Figure 32. Drug reduction strategies and digital scanning score

Data from Duncan (1990)¹²² showing effects of valprioc acid removal on digital scanning score

When considered as a whole, these data do not provide evidence that drug reduction leads to improved memory. Although the data from Duncan, Shorvon, and Trimble¹²² suggest that drug reduction may lead to statistically significant improvements in short-term memory (as measured by digital scanning backwards) in some patients who were removed from sodium valproate, these results may be biased. This is illustrated by Figure 32, which shows that a pretreatment difference in short-term memory existed between patients removed from sodium valproate and patients in the control group. Although this difference was not statistically significant (t = 1.842; p = 0.072; Hedges' d = 0.53; CI: -0.05 to 1.11), it is large enough to have biased the posttreatment between-groups effect size data. Indeed, Duncan, Shorvon, and Trimble¹²² reported that their statistical analyses showed that removal of phenytoin, carbamazepine, or sodium valproate did not lead to improvements in short-term memory.

Figure 33. Drug reduction strategies and tests of psychomotor (more...)

   Figure 33. Drug reduction strategies and tests of psychomotor function

FT: Finger tapping

DH: Dominant hand

NDH: Non-dominant hand

PB: Pegboard

PRF: Pursuit Rotor Failure

PFD: Pursuit Failure Duration

PHT: Phenytoin vs. Control

CBZ: Carbamezapine vs. Control

VPA: Valprioc acid vs. Control

Psychomotor function. All three controlled trials measured psychomotor function before and after drug reduction. May, Bulmahn, Wohlhuter et al.¹²¹ used the pegboard, a pursuit rotor, and tapping. Duncan, Shorvon, and Trimble¹²² used tapping alone, as did Thompson and Trimble.¹²⁶ Data on the effects of drug reduction on psychomotor function in these studies are summarized in Figure 33. These posttreatment, between-groups effect sizes do not provide evidence that implementation of the drug reduction strategy results in improved psychomotor function.

Figure 34. Drug reduction strategies and psychomotor (more...)

   Figure 34. Drug reduction strategies and psychomotor function

Pre-and posttreatment psychomotor function data presented by May (1992)¹²¹

Again, these findings contradict the conclusions drawn by the authors. May, Bulmahn, Wohlhuter et al.¹²¹ reported that their data demonstrated a statistically significant improvement in psychomotor function when measured using finger tapping (with the dominant hand) and pursuit rotor failure (again using the dominant hand). Figure 34 shows graphically the pre-and posttreatment data reported in these studies. These data can lead to different conclusions. As stated above, the discrepancy is due to comparing the changes in psychomotor function from baseline between the two arms of the study instead of comparing the posttreatment data alone. As shown by such a comparison in Figure 34, changes in psychomotor function in the drug reduction arm of the study could reasonably be argued to be due to regression to the mean rather than treatment.

Adverse events

Identification of treatment-related morbidities can only be achieved by comparing reported adverse event rates in patients who underwent drug reduction against a control group comprised of patients who were maintained on their current treatment regimen. Although case series identify possible adverse events that may be associated with a treatment, their data cannot be used to draw evidence-based conclusions about whether these adverse events are a consequence of the drug reduction strategy. As a result, we only considered adverse events data abstracted from controlled trials in this section of the report. However, adverse events data abstracted from the four case series included in the present evidence base are tabled in Evidence Table 100.

One of the three controlled trials reported relevant data on adverse events. Patients included in the drug reduction arm of the study by Duncan, Shorvon, and Trimble¹²² suffered no additional adverse events compared to patients included in the control arm. Thus, although the patients undergoing drug reduction did experience some adverse events, these adverse events cannot be attributed to drug reduction strategy used in this study.

Mortality

No patients were reported to have died during any of the seven included studies. Thus, no evidence exists to suggest that implementation of the drug reduction strategy leads to increased mortality.

Comparisons of Drug Strategies

None of the trials that met the inclusion criteria directly compared the drug strategies. Indirectly comparing two of the drug strategies is possible if the patients enrolled in all of the trials of all strategies were similar. The drug reduction strategy cannot be compared with the other two strategies, because of the differing intentions of investigators in these latter trials. The intent of the trials of polytherapy and sequential monotherapy was to reduce seizures without causing adverse effects, whereas the intent of the trials of drug reduction therapy was to reduce the number of drugs without increasing seizures.

To determine whether trials of polytherapy and sequential monotherapy enrolled similar patients, we compared the number of drugs given to patients receiving these strategies. Differences in the number of drugs likely mean that the severity of epilepsy in patients who received polytherapy was different from that in patients who received sequential monotherapy. Among 11 trials of sequential monotherapy that reported this percentage, two (18 percent) reported that more than half of the patients were receiving two or more prior AEDs. By contrast, among the 18 trials of polytherapy that reported this percentage, 16 (89 percent) reported that more than half of the patients were receiving two or more prior AEDs. These percentages are significantly different (χ²(1) =14.5, p = 0.00014). Thus, patients who received polytherapy had more severe epilepsy compared to patients who received sequential monotherapy. This difference precludes comparison of the quantitative results of the two strategies.

A qualitative comparison, however, suggests that polytherapy is clinically preferable to sequential monotherapy. In the section on sequential monotherapy, the evidence indicated that some patients had harmful increases in seizures as a direct result of the treatment, and whether sequential monotherapy caused any patients to become seizure-free could not be determined. In short, sequential monotherapy appeared more likely to be harmful than beneficial. By contrast, the reverse was true for polytherapy. Adding a drug reduced seizures by 50 percent in many patients, whereas adverse effects causing trial exits were rare. By inference, this suggests that polytherapy is preferable to sequential monotherapy.

Further, as discussed above, patients who received polytherapy had been receiving more drugs before the trial, thus they likely had more severe epilepsy. Patients with more severe epilepsy are, by definition, more difficult to treat. Thus, even in a more difficult-to-treat population, polytherapy helped many patients. This finding underscores the qualitative conclusion that polytherapy is preferable to sequential monotherapy for patients with treatment-resistant epilepsy.

Surgical Interventions

In this section, we address the efficacy of surgical intervention when treatment with AEDs has failed to produce adequate seizure control. Patients who receive surgery have been determined to be treatment-resistant as part of their presurgical evaluation. For most patients, only a single surgical option will be available due to the nature and location of the lesion or condition responsible for generating their seizures. In patients undergoing temporal lobectomy, hemispherectomy, or corpus callosotomy, some variations in procedures are available.

A list of the specific surgical interventions and outcome measures addressed under surgical interventions is presented in the following section.

Number of studies addressing surgical intervention

The order of the material presented in this section differs from that presented in the discussion of other interventions. This change in organization was necessary because we required a minimum 2-year followup period for most outcome measures used to evaluate surgery. The only exceptions were outcomes for mood (depression and psychosis), cognitive function (IQ and memory), and complications and mortality related to surgery. We shortened the minimum required followup time for these outcomes because they may manifest themselves relatively early after surgery.

Different studies make up the evidence base for each outcome. We will separately discuss each outcome and its specific evidence base under each surgical intervention rather than examining all of the studies in the evidence base for a single intervention. This section, number of studies addressing surgical intervention, presents an overview of all of the studies meeting our inclusion criteria for each surgical intervention examined under Key Question #5. The actual evidence base for each intervention and outcome will be discussed separately later in this report.

Table 17. Epilepsy surgery studies

Table 17. Epilepsy surgery studies

Intervention	Total Number of Studiesa
Temporal Lobe Surgery	105
Corpus Callosotomy	26
Frontal Lobe Surgery	18
Hemispherectomy	11
Multiple Subpial Transection	10
Parietal Lobe Surgery	2
Occipital Lobe Surgery	2
Surgical Controls	12

a

Seven studies reported on more than one surgery category and are therefore double counted in this table

One hundred and seventy-nine studies met our inclusion criteria for surgical intervention. We provide a listing of each study meeting the inclusion criteria for each surgical intervention in Evidence Table 102 and a summary in Table 17. Only two studies each were found to meet our inclusion criteria for parietal lobe and occipital lobe surgery for the treatment of epilepsy. Consequently, we did not assess these interventions.

Evidence Tables 103 to 108 provide general information on each of the studies examined in this report organized into tables according to surgical intervention or reporting of control patients. The information in these tables includes the years during which the studies were conducted, the country in which the study was conducted, the primary center where the surgery was performed, if the study was conducted in multiple centers, whether patients were selected retrospectively or prospectively, and the study design.