Presented at the 26th Annual Forum of the American College of Veterinary Medicine, San Antonio, TX, June 2008.
Corresponding author: Karen R. Muñana, Department of Clinical Sciences, College of Veterinary Medicine, North Carolina State University, 4700 Hillsborough Street, Raleigh, NC 27606; e-mail: email@example.com
Background: The placebo effect is a well-recognized phenomenon in human medicine; in contrast, little information exists on the effect of placebo administration in veterinary patients.
Hypothesis: Nonpharmacologic therapeutic effects play a role in response rates identified in canine epilepsy trials.
Animals: Thirty-four dogs with epilepsy.
Methods: Meta-analysis of the 3 known prospective, placebo-controlled canine epilepsy trials. The number of seizures per week was compiled for each dog throughout their participation in the trial. Log-linear models were developed to evaluate seizure frequency during treatment and placebo relative to baseline.
Results: Twenty-two of 28 (79%) dogs in the study that received placebo demonstrated a decrease in seizure frequency compared with baseline, and 8 (29%) could be considered responders, with a 50% or greater reduction in seizures. For the 3 trials evaluated, the average reduction in seizures during placebo administration relative to baseline was 26% (P= .0018), 29% (P= .17), and 46% (P= .01).
Conclusions and Clinical Importance: A positive response to placebo administration, manifesting as a decrease in seizure frequency, can be observed in epileptic dogs. This is of importance when evaluating open label studies in dogs that aim to assess efficacy of antiepileptic drugs, as the reported results might be overstated. Findings from this study highlight the need for more placebo-controlled trials in veterinary medicine.
The placebo effect is a well recognized, but poorly understood phenomenon that involves a nonspecific psychological or physiological therapeutic effect of a medical intervention that lacks specific activity for the condition being treated.1 Early medical practices were based on the placebo effect, wherein placebos were administered with the purpose of producing a desired therapeutic response. More recently, the use of placebos has focused primarily on its role as a control in randomized-clinical trials that allows for an unbiased estimate of the treatment effects of the agent being evaluated.
Results from numerous human trials have demonstrated that placebos can improve subjective and objective outcomes in patients with a wide range of clinical conditions. A beneficial effect of placebo administration has been reported in 60–90% of all human diseases,2 including musculoskeletal, respiratory, cardiac, dermatologic, gastrointestinal, and nervous system disorders. Furthermore, a placebo response rate of approximately 35% is commonly cited in the medical literature,3 although higher rates have been reported and are most frequently seen in diseases with clinical signs that wax and wane, fluctuate, or spontaneously remit.4 Because of the potential magnitude of this effect, placebo-controlled studies are considered necessary to gauge the true efficacy of a novel intervention, and are the basis for drug evaluation and approval in human medicine.
In contrast, the placebo effect has been largely disregarded in veterinary medicine, with only 2 publications identified that address the issue of a placebo effect in animals.1,5 However, with the recent emphasis placed on evidence based medicine in veterinary practice, it seems appropriate to consider the effect of placebos in veterinary patients, particularly the extent to which animals may demonstrate an improvement in disease manifestations that could be because of nonspecific effects of a therapeutic intervention.
The hypothesis tested in this study is that nonpharmacologic therapeutic effects play a role in response rates identified in canine epilepsy trials. The specific aim is to determine the magnitude of the placebo response in randomized-controlled trials evaluating new treatment modalities for refractory canine epilepsy.
Materials and Methods
Data was compiled from 3 clinical trials evaluating the safety and efficacy of novel treatments for refractory canine epilepsy in which a placebo arm was a component of the study protocol. The studies evaluated were performed by 2 of the authors (K.R.M., E.E.P.), enabling easy access to the data necessary to undertake the present analysis. A database search was performed to identify any additional placebo-controlled canine epilepsy trials that might be included in the analysis, but none were found. Inclusion criteria were similar for all 3 studies and consisted of (1) an onset of seizures between 1 and 5 years of age; (2) a normal diagnostic evaluation, including physical examination, neurological examination, CBC, chemistry profile, urinalysis, and bile acid tolerance test; (3) treatment with either phenobarbital and/or potassium bromide at established therapeutic serum levels; (4) a seizure frequency of at least 4 seizures per month or a history of cluster seizures; and (5) a 1 year documented history of seizures. All dogs were classified as having generalized seizures based on owners' descriptions of seizure episodes. Some of the dogs were also reported to have occasional partial seizures. No attempt was made to further characterize the seizure type with video monitoring or electroencephalography. The therapeutic interventions evaluated included a surgical implant,6 dietary modification,a and a novel antiepileptic drug (unpublished data). All studies were placebo controlled and blinded such that both the owner and veterinary personnel involved in evaluating the dog were unaware of the treatment being administered. The novel interventions were evaluated as add-on therapies, and as such, dogs were maintained on their previously prescribed conventional antiepileptic medication throughout the study. The surgical implant and drug study were cross over in design, and consisted of an 8-week baseline period followed by 2 treatment periods of 13–16 weeks in length, with a 4-week washout period between the 2 treatment periods. The diet trial was designed as a parallel study, and included a baseline period of 3–6 months followed by a treatment period of 6 months. In each study, the treatment group received the experimental therapeutic intervention in addition to conventional antiepileptic drug treatment, and the control group received a matching placebo in addition to conventional antiepileptic drug treatment.
Seizure monitoring was based on owner observations. Information was recorded on study forms that were adapted from those used for human seizure monitoring. Owners were instructed to complete a daily diary noting medication administered, the presence of any seizure activity, and whether any signs of illness, change in activity or attitude were noted. Owners were asked to complete an additional form immediately after any observed seizure that captured information on specific manifestations of the episode. Based on these observations, a weekly seizure frequency was compiled for each dog throughout the study. The average weekly seizure frequency for each of the study periods (baseline, placebo administration, and active treatment) was calculated. As is the convention for human epilepsy trials, a positive treatment response was defined as a 50% or greater reduction in seizure frequency compared with baseline. To assess for any changes in drug levels that might affect seizure frequency, mean serum antiepileptic drug concentrations obtained at the end of baseline, placebo administration and active treatment were calculated and compared for each trial.
Log-linear models were developed for each of the study designs (cross-over and parallel) to evaluate weekly seizure frequency during treatment and placebo relative to baseline. Such models account for the treatment effect as well as the effect of time.
For a cross-over study, the study period can be divided into 3 periods: period 1 corresponding to the baseline followed by 2 periods for the treatment and placebo. Denote by Yij the seizure count for the ith dog at the jth week and let μij=E(Yij) be the average seizure count per week. Because the dogs are randomized to the treatment and placebo, we then consider the following log-linear model for μij:
where I(·) is an indicator function. With this model specification, β0 is the log average seizure count per week at period 1 (assumed to be independent of treatment sequence because of randomization), (β1+β2)/2 can be interpreted as the average treatment effect relative to baseline and (β3+β4)/2 is the average placebo effect relative to baseline. Then β1−β3 is the treatment effect relative to placebo at period 2, β2−β4 is the treatment effect relative to placebo at period 3, and (β1−β3+β2−β4)/2 is the average treatment effect relative to the placebo.
For a parallel study, the study period is divided into 2 periods: period 1 for the baseline and period 2 for the treatment or the placebo. Using the above notations, we consider the following log-linear model for μij:
where β0 has the same interpretation as before, β1 is the effect of the treatment relative to the placebo, and β2 is the placebo effect plus the possible time effect of period 2 relative to period 1.
The assumption was made that the variance of Yij has the form of Poisson variance function with a possible overdispersion. To account for the correlation in the longitudinal data in each patient, the inference on effect parameters was carried out by the generalized estimating equation (GEE) methodology with equal correlation matrix as the working correlation matrix.7 A commercially available software programb was used to implement the analysis.
Thirty-four dogs were included in this study: 9 dogs as participants in the study evaluating the surgical implant, 14 as participants in the study evaluating novel drug therapy, and 11 as participants in the study evaluating dietary modification. In the latter study, which was designed as a parallel study, 5 of 11 dogs were randomized to placebo treatment while the remaining 6 dogs were randomized to the active treatment.
Data on seizure frequency during active treatment and placebo administration relative to baseline for dogs in the 3 trials are summarized in Table 1 and Figure 1.
Table 1. Seizure frequency relative to baseline during administration of active treatment and placebo for dogs in epilepsy trials.
Responders: dogs with a ≥ 50% reduction in seizure frequency.
Log-linear analysis of weekly seizure frequency for the population in dogs in each trial identified a reduction in seizures during placebo administration of 26–46%. The 2 cross-over studies involving the surgical implant and the drug trial demonstrated a reduction in seizures during placebo administration compared with baseline of 26% (P= .0018) and 29% (P= .17), respectively. For the dietary modification trial, which was designed as a parallel study, dogs administered placebo had a 46% (P= .01) decrease in seizure frequency from baseline.
There were no significant changes in serum concentrations of either phenobarbital or bromide throughout the study, although dogs in both the surgical implant trial and the drug trial demonstrated a trend toward increasing serum bromide levels during placebo administration relative to baseline that was not associated with a change in potassium bromide dosage. Similarly, mild increases in both phenobarbital and bromide levels from baseline were seen in dogs in the placebo arm of the diet trial (Table 2).
Table 2. Serum antiepileptic drug concentrations during baseline, active treatment, and placebo administration for dogs in epilepsy trials.
Phenobarbital Serum Concentration (μg/mL)
Bromide Serum Concentration (mg/dL)
Values reported as mean (standard error).
Results from this study suggest that a positive response to placebo administration can be observed in epileptic dogs. A decrease in seizure frequency was seen in more than half of the dogs in the trials during treatment with placebo when compared with baseline values. Eight of 28 dogs (29%) evaluated in this study could be classified as responders, based on a 50% or greater reduction in seizure frequency. Futhermore, placebo administration was associated with an estimated reduction in seizure frequency of 26–46% in the population of dogs studied. This response was identified across the different treatment modalities evaluated in these trials, which included a surgical implant, a novel antiepileptic drug, and a dietary modification, suggesting that this effect is not specific with respect to treatment type.
The findings from this study are consistent with results from similar studies evaluating the placebo effect in human epilepsy. A recent meta-analysis of randomized trials of antiepileptic drugs in human patients identified a placebo response rate of 9.3–16.6%,8 while an earlier study comparing data from trials involving 5 antiepileptic drugs demonstrated placebo response rates of 4–18%, with an average rate of 10% over all studies.9
An obvious difference between the current study and those involving human subjects involves the inherent nature of the placebo response. In human studies, placebo effects are primarily because of expectations of an individual as a patient being treated in a double-blind manner. In veterinary studies, the placebo response can be because of effects on the animal, but more importantly may be a result of expectations of the pet owner regarding treatment. It is likely that the expectations of the owner might play a greater role in a study such as this, where the owners are responsible for administration of treatment and outcome measures (ie, seizure frequency) are derived solely from owner observations.
It is possible that the effect observed in the present study is because of nonspecific antiepileptic properties of the placebo, but this is considered unlikely. Although a decrease in seizure frequency was observed during placebo administration, the change in seizure frequency relative to baseline did not assume significance in all of the trials (P > .05). Rather, it is more likely that the observed effect is because of what has been termed the “perceived placebo effect,” or a change that is observed after placebo administration that is not directly caused to placebo.10 The perceived placebo effect includes any changes associated with the administration of placebo, including the effect of time. It is important to note that the statistical models utilized in this study do not differentiate the treatment effect from the effect of time, and consequently the analysis performed provides a measure of multiple potential factors that can contribute to treatment response. These factors may include regression to the mean, investigator bias, client bias, the potential for a higher level of care during the study, and improved adherence to treatment with active medication that is being administered in addition to the placebo during the study.
Regression to the mean is a statistical term used to describe the fluctuations of biological variables that occur over time and take the form of a sine wave around the mean.11 Epilepsy is a waxing and waning disorder, and fluctuations in seizure frequency are common over the course of the disease. Owners are most likely to seek a change in therapy for their pet when seizures are under poor control. Over the short term, improvement in the seizure frequency is probable, regardless of the treatment administered. However, this improvement is often erroneously attributed to a recently instituted change in therapy, whereas in fact it is because of an effect of time.
The potential for bias by individuals involved in the study cannot be overlooked. Even in blinded studies, expectations of a response can influence the subjective interpretation of study data by the investigator. The potential for participant bias, known as the Hawthorne effect, is a well-recognized phenomenon in clinical trials. The act of participating in a trial provides individuals with a better experience because of the focus of interest toward them, which is rewarding for its own sake. Consequently, participants document better results regardless of the change provided or the treatment experienced. Because canine epilepsy trials rely on owner observations to quantify seizure frequency, an owner's expectations of a trial can influence the study results for their dog. Indeed, it has been shown that the size of the placebo effect may be associated with the nature of the outcome measure, with larger effects observed in trials reporting subjective outcomes.12
Finally, it is possible that improved adherence to the established antiepileptic drug regimen during participation in the trial played a role in the present study findings. Although prescribed dosages of conventional antiepileptic medications stayed constant throughout the studies, a slight increase in bromide serum concentrations relative to baseline was identified during placebo administration in all of the trials, and a similar trend in phenobarbital serum levels was observed in the diet trial. The differences in serum drug concentrations were relatively small and did not approach statistical significance; nonetheless this may have contributed to the reduction in seizure frequency identified during placebo administration.
A greater placebo response was identified in the parallel study compared with the cross-over studies. Sixty percent of dogs in the parallel study were classified as placebo responders, compared with response rates of 0% and 36% in the cross-over studies. Similarly, the overall decrease in seizure frequency during placebo administration compared with baseline for the population of dogs was 46 % for the parallel study, compared with values of 26% and 29% for the cross-over studies. Although the total number of dogs in the trials is too low to draw any conclusions about these percentages, it seems plausible that owner bias might assume a greater role in a parallel study design than in a cross-over design. Owners that choose to enroll their dogs in trials tend to have a positive attitude toward the new intervention being evaluated and its potential to help their animal. This desire to have the treatment be effective can influence owners' observations and data collection for the study. The cross-over study design, in which each dog receives the therapeutic agent being tested, might minimize this bias. In contrast, the parallel study design, in which each participant receives either the test agent or placebo, has the potential to highlight this source of bias.
The beneficial effects of placebo administration are well documented in laboratory animals.1 Animal studies have served as the basis of investigation into the mechanism of the placebo effect for most of the past century. However, to the authors' knowledge there is only one other report that specifically addresses the placebo effect in clinical veterinary medicine.5 Jæger and colleagues performed a controlled parallel study to evaluate gold bead therapy as a treatment for canine hip dysplasia. The main outcome parameter in the study was a change in pain signs as assessed by the owner. Owners were also asked to document what kind of treatment they believed had been administered to their dog. Results demonstrated that when owners believed that their dog was receiving the active treatment, they reported a significantly greater improvement in pain signs compared with those who believed their dog was receiving placebo or were uncertain about the treatment administered. The authors concluded that blinding and randomization are necessary in veterinary studies, particularly where the outcome is based on owner's assessment of clinical improvement.
The treatment of canine epilepsy is evolving as several new methods of therapy have recently become available. Nine new antiepileptic drugs have been approved for use in humans since 1990, and the pharmacokinetics of many of these drugs support their use in veterinary patients. Recently, there have been publications reporting on gabapentin,13,14 zonisamide,15,16 and levetiracetam17 as treatment for refractory canine epilepsy. All of these studies evaluated the test drug in an uncontrolled, open label manner, with the drug being used in conjunction with phenobarbital, potassium bromide or both. Response rates of 41–80% were reported across these studies, and in several instances these response rates led to the conclusion that the drug being evaluated was efficacious for refractory canine epilepsy. However, based on the results obtained from the present study, it is likely that these results may be overstated and the true efficacy of these agents remains to be determined. In 2 of the trials reported on here, namely the surgical implant and the drug study, comparison of seizure frequency between the active treatment and baseline demonstrated a 27% and 48% reduction in seizures, both of which would be considered statistically significant on their own (P= .0025 and .017, respectively). However, when the reduction in seizures observed with the active treatment is compared with the change in seizure frequency between placebo administration and baseline, the response to the active treatment becomes insignificant.
The main limitation to the present study is the low number of dogs enrolled in each of the trials evaluated, such that it is not possible to conclusively state the significance of the placebo response observed. The lack of financial and administrative support that is required for large-scale trials is perhaps the most consistent factor that limits the ability to perform controlled clinical trials in veterinary medicine. Nonetheless, findings from this study justify the need for placebo-controlled trials to answer key clinical questions in veterinary practice.
aPatterson EE, Muñana KR, Kirk CA, et al. Results of a ketogenic food trial for dogs with idiopathic epilepsy. J Vet Intern Med 2005; 19:421 (abstract)
bPROC GENMOD, SAS Inc, Cary, NC
The authors thank Julie Nettifee-Osborne for her technical assistance with this study. We also acknowledge the referring veterinarians and clients whose dogs participated in the epilepsy trials evaluated.