Dr. Beinlich has remarked that it is easy to remember the 50% responder rates for new seizure medications: they are always between 30-40%. I was thinking about this observation and I decided to look into a little more. I found as expected that Dr. Beinlich is correct. In a non-scientific manner, I pub-meded phase III epilepsy trials. I looked just for double-blinded placebo controlled trials on adults, and didn’t look at any open-label extensions or meta-analyses. I found 10 studies involving perampanel, eslicarbazepine, brivaracetam, lacosamide, and the one trial with trigeminal nerve stimulation.

Below is a graph of my findings. The 50% responder rates cluster with a mean of 36.1 STD DEV of 0.045, median of 35.5% and an interquartile range of 8%.

The fact that all these trials cluster around this range of response (it is similar if you want to look at median reduction in seizure instead) must tell us something about the nature of epilepsy. You could hypothesize that all these medications despite their different mechanism of actions, probably all affect one underlying process. It might be that this underlying process when modified renders most people completely seizure free, but for the 1/3 of patients that don’t become seizure free the impact is only modest.

One way to look at epilepsy is the concept of the seizure threshold model of epilepsy (illustrated well in Dr. Engel’s book *Seizures and Epilepsy*). In this model a local ictal process needs to overcome a “seizure threshold” for a seizure to occur. Possibly our medications just affect this seizure threshold and this modification is sufficient for most patients, but the refractory patient’s epilepsy could be dominated by the local process hence the modest effect of medications. It would make sense then that surgery or treatments focused on the local process may hold more promise.

I wrote a couple of simple programs (*Disclaimer I have no programing training, so I do not vouch for the syntax/quality of these programs, this is only for fun*) that model drug trials using different underlying assumptions about what the underlying epileptic process might look like.

Probably the simplest model would be to use a Poisson process. This model assumes that there is a finite probability of a seizure over a finite time length (1 min in this model). This probability is independent of the period before. The probability can be varied by an absolute or proportional amount as a theoretic way that a medication or other factor could affect the system.

Another more complex model based more on the seizure threshold model could look like a local Gaussian process with a global seizure threshold. A seizure would occur if the Gaussian crossed the threshold. The drug could affect either the seizure threshold or the local process depending on how you would like to design the trial.

Obviously these are just gross models that make vast assumptions and simplifications and are just meant to encourage thought about these subjects. For people with a serious interest in computer neural modeling there is a great website ModelDB (senselab.med.yale.edu/ModelDB) that has numerous models to chose from.

I copied the Matlab programs I wrote if anyone wants to play with them:

%Poisson Distrubition for probability of seizure for m months with drug

%effect size of de (decreasing the probability of seizure per a min by de proportion)

%m=number of months (30 days) for trial pre and post drug

m=1.5;

%de=drug effect size as proportion decrease in seizure probability

de=0.45;

%da=drug effect as an absolute reduction in seizure probability

da=0.0;

%n=number of subjects

n=150;

%p_pre=probability of seizure in 1 minute (10^-5 to 10^-4)

p_pre=0.00008;

%SD=standard deviation in population of propensity for seizures

SD=0.01;

for w=1:n;

gaus(w)=SD*randn(1);

p_pre_n=(1+gaus(w))*p_pre;

p_drug_n=p_pre_n*(1-de)-da;

H_pre=60*24*p_pre_n*30*m;

H_drug=60*24*30*m*p_drug_n;

sz_number_pre=random(‘poisson’, H_pre)/m;

sz_number_pre_tot(w)=sz_number_pre;

sz_number_drug=random(‘poisson’,H_drug)/m;

sz_number_drug_tot(w)=sz_number_drug;

if sz_number_pre==0;

per_change(w)=0;

else

per_change(w)=(sz_number_pre-sz_number_drug)/sz_number_pre;

end

if sz_number_pre+sz_number_drug==0;

sz_free(w)=1;

else sz_free(w)=0;

end

end

per_change;

num_50=gt(per_change,0.5);

Responder_Fifty=sum(num_50)/n

sz_0=gt(per_change,0.9999999999999999);

Sz_Free=sum(sz_0)/n+sum(sz_free)/n

neg_sz=lt(per_change,0.0000000000000000000001);

Seizure_Worse=sum(neg_sz)/n

Median_per_change=median(per_change)

Mean_Sz_Pre=mean(sz_number_pre_tot)

Mean_Sz_Drug=mean(sz_number_drug_tot)

x=(1:n);

%Shows a scatter plot of the seizures before and after drug treatment

%for each subject

plot(x,sz_number_pre_tot,’x’,x,sz_number_drug_tot,’d’)

For the next model first you have to define a function.

function [f] = dt(y,de,m,base,dy,local)

%=seizure frequency generating function, using a Gaussian with a threshold

d=(24*60*30*m);

do=(de*base);

b=randn(1,d);

k=(y+dy*local*base)*b+(base+do);

u=gt(k,0);

f=sum(u);

end

Then you can write the script.

%Drug trial model using Gaussian and Baseline Model to represent Focal

%Epilepsy

%Independent Variables

%de=effect of drug as proportion of seizure threshold

de=0.04;

%dy=effect of drug in reducing local as a proportion of local

dy=0.0;

%m=time of trial in number of months

m=2;

%n=number of subjects

n=500;

%local=propotion of baseline that will be the standard deviation from

%the Gaussin local process

local=0.245;

%base=baseline “seizure threshold” fixed at -100, all values relative

%to this variable

base=-100;

%pop_dev=the variability in the focal epileptogenic region in the

%population of people with medication refractory focal epilepsy

%represented here with a random number generator

pop_dev=0.03;

for r=1:n;

y(r)=local*(-base)+(pop_dev*(-base))*randn(1);

sz_d=dt(y(r),de,m,base,dy,local);

sz_d_tot(r)=sz_d;

sz_b=dt(y(r),0,m,base,0,local);

sz_b_tot(r)=sz_b;

if sz_b==0;

per_change(r)=0;

else

per_change(r)=(sz_b-sz_d)/sz_b;

end

if sz_b+sz_d==0;

sz_free(r)=1;

else sz_free(r)=0;

end

end

per_change;

num_50=gt(per_change,0.5);

Responder_Fifty=sum(num_50)/n

sz_0=gt(per_change,0.9999999999999999);

Sz_Free=sum(sz_0)/n+sum(sz_free)/n

neg_sz=lt(per_change,0.0000000000000000000001);

Seizure_Worse=sum(neg_sz)/n

Median_per_change=median(per_change)

Mean_Sz_Pre=mean(sz_b_tot)/m

Mean_Sz_Drug=mean(sz_d_tot)/m

x=(1:n);

%Shows a scatter plot of the seizures before and after drug treatment

%for each subject

plot(x,sz_b_tot,’x’,x,sz_d_tot,’d’)

Over the decades, epileptologist have pondered and discussed this phenomenon and a concensus conclusion is usually not reached. First we must consider the clinical trials use a very gross seizure selection criteria. Partial onset seizures do not all behave the same. To include all into one population to study may be one reason for the same responder rate being found across studies. The most obvious example of differential response is convulsion associated with JME do not respond as vigoreously to phenytoin or carbamazepine as valproic acid or lamictal. Only one randomized trial exists in this population which is the one I reported about 1980 comparing valproic acid to phenytoin. When only patients with definite primary generalized epilepsy were analyzed seizure free rate was much higher with valproic. Kifffen Penry was a major advocate of selecting the right therapy for a specific epilepsy type. Again no formal trial was conducted but he present many patient histories who were not controlled with multiple drugs only to have control achieved when some of the AEDs were eliminated (ofter carbamazepine or phenobarbital). From his examples many of us feel some AEDs can either aggrevate or prevent control for some seizure subtypes from being achieved. Carbamazepine is felt to aggrevate myoclonic seizures however again we do not have a controlled clinical trial to prove this. I personal experience I feel control of complex partial seizures originating in the frontal lobe may be prevented by the presence of some AEDs. The picture is complex and we should not lump all “seizures” together in clinical trials which is likely a significant reason for your findings.

Thanks

R Eugene Ramsay, MD

Director, Intl Center for Epilepsy

eramsay@epiworld.com

305-606-3800

Ochsner Health Systems.