Very rare side effects - how are they measured?

[Carn]

Often, when reading the package descipton of some medicine, there are listed among very rare(1/100000) side effects dizziness, headache, general exhaustion or other sort of non specific pain or illness.

I suspect for no medicine were there done enough DBPC studies so a meta analysis would prove the presence of a 1/100000 side effect, so how many of these side effects are realy side effects and how many are just listed, because they conincided a few times someone took the medicine?

Carn

[jasime]

Hello there Carn, this is a good query and in simple terms I would like to know that why do some people encounter side effects that arent even listed in the properties of the medicine? why does this happens is there no such clinical trials that can convincingly make us acquainted with all the side effects that one can experience?

[Carn]

why does this happens is there no such clinical trials that can convincingly make us acquainted with all the side effects that one can experience?

Because human body cannot be throughfully calculated yet and nobody is able to do a DBPC study with 1 billion peope.

But you're question points in a different direction than mine, i wanted to know, how actually someone could get to the sceintific statement, that some medicine has a 1/100000 side effect.

Carn

[Zep]

It's actually a probability statement of the likelihood of a reaction. It does NOT mean that exactly the 100,000th person will get a side effect (and the 200,000th, 300,000th, etc). A common misunderstanding, btw.

Most "common" medicines are trialled over many thousands of people in many countries. And some of them do get adverse reactions. From such huge numbers and so few reactions, it is possible to obtain this statistical estimate of potential side effects.

[Mojo]

They should also get feedback about possible reactions from doctors once the medicine is in use, which might pick up very rare possible side effects.

[Carn]

Most "common" medicines are trialled over many thousands of people in many countries. And some of them do get adverse reactions. From such huge numbers and so few reactions, it is possible to obtain this statistical estimate of potential side effects.

But how is it derived, that the adverse reaction was due to the medicine and how is it calculated, that it is a maximal 1/100000 chance and not a 1/30000?

Are there enough DBPC studies done to get above a few hundred thousand trial subjects?

Because i cannot see how it should be possible to have DBPC studies amounting to 50000 subjects and then conclude, that some effect noted is less than 1/100000 chance.

Carn

[Carn]

They should also get feedback about possible reactions from doctors once the medicine is in use, which might pick up very rare possible side effects.


And how does one know, the patients didn't get a dizzy by chance and the doctor mistakenly thought it's from the medicine?

Carn

[Zep]

It's not a "maximal chance" of 1/100000 or whatever. I think that's the mistake in your understanding you are making here. You are not guaranteed to find one reactive person if you randomly pick an exact 100,000 sample out of a population. You might, but you might not too.

As I understand it, it works like this: If you started picking people at random out of a huge population, it's odds of 1 in 100,000 at each pick of that person being adversely reactive to this medicine.

Think of a trivial parallel example: The odds of selecting a female in any one (random) pick out of a huge population is 1 in 2 (near enough). But if you pick exactly 2 people out of this population randomly, are you guaranteed exactly 1 female only in the result? No - you might, but you might not too. The odds remain 1 in 2 for each pick.

[geni]

The figures from the most part don't come from DBPC trails. They come from stuff such as the yellow card scheme:


http://www.mhra.gov.uk/home/idcplg?IdcService=SS_GET_PAGE&nodeId=287

[Mojo]

And how does one know, the patients didn't get a dizzy by chance and the doctor mistakenly thought it's from the medicine? The doctor is supposed to report anything they suspect of being an adverse reaction. See the link provided by Geni, under "Healthcare professional reporting" > "What to report": Complete a Yellow Card if you have a suspicion that a drug has caused an adverse reaction. But remember - if in any doubt – report today!Once the MRHA have the data, they can then look at whether a particular drug has had more of a particular reaction reported than might be expected if the reaction was just coincident.

[Deetee]

When trials are published, the data usually includes the "side effect" rate in the placebo group, so researchers can get an idea of how much of the reported toxic effect is due to the drug. So if 12% get headache on drug X, but only 9% get headache on placebo, they need to determine if this difference is statistically significant.

If there is a reported unusual side effect - eg a death from severe liver failure in the treatment group as an example, the researchers will have to analyse the clinical data in great detail and determine if this could be due to the drug. Clues might exist if simple liver function test abnormalities are found more frequently in the Drug X group for example, indicating that the chance of an isolated severe reaction is not an implausible outcome. The death might equally be due to some other factor - a different drug, or a deterioration or complication of the underlying medical disorder etc. Each case will be analysed and an estimate of the probability that the outcome was related to the trial drug will be made.

For very rare side effects that are unlikely to be found in the context of a trial, the drug industry and watch-dogs rely on "post-marketing surveillance", as Mojo suggestsed. New drugs are flagged up for special attention after they are approved for general release, and doctors asked to report any suspicions to the relevant authority (Committee for Safety of Medicines/MHRA) - they have a web site here (http://www.mhra.gov.uk/home/idcplg?IdcService=SS_GET_PAGE&nodeId=287). These newly licensed drugs are known as "black triangle" drugs, because formularies have to carry a black triangle marking these drugs so docs know they are still under surveillance.

Alerts concerning suspected rare reactions can be issued, and if it is felt that a new reaction is described which is of sufficient severity (eg death) to warrant withdrawl of the drug, then this will be done even though it may be a very rare side effect. There are several examples where this has happened recently - sometimes a quite unexpected problem has reared its head, on other occasions a problem that was thought to be fairly mild turns out to be severe in some cases.

Common sense is needed - if a new drug for easing symptoms of PMS turns out to cause severe liver problems in 1:10000 people, it will probably get pulled. If a new AIDS drug causes liver problems in 1:100 people, it still might get the OK.

ETA - here is a link (http://www.impostertrial.com/physician.htm#Statin%20associated%20Rhabdomyolysis ) concerning a statin (Cholesterol lowering drug). Cerivastatin was withdrawn after it appeared to cause more muscle damage than expected. All the statins can do this, but in the case of cerivastatin (BAYCOL in the USA) it was a bit too much to allow its continued use, particularly as there were other statins around which could do the same job, but more safely. The tables in the link give ideas of the frequencies involved.
These data excludes a newish statin, rosuvastatin (Crestor) over which there have been concerns as well, but it is still in use.

[Carn]

The figures from the most part don't come from DBPC trails. They come from stuff such as the yellow card scheme:


http://www.mhra.gov.uk/home/idcplg?IdcService=SS_GET_PAGE&nodeId=287

But doesn't this method without further input give a lot aof false reports, due to chance or placebo effect?
Think of all homeopathy fans, who fear conventional medicine, if they have to still take it, they could end up to think every slight misfeeling is a side effect of the medicine. They then go to their homeopathy prescribing doctor, who sends off a report. That happens a few times and suddenly the medicine has a very rare side effect.

Is this somehow prevented or is it actually happening, which then means any package descripton saying "has very rare side effect", often means, "someone thought it had the side effect, but probably imagined it"?

Carn

[Deetee]

The figures from the most part don't come from DBPC trails. They come from stuff such as the yellow card scheme:

http://www.mhra.gov.uk/home/idcplg?IdcService=SS_GET_PAGE&nodeId=287

I hate you geni - you say what I did in one sentence, and because it takes you 1/100th of the time, you get your bit in first..
:p

[geni]

But doesn't this method without further input give a lot aof false reports, due to chance or placebo effect?
Think of all homeopathy fans, who fear conventional medicine, if they have to still take it, they could end up to think every slight misfeeling is a side effect of the medicine. They then go to their homeopathy prescribing doctor, who sends off a report. That happens a few times and suddenly the medicine has a very rare side effect.

Is this somehow prevented or is it actually happening, which then means any package descripton saying "has very rare side effect", often means, "someone thought it had the side effect, but probably imagined it"?

Carn

Well there are various statistical tricks but "very rare side effects" probaly include a few things which were due to chance.

[Carn]

It's not a "maximal chance" of 1/100000 or whatever. I think that's the mistake in your understanding you are making here. You are not guaranteed to find one reactive person if you randomly pick an exact 100,000 sample out of a population. You might, but you might not too.

As I understand it, it works like this: If you started picking people at random out of a huge population, it's odds of 1 in 100,000 at each pick of that person being adversely reactive to this medicine.

Think of a trivial parallel example: The odds of selecting a female in any one (random) pick out of a huge population is 1 in 2 (near enough). But if you pick exactly 2 people out of this population randomly, are you guaranteed exactly 1 female only in the result? No - you might, but you might not too. The odds remain 1 in 2 for each pick.

I very well understand what chance is, and i used "maximal" because nobody could say whether a side occurs with 1/100000, 1/105000 or 1/98000.
So i suspect, that people try to derive from the data just whether its above or below 1/100000 and if above, they place it to rare side effects aka 1/10000.

But for example one cannot prove that a dice has a chance of less or equal to 30% to show 6 by throwing it 3 times. I do not know exactly, but i'd guess you would need something like 20-40 times to be able to reliably tell, that 6 has a less than 30% chance of occuring.

Same with 1/100000 side effects, which do not leave hard evidence behind(e.g. dead, muscle loss), to realy go about proving it occurs due to the medicine and it occurs closer 1/100000 times and not closer 1/10000, one has to have DBPC studies with tens of thousands of subjects, maybe even more than a hundred thousand.

Carn

I think you misunderstood me, i'm trying to understand, whether very rare side effects in package description have the same solid prove as a 1/10 side effect.

[Carn]

Well there are various statistical tricks but "very rare side effects" probaly include a few things which were due to chance.


Starting with which rarity one can safely assume, that the side effects have reliably been checked for?

Carn

[Asolepius]

At market launch, we only have clinical trial data for assessing side effect frequencies. There are well established criteria for grading side effects in trials. In fact they are never called side effects, they are `adverse events' (AEs). This is because the primary criterion is whether it occurred, not what caused it. While it's accepted that there is a fair bit of under-reporting in clinical trials, the situation is getting better. All sorts of clues are available to detect an unreported AE. For example, did the patient take a non-study drug (say for a headache)? Did they turn up at the emergency clinic? Is there something wrong with a lab test result? Once the AE has been nailed down as having happened (with dates and times), it then gets graded for severity and causality. This is getting closer to your OP Carn. One test is whether the AE is related in time to study drug dosing. Not a cast-iron test at all of course, if the AE has a long latency (takes time to appear). Another test is dechallenge - does the AE resolve when the drug is stopped? Most AEs are mild and don't cause patients to drop out of the study, so for those causality is difficult to assess. A more rigorous test is rechallenge - does the AE come back when the study drug is restarted? In 25 years I don't think I have seen anyone do this in a clinical trial! But you often see it in individual case reports published in the journals. Most of the time, causality is a value judgement by the doctor doing the trial. In fact when writing the report we never use the term causality, we only talk about whether the AE was related to treatment.

Now all this has a bearing on such cases as Vioxx. Plaintiffs are trying to prove that Vioxx caused their AEs. I think you can see that this is pretty much impossible to prove for any individual. What's easier to prove is whether Merck knew about the risk and failed to warn patients (via their doctors). That's also going through the courts of course.

The upshot of all this is that pharmacovigilance is an imprecise science. Only when we get really large numbers of patients treated with the drug, do the errors start to dilute out. With big numbers, associations between certain AEs and certain drugs get more concrete and can be considered more as causal. But there are no absolutes in science, only diminishing uncertainty as knowledge progresses.

Sorry, I have rambled on a bit :) Does this help?

[Capsid]

Further to Asolepius' causality remarks I have an amusing anecdote I'd like to share in which one of the clinical trial subjects enrolled into a vaccine study suffered a gunshot wound. This was reported as an adverse event but was not considered to be related to the vaccine treatment :)

Further to Deetee's comments on unexpected events, the first rotavirus vaccine (Rotashield) was withdrawn because it caused intussuception in a very small number of children. The clinical trials were not large enough to reveal this rare event. The new rotavirus vaccine recently launched was evaluated with very large numbers of subjects to address this issue specifically.

[Mongrel]

Would they also use data from chemically similar drugs to predict AEs?

[Rolfe]

Capsid can spell "intussusception" - respect! (oops, no he can't, I can't read!)

I saw a packet of an HRT product that listed a whole load of things "patients taking this medicine had reported", then let the cat out of the bag by remarking that most of them were probably just menopausal symptoms anyway.

Rolfe.

[Capsid]

Capsid can spell "intussusception" - respect! (oops, no he can't, I can't read!)
Pah! I knew I should have checked.

[Asolepius]

Would they also use data from chemically similar drugs to predict AEs?
Yes. An essential document for clinical trials is the investigator brochure. This summarises all data so far on the drug, and will commonly discuss safety data in relation to the class of drugs (if the drug is a `me too'). AEs which are completely unexpected in relation to the mode of action and what is known of the drug class require expedited (ie rapid) reporting. Actually this category of AEs is called `serious and/or unexpected' (SAEs). Even if not serious (and that doesn't mean severe necessarily), if unexpected they get special processing. Non-SAEs get reported at the end of the trial, SAEs must be reported within 24 hours initially, with full data within a few days depending on the regulatory authority.

Sorry, I am getting sidetracked again. I'm just trying to get across how hard the drug industry and the regulators try to look out for the unexpected, and that means they have to define the expected first. Thalidomide caused lots of problems because the side effects were unexpected and nobody was looking for them. Much the same happened with praxolol in the 60s. It caused retroperitoneal fibrosis and nobody expected a heart drug to do that.