Metronidazole - shouldn't we be concerned that the question of carcinogenic potential hasn't been settled?

[StringJunky]

Ranitidine, amoxicillin plus metronidazole  are what discoverer of the H. pylorii cause of peptic used in his pioneering work. It has clearly continued to have efficacy over forty years later.

Quote

SUMMARY
After preliminary studies in 1981, Marshall and Warren conducted a study in
which the new bacterium, Helicobacter pylori, was cultured. In that series, 100%
of 13 patients with duodenal ulcer were found to be infected. The hypothesis
that peptic ulcer was caused by a bacterial infection was not accepted without
a fi ght. Most experts believed that Helicobacter was a harmless commensal infecting people who had ulcers for some other reason. In response, Marshall
drank a culture of Helicobacter to prove that the bacteria could infect a healthy
person and cause gastritis. The truth behind peptic ulcers was revealed; i.e.
very young children acquired the Helicobacter organism, a chronic infection
which caused a lifelong susceptibility to peptic ulcers. Marshall developed
new treatments for the infection and diagnostic tests which allowed the
hypothesis to be evaluated and proven. After 1994 Helicobacter was generally
accepted as the cause of most gastroduodenal diseases including peptic ulcer
and gastric cancer. As a result of this knowledge, treatment is simply performed and stomach surgery has become a rarity.

https://www.nobelprize.org/uploads/2018/06/marshall-lecture.pdf

 

[Alfred001]

The claim about other ABs causing cancer is based on two studies alone. The nurse study and the clarithromycin study. The CLA study looked at all cause mortality, not cancer and median followup was 3 years, so you're not gonna detect any cancers with that.

The epidemiological studies they reference lack data on dose and duration, so I don't know what the basis is for the claim about the safety of a short course of metro.

EDIT: This is in response to Charon's post form the previous page.

Edited December 3, 2023 by Alfred001

[StringJunky]

1 minute ago, Alfred001 said:

The claim about other ABs causing cancer is based on two studies alone. The nurse study and the clarithromycin study. The CLA study looked at all cause mortality, not cancer and the epidemiological studies they reference lack data on dose and duration, so I don't know what the basis is for the claim about the safety of a short course of metro.

Can I ask why you are so invested in this subject? It seems to be beyond intellectual curiosity.

[CharonY]

1 hour ago, Alfred001 said:

The claim about other ABs causing cancer is based on two studies alone. The nurse study and the clarithromycin study. The CLA study looked at all cause mortality, not cancer and median followup was 3 years, so you're not gonna detect any cancers with that.

No, there is far more evidence of that, dating back to the 80s. The effect size is overall weak, but shows up fairly persistently in multiple human cohorts. 

A review summarizing some of those studies:

https://doi.org/10.3390/cancers11081174

A random selection of papers:

Breast cancer and antibiotics early study here:  doi:10.1001/jama.291.7.827 Other studies found mild effects, but there are mechanistic hypotheses underpinning this relationship: https://doi.org/10.3390/cells8121642

Relationship of AB use and colon cancer: http://dx.doi.org/10.1136/gutjnl-2016-313413; https://doi.org/10.1007/s10620-015-3828-0; 

Discussion of the role of microbiota, antibiotics and cancer https://doi.org/10.1016/j.ejca.2015.08.015

And the list goes on. 

Stating that there are only two papers are seriously misunderstanding the literature information. Also considering that these effects keep popping up in various studies, the link between AB and cancer in humans is far stronger than any short-term effect exclusive to metronidazole. 

1 hour ago, StringJunky said:

Can I ask why you are so invested in this subject? It seems to be beyond intellectual curiosity.

I echo this sentiment. It is unclear how this particular AB is assumed to be vastly different in risk compared to all the others.

 

[John Cuthber]

12 hours ago, Alfred001 said:

Ok, they say later on in the paper that they think there is good evidence that it doesn't have a significant carcinogenic effect in man.

Which isn't not to say there isn't one:

Which is what I have been saying all along.

You seem to have missed this.

12 hours ago, John Cuthber said:

If, as you suggest, the stuff is causing significant harm, how come things like the yellow card scheme (not to mention a stack of ambulance chasing lawyers) have not noticed it?

Can you let us know your explanation?

[Alfred001]

13 hours ago, CharonY said:

No, there is far more evidence of that, dating back to the 80s. The effect size is overall weak, but shows up fairly persistently in multiple human cohorts. 

A review summarizing some of those studies:

https://doi.org/10.3390/cancers11081174

A random selection of papers:

Breast cancer and antibiotics early study here:  doi:10.1001/jama.291.7.827 Other studies found mild effects, but there are mechanistic hypotheses underpinning this relationship: https://doi.org/10.3390/cells8121642

Relationship of AB use and colon cancer: http://dx.doi.org/10.1136/gutjnl-2016-313413; https://doi.org/10.1007/s10620-015-3828-0; 

Discussion of the role of microbiota, antibiotics and cancer https://doi.org/10.1016/j.ejca.2015.08.015

And the list goes on. 

Stating that there are only two papers are seriously misunderstanding the literature information. Also considering that these effects keep popping up in various studies, the link between AB and cancer in humans is far stronger than any short-term effect exclusive to metronidazole. 

I echo this sentiment. It is unclear how this particular AB is assumed to be vastly different in risk compared to all the others.

 

I will respond to this as soon as I have the time to read the studies, but in the meantime, does anyone have the answer to these questions? I really want to understand that study.

18 hours ago, Alfred001 said:

However, can someone explain what I'm not getting in this part:

1,336 and 564 cancers among users and non-users in at least 15 years of followup? Doesn't that mean there were 1900 cancers in 1219 people??? More cancers than people? I mean I know a person can get cancer multiple times, but THAT many multiple times? I doubt they counted recurrences as individual cancers and even if they did... Or am I completely misinterpreting things?

Also, incidence of cancer in 15 years of CANCER-FREE followup... what? What am I not getting here?

And then thirdly, 2.38x more cancer among metro users, how is that not significant? Ok, I see that the CI ranges from sub-1 to 6.12, but isn't that CI so wide as to be meaningless? And how likely is it that a 2.38x difference in 15 years is just down to chance???

Also, the limitations here are a bit worrisome, especially the absence of data on dose, duration and compliance.

 

 

 

 

 

[CharonY]

3 hours ago, Alfred001 said:

1,336 and 564 cancers among users and non-users in at least 15 years of followup? Doesn't that mean there were 1900 cancers in 1219 people???

No you misread the metric. The measure is per 100,000 person-years, not persons. 

The second thing you likely missed is reading through the study design, where they describe their follow up. Specifically the start date is when they get their first dose dispensed (for the user group). The follow-up ends either with the latest known consultation or the first diagnosed case of cancer. 

3 hours ago, Alfred001 said:

And then thirdly, 2.38x more cancer among metro users, how is that not significant? Ok, I see that the CI ranges from sub-1 to 6.12, but isn't that CI so wide as to be meaningless? And how likely is it that a 2.38x difference in 15 years is just down to chance???

Because the result was statistically insignificant (.11). The statistical power of that cohort (i.e. folks that were cancer-free for over 15 years and remained enrolled in the program) is just too low to be sure that it was not a statistical fluke. 

The caveats are pretty much standard, having more data is of course better, but often not feasible and often nearly impossible for multi-year studies. Keeping folks in these programs is very, very difficult.

 

 

[Alfred001]

6 minutes ago, CharonY said:

No you misread the metric. The measure is per 100,000 person-years, not persons. 

Yes, I understood that part. It's still the people getting the cancers. Doesn't change what's strange about it.

6 minutes ago, CharonY said:

The second thing you likely missed is reading through the study design, where they describe their follow up. Specifically the start date is when they get their first dose dispensed (for the user group). The follow-up ends either with the latest known consultation or the first diagnosed case of cancer. 

I understood that as well, I'm not sure what you're getting at with this part.

7 minutes ago, CharonY said:

Because the result was statistically insignificant (.11). The statistical power of that cohort (i.e. folks that were cancer-free for over 15 years and remained enrolled in the program) is just too low to be sure that it was not a statistical fluke. 

Yes, but that's what I'm asking, how is it possible to have as large a sample as they did and find as large an effect as they did and for it to be down to chance? 608 pairs in 15+ years and over twice as many cancers in the metro group, how could that be chance?

Even if we accept their confidence interval, the value is 0.82-6.12. Most of the range of the interval covers a greater incidence of cancer by a factor of up to 6. At best, based on this, we can say it's unclear whether there is an effect and that the effect could be large (granted, also could be no effect), but we can't say there is no effect.

Does anyone know how they would have calculated that CI and that p value? I mean in concrete terms, with these specific numbers.

17 hours ago, CharonY said:

No, there is far more evidence of that, dating back to the 80s. The effect size is overall weak, but shows up fairly persistently in multiple human cohorts. 

A review summarizing some of those studies:

https://doi.org/10.3390/cancers11081174

A random selection of papers:

Breast cancer and antibiotics early study here:  doi:10.1001/jama.291.7.827 Other studies found mild effects, but there are mechanistic hypotheses underpinning this relationship: https://doi.org/10.3390/cells8121642

Relationship of AB use and colon cancer: http://dx.doi.org/10.1136/gutjnl-2016-313413; https://doi.org/10.1007/s10620-015-3828-0; 

Discussion of the role of microbiota, antibiotics and cancer https://doi.org/10.1016/j.ejca.2015.08.015

And the list goes on. 

Stating that there are only two papers are seriously misunderstanding the literature information. Also considering that these effects keep popping up in various studies, the link between AB and cancer in humans is far stronger than any short-term effect exclusive to metronidazole. 

Point taken. When I said two papers, I was referring to what the Japanese study provided as references for their claim.

However, the first study is talking about all ABs and different classes of ABs in aggregate. I would be interested in the safety data on the individual antibiotics that are used in H pylori therapy and therefore represent alternatives to metro for that indication. Those are clarithromycin, tetracycline, bismuth (I guess in some sense not an AB, but has the effect and used for the purpose), amoxicillin, levofloxacin and a few less commonly used ones such as moxifloxacin and doxycycline.

The FDA issued a fairly stern warning for quinolones (levofloxacin, moxifloxacin...) in 2016. for some severe side effects, including potentially permanent ones, but I don't know how common those are.

I don't know much about the safety profile of the other drugs, outside of penicillin allergy in connection with amoxicillin.

[CharonY]

22 hours ago, Alfred001 said:

Yes, I understood that part. It's still the people getting the cancers. Doesn't change what's strange about it.

It is not strange, do you understand what person-years are and why that number is going to be larger than persons (specifically look how many person years are there in aggregate and what they normalized it against).

 

22 hours ago, Alfred001 said:

I understood that as well, I'm not sure what you're getting at with this part.

If you did, then there is no reason to be confused about it. So either you did not understand and got confused, or you understood and pretend to be confused. Which is it?

 

22 hours ago, Alfred001 said:

Yes, but that's what I'm asking, how is it possible to have as large a sample as they did and find as large an effect as they did and for it to be down to chance? 608 pairs in 15+ years and over twice as many cancers in the metro group, how could that be chance?

Again, these are person-years not numbers of cancer incidences. What they calculate are proportionate hazard ratios and likely age stratified (I do not recall). So the attributable risk ratio (as you acknowledge) had a huge spread, looking at the CI. So obviously the P is going to be high. This is the whole purpose of statistical tests, so that we just don't just look at the higher number and make inappropriate assumptions. Remember, these are matched pairs, and what it suggests is that there is going to be a big spread, which is really expected for rare conditions such as cancer (with a big impact on the age bracket of the matched pairs).

But if you are genuinely interested in statistical analyses I suggest to dig out a good textbook (and to be honest in order to fully recapitulate the methodology I would need to so, too to avoid errors- I rarely use risk ratio calculations in matched cohorts).

 

With regard to the other antibiotics, tetracycline, penicillin and nitrofuran have have some history in being associated with breast cancer specifically. There have been discussions of how microflora disruptions can influence the immune system and modulate estrogen levels.  But as mentioned before, there is also increasing awareness that especially long-term disruptions of gut microbiota, really with any antibiotic, are likely to have some impact (though the effects are likely to be very complicated, in some cases antibiotics are part of the cancer therapy).

Ultimately, you won't find a perfectly safe antibiotic, they all carry risks. And trying to quantify them is not going to be terribly useful, unless there is a huge effect size to be measure. The reason is also clear, our bodies (and microbiota) are dependent on an uncountable number of things that we accumulate over our lifetime. Some antibiotics might be fairly safe for some individuals but if the same individuals take a certain drug, have a certain lifestyle or happened to have a specific type of infections, the risk on the individual level might skyrocket. There is simply no reasonable way to capture all this diversity.

So all we can do is looking at rough aggregates and there, small differences rarely matter as the spread (or CI) is going to be very broad, anyway. With regard to AB, the most important aspect is whether they work in the first place. I.e. folks look at local resistance profiles and prescribe ABs that work. The secondary aspect is then to look whether the patient has any immediate adverse reactions to them. Long-term concerns are not unimportant, but are generally secondary unless a smoking gun study emerges. But that will take time. And if we wait for them before treating immediate issues, we will do more harm then good.

[Alfred001]

20 hours ago, CharonY said:

It is not strange, do you understand what person-years are and why that number is going to be larger than persons (specifically look how many person years are there in aggregate and what they normalized it against).

So it's actual incidence of cancer, divided by person years in the study and then multiplied by 100 000? I'm not sure what you mean by normalization.

20 hours ago, CharonY said:

If you did, then there is no reason to be confused about it. So either you did not understand and got confused, or you understood and pretend to be confused. Which is it?

Why not just explain what you mean instead of getting petty?

20 hours ago, CharonY said:

Again, these are person-years not numbers of cancer incidences. What they calculate are proportionate hazard ratios and likely age stratified (I do not recall). So the attributable risk ratio (as you acknowledge) had a huge spread, looking at the CI. So obviously the P is going to be high. This is the whole purpose of statistical tests, so that we just don't just look at the higher number and make inappropriate assumptions. Remember, these are matched pairs, and what it suggests is that there is going to be a big spread, which is really expected for rare conditions such as cancer (with a big impact on the age bracket of the matched pairs).

Ok, but the most we can say from this study, based on that CI, is that we don't know whether there is a greater risk in metro users, not that there isn't one, and in fact, given that the great majority of the interval lies above 1, it seems much more likely that there is one than that there isn't one, no?

Edited December 5, 2023 by Alfred001

[CharonY]

1 hour ago, Alfred001 said:

So it's actual incidence of cancer, divided by person years in the study and then multiplied by 100 000? I'm not sure what you mean by normalization.

Yes. Normalization generally means adjusting the value to a certain scale. In this case to 100,000 person-years.

 

1 hour ago, Alfred001 said:

 

1 hour ago, Alfred001 said:

Why not just explain what you mean instead of getting petty?

Because you constantly claim that you understood things perfectly, yet your questions clearly show that you don't (especially basic definitions). While I am happy to teach, it is very difficult if you do not realize that you have to revise your basic assumptions. And frankly I do have enough entitlement from my students, and a direct challenge often shortens things a fair bit.

 

1 hour ago, Alfred001 said:

Ok, but the most we can say from this study, based on that CI, is that we don't know whether there is a greater risk in metro users, not that there isn't one, and in fact, given that the great majority of the interval lies above 1, it seems much more likely that there is one than that there isn't one, no?

No, if a difference is not significant it means that the distributions are not distinguishable from each other. It does not matter if the means or CI is skewed in one direction or another, that is not something the test can tell us.

If there was a trend, the statistical power of the cohort is insufficient to show it (and/or the effect size is too small). Also, one thing to consider is that the cohort over 15 years is likely older, and increasingly other confounders influence cancer risk, as acknowledged in the study.

As noted, there are not really many studies that set up to prove a non-effect (safety is usually assessed in clinical trials) and there is basically no way for any treatment to do that conclusively, especially when looking at long-term effects. What studies can do, is try to see effects (as this one does) while controlling for a set number of factors. The complexity of the matter is also why we do not have figured out the perfect healthy food, for example. Likewise, there won't be risk-free medication. All we have is the weight of available evidence and never certainty. Also, it is often the case for weak effects that some studies show an effect and others don't. So evidence of one or the other side of argument will pile up over time until a tipping point for action is reached. So far the available studies show no outsized role of metronidazole in short-term harm (compared to other antibiotics), increasing evidence of general carcinogenic effects of long-term treatment with antibiotics, but also no true alternatives to antibiotic treatment.

[Alysdexic]

Concern exists. Metronidazole, a chemotherapeutic agent, shows potential as a carcinogen; rat and mouse studies find increased rates of lung tumors and lymphoma in each. Other studies show potential DNA damage in human cells. But there are no sheer or proven human carcinogenicity data, warranting excess worry.

The US FDA has a warning about metronidazole and WHO lists metronidazole in their Group 2B, "possibly carcinogenic to humans." Patients generally seem to find more benefit in treatment by metronidazole during severe bacterial and protozoan infections than concern about risks, and acute exposure to metronidazole is not linked to cancer in humans (yet or maybe ever). Risk-benefit analýsis is central to health decisions, in any and each case.

[Alfred001]

1 hour ago, CharonY said:

Because you constantly claim that you understood things perfectly, yet your questions clearly show that you don't (especially basic definitions). While I am happy to teach, it is very difficult if you do not realize that you have to revise your basic assumptions. And frankly I do have enough entitlement from my students, and a direct challenge often shortens things a fair bit.

Ok, can you explain what you meant by that part?

1 hour ago, CharonY said:

No, if a difference is not significant it means that the distributions are not distinguishable from each other. It does not matter if the means or CI is skewed in one direction or another, that is not something the test can tell us.

Doesn't the CI mean that the actual effect is somewhere in that range? If so, why is it then not more likely, since so much of the range is above 1 that there is an effect?

In fact, the CI for the 12< year follow up group is 0.92-2.20 - almost entirely 1<. Can't we say there that an effect existing is significantly more likely than not existing, although it not existing (or a protective effect existing) is a possibility as well?

2 hours ago, CharonY said:

If there was a trend, the statistical power of the cohort is insufficient to show it (and/or the effect size is too small).

Right, in other words, we cannot know from the study whether there's an effect.

2 hours ago, CharonY said:

As noted, there are not really many studies that set up to prove a non-effect (safety is usually assessed in clinical trials) and there is basically no way for any treatment to do that conclusively, especially when looking at long-term effects.

Why wouldn't it be if the effect is strong enough or the sample large enough?

I mean the CI for the 12 year group is fairly narrow and almost entirely above 1. If you added some more pairs and narrowed the CI further maybe you'd get entirely above 1 and be sure that there's an effect.

And to begin with, the CI could have been 2-3, if this were some other drug with a stronger effect.

[CharonY]

16 hours ago, Alfred001 said:

Doesn't the CI mean that the actual effect is somewhere in that range? If so, why is it then not more likely, since so much of the range is above 1 that there is an effect?

In fact, the CI for the 12< year follow up group is 0.92-2.20 - almost entirely 1<. Can't we say there that an effect existing is significantly more likely than not existing, although it not existing (or a protective effect existing) is a possibility as well?

You can read up the interpretation of CIs here https://en.wikipedia.org/wiki/Confidence_interval

Specifically:

Quote

• A 95% confidence level does not mean that for a given realized interval there is a 95% probability that the population parameter lies within the interval (i.e., a 95% probability that the interval covers the population parameter).^[18] According to the frequentist interpretation, once an interval is calculated, this interval either covers the parameter value or it does not; it is no longer a matter of probability. The 95% probability relates to the reliability of the estimation procedure, not to a specific calculated interval.^[19] Neyman himself (the original proponent of confidence intervals) made this point in his original paper:^[10]

  It will be noticed that in the above description, the probability statements refer to the problems of estimation with which the statistician will be concerned in the future. In fact, I have repeatedly stated that the frequency of correct results will tend to α. Consider now the case when a sample is already drawn, and the calculations have given [particular limits]. Can we say that in this particular case the probability of the true value [falling between these limits] is equal to α? The answer is obviously in the negative. The parameter is an unknown constant, and no probability statement concerning its value may be made...

• A 95% confidence level does not mean that 95% of the sample data lie within the confidence interval.
• A 95% confidence level does not mean that there is a 95% probability of the parameter estimate from a repeat of the experiment falling within the confidence interval computed from a given experiment.^[16]

 

16 hours ago, Alfred001 said:

Why wouldn't it be if the effect is strong enough or the sample large enough?

Because a) in terms of safety we only look for certain defined endpoints (e.g. death, cancer, etc.) so potential other effects can be easily missed,  and b) experiments are set up to test the null (i.e. no effect) so it is not really possible to calculate the likelihood of no effect.

For the extremes and for short term you can establish a measure of safety (i.e. no one dying within 6 months of taking a medication). But if you want to look all effects (liver, kidney, inflammation, immune modulation, cardiovascular health, and so on) or for effects in the long term, confounders will have an increasingly bigger role (such as diet, lifestyle, age, health status etc.). Controlling for all these factors is near impossible (there would be a near infinite list to track for each person). I brought up the issue of diet, which had over the years huge cohorts and long-time data, but the effects have not been reproducible.

[Alfred001]

On 12/6/2023 at 6:14 PM, CharonY said:

Because a) in terms of safety we only look for certain defined endpoints (e.g. death, cancer, etc.) so potential other effects can be easily missed,

But that's all we want to look at in this situation. Cancer.

On 12/6/2023 at 6:14 PM, CharonY said:

b) experiments are set up to test the null (i.e. no effect) so it is not really possible to calculate the likelihood of no effect.

Not sure what you mean by this, but the experiment is set up to see whether people who take metro have more cancers than the people who don't, I don't see why that would be impossible to determine, provided an adequately large sample to factor out chance.

On 12/6/2023 at 6:14 PM, CharonY said:

A 95% confidence level does not mean that for a given realized interval there is a 95% probability that the population parameter lies within the interval (i.e., a 95% probability that the interval covers the population parameter).

I could have sworn that this is how it was explained in a number of youtube videos I watched on the topic! Ok, I definitely had a misconception about what a CI is.

Nevertheless, that study still doesn't allow us to say whether there's an effect or not, right? The upper bounds of the CIs are 6+ and 2+, which means the effect could be of that magnitude.

[John Cuthber]

On 12/4/2023 at 9:12 AM, John Cuthber said:

Which is what I have been saying all along.

You seem to have missed this.

On 12/3/2023 at 8:19 PM, John Cuthber said:

If, as you suggest, the stuff is causing significant harm, how come things like the yellow card scheme (not to mention a stack of ambulance chasing lawyers) have not noticed it?

Can you let us know your explanation?

Still waiting.

[guidoLamoto]

There's a couple mistakes in logic in this discussion common to those who are not familiar with research data collection and analysis. 

For example, what do we mean by "causes?" Eg- Everyone "knows" that smoking causes cancer, right? Then how do you explain the fact that less than 15% of long time smokers ever get cancer?...It's because there are other factors (genetic and just plain luck) involved.

Then there's the problem of the logistics of doing sutdies-- Studies are usually only funded for relatively short periods of time- a few years, while diseases often require decades to show themselves and evolve. The researchers try to compensate for this by including large numbers of subjects and calculating "patient years."..But it's easy to see that following 10 pts for 20 yrs (200 pt-yrs) is not the same as following  100 pts for 2 yrs (also 200 pt-yrs). Not much may happen in 2 yrs, but a lot of things coould develop over 20 yrs.

Medical studies cannot be viewed in a mechanistic, cause-and-effect way. It's more a matter of probabiities. Interpreting these probablities is even more difficult because often the rates involved are low.  Let's say we're looking at lymphoma in farmers exposed to glyphosate. Several studies show that about 3 cases of lymphoma appear over a course of 10,000 pt-yrs. A few show 3 -6 cases, a couple 1 or 2 cases. They average out to 3/100,000...That variation from study to study, 1- 6 cases/100k is called statisrical wandering (error)- a well known phenomenon...Then they followed the non-farmers (no exposure) and found the same wandering and the same averages. The only scientific conclusion we can draw is that the chemical is not responsible for an increase in lymphoma rates....BUT- we can't really conclude that no person is safe from exposure. ...Statistics only apply to the group- NEVER the individual. Willy Mays, a lifetiine .300 hitter, can't get 3/10ths of a hit in his next At Bat.

Then there's the problem of risk to benefit ratio. The FDA wouldn't license a drug for treating the common cold if it had the dangerous side effect profile of some of the nastier chemo- agents for cancer.

The short answer to the original question is that the benefits of metronidazole far outweigh the risks of taking it.