thank you for this. I don't take gabapentin but another pain med to manage my lower back pain so the topic is of get interest to me. It is really helpful for how you laid this out and used such clear examples. in the future when they come out w/ other "flavors" of this - which you know they will - it will be easier to see the flaws in reading those reports as well.
while reading this i couldn't help but wonder about the study subjects and how they were recruited and the composition of the sample. What is the power to detect a difference? and is it realistic to expect enough to develop signs of dementia in 10 yrs? those questions are perhaps mute given that a positive association was reported however it might provide additional insight into how much weight to give the results and the paper in general
Those are all great questions, particularly the issue of whether 10 years of follow-up is enough. And especially since the study population included people as young as 18!
Your take, Dr. Murray, is exactly what I found on reading it after Dr. Topol posted it on Twitter, to which I replied with a snarky comment "Meh. New math? In the intro: 'However, a recent analysis indicated that long-term use of gabapentin for chronic pain does not increase the risk of dementia, regardless of dosage, age, or gender.3'"
I hate these studies. High risk of publication bias, not to mention other kinds of bias, and they freak patients (and providers) out, probably needlessly. See PPIs and every kind of outcome you can think of. Looking forward to the next installment.
Looking forward to part 2. Correlation is not causation of course. But I also suspect that people with chronic back pain have a higher incidence of dementia because they are more likely to not engage in activities as much and become withdrawn and isolated. Also, I was prescribed gabapentin once for back pain but it was not the first thing I tried. It was perhaps #5. So people who take this drug for back pain may be far into their treatment. By the way, although my dose was increased to very high levels, it had zero effect on my pain.
To call gabapentin a "pain med" is a bit of a stretch. It's prescribed as a pain med, but whether it works on pain is another issue. Prescribing it for back pain is an off-label use, with little evidence to support it. I hope Part 2 looks back at gabapentin's sordid history, not just this study (which certainly has it flaws).
Thank you, Ellie. The issue you raise - What causal question is a study trying to address? - is a very common problem in many papers. And, as you imply, many papers that appear to draw causal conclusions were not actually studying causality at all. They use selected words like "association", which usually applies, and "risk" and "effect size" etc. which imply causality even if they didn't look at it.
And it gets worse. Many papers show "adjusted" results, meaning they included suspected confounders in the analysis. But they rarely show that the covariates they adjusted for are actually confounders. A valid confounder needs to be a *cause* of the outcome of interest while not itself being caused by the exposure nor cause the variable.
What many papers do instead is try a huge grab bag of miscellaneous variables and see whether including each one in the analysis reduces the size of the estimate. If yes, it's included. Never mind whether it's a valid confounder. That common, faulty process tends to produce under-estimates of the association they are looking for. Even worse, faulty treatment of confounders can lead to claims that there is NO association between the exposure and the outcome when there might actually be one. All because of treating variables as confounders when they are not.
Those who are not steeped in this topic would be amazed how often this happens.
thank you for this. I don't take gabapentin but another pain med to manage my lower back pain so the topic is of get interest to me. It is really helpful for how you laid this out and used such clear examples. in the future when they come out w/ other "flavors" of this - which you know they will - it will be easier to see the flaws in reading those reports as well.
while reading this i couldn't help but wonder about the study subjects and how they were recruited and the composition of the sample. What is the power to detect a difference? and is it realistic to expect enough to develop signs of dementia in 10 yrs? those questions are perhaps mute given that a positive association was reported however it might provide additional insight into how much weight to give the results and the paper in general
Those are all great questions, particularly the issue of whether 10 years of follow-up is enough. And especially since the study population included people as young as 18!
Your take, Dr. Murray, is exactly what I found on reading it after Dr. Topol posted it on Twitter, to which I replied with a snarky comment "Meh. New math? In the intro: 'However, a recent analysis indicated that long-term use of gabapentin for chronic pain does not increase the risk of dementia, regardless of dosage, age, or gender.3'"
You’ve laid out how to think about and assess studies like this really well, in terms a non-clinician can understand. Thank you!
I hate these studies. High risk of publication bias, not to mention other kinds of bias, and they freak patients (and providers) out, probably needlessly. See PPIs and every kind of outcome you can think of. Looking forward to the next installment.
Yes, and they always get so much more media attention than they really deserve!
Looking forward to part 2. Correlation is not causation of course. But I also suspect that people with chronic back pain have a higher incidence of dementia because they are more likely to not engage in activities as much and become withdrawn and isolated. Also, I was prescribed gabapentin once for back pain but it was not the first thing I tried. It was perhaps #5. So people who take this drug for back pain may be far into their treatment. By the way, although my dose was increased to very high levels, it had zero effect on my pain.
To call gabapentin a "pain med" is a bit of a stretch. It's prescribed as a pain med, but whether it works on pain is another issue. Prescribing it for back pain is an off-label use, with little evidence to support it. I hope Part 2 looks back at gabapentin's sordid history, not just this study (which certainly has it flaws).
Thank you, Ellie. The issue you raise - What causal question is a study trying to address? - is a very common problem in many papers. And, as you imply, many papers that appear to draw causal conclusions were not actually studying causality at all. They use selected words like "association", which usually applies, and "risk" and "effect size" etc. which imply causality even if they didn't look at it.
And it gets worse. Many papers show "adjusted" results, meaning they included suspected confounders in the analysis. But they rarely show that the covariates they adjusted for are actually confounders. A valid confounder needs to be a *cause* of the outcome of interest while not itself being caused by the exposure nor cause the variable.
What many papers do instead is try a huge grab bag of miscellaneous variables and see whether including each one in the analysis reduces the size of the estimate. If yes, it's included. Never mind whether it's a valid confounder. That common, faulty process tends to produce under-estimates of the association they are looking for. Even worse, faulty treatment of confounders can lead to claims that there is NO association between the exposure and the outcome when there might actually be one. All because of treating variables as confounders when they are not.
Those who are not steeped in this topic would be amazed how often this happens.
.