In the March issue’s discussion of benzodiazepines, there was no discussion of possible risk for dementia. What does PL think?
The potential association between benzodiazepine use and dementia was highlighted by a recent large epidemiological study (Billioti de Gage et al, BMJ, Sept 2014). That report was prominently published, and has had a good deal of attention. The problem with this kind of research, as highlighted in the April PL issue, is in the Statistics column on confounding bias. These non-randomized epidemiological studies always suffer from some confounding bias. This is why readers should never take the results of such epidemiological reports at face value.
The best way to reduce confounding bias in such non-randomized studies is to use "regression modeling". This means that some potential confounding factors - which could influence the results - are measured and "adjusted for" in the statistical model. In this case, the question was whether benzodiazepines increase the risk of dementia. Well, many things increase the risk of dementia, like age, diabetes, depression, anxiety, substance abuse, hypertension, and other factors.
The problem with large epidemiological studies is that people are impressed with largeness, rather than quality. But the larger the study, the more common it is that confounding factors aren't measured or adequately adjusted. When you have huge samples, you can't interview each person directly to know how much depression or anxiety they had in their lifetime, or to identify a host of other medical or psychiatric risk factors. In other words, huge samples have the advantage of hugeness, but the disadvantage of not characterizing clinical features in much detail.
All that being said, this epidemiological study was adjusted for some important confounding factors. Besides the usual easily measurable factors of age and gender, a regression model adjusted for some medical illnesses (like diabetes and hypertension) and for depression, defined as the diagnosis of "major depressive disorder" in medical charts.
This is better than nothing, but whether or not the study adequately adjusts for the presence and severity of depressive illness fully relies on whether the treating clinicians in this large sample had accurately and adequately identified and documented depressive symptoms.
So there is some room for doubt as to whether the study adequately adjusted for depression, at least. Further, one might ask a question that has to do with "confounding by indication”. Benzodiazepines are used frequently for anxiety. Anxiety increases adrenal hormone activity, which increases the risk for dementia. How do we know that the association between benzodiazepine use and dementia wasn't a classic case of confounding bias, where the third factor of anxiety, associated with benzodiazepine use, directly causes dementia?
One can’t rule out this possibility from this analysis because anxiety diagnoses or symptoms weren't adjusted in the regression model, simply because the data weren't collected as part of the routine clinical practice which was the basis for the data used in the study.
The conclusion from this long discussion is that we simply can't accept the results at face value. Randomized studies would be much more definitive but they haven't been done. In the meantime, it's worthwhile noting that some animal studies show that benzodiazepines are neuroprotective, keep neurons alive, in human and animal studies of stroke (WS Huang et al, Psychiatry Clin Neurosci. 2014;68:255-62). Thus, there are some biological data to counter this clinical hypothesis that benzodiazepines might increase dementia risk. There are other clinical studies which also don't find increased risk of dementia with benzodiazepines. As with many medical topics, the question remains to be answered with reasonable confidence. But we can say these data as they stand don't prove the claim that benzodiazepines increase risk of dementia.