Letters Benzodiazepines and dementia

Authors’ reply to Coyle-Gilchrist and colleagues and Bocti and colleagues

BMJ 2012; 345 doi: http://dx.doi.org/10.1136/bmj.e7993 (Published 26 November 2012) Cite this as: BMJ 2012;345:e7993
  1. Sophie Billioti de Gage, PhD student1,
  2. Bernard Bégaud, professor1,
  3. Antoine Pariente, associate professor1
  1. 1Université Bordeaux Segalen, INSERM U657, 146 rue Léo Saignat, F-33076, Bordeaux cedex, France
  1. sophie.billiotidegage{at}u-bordeaux2.fr

Coyle-Gilchrist and colleagues and Bocti and colleagues are directly or indirectly concerned with the possible non-causal nature of the association we found between benzodiazepines and dementia.1 2 3 We emphasise that we never concluded that this association was causal. We designed our study to go further than previously published studies by minimising possible confounding and reverse causation biases. For this reason, among others, we analysed only actual incident users and not prevalent users at the expense of a drastic reduction in sample size.

Both sets of authors point out that the symptoms that are possible indications for benzodiazepine prescription (anxiety, sleep disorders, and depression) may precede dementia by years, possibly 10 or more, whereas the median duration of follow-up in our study was 6.2 years. We obviously agree but make three points.

Firstly, despite being about seven years on average, the duration of follow-up was longer in a significant number of subjects, which made it possible to show that the excess risk in exposed subjects was mainly observed after seven or eight years, with a stronger trend after 10 years.

Secondly, we found a significant association in the case-control analysis for the most distant exposures (starting eight years or more before dementia) and not for recent ones.

Finally, studies conducted on symptoms observed during the prodromal phase of dementia clearly show that their frequency increases when approaching dementia onset.4 Therefore, if benzodiazepine prescription was mainly justified by such prodromes, the association would be expected to be stronger for treatments started a few years before dementia and not the reverse, as shown by our results.


Cite this as: BMJ 2012;345:e7993


  • Competing interests: None declared.