Article Text


Benzodiazepine use and risk of dementia: evidence from the Caerphilly Prospective Study (CaPS)
  1. John Gallacher1,
  2. Peter Elwood1,
  3. Janet Pickering1,
  4. Antony Bayer2,
  5. Mark Fish3,
  6. Yoav Ben-Shlomo4
  1. 1Department of Primary Care and Public Health, Cardiff University, Cardiff, UK
  2. 2Department of Geriatric Medicine, Cardiff University, Cardiff, UK
  3. 3Department of Neurology, Musgrove Park Hospital, Taunton, Somerset, UK
  4. 4School of Social and Community Medicine, University of Bristol, UK
  1. Correspondence to Dr John Gallacher, Department of Primary Care and Public Health, Cardiff University, Neuadd Meirionnydd, University Hospital of Wales, Heath Park, Cardiff CF14 4XY, UK; gallacher{at}


Background Benzodiazepine use is widespread in older people, although its benefit is uncertain.

Aim To investigate the long-term effect of benzodiazepine use upon dementia risk.

Methods A prospective cohort of men seen on five occasions over 22 years with full medication histories, repeat measures of cognitive function and a clinical diagnosis of dementia.

Results Of 1134 men with complete data, 103 (9.1%) had been taking benzodiazepines regularly at one or more phases. These men showed a marked increased incidence of dementia (OR=3.50, 95% CI 1.57 to 7.79, p=0.002), which persisted despite adjustment for psychological distress and other covariates. Men exposed in earlier phases showed a greater association than more recent exposure, counter to what one would expect if this was due to reverse causation, though we failed to demonstrate a dose–response effect with drug duration.

Conclusion The taking of benzodiazepines is associated with an increased risk of dementia.

  • Dementia
  • benzodiazepines
  • cognitive function
  • epidemiology
  • ageing
  • psychosocial factors
  • public health

Statistics from


With the ageing of the population, dementia has become a major public health problem. Many drugs, and especially benzodiazepines, are believed to cause cognitive impairment, yet the use of these drugs appears to be widespread. Puustinen et al1 found that ‘nearly every second patient’ in an acutely hospitalised population in Finland were taking benzodiazepines, and Paterniti et al2 estimated that >1 million French people aged 60 years and older are chronic users. In the UK, 11.7 million prescriptions were issued for benzodiazepines in 2007.3 Published evidence on the relationship between benzodiazepines and cognitive decline or dementia is inconsistent. While a number of studies report an increase in cognitive decline in benzodiazepine users,2 ,4 others find no such evidence.5 ,6 Evidence on dementia is more limited. Two small prospective studies found no evidence of harm over 3 and 5 years, respectively,7 ,8 while two case–control studies report an increased risk of dementia with benzodiazepine use.9 ,10 In a review of the evidence, Verdoux et al11 commented on the ‘discrepant findings’ on the risk of cognitive decline and concluded that ‘the hypothesis that long-term exposure to benzodiazepines may induce permanent brain damage is merely speculative’. A third strain of evidence comes from pharmacological studies that have identified acute adverse cognitive effects of benzodiazepine use in older persons,1 ,12 which may persist for over 6 months after withdrawal of medication.13 Further evidence is therefore required to establish whether benzodiazepine use has any long-term adverse cognitive implications. In the present study, we report the first long-term prospective evidence on the association between benzodiazepine use and risk of dementia.



The Caerphilly Prospective Study is based on a representative population sample of men born between 1920 and 1939, resident in a typical small town in South Wales. The men were first seen between 1979 and 1983 when aged 45–64 years. The initial response rate was 89%, and the survivors have been re-examined on four occasions: 1983–1988, 1989–1991, 1993–1995 and 2002–2004 (phases 2–5). Ethical approval has been obtained for each phase of the study with the most recent approval being obtained from the South East Wales Research Ethics Committee.


Complete lists of drugs taken ‘regularly’ by each man were recorded at each examination. Two measures of benzodiazepine use are used in what follows: first, the reporting of the use of the drug at any phase of the study. Second, an estimate of the likely duration of use of the drug was made by identifying men who reported its use at only one examination (referred to as ‘4 years or less’ in what follows) and those who reported its use at two or more examinations (referred to as ‘>4 years’).

In phase 3, when the men were aged 55–74 years, several tests of cognitive function, including the AH4, National Adult Reading Test (NART) and four choice reaction time task, the Mini-Mental State Examination (MMSE) and Cambridge Cognitive Examination (CAMCOG), were completed by each man.14 The AH4 is a test of fluid IQ, while NART is a test of crystallised IQ and is considered to estimate premorbid cognitive ability. The MMSE and CAMCOG are tests of global impairment. Each man also completed the General Health Questionnaire (GHQ-30),15 and the standard cut-off of a score of >4 was used as an indication of psychological distress, which involves both anxiety and depression. Data on trait anxiety16 were also available from the examination 5 years prior to this (phase 2), and we have used a log-transformed anxiety score due to its skewed distribution. Data were also collected at phase 3 for a number of aspects of sleep, including daytime sleepiness.17

In the fifth examination, the men being aged 65–84 years, tests of cognitive function were repeated, and those men who had a score of <83 on the CAMCOG or a decline in CAMCOG score of ≥10, together with subjects who failed to complete the CAMCOG, were selected for a clinical assessment (by MF).18

Full details of the clinical assessment of the selected men are reported elsewhere.18 In brief, this included a modified CAMDEX interview of subject and informant,19 the Rosen-revised Hachinski Ischaemic score,20 neurological examination with Frontal Assessment Battery,21 Clinical Dementia Rating22 and the Informant Questionnaire on Cognitive Decline in the older people.23 Available medical records were reviewed for details of relevant medical history, evidence of functional impairment due to cognitive impairment and results of neuroimaging and other relevant investigations.

Subjects diagnosed with vascular dementia fulfilled the NINCDS-AIREN criteria for possible or probable vascular dementia.24 Subjects were classified as non-vascular dementia if they fulfilled DSM-IV criteria for dementia25 and had no clinical features to suggest cerebrovascular disease operationalised as a Hachinski Ischaemic score ≤2 and an absence of cerebral infarction or significant white matter change on available neuroimaging. Most of these subjects fulfilled the NINCDS-ADRDA criteria for probable Alzheimer's disease,26 but due to small numbers, all non-vascular conditions were combined in the analyses which follows. Subjects who had screened positive but with insufficient impairment to warrant dementia diagnosis on clinical assessment were classified as cognitive impairment not dementia (CIND).


We first present the relationships between benzodiazepine taking at any phase of the study and the subsequent development of dementia. We present ORs (95% CIs and p values) from a logistic regression model adjusted for confounding at baseline by age, social class, education, smoking, cardiovascular disease and at phase 3 of the study for a number of cognitive tests as factors that may determine both risk of dementia and use of benzodiazepines. Three further models were tested with further adjustment for psychological distress, trait anxiety and daytime sleepiness and more proximal determinants of benzodiazepine use. We investigated a dose–response relationship by grouping men according to duration of drug use as defined above. To investigate reverse causation, that is, that any association between benzodiazepines and dementia reflects premorbid changes that increase the risk of being prescribed benzodiazepines secondary to the disease process, we repeated the analyses comparing more recent exposure to benzodiazepines, within 12 years of outcome (phases 4 or 5) with earlier exposure and between 13 and 22 years prior to outcome (phase 1, 2 or 3). If associations are due to reverse causation, we would hypothesise that associations would be stronger for more recent exposure than past exposure.


At the time of the most recent re-examination of the Caerphilly cohort, there were 1634 surviving eligible men of whom 1134 (70%) provided complete data and are the subject of this analysis. One hundred and three men (9.1%) reported taking benzodiazepines regularly at some time. Of these, 41 men reported benzodiazepine use at only one phase of the study (shown as ‘≤4 years’) and 62 men reported at more than one phase (shown as ‘>4 years’). Mean follow-up was 22 years (range 19–24 years).

Men who were not followed were more likely to be older, have manual occupations, hold no educational qualifications and be current smokers at baseline (table 1). They were less likely to be married and slightly more anxious, although less likely to use benzodiazepines. Not followed men had poorer cognitive function. The differences in age and marital status were small. No difference in psychological distress was detected.

Table 1

Sample characteristics according to follow-up status

A comparison between men who had and had not taken benzodiazepines found men who took benzodiazepines were more likely to be psychologically distressed and have higher levels of trait anxiety and daytime sleepiness and a slightly lower body mass index (table 2). Men who had ever taken benzodiazepines had slightly worse cognitive function performance except for the NART score. Although there was no association of NART with benzodiazepine use, there was an association of education with benzodiazepine use.

Table 2

Use of benzodiazepines and possible confounding factors at baseline

When the men were aged 65–84 years, 268 (24%) were selected by the criteria described above for a clinical examination and 93 were found to have dementia (table 3). In 44 of the men with dementia, this was judged to be due to vascular disease processes, and in 49, the dementia was judged to be due to non-vascular disease, mostly Alzheimer's disease. There were associations for both vascular dementia (OR=3.10, 95% CI 0.98 to 10.72) and non-vascular dementia (OR=3.34, 95% CI 1.10 to 10.18). In contrast, there was no evidence of an association with CIND (OR=0.63, 95% CI 0.27 to 1.48). Further adjustment for psychological distress, trait anxiety and daytime sleepiness had little effect on these associations, although NART score was not related to dementia independently of benzodiazepine use.

Table 3

OR for dementia and for cognitive impairment not dementia (CIND) in men who reported taken benzodiazepine compared with those who reported never taking the drug

No evidence for a dose–response relationship was found (table 4). Men who were exposed to benzodiazepines for ≤4 years showed a much higher risk of dementia (OR=4.38, 95% CI 1.15 to 16.75) than those who took them for >4 years (OR=2.31, 95% CI 0.74 to 7.20). This pattern was found for both vascular and non-vascular dementia.

Table 4

OR for cognitive impairment in 93 men by estimated duration of taking benzodiazepine

Evidence for reverse causation was sought by comparing earlier (13–22 years prior to outcome: study phases 1, 2 and 3) and more recent benzodiazepine use (within 12 years of outcome: study phases 4 and 5) (table 5). A recency effect was not found. For earlier use of benzodiazepines, evidence of an association was strongest for non-vascular dementia (OR=4.19, 95% CI 0.90 to 19.49). Raised ORs were also found for later use, but these were smaller than for earlier use and, possibly due to small numbers, statistical significance was not achieved. Counter to a reverse causation hypothesis, the association between benzodiazepine use and dementia was stronger with a longer latency period than for those men who had commenced therapy more recently.

Table 5

OR for dementia according to benzodiazepine use in phase of study


In a representative population sample of men with high follow-up rates over 22 years, the risk for dementia associated with the use of benzodiazepines is high and if causal would be alarming for what may be an iatrogenic cause. Although an element of reverse causality cannot be totally excluded, this is unlikely to be a complete explanation as there was only a modest difference in cognitive function when assessed at phase 3 of the study; the association was sustained once men with more recent exposure to benzodiazepines were excluded and the ORs were more marked for men only exposed in earlier than later phases of the study.

We observed little evidence that our association differed by whether the dementia was thought to be of vascular as compared with non-vascular origin. This may reflect difficulties in clinically differentiating these subtypes, and it is well recognised that mixed dementia with both Alzheimer's and vascular pathology is under diagnosed. Prior reports have noted a reduction in benzodiazepine receptors in various brain regions at autopsy of Alzheimer's disease patients.27

Previous literature

It is difficult to compare the results we present with those in the literature. First, most of the published studies relate benzodiazepine taking to cognitive decline and not to clinical dementia. In fact, if we ignore the distinction between dementia and CIND in our data, the adjusted OR with benzodiazepine taking which we obtain is 1.60 (95% CI 0.88 to 2.92). This finding is reasonably similar to the findings of Paterniti et al2 in a prospective 4-year study (OR 1.9, 95% CI 1.0 to 3.6). A number of other retrospective case–control studies reported finding no significant effect on cognitive function.1 ,4 ,6

Only a few studies appear to have reported on dementia, rather than cognitive decline. Lagnaoui et al9 conducted a case–control study with 150 patients with dementia and reported an OR of 2.3 (95% CI 1.2 to 4.5) for the use of benzodiazepine. Wu et al10 compared benzodiazepine use in a case–control study of 779 patients with dementia and reported an OR of 2.37 (p<0.001) in subjects who had taken benzodiazepine for >180 days within a 1-year period, together with a significant relationship between dementia and the cumulative dose of the drug taken. On the other hand, Fastbom et al7 followed a sample of 242 subjects with low cognitive function test results for 3 years and found a reduced incidence of dementia in those who used benzodiazepine (9%) compared with the incidence in non-users of the drug (23%).

The division we make by duration of drug taking is crude, and yet a difference in the incidence of dementia was found, and is consistent with the previous finding of Lagnaoui et al,9 who found a higher risk of dementia in former (OR=2.3, 95% CI 1.2 to 4.5) than current (OR=1.0, 95% CI 0.6 to 1.6) benzodiazepine users. One would normally expect to see a positive dose–response relationship so that greater duration of exposure is associated with greater risk, but a contrary pattern shown by two studies deserves closer attention.

Strengths and limitations

The main strength of our data comes from their prospective nature and the long follow-up period of 22 years. This period is more than double the duration of all previous studies. In addition, we were able to control for the major confounders (depression and anxiety), which are strong determinants of benzodiazepine prescribing behaviour as well as risk factors for dementia. Further adjustment was also made for socioeconomic indicators (social class, education, marital status). Drug taking was carefully and repeatedly recorded at each 5-year phase of the study, although the reasons for prescribing benzodiazepines was not recorded, and dementia was diagnosed clinically at the end of the follow-up period by two observers, who were blind with regard to previous drug taking habits removing any possibility of recall or observer bias. A major limitation of the study is, however, the small number of subjects who reported the taking of benzodiazepine and hence our relatively imprecise estimates of risk. The division into vascular and non-vascular dementia is known to have considerable uncertainty when validated against postmortem verification.28 Nevertheless, we have already shown within the same cohort of men that sleep disturbance, anxiety and haemostatic factors are differentially predictive of the two types of dementia.29–31


We think it is unlikely that reverse causation (also known as ‘protopathic bias’) explains our findings and believe that these data are cause for concern. Although evidence of an association with recent benzodiazepine use was found, indicating an effect of prodromal or early dementia on benzodiazepine prescribing, this did not account for the association of dementia with benzodiazepine exposure 13–22 years prior to outcome. Furthermore, apart from the modest differences in cognitive function at phase 3 of the study, NART score, which is an estimate of premorbid cognitive function, did not show a difference with benzodiazepine exposure, hence making ‘confounding by indication’ unlikely. Although measures of sleepiness and psychological state were not available at baseline, when adjustment was made for daytime sleepiness and psychological state assessed at phase 3 of the study, no attenuation of effect was found. The absence of an association with CIND requires consideration. It either suggests that the association with dementia is spurious or that any harm due to benzodiazepine use was advanced at the point of CIND assessment. This latter explanation would be consistent with finding of an association of dementia with earlier benzodiazepine use.

It is possible that this is a chance phenomenon or due to residual confounding, although where numbers allowed, the associations found were reasonably strong and adjustment was made for socioeconomic indicators. An alternative explanation is that men who are more sensitive to drug-related side effects (and who stop taking benzodiazepines after a short time) are either more susceptible to dementia or their risk of dementia is increased due to benzodiazepines. Men who do not experience detectable side effects of the drug, however, are either less susceptible to dementia or suffer less cerebral harm from taking the drug. In this case, benzodiazepines may either be a pharmacological biomarker or a causal agent for dementia or both.


In conclusion, we have provided long-term prospective evidence of a possible adverse effect of benzodiazepines on the development of dementia. This is consistent with previous findings on dementia but is based on a far greater follow-up period reducing the likelihood of reverse causation. This association was observed for both vascular and non-vascular dementia. Given the widespread use of these drugs and the ageing population, it is important that other studies with better data on the determinants of benzodiazepine prescribing, side effects and reasons for drug cessation examine this association. Furthermore, it is important to examine if these associations are specific for benzodiazepines or are seen with other types of hypnotic drugs or anxiolytic therapies. In view of the evidence now available, it is doubtful if randomised trials of benzodiazepine would be ethically acceptable.

What is already known on this subject

Benzodiazepines are prescribed widely for older people and can affect cognitive function. The effect of benzodiazepines on dementia is unclear due, in part, to a dearth of long-term follow-up data.

What this study adds

From this study, we know that benzodiazepine use is associated with dementia in the long term (22 years) and that this is unlikely to be due to protopathic bias or confounding by indication. The absence of a dose (duration)–response relationship suggests that any effect is limited to a susceptible subgroup rather than widespread. Whether benzodiazepines are an iatrogenic cause of dementia or a biomarker for dementia risk is unclear. These findings indicate that great care should be taken upon the beginning of benzodiazepines with middle-aged and older people.


We would like to thank Truda Bell and her field team for their excellent work in collecting these data along with members of the cohort for their generous co-operation.


View Abstract


  • Baseline data were obtained from the Caerphilly Study Archive, Department of Social Medicine, Bristol University, UK.

  • Funding This work was funded by the Alzheimer's Society. The Caerphilly Study was initiated by the former MRC Epidemiology Unit (South Wales) and was funded by the Medical Research Council of the United Kingdom.

  • Competing interests None.

  • Patient consent Obtained.

  • Ethics approval This study was approved by South east Wales Research Ethics Committee.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.