Benzodiazepine use in relation to long-term dementia risk and imaging markers of neurodegeneration: a population-based study – BMC Medicine

Study population

Data was drawn from the Rotterdam Study, of which details have been described elsewhere [15]. In short, the Rotterdam Study is an ongoing prospective population-based cohort study, which started in 1990 with the aim to investigate the occurrence and determinants of common diseases in the elderly [15]. In 1990, the Rotterdam Study started with an original cohort of 7983 participants aged 55 years and older (RS-I). In 2000, this cohort was extended with 3011 participants who had reached age 55 or moved into the study area (RS-II). In 2006, an additional 3932 participants aged 45 years and over were included (RS-III), which resulted in a total study population of 14,926 participants. Participants undergo follow-up examinations every 4 years at a dedicated research center. For the incident dementia analyses in the current study, we included participants aged 60 years or older who took part in the fourth visit of RS-I (2002–2004), the second visit of RS-II (2004–2005), or the first visit of RS-III (2006–2008). Of 6258 eligible participants, we excluded those with cognitive impairment (Mini-Mental State Examination (MMSE) N = 806), those with missing pharmacy data (N = 8), and those who withdrew informed consent for dementia follow-up (N = 1), which resulted in the inclusion of 5443 participants. An overview of the inclusion of participants is presented in Additional file 1: Figure S1. Brain magnetic resonance imaging (MRI) was incorporated in the Rotterdam Study protocol from 2005 onwards. For the MRI analyses, we included cognitively healthy (MMSE ≥ 26) participants who underwent brain MRI between 2005 and 2015. Given the long preclinical phase of neurodegenerative disease, all persons > 45 years were allowed to participate. Of 4956 eligible participants who underwent brain MRI, 4836 had at least one scan that passed quality control.

Use of benzodiazepines and Z-drugs

Information on benzodiazepine use was available through pharmacy dispensing records from 1991 onwards for cohort RS-I, and from 1995 onwards for cohort RS-II and cohort RS-III, classified according to the Anatomical Therapeutic Chemical (ATC) code. We extracted all filled prescriptions of benzodiazepines from inception of pharmacy records to study baseline (2002–2008). For each prescription, we extracted the prescription date, duration of use, and strength in defined daily dosage (DDD), as defined by the World Health Organization [16]. Benzodiazepine use was defined as prescription of anxiolytics (ATC-code: N05BA) or sedative-hypnotics (ATC-code: N05CD) between inception of pharmacy records and study baseline. We further calculated cumulative defined daily dose from inception of pharmacy records to baseline and whether participants were still using at baseline or discontinued use prior to baseline. Similarly, information on the use of Z-drugs (ATC-code: N05CF) was extracted.

Dementia screening and surveillance

Participants were screened for dementia at each center visit, using the Mini-Mental State Examination (MMSE) and the Geriatric Mental Schedule (GMS). Those with MMSE  0 underwent further investigation, including an informant interview and the Cambridge Examination for Mental Disorders of the Elderly (CAMDEX). In addition, the entire cohort was continuously under surveillance for dementia through electronic linkage with medical records from general practitioners and the regional institute for outpatient mental health care. All cases suspect for dementia were reviewed by a consensus panel, led by a consultant neurologist, which applied standard criteria for dementia (Diagnostic and Statistical Manual of Mental Disorders (DSM)-III-R) to come to a final diagnosis. Participants were censored at date of dementia diagnosis, date of death, date of loss to follow-up, or January 1, 2020, whichever came first [17]. Follow-up for dementia until 1 January 2020 was complete for 93.9% of the potential person years.

MRI protocol and image processing

MRI of the brain was performed on a 1.5 T scanner (General Electric Healthcare, Milwaukee, WI) using an 8-channel head coil. Imaging acquisition included a high-resolution 3D T1-weighted, proton density-weighted, and a fluid-attenuated inversion recovery (FLAIR) sequence. A detailed scan protocol of the Rotterdam Study is described elsewhere [18]. Volumes in milliliters (mL) of the total brain, grey matter, and white matter were obtained by automated tissue segmentation based on a k-nearest neighbor algorithm. All segmentations were visually inspected and manually corrected when necessary. Volumes of subcortical structures involved in memory and mood regulation (i.e., the hippocampus, thalamus and amygdala) were obtained by processing T1-weighted images with FreeSurfer (version 6.0) [19].

Other measurements

Information on age, sex, educational attainment (primary, lower, intermediate or higher education), smoking habits (never, current, or former), and alcohol use (grams/day) was ascertained during a home interview. Prevalence of stroke, cancer, coronary heart disease, congestive heart failure, atrial fibrillation, and chronic obstructive pulmonary disease was assessed by interview at baseline and verified in medical records. The Composite International Diagnostic Interview (CIDI) [20] was used for the assessment of symptoms of anxiety; the Pittsburg Sleep Quality Index (PSQI) [21] was used to assess sleep quality. Presence of depressive symptoms was defined as a score of > 15 on the Center for Epidemiology Depression Scale (CES-D) [22] or the use of antidepressants. During baseline center visit, blood pressure was measured in sitting position using a random-zero sphygmomanometer; hypertension was defined as a systolic blood pressure > 140 mmHg, a diastolic blood pressure > 90 mmHg, or the use of blood pressure-lowering medication. The estimated glomerular filtration rate (eGFR) was calculated using the Chronic Kidney Disease Epidemiology Collaboration equation, based on creatinine concentrations in fasting blood samples [23]. Diabetes was defined as fasting blood glucose > 7.0 mmol/L or use of antidiabetic medication. Total fat mass was obtained using dual-energy X-ray absorptiometry (DXA) scans.

Statistical analyses

Missing covariate data were imputed using tenfold imputation. Distribution of variables was similar in the imputed and non-imputed datasets. Percentages of missing data are shown in the footnote of Table 1. Data on all variables were at least 90% complete, except for fat mass (73%). For the main analyses, ever use of benzodiazepines was compared to never use. In secondary analyses, we distinguished former from current use, stratified by the median cumulative defined daily dose, and differentiated anxiolytic from sedative-hypnotic benzodiazepines.

First, we determined the association between benzodiazepine use and risk of all-cause dementia using Cox proportional hazards regression models. All analyses were adjusted for age, sex, education, and time between inception of pharmacy records and baseline visit (model 1) and additionally for smoking habits, alcohol use, total fat mass, eGFR, presence of symptoms of anxiety, sleep problems or depression, and prevalence of diabetes, stroke, atrial fibrillation, congestive heart failure, coronary heart disease, cancer, and chronic obstructive pulmonary disease (model 2). In various sensitivity analyses, we then (i) stratified by presence of at least one anxiety disorder according to a score above the cut-off on the CIDI; (ii) stratified by the presence of sleep problems according to the PSQI; (iii) stratified on high vs. low alcohol use, where high alcohol use was defined as average consumption of more than 2 units (10 g) per day; and (iv) included benzodiazepine use as a time-varying variable up till dementia diagnosis. We compared the associations of oxazepam (t_1/2 = 5–15 h) and diazepam (t_1/2 = 20–70 h) with dementia risk to assess the effect of drug half-life. Finally, we assessed the effect of use of Z-drugs on dementia risk.

Next, we determined baseline differences in brain volumes between benzodiazepine users and non-users using linear regression models and applied linear mixed models to determine the association between benzodiazepine use and change in brain volumes over time of the total brain, white matter, grey matter, hippocampus, amygdala, and thalamus. Adjustments were similar to the dementia models, with the addition of total intracranial volume. To account for possible nonlinear trajectories, we included splines of follow-up time, with knots at the median follow-up duration of 3.3 years. An interaction of follow-up time with age was included, to allow for slope differences in the relationship with age.

Analyses were done using SPSS version 28 [24] and R version 4.1.3 (packages: “Mice,” “nlme”).

Patient and public involvement

Participants of the Rotterdam Study are represented through a panel that is consulted on a regular basis about study management and results. All participants are informed on results and publications of the Rotterdam Study through newsletters. In the current manuscript, participants were not involved in the development of research questions or study design.