Neurology · Medical Literature
Use of Sedative-Hypnotic Drugs and the Risk of Developing Alzheimer's Disease: A Systematic Review, Meta-Analysis and Meta-Regression.
Abstract
Sedative-hypnotics, including benzodiazepines (BZDs) and non-benzodiazepine hypnotics (Z-drugs), are widely prescribed for insomnia and anxiety, particularly in older adults. Their long-term cognitive safety and potential association with Alzheimer's disease (AD) remain uncertain. We examined whether use of BZDs and Z-drugs is associated with incident AD and assessed variation by drug class, pharmacokinetics, and methodological factors. PubMed, Embase, and the Cochrane Central Register of Controlled Trials were searched from inception to 16 August 2025 without language restrictions. Reference lists of eligible articles and reviews were screened. We included observational cohort and nested case-control studies enrolling adults without dementia at baseline that compared BZD or Z-drug users with non-users and reported incident AD diagnosed using validated clinical or administrative criteria (e.g., ICD-9/10, NINCDS-ADRDA, or NIA-AA). We excluded reviews, case reports, conference abstracts, studies with overlapping populations, and studies without extractable effect estimates. Two reviewers independently screened studies, extracted data, and assessed risk of bias using ROBINS-E. Random-effects meta-analyses were performed separately for odds ratios (ORs) and hazard ratios (HRs). Heterogeneity was quantified with I2. Publication bias was evaluated with funnel plots and Egger test when applicable. Subgroup and meta-regression analyses assessed clinical and methodological modifiers. Certainty of evidence was rated using GRADE. The protocol was prospectively registered (PROSPERO CRD420251141623). Thirteen studies (N = 721,354 subjects) were included. Overall sedative-hypnotic use was associated with higher odds of AD (OR 1.29; 95% CI, 1.10-1.53; I2 = 86.5%). Estimates restricted to HRs were attenuated and not statistically significant (HR 1.17; 95% CI, 0.87-1.58; I2 = 73.1%). In subgroup analyses, BZDs overall (OR 1.21; 95% CI 1.07-1.36), Z-drugs (OR 1.14; 95% CI 1.10-1.18; I2 = 0%), and short-acting agents (OR 1.19; 95% CI 1.04-1.36) were associated with higher odds of AD, whereas broad-acting BZDs were not (OR 1.01; 95% CI 0.98-1.05). Long-acting agents showed a borderline estimate (OR 1.44; 95% CI 0.99-2.09). Age-stratified analyses showed higher odds in individuals aged <75 years (OR 1.36; 95% CI 1.24-1.49), but not in those aged ≥75 years (OR 1.14; 95% CI 0.61-2.11). Estimates were also higher in studies using ICD-based definitions (OR 1.47; 95% CI 1.16-1.86) than in those using clinical criteria (OR 1.13; 95% CI 0.84-1.52). Meta-regression identified drug class and publication year as significant moderators. Risk of bias was rated moderate to serious in several studies, mainly due to residual confounding and exposure misclassification. Certainty of evidence ranged from very low to moderate. Use of BZDs and Z-drugs was associated with increased odds of AD, with variation across drug classes and pharmacokinetic profiles. Short-acting agents, BZDs overall, and Z-drugs were associated with higher risk, whereas broad-acting BZDs were not; this finding should be interpreted with caution given subgroup heterogeneity and limited statistical power. Residual confounding and reverse causation limit causal inference. These results support careful prescribing and the need for prospective studies with detailed characterization of exposure, dose, duration, and clinical indication to clarify whether observed associations reflect drug-related effects or underlying disease processes. PROSPERO protocol number: CRD420251141623.
Clinical relevance
Is the use of benzodiazepines (BZDs) or non-benzodiazepine hypnotics (Z-drugs) associated with an increased risk of incident Alzheimer's disease (AD), and does this association vary by drug class, pharmacokinetic profile (e.g., half-life), or methodological characteristics of included studies? __SNAPSHOT__ {"studyDesign":"Systematic review, random-effects meta-analysis, and meta-regression of observational studies (cohort and nested case-control designs)","setting":"International; PubMed, Embase, and Cochrane Central Register of Controlled Trials searched from inception to 16 August 2025, without language restrictions","sampleSize":"13 studies included; N = 721,354 subjects","population":"Adults without dementia at baseline who were users of BZDs or Z-drugs compared to non-users; specific age ranges and comorbidity profiles varied across included primary studies","intervention":"Exposure to BZDs and/or Z-drugs (e.g., zolpidem, zaleplon, zopiclone) prescribed for insomnia or anxiety","comparator":"Non-users of sedative-hypnotic medications","primaryOutcome":"Incident Alzheimer's disease, diagnosed using validated clinical or administrative criteria (ICD-9/10, NINCDS-ADRDA, or NIA-AA guidelines)","followUp":"Not reported uniformly; varied across the 13 included primary studies","fundingSource":"Not reported in the available abstract"}
Key findings
Across 13 observational studies enrolling 721,354 subjects, overall sedative-hypnotic use was associated with statistically significantly higher odds of incident AD in the pooled OR analysis (OR 1.29; 95% CI 1.10–1.53; I² = 86.5%), indicating very high statistical heterogeneity. However, when the analysis was restricted to studies reporting hazard ratios — a metric better suited to time-to-event outcomes in cohort designs — the pooled estimate was attenuated and no longer statistically significant (HR 1.17; 95% CI 0.87–1.58; I² = 73.1%). In drug class–specific subgroup analyses, BZDs overall (OR 1.21; 95% CI 1.07–1.36), Z-drugs (OR 1.14; 95% CI 1.10–1.18; I² = 0%), and short-acting agents (OR 1.19; 95% CI 1.04–1.36) were each associated with statistically significantly higher odds of AD. Notably, the Z-drug subgroup showed no statistical heterogeneity (I² = 0%), suggesting a more consistent signal for this class. Broad-acting BZDs were not significantly associated with AD (OR 1.01; 95% CI 0.98–1.05), and long-acting agents showed only a borderline non-significant estimate (OR 1.44; 95% CI 0.99–2.09). Age-stratified analyses demonstrated higher odds in individuals aged <75 years (OR 1.36; 95% CI 1.24–1.49), with no significant association in those aged ≥75 years (OR 1.14; 95% CI 0.61–2.11), though the wide CI in the older age stratum reflects limited precision. Studies using ICD-based AD definitions yielded higher pooled estimates (OR 1.47; 95% CI 1.16–1.86) than studies relying on clinical criteria (OR 1.13; 95% CI 0.84–1.52), raising concern for diagnostic misclassification as a source of heterogeneity. Meta-regression identified drug class and publication year as statistically significant moderators of the pooled effect. NNT/NNH was not calculable from reported data. No specific adverse event data were reported in the available abstract.
Practice implications
This meta-analysis established a statistically significant association between sedative-hypnotic use (BZDs and Z-drugs) and incident AD when pooling OR-based estimates across 13 observational studies. However, it did NOT establish causation. The most clinically informative pooled estimate — the HR-based analysis, which better accounts for time-to-event structure and cohort designs — was non-significant (HR 1.17; 95% CI 0.87–1.58), substantially attenuating the apparent association. The certainty of evidence, rated using GRADE, ranged from very low to moderate, and risk of bias was rated moderate to serious in several studies, primarily due to residual confounding and exposure misclassification. Critical unmeasured confounders include indication bias (anxiety, insomnia, and depression are themselves AD risk factors and prodromal features) and reverse causation (prescriptions may reflect early, undiagnosed neurodegenerative symptoms rather than causing them). The discrepancy between ICD-based and clinical-criteria-based diagnostic outcomes strongly suggests outcome ascertainment heterogeneity. Findings may be most applicable to adults under age 75 with long-term sedative-hypnotic use, though subgroup analyses were exploratory. Elderly patients ≥75 years — who constitute the majority of clinical concern — were not adequately represented with a precise estimate. Patients with extensive comorbidities, concurrent polypharmacy, or established cognitive reserve differences were not separately analyzed. These results should not be used to make individualized prescribing decisions, and practice change is premature. However, the consistent direction of association across drug classes, combined with existing guideline caution regarding long-term BZD use in older adults (e.g., Beers Criteria), reinforces the importance of minimizing unnecessary chronic prescribing. Only available as an abstract and structured summary; full methodology and sensitivity analysis details were not directly accessible. __LIMITATIONS__ Substantial and unexplained statistical heterogeneity in the primary (OR-based) pooled estimate (I² = 86.5%), limiting confidence in the summary effect size,The HR-based pooled estimate, more appropriate for time-to-event analyses, was non-significant — a critical discrepancy that undermines causal inference,All included studies are observational (cohort or nested case-control), precluding causal inference regardless of pooling methodology,Residual confounding and confounding by indication: anxiety, depression, and insomnia — indications for sedative-hypnotic prescribing — are independent risk factors for AD and are difficult to fully adjust for,Reverse causation: prodromal AD may drive sedative-hypnotic prescriptions for early neuropsychiatric symptoms, inflating apparent exposure–outcome associations,Heterogeneity in AD outcome ascertainment: ICD-based administrative versus clinical/biomarker-based diagnostic criteria yielded divergent pooled estimates, suggesting outcome misclassification,Exposure misclassification: reliance on prescription records rather than actual medication use; dose, duration, and cumulative exposure not uniformly reported or pooled,GRADE certainty ranged only from very low to moderate, limiting translation to clinical practice,Risk of bias rated moderate to serious in several primary studies (ROBINS-E), predominantly due to residual confounding and exposure misclassification,Limited subgroup precision for adults ≥75 years (OR 1.14; 95% CI 0.61–2.11), the population most relevant to clinical prescribing decisions,Publication bias: assessed but likely present given the known tendency to publish positive associations in pharmacoepidemiology,Full funding sources across the 13 included primary studies not reported in the available abstract or summary, limiting assessment of potential industry influence,Only abstract and a structured journal summary were directly accessible; full sensitivity and meta-regression tables could not be independently verified __BOTTOM_LINE_CATEGORY__ potentially_practice_informing __BOTTOM_LINE__ This meta-analysis of 13 observational studies (N = 721,354) found a statistically significant association between sedative-hypnotic use and incident Alzheimer's disease on OR-based pooling (OR 1.29; 95% CI 1.10–1.53), but the more methodologically appropriate HR-based pooled estimate was non-significant (HR 1.17; 95% CI 0.87–1.58), and GRADE certainty was very low to moderate throughout. Residual confounding, indication bias, reverse causation, and high heterogeneity (I² = 86.5%) preclude causal attribution, and these findings alone do not justify changes to prescribing practice. Clinicians should nonetheless continue to follow existing guideline recommendations — including Beers Criteria — to limit long-term, unnecessary sedative-hypnotic use in older adults, while adequately treating underlying insomnia and anxiety through preferred non-pharmacological approaches.