Anticholinergic Drugs and Incident Dementia, Mild Cognitive Impairment and Cognitive Decline

A Meta-analysis

Nina T. Pieper; Carlota M. Grossi; Wei-Yee Chan; Yoon K. Loke; George M. Savva; Clara Haroulis; Nicholas Steel; Chris Fox; Ian D. Maidment; Antony J. Arthur; Phyo K. Myint; Toby O. Smith; Louise Robinson; Fiona E. Matthews; Carol Brayne; Kathryn Richardson


Age Ageing. 2020;49(6):939-947. 

In This Article



The study protocol was registered with PROSPERO (Registration:CRD42016039289). This systematic review and meta-analysis was reported according to the Meta-analyses of Observational Studies in Epidemiology guidelines.[18]

Search Strategy

An updated search from our 2014 review[16] was undertaken by one researcher (YKL) using Ovid SP MEDLINE and EMBASE between January 2013 and March 2016 (search terms given in Appendix 1), using recommended methods for updating searches.[19] Further studies identified using this search were automatically forwarded between March 2016 and April 2018. There were no language restrictions. The previous review search began in 2002 to capture studies using anticholinergic scales, with studies published before 2002 generally using serum anticholinergic activity (SAA).[20] We also re-evaluated all studies included in the previous review for inclusion in the current review[16] and re-screened abstracts from non-English and retrospective studies excluded from that review. References in published systematic reviews were hand searched, and we contacted experts within the field for further eligible studies.

Selection Criteria

Abstracts were independently assessed for inclusion by two researchers (NP and WYC or CH). Inclusion criteria were randomised controlled trials (RCTs) or observational studies investigating anticholinergic effects on human adults (using an anticholinergic scale[8,21,22] or specific anticholinergic drugs), on the following outcomes: (i) dementia, (ii) MCI or (iii) cognitive decline.

Exclusion criteria were <12 weeks follow-up between measurement of drug exposure and outcome; cross-sectional studies, case reports, literature reviews, clinical audits, editorials and conference abstracts; mean participant age less than 50 years; anticholinergic exposure based on SSA alone (due to inconsistent relationships with cognitive outcomes[23]); and studies including mostly participants with existing dementia.

Data Extraction

The following were independently extracted by two researchers (NP and WYC or CH): study design, data source, country, proportion of male participants, mean participant age, number of participants, definition of anticholinergic drug, primary exposure measure, length of follow-up time, effects on cognitive outcomes (as odds ratios [OR], hazard ratios or raw data that could be converted into an OR) and covariates included in multivariable analysis. Authors were contacted for additional data when studies provided insufficient data for calculating an OR. The extracted information was reviewed by two statisticians (KR and CG) and discrepancies resolved by consensus.

Risk of Bias Assessment

Risk of bias with respect to estimating causal effects was independently assessed for each study effect by two researchers (WYC and NP) using the Cochrane Risk Of Bias In Non-randomized Studies—of Interventions (ROBINS-I) tool.[24] Discrepancies were resolved through consensus.

Data Synthesis

Results were pooled using random-effects meta-analysis and the inverse variance method where studies used similar definitions of both drug exposure and outcome. Findings for dementia or MCI outcomes were pooled separately. The relative risk was assumed to approximate the OR as dementia and MCI were sufficiently rare events. ORs were pooled separately for studies reporting any (≥1 day), at least short-term (≥90 days) and long-term (≥365 days, or at baseline and 1- or 2-year follow-up for studies using patient interviews) anticholinergic use. Only the effects of drugs with definite/strong anticholinergic activity (scores 2 or 3 on the ACB scale or an equivalent definition) were included. Many drugs are considered to have mild anticholinergic activity (scored 1 on anticholinergic scales), i.e. serum anticholinergic activity or in vitro affinity to muscarinic receptors, but no known clinically relevant cognitive effects, and are excluded from this review. Some studies estimate the 'anticholinergic load' by summing the individual anticholinergic scale scores for each drug participants are taking; however, there were too few studies reporting these associations to include in the meta-analysis.

For each study that reported decline in global cognition measured as a continuous outcome, we derived the standardised mean difference (SMD). This was estimated as the mean difference in decline between exposed and non-exposed groups, divided by the standard deviation of the change scores. For studies that only reported an OR for decline following dichotomisation of change scores, we assumed that cognitive decline was normally distributed and converted this to the SMD by dividing the log-odds ratio by 1.81.[25] Estimated SMDs were then pooled using the random-effects inverse variance method.

We measured statistical heterogeneity using the I2 statistic. The following sources of heterogeneity were assessed using random-effects meta-regression: mean participant age, proportion female, mean baseline Mini-Mental State Examination (MMSE) score (where recorded), study follow-up time, population type (community versus care home) and patient disease group (general, specific condition). We report subgroup results for characteristics associated with the effect estimate at P < 0.10. We also performed three post hoc sensitivity analyses: first excluding studies with a critical risk of bias, second excluding studies only examining a single drug class and third including only studies using the ACB scale,[8] as the most common scale used aimed at central anticholinergic effects. Data was analysed using Stata version 14.0 (StataCorp, College Station, TX).