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


Study Selection

We screened 521 study abstracts: 347 identified through the new literature search, 46 from the previous review, 42 retrospective studies, 74 non-English studies excluded from the previous review and 12 via other sources (Appendix Figure 1). Full text was extracted for 79 studies, with 26 studies meeting our inclusion criteria; 23 cohort studies;[10,11,14,15,26–44] and three case-control studies.[9,45,46] Nineteen studies are included in the meta-analyses.


The 26 identified studies included 621,548 participants with mean study duration of 73 months (range 3–241 months, Table 1, Appendix Tables 1 and 2). Studies were conducted in Europe, North America or Taiwan and were mainly community based, except for three studies of outpatients and two studies of care home residents. The mean proportion of female participants was 60%, and mean participant age was 74 years (range 52–86 years).

Drug use was assessed by patient interview, self-reported written survey and review of prescriptions and drug containers or directly from insurance claims, pharmacy or primary care records. Studies varied according to how they classified anticholinergic drug exposure: 14 studies used a published anticholinergic rating scale (ACB or ADS), four used a literature review and/or expert panel analysis, one used their own validated clinician-rated scale and two studies used national drug/therapeutic formulary classifications. The scales generally aimed to examine central anticholinergic effects, but some were more likely to also capture peripheral effects (Appendix Table 1). One study specifically tested selective serotonin reuptake inhibitors (SSRIs) only and three examined bladder anticholinergics only.

Risk of Bias

Findings from six studies were rated as having a critical risk of bias, 19 as serious, one as moderate and none as low (Appendix Tables 3 and 4). Of the risk of bias subsections: 14 (54%) had a serious or critical risk of bias for confounding; 23 (88%) had a serious risk of bias for participant selection; and 10 (40%) had a serious risk of bias for missing data. Although many studies controlled for age, sex and education, few accounted for anticholinergic drug indications such as depression (Appendix Table 2). However, 24 (92%) had a low risk of bias for outcome measurement, and 23 (88%) had a low or moderate risk of bias for selection of the reported result.

Anticholinergics and Dementia

Eleven of the 12 studies reporting dementia as an outcome were included in the quantitative analyses. The pooled OR for any use of drugs with definite anticholinergic activity and incident dementia was 1.20 (95% confidence interval [CI] 1.09–1.32) from seven studies (Figure 1). There was substantial heterogeneity (I2 = 86%); however, this is likely influenced by the inclusion of three large studies with small variances.[47] No study-level characteristics examined using meta-regression were significantly associated with the OR for dementia. Three large studies dominated the analysis and were similarly derived from US and UK population-based electronic health records.[9,10,45] All had good confounding control: two studies adjusted for a wide range of confounders,[9,10] and the other restricted to the main indication of depression.[45] The pooled OR for anticholinergic use for ≥90 days and ≥365 days and dementia were 1.23 (95% CI 1.17–1.29) and 1.50 (95% CI 1.22–1.85) from three and six studies, respectively (Figure 1). There was little heterogeneity (I2 = 2%) for ≥90 days use, but substantial heterogeneity for ≥365 days use (I2 = 90%). One further cohort study reported no association between anticholinergic use and dementia, reporting an OR of 0.67 (95% CI 0.40–1.15).[39] However, the definition of anticholinergics was broad, being dominated by drugs with 'mild' anticholinergic activity, and so findings were not comparable with the other studies.

Figure 1.

Meta-analysis of odds ratios for dementia by any, at least short-term and long-term definite anticholinergic use versus no use. ^OR (95% CI) estimated as the inverse variance weighted average of the published adjusted ORs for exposures of 1–90, 91–365, 366–1095 and >1095 daily doses for any use, of 91–365, 366–1095 and >1095 daily doses for short-term use (90+ days) and of 366–1095 and >1095 daily doses for long-term use (365+ days). ‡OR (95% CI) estimated as the inverse variance weighted average of the published adjusted ORs for exposures of 90–364, 365–1459 and >1460 daily doses for short-term use (90+ days) and of 365–1459 and >1460 daily doses for long-term use (365+ days). *The Cai 2013 estimate is for 60+ days use versus <60 days, Ancelin 2006 estimated long-term use (365+ days) as use at baseline and at 1-year follow-up and Gomm 2016 estimated long-term use (365+ days) as a prescription every quarter for 6 consecutive quarters. **OR (95% CI) estimated as the inverse variance weighted average of the published adjusted ORs for exposures of oxybutynin, solifenacin and tolterodine. Abbreviations: n, number of dementia cases; N, number of participants.

Anticholinergics and MCI

Six cohort studies estimated the effect of drugs with definite anticholinergic activity and incident MCI. Three studies estimated the effect of any definite anticholinergic use (pooled OR 1.24; 95% CI 0.97–1.59; I2 = 0%) (Figure 2). One study reported an OR of 1.70 (95% CI 0.52–5.57) for ≥60 days use,[26] and we estimated a pooled OR of 2.52 (95% CI 0.71–8.95) from two studies examining long-term use, albeit with substantial heterogeneity (I2 = 80%). A further study reported an OR of 1.15 (95% CI 1.01–1.31) per year of standardised daily doses and incident MCI and hence could not be numerically combined with the other exposure classifications.[43] For a further cohort study of Parkinson's disease patients, we estimated an unadjusted OR of 1.17 (95% CI 0.66–2.06) between anticholinergic use including mild anticholinergics and MCI.[42]

Figure 2.

Meta-analysis of odds ratios for mild cognitive impairment by any, at least short-term and long-term definite anticholinergic use versus no use. *Campbell 2010 estimated long-term use (365+ days) as use at all participating waves (baseline, 3-year and 6-year follow-up).

Anticholinergics and Cognitive Decline

Cognitive decline was reported on by 14 studies. We calculated the SMD of global cognitive decline for six studies:[11,14,31,36,38,46] five studies reported MMSE, while one calculated a composite score from 19 cognitive tests.[38] All but one study adjusted for some potential confounders.[31] Greater cognitive decline was consistently observed among patients taking anticholinergic drugs (SMD 0.15; 95% CI 0.09–0.21, I 2 = 3%) (Figure 3). There was some evidence that studies with longer follow-up reported greater cognitive decline (P = 0.08 from meta-regression) (Appendix Figure 2).

Figure 3.

Meta-analysis of standardised mean differences in global cognitive decline by any definite anticholinergic use versus no use. *Standardised mean difference (95% CI) estimated as the inverse variance weighted average of the estimated standardised mean difference for prevalent and incident users. Decline in global cognition was defined as the change in mean z-score across 19 cognitive tests. Abbreviations: d, standardised mean difference.

A further seven cohort studies were excluded from the meta-analysis. Two studies reported cognition at follow-up and not decline since baseline.[27,35] The Scottish Mental Survey of 1932 provided insufficient detail for pooling, but reported lower mean MMSE scores with anticholinergic use.[39] Finally, four studies used incompatible definitions of anticholinergic use. We were able to calculate SMDs in MMSE scores with anticholinergic use of 0.26 (95% CI −0.02, 0.55) over 18-month follow-up and 0.55 (95% CI 0.17, 0.92) over 6-month follow-up within a Parkinson's disease study and Taiwan Veterans care home study, respectively.[40,42] However, both studies included drugs with mild anticholinergic activity. The SMD in the decline in the Health and Retirement Study (HRS) 27-point Cognition Measure was 0.02 (95% CI −0.08, 0.12) over a 6-year follow-up for a total anticholinergic load of ≥3 in US HRS study.[37] A Canadian cohort study of urinary incontinence clinic outpatients reported no significant decline in MMSE with increased total anticholinergic load, but provided insufficient data for pooling.[26]

Sensitivity Analyses

None of the sensitivity analyses had a substantial effect on the pooled estimates (Appendix Figures 3–11).

Updated Search

Further automated study searches between April 2018 and November 2019 identified eight additional studies meeting our inclusion criteria.[48–55] Two studies focussed on depression,[48,49] and one on Parkinson's disease,[50] with the others having no specific disease focus. Five studies would be excluded from meta-analysis due to incompatible definitions of anticholinergic exposure, cognition or by patient duplication.[48–52] Using ROBINS-I, all study estimates were at serious/critical risk of bias and consistent with our review.[24] One large study replicated findings from the UK primary care population,[9] in another UK primary care cohort,[55] confirming that associations with dementia are limited to certain classes of anticholinergic medications. Similarly, a cohort study from the Netherlands reported greater dementia incidence with a higher anticholinergic burden score, but not once excluding antipsychotics and antidepressants.[51]