Association between anticholinergic burden and dementia in UK Biobank

Abstract Background Previous studies on the relationship between anticholinergic drugs and dementia have reported heterogeneous results. This variability could be due to different anticholinergic scales and differential effects of distinct classes of drugs. Methods Using Cox proportional hazards models, we computed the association between annual anticholinergic burden (AChB) and the risk of dementia in UK Biobank with linked general practitioner prescription records between the years 2000 and 2015 (n = 171,775). Results AChB according to most anticholinergic scales (standardized odds ratio range: 1.027–1.125) and the slope of the AChB trajectory (hazard ratio = 1.094; 95% confidence interval: 1.068–1.119) were predictive of dementia. However, the association between AChB and dementia held only for some classes of drugs, especially antidepressants, antiepileptics, and antidiuretics. Discussion The heterogeneity in previous findings may partially be due to different effects for different classes of drugs. Future studies should establish differences in more detail and further examine the practicality of a general measure of AChB relating to the risk of dementia.


Hypotheses
We expected AChB to be positively associated with dementia across anticholinergic scales and the association to be stronger than the association between dementia and polypharmacy. Furthermore, we anticipated the increase of AChB over time to be positively associated with dementia. Finally, based on previous studies, 6,7 we hypothesized that

Sample
UK Biobank is a prospective study of > 500,000 participants that were recruited across the UK from 2006 to 2010. 14

Anticholinergic burden and drug class
Eleven anticholinergic scales [16][17][18][19][20][21][22][23][24][25][26] were chosen as previously identified 27 and two 28,29 were identified through a recent systematic review. 30 All anticholinergic scales used in this study, including full names and potential reasons for exclusion from the analyses, are listed in Table S2 in supporting information. One scale 25 was modified to include newer drugs as before; 31 for two scales, 17,19 updated versions were used (Aging Brain Care; 32  drugs classified by the authors as having "improbable anticholinergic action" were assigned an anticholinergic burden of 0.5 (between "no anticholinergic potency" and "weak anticholinergic potency") as has been done before. 27 Using the British National Formulary (https://bnf.nice.org.uk 33 ), brand names of anticholinergic drugs in the sample were substituted with generic names. Combination prescriptions containing several anticholinergic compounds were separated into multiple entries, each containing a single anticholinergic compound.
Each prescription was assigned anticholinergic scores based on the ratings from anticholinergic scales. Prescriptions of drugs with ophthalmic, otic, nasal, or topical routes of administration were assigned an anticholinergic score of 0, as before. 23

Covariates and statistical analysis
The predictor in most models was the cumulative AChB in year 0 (the sum of anticholinergic scores of prescriptions for a participant). Due to the low ascertainment of prescriptions in the early years of sampling, 27 year 0 was for each participant defined as the first full year of having been included in the prescriptions' register after the year 1999.
Because the rate of dementia increases with age, participants younger than 60 years at the time of diagnosis or at the end of the prescriptions sampling period (June 30, 2020)-whichever came firstwere excluded from the analyses. Additionally, participants who before year 0 or within a cut-off period after year 0, had been diagnosed with dementia or prescribed a cholinesterase inhibitor (donepezil, galantamine, or rivastigmine) or memantine were excluded from the analyses. For all analyses in the main text, the cut-off period above was 1 year. Based on comments by the reviewers, we varied this cut-off and repeated the analysis on the association between AChB according to the scale by Dúran et al. 21 and dementia for every possible value of this cut-off (1 year to 20 years; Figure S4 in supporting information). People diagnosed with certain disorders are more likely to develop dementia. For this reason, we also excluded participants diagnosed at any point with Parkinson's disease, Huntington disease, Creutzfeldt-Jacob disease, or multiple sclerosis from our analyses. Finally, the prescribing period after the year 2015 was incomplete 27 and was removed.
The data cleaning process is described in Figure S1 in supporting information. For the association between AChB and dementia, Cox proportional hazards models were used, and effects are expressed as hazard ratios (HRs) with accompanying 95% confidence intervals (CIs). For studying time-to-event latencies, logistic regression was used, and effects are expressed in odds ratios (ORs). The association between the longitudinal evolution of AChB and dementia accounted for the competing risk of death and was assessed with the joint model for longitudinal and time-to-event data using the R library JM. 37 For all other analyses using only a single anticholinergic scale, the value-based scale by Durán et al. 21 was used, as it exhibited the strongest association with dementia. Models for which AChB was the main predictor were additionally controlled for polypharmacy. The two models for which polypharmacy was the main predictor differed from each other in the included covariates: (1) one was controlled for all covariates described above except for polypharmacy, and (2)

Characteristics of the sample
After data cleaning, the final sample consisted of 171,775 participants.
Among the participants, 2124 (1.2%) were diagnosed with dementia (Table S4 in Table 1 and Table   S5 in supporting information. Depending on the scale used, anticholinergic drugs constituted between 2.5% and 21.8% of all prescriptions  (Table S6 in supporting information).
The characteristics of anticholinergic prescribing in UK Biobank have been described in greater detail elsewhere. 27  modeled as a competing outcome, a one standard deviation increase in AChB was associated with a 12.0% (95% CI: 7.1%-17.2%) increase in the incidence of dementia, and a 6.0% (95% CI: 3.5%-8.5%) increase in the incidence of all-cause mortality.

Time-to-event latency
We compared the risk of dementia occurring within 12 years, between 12 and 14 years, between 14 and 16 years, 16 and 18 years, or more

Interpretation of the findings
In this study, we used electronic prescription data from 171,775 participants in UK Biobank to study the relationship between AChB and dementia risk. In line with our hypotheses, AChB was associated with dementia across most anticholinergic scales and the best effect estimate for most scales tended to be greater than that for polyphar-macy. The data also supported our hypothesis that the trajectory of AChB over time was predictive of dementia, even after accounting for the competing risk of death. The hypotheses regarding classspecific effects were mostly upheld, with AChB due to antidepressants, antiepileptics, and antihistamines positively associated with dementia risk. However, the effects for antipsychotics and for urological drugs were not significant. We also found associations between additional classes of drugs and risk of dementia, especially high-ceiling diuretics (furosemide). Finally, the strength of the association between AChB and dementia remained unchanged, regardless of the latency between time of measurement and time of diagnosis.
Our results support an association between AChB and dementia across anticholinergic scales, a finding observed previously using selfreported medicine use in UK Biobank. 36 This relationship persisted after controlling for several covariates. Across most anticholinergic scales, AChB was a stronger predictor than the total number of pre- limited to AChB attributable to certain classes of drugs. This is consistent with previous findings 6,7 that reported that AChB attributable to antidepressants, antihistamines, and antiepileptic drugs was associated with dementia; this consistency was not found for antipsychotics and urological drugs. Third, findings here and elsewhere 7 indicate that a higher anticholinergic potency of a drug does not always correspond to a higher risk of dementia.
The consistency in effect sizes for the association between AChB and dementia for different time-to-event latencies has been observed before 6,7

Strengths and weaknesses
The main strengths of our study are the size of the sample, the depth of available data, and the high accuracy of UK Biobank for ascertainment of dementia. 44 Furthermore, our analyses examined AChB from multiple perspectives, including comparing different scales and drug classes.
However, we acknowledge several limitations. The participants in UK Biobank are on average healthier and live in less deprived areas than the UK population. 45 Additionally, linked data do not include information on over-the-counter drugs and dietary supplements. Thus, AChB in the UK is likely higher than estimated in our study. Also, due to the low average age of the participants, UK Biobank has relatively few cases of dementia. Next, our analytical approach exhibits weaknesses. First, the dosages and quantities of medicines used in the calculation of the dosage-and quantity-adjusted scales required substantial manual cleaning and may not have been completely accurate. Second, the assumption of linearity between the predictor and the log hazard was sometimes not satisfied and transformations of the data were required to reliably run the models. Third, comparing the effects of different potencies of anticholinergic drugs, prescriptions with the highest potency were much less common than other groups of drugs. This could have affected the accuracy of our estimate.

CONCLUSIONS AND FUTURE DIRECTIONS
Inconsistencies in the literature, uncertainty of dose-response-or potency-response relationships, a strong drug-class dependency, and the difficulty of excluding confounding by indication, have led some 46 to suggest that a different common denominator-other than anticholinergic effect-is responsible for the observed association between anticholinergic drugs and dementia. If correct, the first goal should be the elucidation of the proposed association. Instead of studying the relationship of a general measure of AChB and cognitive decline, researchers could specify and describe the role of distinct classes of anticholinergic medicines-or even individual drugs.
Considering the role of the cholinergic system in the development of AD, 3 a biological underpinning for the effect of anticholinergic drugs in dementia is intuitive. However, further evidence is needed to determine the brain regions associated with the action of these drugs and the biological pathways likely involved in their proposed effects.