https://orcid.org/0000-0002-9774-6751
Figures
Abstract
Background
Anticholinergic burden has been associated with adverse outcomes such as falls. To date, no gold standard measure has been identified to assess anticholinergic burden, and no conclusion has been drawn on which of the different measure algorithms best predicts falls in older patients from general practice. This study compared the ability of five measures of anticholinergic burden to predict falls. To account for patients’ individual susceptibility to medications, the added predictive value of typical anticholinergic symptoms was further quantified in this context.
Methods and findings
To predict falls, models were developed and validated based on logistic regression models created using data from two German cluster-randomized controlled trials. The outcome was defined as “≥ 1 fall” vs. “no fall” within a 6-month follow-up period. Data from the RIME study (n = 1,197) were used in model development, and from PRIMUM (n = 502) for external validation. The models were developed step-wise in order to quantify the predictive ability of anticholinergic burden measures, and anticholinergic symptoms. In the development set, 1,015 patients had complete data and 188 (18.5%) experienced ≥ 1 fall within the 6-month follow-up period. The overall predictive value of the five anticholinergic measures was limited, with neither the employed anticholinergic variable (binary / count / burden), nor dose-dependent or dose-independent measures differing significantly in their ability to predict falls. The highest c-statistic was obtained using the German Anticholinergic Burden Score (0.73), whereby the optimism-corrected c-statistic was 0.71 after interval validation using bootstrapping and 0.63 in the external validation. Previous falls and dizziness / vertigo had the strongest prognostic value in all models.
Citation: Dinh TS, Meid AD, Rudolf H, Brueckle M-S, González-González AI, Bencheva V, et al. (2023) Anticholinergic burden measures, symptoms, and fall-associated risk in older adults with polypharmacy: Development and validation of a prognostic model. PLoS ONE 18(1): e0280907. https://doi.org/10.1371/journal.pone.0280907
Editor: Alessandra Coin, Clinca Geriatrica, ITALY
Received: May 30, 2022; Accepted: December 22, 2022; Published: January 23, 2023
Copyright: © 2023 Dinh et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: Specific legal and ethical restrictions prohibit public sharing of the data set, as data contain potentially identifying and sensitive patient information (date of birth, diagnoses and medication data in individual patterns that may allow the identification of vulnerable study participants in the regions under study (Witten/Herdecke and Hesse)). Data will be provided upon reasonable request, e.g. to allow reproducibility of the results (contact: Institute of General Practice, Goethe-University Frankfurt am Main, heading: "PROPERmed", Theodor-Stern-Kai 7, 60590 Frankfurt am Main, Germany, Tel. +49 69 6301 84483, Fax: +49 69 6301 6428).
Funding: PROPERmed was funded by the German Innovation Fund (grant number 01VSF16018). RIME was funded by the German Federal Ministry of Education and Research (grant number 01ET1005A). PRIMUM was funded by the German Federal Ministry of Education and Research (grant number 01GK0702). The funder had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Drugs with anticholinergic (ACh) properties are the most frequently prescribed potentially inappropriate medications in older adults [1], irrespective of whether the ACh effects are desired (as with spasmolytics), or are merely a side effect (as with some antipsychotics). They have been associated with a variety of adverse drug reactions (e.g., dry mouth, blurred vision, and drowsiness) and adverse outcomes, most notably including delirium, cognitive decline and falls [2–6]. Falling, in particular, is one of the main causes of disability, injuries and death in older patients, and is associated with hospital admissions, reduced health-related quality of life and increased health care costs [7–9].
Over the past 20 years, more than 20 measures have been developed to quantify ACh burden and help clinicians reduce their adverse effects. No gold standard measure yet exists, presumably because the employed metrics differ considerably in terms of, for example, the (number of) included drugs and the way (cumulative) ACh burden is calculated [3, 5, 10–14]. For instance, the same drug (e.g., furosemide) can be rated as having low or high activity, depending on the measure’s definition [15]. While most measures use a scoring system from 0 (none) to 3 (high) to describe pharmacological interactions with muscarinic receptors, some exceptions [16–19] also consider drug dosage. Many studies have examined the ability of these ACh burden measures to predict adverse outcomes, including falls [3, 5, 10–14].
However, the association between specific measures of ACh burden and falls has been inconsistently described in the literature [3, 12, 20, 21]. While the Drug Burden Index (DBI) [16] and the Anticholinergic Risk Scale (ARS) [22] have shown some degree of association with falls [3, 12, 21, 22], it has not yet been possible to draw a general conclusion on which ACh measure best predicts falls in older patients [21]. Older patients, in particular, are at increased risk of the accumulating consequences of multimorbidity and subsequent polypharmacy [16, 23–26]. In addition, unspecific ACh effects such as dizziness and blurred vision may be overlooked by clinicians and mistakenly interpreted as age-related symptoms [27]. Even though these symptoms may not directly cause substantial harm, they can indirectly lead to falls and other adverse outcomes [28]. It remains unclear which, if any, existing ACh measures can help predict falls in clinical practice, whereby it is also important that ACh symptoms are taken into account.
The aim of this study was therefore to compare the ability of five ACh burden measures to predict falls (and, if possible, to identify the one that performs best), and thereby to quantify the added predictive value of ACh symptoms.
Methods
The rationale behind and methodology used in this study have been described in detail in a study protocol [29]. We therefore provide only a brief summary here, in which we also describe necessary adaptions (a summary of adaptions to the study protocol is included in the S1 Text).
Source of data
Data from the binational PROPERmed database were used in model development and validation. In PROPERmed, individual participant data from five German and Dutch cluster-randomized controlled trials (cRCTs) were combined for pooled modelling purposes [30–32]. For this study, variables on ACh symptoms, a history of falls, and cognitive function were additionally taken from the studies. The RIME study (Reduction of potentially Inadequate Medication in the Elderly [33, 34]) and the PRIMUM study (PRIoritizing Multimedication in Multimorbidity [35]) were the only two cRCTs from PROPERmed to include this data and were therefore considered in this analysis. Both trials aimed to optimize medication in older German general practice (GP) patients. Data from the RIME trial was chosen for model development due to its larger sample size and higher number of events, and data from PRIMUM was used for external validation.
Participants
RIME included 1,197 older patients from 139 GP practices, and PRIMUM 502 patients from 72 GP practices. In RIME, patients were included if they were ≥ 70 years old and were prescribed ≥ 6 chronic medications. PRIMUM included patients aged ≥ 60 years, with ≥ 3 chronic conditions and prescriptions for ≥ 5 chronic medications. Patients with dementia, cognitive impairment, or a reduced life expectancy (RIME: ≤ 6 months, PRIMUM: ≤ 12 months), were excluded from participation.
Outcome
The study outcome was a binary indicator defined as “≥ 1 fall(s)” vs. “no fall(s)” within a 6-month follow-up period. Information on falls was self-reported and collected during patient interviews.
Predictors
For this study, candidate predictors were pre-selected based on a literature review, predictor availability, and clinical reasoning [29]. Candidate predictors were collected at baseline and included variables associated with:
- Sociodemographics and lifestyle (age, sex, living situation, educational level [36], and smoking status);
- Morbidity (in accordance with a list of 24 chronic conditions [30], number of chronic conditions);
- Health-status and well-being (pain, health-related quality of life [37], functional status [38], cognitive function, all-cause hospital admissions, and history of falls);
- Medication (number of drugs and variables to characterize ACh medications (see below));
- ACh symptoms (list of symptoms).
ACh-related symptoms were measured as binary indicators in both trials. The development and validation datasets shared three ACh symptoms (dizziness / vertigo, problems urinating, and stomach pain), while four further ACh symptoms were only present in the development dataset (drowsiness / fatigue, dry mouth, itching, and constipation), and one in the validation dataset (palpitations). In both trials, the history of falls was documented for 6-months before baseline. In RIME, cognitive function was measured using a word list that participants were asked to reproduce, while in PRIMUM, a verbal fluency test was employed [39]. To harmonize the different cognitive assessments in the two studies, quartiles were calculated and cognitive impairment was interpreted as no / mild / medium / severe.
Anticholinergic burden was calculated based on ATC codes and according to five different scales / equations:
- Anticholinergic Risk Scale (ARS) [22];
- Anticholinergic Drug Scale (ADS) [40];
- German Anticholinergic Burden Score (GerABS) [41];
- Muscarinic Acetylcholinergic Receptor ANTagonist Exposure scale (MARANTE) [17];
- German Drug Burden Index (GerDBI).
Of these five ACh measures, three used scores ranging from 0 to 3 [22, 40, 41], and two used equations that took drug dosage into account ([17] and the GerDBI). Details on the GerDBI, which was recently developed as part of the COFRAIL-study [42] and is based on the Drug Burden Index by Hilmer et al. [16, 43], will be published elsewhere. The scales were developed in Germany ([41] and GerDBI), the United States [22, 40], and Belgium [17]. In developing the model, three variables associated with ACh burden were calculated for each of the included scales / equations: a binary variable (prescription of ≥ 1 ACh medication(s)), a “count” variable (number of ACh medications per patient), and a “burden” variable (cumulative ACh burden / load for all of a patient’s medications).
Statistical analysis
Model development and performance.
Logistic regression analysis was used to develop the model, whereby our intention was to quantify the predictive ability of ACh burden measures, and ACh symptoms. We therefore developed the prognostic model stepwise (see below). We used backwards selection based on Akaike’s information criterion (AIC) and determined the suitable functional form for continuous variables using the multivariable fractional polynomial (MFP) approach [44, 45]. Variables (age and sex) that we considered clinically relevant were forced back into the model. The model was developed as follows:
In step 1, a base model (Model 1) was built to predict falls within a 6-month follow-up period using variables associated with sociodemographic / lifestyle, morbidity, health-status / well-being, and number of drugs.
In step 2, the aforementioned variables associated with ACh drug burden use (binary indicator / total (ACh drug) count / cumulative ACh burden) were separately added to the base model, bringing the number of different models to 15 (Models 2.1–2.15). In addition to age and sex, the ACh variable of interest was also considered mandatory in the corresponding models.
In step 3, ACh symptom variables were added to the 15 models developed in step 2 to quantify their additional predictive value. As some symptoms were not available in both development and validation datasets, two different types of model were developed at this stage:
- Step 3A: models that considered only shared symptoms (Models 3.1–3.15), and
- Step 3B: models that included all symptoms from the development dataset (Models 3.16–3.30).
An additional Model 4 (base model + ACh symptoms) was developed in step 4, with the aim to determine the added predictive value from ACh symptoms separately. Step 3B models and an additional Model 4 were only developed for exploratory purposes.
After developing the models, the best-performing step 3A model was selected for internal and external validation (based on the highest c-statistic and lowest AIC) [44]. The c-statistic and the model fit using AIC (in-sample) were used to assess the performance of models and to discriminate between them [44]. Added predictive value was quantified by comparing differences in area under the curve (AUC) and using integrated discrimination improvement (IDI) [46].
Internal validation.
The selected model was internally validated using bootstrapping. This involved creating a bootstrap sample and using it to develop the model and determine the predictive performance in both the bootstrap and the original sample. These steps were repeated 100 times and optimism was estimated. Mean optimism was then subtracted from the apparent performance of the original model to obtain optimization-adjusted performance estimates [47]. To adjust for overfitting, we determined the uniform shrinkage factor by shrinking the regression coefficients. The performance and discriminative ability of the models were compared before and after internal validation.
External validation.
The selected 3A model was externally validated using data from the PRIMUM study. Discriminative ability was assessed using c-statistic, and calibration of actual and predictive risk by estimating the expected / observed ratio. In addition, the calibration slope and calibration-in-the-large were calculated and a calibration plot was produced [48].
Sample size
We used the pmsampsize package in R to calculate the minimum sample size [49, 50]. Due to the nature of the study, this calculation was performed retrospectively and had only exploratory character. Based on the number of candidate predictors, empirical c-statistics (0.87) [51] and the given prevalence in the complete-case population (18.5%), the minimum sample size required to minimize the model’s potential for overfitting was calculated to be n = 1,131 with 210 events. We consider this acceptable with view of the size of the complete-case population of the development cohort (n = 1,015 with 188 events).
Missing data
Multiple imputation techniques were used to handle missing data [47, 52]. The nine imputed datasets corresponded to the percentage of incomplete observations [53, 54], which was approximately 8.5% in the development dataset. The selected model was developed for each of the nine multiply imputed datasets, and pooled estimates were obtained and compared with the results of the complete-case analysis [55].
Technical information and reporting
Statistical analyses were conducted using R version 4.1.2 (R Foundation for Statistical Computing, Vienna, Austria). The present manuscript follows the Transparent Reporting of a multivariable model for Individual Prognosis or Diagnosis (TRIPOD) statement [56].
Results
Participants
The complete-case populations of RIME and PRIMUM included 1,015 and 348 patients respectively. In RIME patients, mean age was 77 years, 50% were female, and 92% had a low or medium level of education. In PRIMUM participants, mean age was 72 years, 55% were female, and 90% had a low or medium level of education. The average number of chronic conditions was six in RIME and five in PRIMUM patients. RIME participants were taking an average of nine chronic medications and had two ACh symptoms, while PRIMUM patients used eight medications and reported one ACh symptom. In RIME, between 12.3% and 78.0% of patients used ≥ 1 ACh medication depending on which scale was used (ARS 12.3%, MARANTE 19.5%, ADS 37.9%, GerDBI 52.3%, and GerABS 78.0%). Anticholinergic drug use in PRIMUM varied between 8.3% and 72.7% (ARS 8.3%, MARANTE 17.8%, ADS 30.2%, GerDBI 39.1%, and GerABS 72.7%). The S1 Table provides a detailed overview of the characteristics of the complete-case population including the percentage of missing values, odds ratios, and confidence intervals from unadjusted bivariate analyses.
Model development
Of the complete-case population of the development set, 188 (18.5%) patients experienced at least one fall within the 6-month follow-up period. Table 1 shows overall prognostic variables stratified per observed outcome for both the development and validation sets. Bivariate analyses from the development set showed that patients that fell tended to have a history of falls (OR = 5.9), to suffer from pain (OR = 2.8) or dizziness (OR = 2.3), to have a higher number of ACh symptoms (OR = 1.2), and to have reduced functional status (OR = 1.2) (see S1 Table). Other predictors that were significantly associated with falls were, for example, the number of chronic conditions, number of chronic medications, health-related quality of life, female sex, and a previous admission to hospital. Anticholinergic drug use was higher in fallers measured with all five ACh scales / equations. Apart from the ARS, bivariate analyses showed an association between falls and both ACh “count” and “burden” variables from all ACh measures, while, except for MARANTE, no binary ACh variables were associated with falls.
Model development and internal validation
When the base model for falls was developed using candidate predictors from the categories of sociodemographics / lifestyle, morbidity, health-status / well-being, and number of drugs (step 1), variable selection using MFP and AIC yielded the best-performing Model 1 with a c-statistic of 0.712 (see S2 Table). The addition of ACh variables (step 2) brought the number of different models to 15, of which the c-statistics ranged from 0.712 to 0.714 (S3 Table). When the 15 models were extended to include ACh symptom variables that were present in both trials (step 3A), they produced c-statistics between 0.724 and 0.732 (see Table 2).
The best-performing model included the binary variable of the German Anticholinergic Burden Score by Kiesel et al. [41]. Based on the AUC, the discrimination performance of the model was 0.732. Bootstrap resampling resulted in a uniform shrinkage factor of 0.86 (average calibration slope from bootstrap samples). This was applied to the model to adjust for overfitting, and resulted in an optimism-adjusted AUC of 0.705 and optimism-adjusted regression coefficients (see Table 3).
External validation
After external validation, c-statistic decreased from 0.732 to 0.632. The expected / observed ratio was 1.09, indicating that the model over-predicts the total number of events. This was supported by a calibration slope of 0.716. The average predicted risk was 17.3% while the observed risk was 16.4%, showing that the model over-estimated the incidence by 0.9% (calibration-in-the-large). A calibration plot is shown in Fig 1.
Additional analyses
After multiple imputation of missing data, pooled estimates did not reveal substantial differences to the results of the complete-case analysis (see S4 Table). Exploratory analyses showed that (1) in terms of the c-statistic and AIC, models that included all symptoms (step 3B) did not significantly differ from models based on shared symptoms, and (2) the development of a model without variables on ACh use (Model 4, see S5 Table) resulted in a model of almost identical discriminative ability to the selected model that included them (AUCw/o ACh variables = 0.730 vs. AUCwith ACh variables = 0.732). Fig 2 shows the ROC curves of the best-performing models for all steps, and reveals minor differences in the discriminative ability of models with and without ACh variables (Model 3.10 vs. Model 4). An overview of the increase in discrimination in terms of Δ AUC and IDI is presented in Table 4.
Sensitivity analysis explored the effect of the intervention on the model’s outcome. The results of sensitivity analysis, in which intervention status was excluded from model development and validation, indicate that the c-statistic decreased from 0.712 to 0.710 for the base model, and from 0.732 to 0.727 for the full model, while the c-statistic of the validated model increased from 0.632 to 0.635 (see S5 Table).
Discussion
Our findings indicate that there are no significant differences in the ability of ACh scales / equations to predict falls. Furthermore, neither the type of ACh variable that was operationalized (i.e., binary indicator / total (ACh drug) count / cumulative ACh burden), nor the use of dose-dependent or dose-independent calculations, improved discrimination. The results indicate that measures of ACh burden hardly improve model performance at all, while measures of ACh symptoms do, at least marginally. In this respect, dizziness / vertigo may be interpreted as a typical side effect of ACh drugs and as such a good predictor of falls. A fundamental distinction must be made between association and prediction. While ACh burden measures may be more or less strongly associated with falls, their added predictive value beyond other significant predictors appears to be limited.
The use of ACh medications varied greatly in our study depending on the ACh metric used (12.3%– 78.0%). This result agrees with previous findings reporting that ACh use in a variety of clinical settings and patient populations ranged from 9%– 80% [12, 57–60] and are presumably caused by differences in the scales’ assessment of ACh burden [3, 5, 10–14]. Compared to international scales, both German indicators (GerDBI and GerABS) indicated higher use of ACh medications in our study population. Despite the observed variation in prevalence, no differences were seen in the association between the various ACh burden measures and falls.
Studies comparing more than two measures of ACh burden and their relationship to clinical outcomes are rare. A recent systematic review by Lisibach et al. [5] identified only two such studies, one of which included falls as an outcome of interest [12]. Of nine ACh burden measures, four were significantly linked to falls, whereby the strongest association was found for the DBI, followed by the ARS, and the scales of Chew et al. and Sittironnarit et al. [61, 62]. While the significant association between the DBI and falls, and especially the advantage it offers of considering dose adjustments, has been described by other researchers [11, 15, 63], dose-dependent ACh measures included in our study (GerDBI and MARANTE) did not predict falls more accurately than any other scales. However, this conclusion is drawn based on the results of the complete-case analysis and should therefore be interpreted with caution as missing data on dosage have prevented us from calculating dose-dependent ACh burden for all participating patients. It should be further noted that the GerDBI includes both medication with anticholinergic and sedative activities.
A systematic review by Stewart et al. [21] compared eight studies of the relationship between the ARS and the Anticholinergic Cognitive Burden Scale (ACBS) [64] with falls. Consistent with our findings, the authors concluded that neither of the ACh measures could be explicitly identified as the best predictor of falls in older adults. Furthermore, results from a study by Ruxton et al. [65] indicate that the intake of individual medications such as imipramine or amitriptyline is more strongly associated with an increased risk of falls than ACh burden, as measured using scales. It is worth noting that the results of the studies described here showed great differences depending on the setting (e.g., primary care, nursing home, insurance database), follow-up duration (3–38.5 months), and the employed definition of falls (e.g., self-reported, falls reported in medical records) [12, 21, 65]. The limited comparability of our findings with previous research should therefore be taken into account when interpreting our results.
From a practical point of view, the results of our study call into question the use of ACh scales / equations to predict falls. On the other hand, although small, the added value of using symptoms to predict falls was nevertheless noteworthy. Attention should therefore be paid to the presence of adverse effects in patients taking an ACh medication, as they may indicate whether a patient is particularly sensitive. While the number of symptoms only showed a significant association in bivariate analyses, the association with dizziness / vertigo was statistically significant and present in all multivariate models. The strong association between previous falls and dizziness / vertigo and future falls has been identified in numerous other studies [66–70]. Dizziness / vertigo and balance problems have previously been linked to ACh burden, and can be clinically relevant when they increase the risk of falling [20]. Symptoms may therefore help to operationalize patients’ susceptibility and may be of interest in future research in this context. Further investigations should also consider the role of symptoms in relation to the intake of specific ACh medications and such fall-related outcomes as physical decline and fractures [71]. This is also true of other drugs that raise the risk of falls [72–74], even though most of them have been characterized to be of ACh nature [75]. An individual evaluation of patients’ medications (e.g., through a structured medication review [76]) in conjunction with a previous history of falls and dizziness / vertigo (as the strongest predictors identified in our study) should therefore be further explored.
To the best of our knowledge, this is the first study to explicitly investigate whether an individual’s risk of falling can be better predicted when ACh symptoms are considered in addition to qualitative and quantitative measures of ACh burden. With few exceptions, most studies examining the link between ACh measures and falls were conducted outside Germany [3, 5, 77] and used scales that had not been adapted for use in the German drug market. In our study, differences in the performance of the various tools were small, but the metric which performed best was one of the two German tools.
The likelihood of falls in older persons is dependent on an almost infinite number of factors. Risk and protective factors depend on a complex interplay between individual characteristics, such as age, morbidity, functional status, medication and behavioral patterns, and social and living environments. This poses limits to any attempt to explore causal and / or prognostic associations. The unavailability of data prevented us from considering known risk factors for falls in older patients such as gait problems or muscle weakness [67, 70], or from drawing conclusions for specific patient populations, such as people living with HIV [78]. It is also unclear whether a longer follow-up period or a different way of operationalizing falls (continuous count) would have resulted in different findings. Although the follow-up duration of 6 months was shorter than in most other studies [79, 80], another limitation was nonetheless the risk of recall bias when patients self-reported falls [81]. Furthermore, information on falls used in other analyses was mainly based on claims data and medical records, in which severe falls with serious consequences are preferentially recorded.
From a methodological point of view, the best-performing model yielded a c-statistic of 0.732, which indicates that the model had acceptable discrimination. External validation resulted in a decrease in the model’s discriminative ability, and calibration measures indicated that, to some degree, the model over-predicted and over-estimated the risk of falls. Sensitivity analysis indicated that intervention status had an impact on the model’s outcome. A general challenge to external model validation in different study cohorts has already been identified in the prediction of hospital admissions in PROPERmed [32]. In our case, this may be partly explained by differences in the inclusion criteria and in the baseline risk of patients from the development and validation datasets. On average, RIME patients were older (77 vs. 72 years), more frequently hospitalized, had more chronic conditions, took more drugs, and reported more symptoms than PRIMUM patients [30]. Other potential limitations relate to the harmonization of predictors, the insufficient sample sizes of the development and validation cohorts, and the low numbers of events (n = 188 in the development cohort and n = 55 in the validation cohort). As a rule-of-thumb, a minimum number of 100 events has been recommended for the external validation of a prognostic model [82]. It is also worth noting that our results reflect the data, ACh burden measures, and outcomes, selected for this study. It should therefore be borne in mind that our findings may have been different if we had chosen different measures of ACh burden or outcomes other than falls.
In conclusion, the findings of this study call into question the added value of using measures of ACh burden for risk classification in fall-prediction models. On the other hand, our findings indicate that the inclusion of symptom variables improve model performance. Medications that pose a risk of falling are a risk factor that can be directly modified, for example by changing the medication, or adjusting the dose. The prescription of medications with ACh properties in older patients should therefore be carefully evaluated, especially in patients with a history of falls and existing symptoms of dizziness / vertigo.
Supporting information
S1 Table. Patient characteristics and results from the bivariate analysis of the complete-case population of the development set (n = 1,015).
Abbreviations: ARS–Anticholinergic Risk Scale (23); ADS–Anticholinergic Drug Scale (41); GerABS–German Anticholinergic Burden Score (42); MARANTE–Muscarinic Acetylcholinergic Receptor ANTagonist Exposure Scale (18); GerDBI–German Drug Burden Index.
https://doi.org/10.1371/journal.pone.0280907.s002
(DOCX)
S2 Table. Base model for falls within 6-months of follow-up (Model 1).
https://doi.org/10.1371/journal.pone.0280907.s003
(DOCX)
S3 Table. Comparison of model for falls within 6-months of follow-up from step 2.
Abbreviations: ACh–anticholinergic; ARS–Anticholinergic Risk Scale (23); ADS–Anticholinergic Drug Scale (41); AIC–Akaike Information Criterion; GerABS–German Anticholinergic Burden Score (42); MARANTE–Muscarinic Acetylcholinergic Receptor ANTagonist Exposure Scale (18); GerDBI–German Drug Burden Index.
https://doi.org/10.1371/journal.pone.0280907.s004
(PDF)
S4 Table. Pooled estimates after multiple imputation.
Abbreviation: GerABS–German Anticholinergic Burden Score (42).
https://doi.org/10.1371/journal.pone.0280907.s005
(DOCX)
S5 Table. Base model incl. symptoms for falls within 6-months of follow-up (Model 4).
https://doi.org/10.1371/journal.pone.0280907.s006
(DOCX)
S6 Table. Sensitivity analysis.
Abbreviation: CI - confidence interval.
https://doi.org/10.1371/journal.pone.0280907.s007
(PDF)
Acknowledgments
The authors would like to thank Kiran Chapidi for support in data management. The authors are also grateful to all members of the PROPERmed, PRIMUM, and RIME study groups and Phillip Elliott for conducting a language review.
References
- 1. Hukins D, Macleod U, Boland JW. Identifying potentially inappropriate prescribing in older people with dementia: A systematic review. Eur J Clin Pharmacol 2019; 75(4):467–81. pmid:30610274
- 2. Mintzer J, Burns A. Anticholinergic side-effects of drugs in elderly people. J R Soc Med 2016; 93(9):457–62.
- 3. Welsh TJ, van der Wardt V, Ojo G, Gordon AL, Gladman JRF. Anticholinergic drug burden tools/scales and adverse outcomes in different clinical settings: A Systematic Review of Reviews. Drugs Aging 2018; 35(6):523–38. pmid:29736815
- 4. Ness J, Hoth A, Barnett MJ, Shorr RI, Kaboli PJ. Anticholinergic medications in community-dwelling older veterans: Prevalence of anticholinergic symptoms, symptom burden, and adverse drug events. Am J Geriatr Pharmacother 2006; 4(1):42–51. pmid:16730620
- 5. Lisibach A, Benelli V, Ceppi MG, Waldner-Knogler K, Csajka C, Lutters M. Quality of anticholinergic burden scales and their impact on clinical outcomes: a systematic review. Eur J Clin Pharmacol 2021; 77(2):147–62. pmid:33011824
- 6. Nagai J, Ishikawa Y. Analysis of anticholinergic adverse effects using two large databases: The US Food and Drug Administration Adverse Event Reporting System database and the Japanese Adverse Drug Event Report database. PLoS One 2021; 16(12):e0260980. pmid:34855908
- 7.
National Institute for Health and Care Excellence (NICE). Falls: Assessment and prevention of falls in older people; 2013 [cited 2020 Apr 14]. Available from: URL: https://www.ncbi.nlm.nih.gov/books/NBK258885/pdf/Bookshelf_NBK258885.pdf.
- 8. Verma SK, Willetts JL, Corns HL, Marucci-Wellman HR, Lombardi DA, Courtney TK et al. Falls and fall-related injuries among community-dwelling adults in the United States. PLoS ONE 2016; 11(3):e0150939. pmid:26977599
- 9. Jong MR de, van der Elst M, Hartholt KA. Drug-related falls in older patients: implicated drugs, consequences, and possible prevention strategies. Ther Adv Drug Saf 2013; 4(4):147–54. pmid:25114778
- 10. Cardwell K, Hughes CM, Ryan C. The Association between anticholinergic medication burden and health related outcomes in the ’oldest old’: A Systematic Review of the Literature. Drugs Aging 2015; 32(10):835–48. pmid:26442862
- 11. Villalba-Moreno AM, Alfaro-Lara ER, Pérez-Guerrero MC, Nieto-Martín MD, Santos-Ramos B. Systematic review on the use of anticholinergic scales in poly pathological patients. Arch Gerontol Geriatr 2016; 62:1–8. pmid:26518612
- 12. Mayer T, Meid AD, Saum K-U, Brenner H, Schöttker B, Seidling HM et al. Comparison of nine instruments to calculate anticholinergic load in a large cohort of older outpatients: Association with Cognitive and Functional Decline, Falls, and Use of Laxatives. Am J Geriatr Psychiatry 2017; 25(5):531–40. pmid:28233606
- 13. Salahudeen MS, Duffull SB, Nishtala PS. Anticholinergic burden quantified by anticholinergic risk scales and adverse outcomes in older people: A systematic review. BMC Geriatr 2015; 15:31. pmid:25879993
- 14. Salahudeen MS, Hilmer SN, Nishtala PS. Comparison of anticholinergic risk scales and associations with adverse health outcomes in older people. J Am Geriatr Soc 2015; 63(1):85–90. pmid:25597560
- 15. Villalba-Moreno AM, Alfaro-Lara ER, Santos-Ramos B. Anticholinergic risk: Use and limitations of anticholinergic scales. Eur J Intern Med 2015; 26(10):e65–6. pmid:26365371
- 16. Hilmer SN, Mager DE, Simonsick EM, Cao Y, Ling SM, Windham BG et al. A drug burden index to define the functional burden of medications in older people. Arch Intern Med 2007; 167(8):781–7. pmid:17452540
- 17. Klamer TT, Wauters M, Azermai M, Durán C, Christiaens T, Elseviers M et al. A Novel Scale Linking Potency and Dosage to Estimate Anticholinergic Exposure in Older Adults: the Muscarinic Acetylcholinergic Receptor ANTagonist Exposure Scale. Basic Clin Pharmacol Toxicol 2017; 120(6):582–90. pmid:28090742
- 18. Nguyen PV-Q, Pelletier L, Payot I, Latour J. The Delirium Drug Scale is associated to delirium incidence in the emergency department. Int Psychogeriatr 2018; 30(4):503–10. pmid:29208069
- 19. Cao Y-J, Mager DE, Simonsick EM, Hilmer SN, Ling SM, Windham BG et al. Physical and cognitive performance and burden of anticholinergics, sedatives, and ACE inhibitors in older women. Clin Pharmacol Ther 2008; 83(3):422–9. pmid:17713474
- 20. Phillips A, Heier M, Strobl R, Linkohr B, Holle R, Peters A et al. Exposure to anticholinergic and sedative medications using the Drug Burden Index and its association with vertigo, dizziness and balance problems in older people—Results from the KORA-FF4 Study. Exp Gerontol 2019; 124:110644. pmid:31260723
- 21. Stewart C, Taylor-Rowan M, Soiza RL, Quinn TJ, Loke YK, Myint PK. Anticholinergic burden measures and older people’s falls risk: a systematic prognostic review. Ther Adv Drug Saf 2021; 12:20420986211016645. pmid:34104401
- 22. Rudolph JL, Salow MJ, Angelini MC, McGlinchey RE. The anticholinergic risk scale and anticholinergic adverse effects in older persons. Arch Intern Med 2008; 168(5):508–13. pmid:18332297
- 23. Thomas EJ, Brennan TA. Incidence and types of preventable adverse events in elderly patients: population based review of medical records. BMJ 2000; 320(7237):741–4. pmid:10720355
- 24. Delafuente JC. Pharmacokinetic and pharmacodynamic alterations in the geriatric patient. Consult Pharm 2008; 23(4):324–34. pmid:18454589
- 25. Gnjidic D, Hilmer SN, Blyth FM, Naganathan V, Waite L, Seibel MJ et al. Polypharmacy cutoff and outcomes: Five or more medicines were used to identify community-dwelling older men at risk of different adverse outcomes. J Clin Epidemiol 2012; 65(9):989–95. pmid:22742913
- 26. Kouladjian O’Donnell L, Gnjidic D, Nahas R, Bell JS, Hilmer SN. Anticholinergic burden: Considerations for older adults. J Pharm Pract Res 2017; 47(1):67–77.
- 27. Strobach D. Anticholinerge Arzneistoffe: Erkennen, erklären, ersetzen. Pharmazeutische Zeitung 2013; (41).
- 28. Chimirri S, Aiello R, Mazzitello C, Mumoli L, Palleria C, Altomonte M et al. Vertigo/dizziness as a Drugs’ adverse reaction. J Pharmacol Pharmacother 2013; 4(Suppl 1):S104–9. pmid:24347974
- 29. Dinh TS, González-González AI, Meid AD, Snell KIE, Rudolf H, Brueckle M-S et al. Are Anticholinergic Symptoms a Risk Factor for Falls in Older General Practice Patients With Polypharmacy? Study Protocol for the Development and Validation of a Prognostic Model. Front Pharmacol 2020; 11:577747. pmid:33519441
- 30. González-González AI, Dinh TS, Meid AD, Blom JW, van den Akker M, Elders PJM et al. Predicting negative health outcomes in older general practice patients with chronic illness: Rationale and development of the PROPERmed harmonized individual participant data database. Mech Ageing Dev 2021; 194:111436. pmid:33460622
- 31. González-González AI, Meid AD, Dinh TS, Blom JW, van den Akker M, Elders PJM et al. A prognostic model predicted deterioration in health-related quality of life in older patients with multimorbidity and polypharmacy. J Clin Epidemiol 2021; 130:1–12. pmid:33065164
- 32. Meid AD, Gonzalez-Gonzalez AI, Dinh TS, Blom J, van den Akker M, Elders P et al. Predicting hospital admissions from individual patient data (IPD): an applied example to explore key elements driving external validity. BMJ Open 2021; 11(8):e045572. pmid:34348947
- 33. Rudolf H, Thiem U, Aust K, Krause D, Klaaßen-Mielke R, Greiner W et al. Reduction of Potentially Inappropriate Medication in the Elderly–Results of a Cluster- Randomized, Controlled Trial in German Primary Care Practices (RIME). Dtsch Arztebl Int 2021; (Forthcoming).
- 34. Thiem U, Wilm S, Greiner W, Rudolf H, Trampisch H-J, Müller C et al. Reduction of potentially inappropriate medication in the elderly: design of a cluster-randomised controlled trial in German primary care practices (RIME). Ther Adv Drug Saf 2020; 12:2042098620918459. pmid:32435445
- 35. Muth C, Uhlmann L, Haefeli WE, Rochon J, van den Akker M, Perera R et al. Effectiveness of a complex intervention on Prioritising Multimedication in Multimorbidity (PRIMUM) in primary care: results of a pragmatic cluster randomised controlled trial. BMJ Open 2018; 8(2):e017740. pmid:29478012
- 36. Eurostat Statistics Explained. International Standard Classification of Education (ISCED).; 2020 [cited 2021 Dec 29]. Available from: URL: https://ec.europa.eu/eurostat/statistics-explained/index.php?title=International_Standard_Classification_of_Education_(ISCED).
- 37. EuroQol—a new facility for the measurement of health-related quality of life. Health Policy 1990; 16(3):199–208. pmid:10109801
- 38. Saliba D, Elliott M, Rubenstein LZ, Solomon DH, Young RT, Kamberg CJ et al. The Vulnerable Elders Survey: a tool for identifying vulnerable older people in the community. J Am Geriatr Soc 2001; 49(12):1691–9. pmid:11844005
- 39. Luck T, Riedel-Heller SG, Wiese B, Stein J, Weyerer S, Werle J et al. CERAD-NP-Testbatterie: Alters-, geschlechts- und bildungsspezifische Normen ausgewählter Subtests. Ergebnisse der German Study on Ageing, Cognition and Dementia in Primary Care Patients (AgeCoDe). Z Gerontol Geriatr 2009; 42(5):372–84.
- 40. Carnahan RM, Lund BC, Perry PJ, Pollock BG, Culp KR. The Anticholinergic Drug Scale as a measure of drug-related anticholinergic burden: associations with serum anticholinergic activity. J Clin Pharmacol 2006; 46(12):1481–6. pmid:17101747
- 41. Kiesel EK, Hopf YM, Drey M. An anticholinergic burden score for German prescribers: score development. BMC Geriatr 2018; 18(1):239. pmid:30305048
- 42. Mortsiefer A, Wilm S, Santos S, Löscher S, Wollny A, Drewelow E et al. Family conferences and shared prioritisation to improve patient safety in the frail elderly (COFRAIL): study protocol of a cluster randomised intervention trial in primary care. Trials 2020; 21(1):285. pmid:32197631
- 43. Hilmer SN, Mager DE, Simonsick EM, Ling SM, Windham BG, Harris TB et al. Drug burden index score and functional decline in older people. The American Journal of Medicine 2009; 122(12):1142–1149.e1-2. pmid:19958893
- 44. Sauerbrei W, Perperoglou A, Schmid M, Abrahamowicz M, Becher H, Binder H et al. State of the art in selection of variables and functional forms in multivariable analysis-outstanding issues. Diagn Progn Res 2020; 4:3. pmid:32266321
- 45. Royston P, Altman DG. Regression Using Fractional Polynomials of Continuous Covariates: Parsimonious Parametric Modelling. Applied Statistics 1994; 43(3):429.
- 46. Pencina MJ, D’Agostino RB, Demler OV. Novel metrics for evaluating improvement in discrimination: net reclassification and integrated discrimination improvement for normal variables and nested models. Statist. Med. 2012; 31(2):101–13. pmid:22147389
- 47. Moons KGM, Kengne AP, Woodward M, Royston P, Vergouwe Y, Altman DG et al. Risk prediction models: I. Development, internal validation, and assessing the incremental value of a new (bio)marker. Heart 2012; 98(9):683–90. pmid:22397945
- 48. Steyerberg EW, Vickers AJ, Cook NR, Gerds T, Gonen M, Obuchowski N et al. Assessing the performance of prediction models: a framework for traditional and novel measures. Epidemiology 2010; 21(1):128–38. pmid:20010215
- 49. Riley RD, Ensor J, Snell KIE, Harrell FE, Martin GP, Reitsma JB et al. Calculating the sample size required for developing a clinical prediction model. BMJ 2020; 368:m441. pmid:32188600
- 50. Riley RD, Snell KI, Ensor J, Burke DL, Harrell FE, Moons KGM et al. Minimum sample size for developing a multivariable prediction model: PART II—binary and time-to-event outcomes. Stat Med 2019; 38(7):1276–96. pmid:30357870
- 51. Smith MI, Lusignan S de, Mullett D, Correa A, Tickner J, Jones S. Predicting Falls and When to Intervene in Older People: A Multilevel Logistical Regression Model and Cost Analysis. PLoS One 2016; 11(7):e0159365. pmid:27448280
- 52.
Harrell FE. Regression Model Strategies: with Applications to Linear Models, Logistic and Ordinal Regression, and Survival Analysis. 2nd Edition Cham. Heidelberg, London: Springer.
- 53. White IR, Royston P, Wood AM. Multiple imputation using chained equations: Issues and guidance for practice. Statist. Med. 2011; 30(4):377–99. pmid:21225900
- 54. Hippel PT von. How to Impute Interactions, Squares, and other Transformed Variables. Sociological Methodology 2009; 39(1):265–91.
- 55. Sterne JAC, White IR, Carlin JB, Spratt M, Royston P, Kenward MG et al. Multiple imputation for missing data in epidemiological and clinical research: potential and pitfalls. BMJ 2009; 338:b2393. pmid:19564179
- 56. Collins GS, Reitsma JB, Altman DG, Moons KGM. Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD): the TRIPOD statement. Ann Intern Med 2015; 162(1):55–63. pmid:25560714
- 57. Naples JG, Marcum ZA, Perera S, Gray SL, Newman AB, Simonsick EM et al. Concordance Between Anticholinergic Burden Scales. J Am Geriatr Soc 2015; 63(10):2120–4. pmid:26480974
- 58. Turró-Garriga O, Calvó-Perxas L, Vilalta-Franch J, Blanco-Silvente L, Castells X, Capellà D et al. Measuring anticholinergic exposure in patients with dementia: A comparative study of nine anticholinergic risk scales. Int J Geriatr Psychiatry 2018; 33(5):710–7. pmid:29292848
- 59. Chatterjee S, Mehta S, Sherer JT, Aparasu RR. Prevalence and predictors of anticholinergic medication use in elderly nursing home residents with dementia. Drugs Aging 2010; 27(12):987–97.
- 60. McNeely SS, Bhattacharya R, Aparasu RR. Prevalence of anticholinergic use among older home health patients. J Clin Nurs 2013; 22(1–2):285–8. pmid:23216557
- 61. Chew ML, Mulsant BH, Pollock BG, Lehman ME, Greenspan A, Mahmoud RA et al. Anticholinergic activity of 107 medications commonly used by older adults. J Am Geriatr Soc 2008; 56(7):1333–41. pmid:18510583
- 62. Sittironnarit G, Ames D, Bush AI, Faux N, Flicker L, Foster J et al. Effects of anticholinergic drugs on cognitive function in older Australians: results from the AIBL study. Dement Geriatr Cogn Disord 2011; 31(3):173–8. pmid:21389718
- 63. Wilson NM, Hilmer SN, March LM, Cameron ID, Lord SR, Seibel MJ et al. Associations between drug burden index and falls in older people in residential aged care. J Am Geriatr Soc 2011; 59(5):875–80. pmid:21539525
- 64. Boustani M, Campbell N, Munger S, Maidment I, Fox C. Impact of anticholinergics on the aging brain: a review and practical application. Aging Health 2008; 4(3):311–20.
- 65. Ruxton K, Woodman RJ, Mangoni AA. Drugs with anticholinergic effects and cognitive impairment, falls and all-cause mortality in older adults: A systematic review and meta-analysis. Br J Clin Pharmacol 2015; 80(2):209–20. pmid:25735839
- 66. Tromp A, Pluijm S, Smit J, Deeg D, Bouter L, Lips P. Fall-risk screening test. J Clin Epidemiol 2001; 54(8):837–44.
- 67. Ambrose AF, Paul G, Hausdorff JM. Risk factors for falls among older adults: A review of the literature. Maturitas 2013; 75(1):51–61. pmid:23523272
- 68. Schlick C, Schniepp R, Loidl V, Wuehr M, Hesselbarth K, Jahn K. Falls and fear of falling in vertigo and balance disorders: A controlled cross-sectional study. J Vestib Res 2016; 25(5–6):241–51. pmid:26890425
- 69. Kalula SZ, Ferreira M, Swingler GH, Badri M. Risk factors for falls in older adults in a South African Urban Community. BMC Geriatr 2016; 16:51. pmid:26912129
- 70. Deandrea S, Lucenteforte E, Bravi F, Foschi R, La Vecchia C, Negri E. Risk factors for falls in community-dwelling older people: A systematic review and meta-analysis. Epidemiology 2010; 21(5):658–68. pmid:20585256
- 71. Valladales-Restrepo LF, Duran-Lengua M, Castro-Osorio EE, Machado-Alba JE. Consistency between anticholinergic burden scales in the elderly with fractures. PLoS One 2020; 15(2):e0228532. pmid:32092055
- 72. Vries M de, Seppala LJ, Daams JG, van de Glind EMM, Masud T, van der Velde N. Fall-Risk-Increasing Drugs: A Systematic Review and Meta-Analysis: I. Cardiovascular Drugs. J Am Med Dir Assoc 2018; 19(4):371.e1-371.e9. pmid:29396189
- 73. Seppala LJ, Wermelink AMAT, Vries M de, Ploegmakers KJ, van de Glind EMM, Daams JGet al. Fall-Risk-Increasing Drugs: A Systematic Review and Meta-Analysis: II. Psychotropics. J Am Med Dir Assoc 2018; 19(4):371.e11-371.e17. pmid:29402652
- 74. Seppala LJ, van de Glind EMM, Daams JG, Ploegmakers KJ, Vries M de, Wermelink AMATet al. Fall-Risk-Increasing Drugs: A Systematic Review and Meta-analysis: III. Others. J Am Med Dir Assoc 2018; 19(4):372.e1-372.e8.
- 75. Mehdizadeh D, Hale M, Todd O, Zaman H, Marques I, Petty D et al. Associations Between Anticholinergic Medication Exposure and Adverse Health Outcomes in Older People with Frailty: A Systematic Review and Meta-analysis. Drugs Real World Outcomes 2021; 8(4):431–58. pmid:34164795
- 76. Dinh TS, Brueckle M-S, González-González AI, Fessler J, Marschall U, Schubert-Zsilavesz M et al. Evidence-Based Decision Support for a Structured Care Program on Polypharmacy in Multimorbidity: A Guideline Upgrade Based on a Realist Synthesis. J Pers Med 2022; 12(1). pmid:35055383
- 77. Barnett K, Mercer SW, Norbury M, Watt G, Wyke S, Guthrie B. Epidemiology of multimorbidity and implications for health care, research, and medical education: a cross-sectional study [The Lancet, 380(9836), 37–43]. The Lancet 2012; 380(9836):37–43. pmid:22579043
- 78. Mazzitelli M, Milinkovic A, Pereira B, Palmer J, Tong T, Asboe D et al. Polypharmacy and evaluation of anticholinergic risk in a cohort of elderly people living with HIV. AIDS 2019; 33(15):2439–41. pmid:31764110
- 79. Squires P, Pahor M, Manini TM, Vouri S, Brown JD. Impact of Anticholinergic Medication Burden on Mobility and Falls in the Lifestyle Interventions for Elders (LIFE) Study. J Clin Med 2020; 9(9). pmid:32947839
- 80. Richardson K, Bennett K, Maidment ID, Fox C, Smithard D, Kenny RA. Use of medications with anticholinergic activity and self-reported injurious falls in older community-dwelling adults. J Am Geriatr Soc 2015; 63(8):1561–9. pmid:26200894
- 81. Gillespie LD, Robertson MC, Gillespie WJ, Sherrington C, Gates S, Clemson LM et al. Interventions for preventing falls in older people living in the community. Cochrane Database Syst Rev 2012; (9):CD007146. pmid:22972103
- 82. Collins GS, Ogundimu EO, Altman DG. Sample size considerations for the external validation of a multivariable prognostic model: a resampling study. Stat Med 2016; 35(2):214–26. pmid:26553135