For this analysis, we used data from the MultiCare Cohort Study,19 a prospective observational cohort study in primary care registered as ISRCTN89818205. The MultiCare Cohort Study was approved by the ethics committee of the medical association of Hamburg on 14 February 2008 and amended on 17 November 2008 (approval no. 2881). All participating patients gave written informed consent. This manuscript was drafted following the STROBE guidelines.20

In the MultiCare Cohort Study, patients from 158 general practices located in Germany were recruited in the major cities Bonn, Dusseldorf, Frankfurt/Main, Hamburg, Jena, Leipzig, Mannheim and Munich. From each practice, up to 50 eligible patients were randomly selected and invited by mail to participate in the study. Patients interested in participating consulted their GPs. The GPs informed them about the study and obtained informed consent. The project staff then contacted participating patients and conducted personal interviews at the patients’ home. Additionally, GPs were interviewed in their practices. After baseline assessments, three follow-ups were conducted, each after 15 months. Data collection took place between 21 July 2008 and 4 December 2013.

In each practice, electronic medical records were screened for patients with multimorbidity who were 65 to 85 years old and visited their GPs at least once in the last completed quarter (3 months accounting period). Multimorbidity was operationalised by three diseases out of a list of 29 chronic conditions. Patients were excluded if they were treated on behalf of other GPs, if they were unable to participate in interviews (eg, blind or deaf), if they had insufficient German language skills or were unable to read or write. Moreover, patients were excluded if they lived in a nursing home, if they had a severe disease probably leading to death within 3 months, if they had no capacity to consent (eg, dementia) and if they participated in other medical studies at the time of recruitment.

Independent variables were age, sex, educational level, number of diseases weighted by severity, depressiveness, cognitive function and health-related quality of life.

Personal interviews with patients and their GPs were used to assess socio-demographic data as well as physical and psychological conditions. Sociodemographic data included age, sex and educational level documented pursuant to Comparative Analysis of Social Mobility in Industrial Nations, a classification adapted to the German education system, that categorises three education levels: ’primary or below’, uncompleted, general elementary or basic vocational education; ‘secondary’, secondary school certificate or ‘A’ level equivalent; and ‘tertiary’ higher or lower tertiary education.21

Information about morbidity was provided by GPs. The number of diseases was calculated using a standardised instrument including 46 chronic conditions selected due to high prevalence and chronicity.22 This instrument also included a severity rating based on five levels (from 0=trivial to 4=very severe), which were used as weights for calculating the disease score. All other information was obtained from patient interviews.

The Geriatric Depression Scale (GDS-15) was used to measure depressive symptoms.23 The scale consists of 15 items that can be answered either with ‘no’ or ‘yes’, from which a sum score is calculated. More than five points are considered as depressed. Cognitive function was tested via the Letter Digit Substitution Test (LDST).24 In this test, patients replace letters by numbers in a certain time and show their individual speed-dependent cognitive skills.

The health-related quality of life questionnaire (EQ5D-3L,25 value set UK)26 consists of questions in the five dimensions mobility, self-care, usual activities, pain and anxiety/depression. The value one represents the best health conditions. From these values, specific subtrahends can be subtracted based on the severity of limitations in the respective dimension.

Missing values in independent variables were imputed by hot deck imputation based on the Gower distance measure. This procedure was described in another publication.22 In our analyses, the proportion of missing values was 0.2% to 0.3% for EQ5D-5L, value set UK, 0.4% to 0.8% for GDS, 0% to 6.7% for the number of diseases weighted by severity and 8.1% to 10.9% for LDST. There was no other item non-response in the analysed data.

To investigate the change of ACB over time, the German anticholinergic burden score (ACB) was calculated and used as the dependent variable.27

The ACB developed by Kiesel et al evaluated 504 drugs that are available in Germany.27 Drugs were scored from 0 to 3 as having no, weak, moderate or strong anticholinergic effects. 356 drugs were categorised with no anticholinergic effects (ACB=0), 104 drugs were scored as weak (ACB=1), 18 as moderate (ACB=2) and 29 with strong anticholinergic effects (ACB=3).

In the MultiCare Cohort Study, medication was assessed in brown bag reviews conducted at the patients' homes. We calculated and analysed the ACB by including all medications with anticholinergic effects (including over-the-counter medications) prescribed to participants during the study period.

To examine the changes in the ACB over time in relation to participants' health characteristics, we calculated the ACB for each participant individually. We excluded inhalation drugs (tiotropium and ipratropium) because of their low systemic exposure resulting in lower levels of side effects. Also, we excluded prednisolone due to the different ways of application (dermal, ocular and oral) and unknown duration of application. Theophylline and triamterene are not in common use anymore and were therefore not analysed.

To describe the potential changes in the number of prescriptions over time, we selected five drugs from the baseline top ten anticholinergic drugs. These drugs were metoprolol (ACB=1), metformin (ACB=1), furosemide (ACB=1), tramadol (ACB=2) and digitoxin (ACB=1).

Descriptive data were reported as percentages and means with SD. The association between independent variables and anticholinergic drug burden was analysed at baseline and longitudinally. In the longitudinal analyses, independent variables comprised the first three waves of data collection, and dependent variables comprised the respective subsequent follow-ups. Thus, change in ACB scores was modelled as the difference between two subsequent waves of data collection. For example, we analysed the association of independent variables at baseline (including baseline ACB scores) with ACB scores at follow-up 1.

We conducted multilevel mixed-effects multivariable linear regression analyses adjusted for random effects on study centre, GP practice within study centre, and—for longitudinal analyses—participant within GP practice within study centre levels. Longitudinal analyses were baseline-adjusted and controlled for the respective follow-up. In order to represent realistic values of independent variables, we standardised continuous independent variables based on rounded SD, for example, three points difference for depressiveness.

We conducted a sensitivity analysis including only patients who participated throughout all four waves of data collection. In addition, we also added a second sensitivity analysis for which the ACB score was reduced to five levels. In this analysis, the levels 0 through 3 represent the respective ACB scores, and level 4 represents ACB scores of four through 13 (maximum). In order to make results comparable, the sensitivity analyses used the same statistical methods and the same statistical model as the main analysis. An alpha level of 5% (p<0.05) was defined as statistically significant. The statistical analyses were performed using Stata 15.1.

A total of 24 862 patients were screened for inclusion and exclusion criteria (cf. figure 1). The resulting 7172 eligible patients were invited for participation in the study, and 3317 patients (46.2%) gave informed consent. Retrospectively, 128 patients had to be excluded, because they died or exclusion criteria were fulfilled without the knowledge of the respective GP. At each additional assessment, some participants from the previous follow-up refused to continue participation in the study, and some could not be contacted anymore. The final sample size was 3189 at baseline, 2746 at follow-up-1, 2375 at follow-up-2 and 2068 at follow-up 3. Due to missing previous observations, 44 observations at follow up-2 and 77 observations at follow up-3 had to be excluded. Therefore, at baseline, 3189 observations, and between follow-up 1 and follow-up 3, 7068 observations could be analysed.

An overview of baseline characteristics of the study population is given in table 1. On average, patients were 74.4 years old, 59.2% were female, 26.8% had secondary education and 10.9% tertiary. Pursuant to ACB, 36.1% of the patients had no drugs with anticholinergic effects.

The mean ACB score was 1.5 at baseline and showed little change over subsequent follow-ups (cf. table 2). In the cross-sectional analyses at baseline, lower age (coefficient: −0.07, 95% CI: -0.12/–0.01, p=0.026), female sex (0.18, 0.05/0.30, p=0.004), higher severity-weighted number of diseases (0.23, 0.17/0.29, p<0.001), higher number of depressive symptoms (0.22, 0.14/0.29, p<0.001), worse cognitive function (−0.10 to –0.16/−0.04, p=0.001) and worse health-related quality of life (−0.27 to –0.34/−0.21, p<0.001) were associated with higher scores in ACB (cf. table 3).

Of all anticholinergic medications (including over-the-counter drugs), the most commonly prescribed anticholinergics at baseline were metoprolol (20.7%), metformin (13.7%), furosemide (5.8%), tramadol (3.3%) and digitoxin (2.9%) (cf. table 2).

In the longitudinal analyses, higher severity-weighted number of diseases (coefficient: 0.08, 95% CI: 0.05/0.10, p<0.001), higher number of depressive symptoms (0.04, 0.004/0.08, p=0.030), worse cognitive function (−0.03 to –0.06/−0.001, p=0.044) and lower health-related quality of life (−0.05 to –0.08/−0.01, p=0.006) were associated with increasing ACB score over time. Additionally, the ACB score was associated with the score at the previous follow-up (1.30, 1.26/1.34, p<0.001) and the observation time (0.29, 0.22/0.36, p<0.001 for follow up-2 and 0.11, 0.04/0.18, p=0.003 for follow up-3 compared with follow up-1) (cf. table 4).

The sensitivity analyses can be found in online supplemental tables S1 and S2 in the supplement. Results were similar when the ACB score was reduced to five levels. The number of diseases (0.07, 0.04/0.09, p<0.001), depressiveness (0.03, 0.001/0.07, p=0.041), cognitive function (−0.03 to –0.06/−0.007, p=0.012) and health-related quality of life (−0.03 to –0.06/−0.01, p=0.018) were still associated with change in ACB score. In contrast, in the analysis including only patients who participated in all four waves of data collection (n=1991), the number of diseases was still associated with change in ACB score (0.09, 0.06/0.12, p<0.001), while depressiveness (p=0139), cognitive function (p=0.060) and health-related quality of life (p=0.153) lost their statistical significance.

Lower age, female sex, higher severity-weighted number of diseases, higher number of depressive symptoms, worse cognitive function and worse health-related quality of life were associated with higher German ACB at baseline. Over a 5 year period, no significant increase in ACB was found in the overall sample. This might be an effect of selective survival as the average health condition of the population deteriorated during this period, and higher severity-weighted number of diseases, higher number of depressive symptoms, worse cognitive function and worse health-related quality of life were associated with an increase in ACB over time.

Our study participants were 65 to 85 years old, and 36.1% of them had no medication with anticholinergic effects according to the German ACB score (ie, ACB=0). At baseline, ACB was significantly higher in women than in men. These results were consistent with the findings of the study by Reinold et al in which the ACB was described in a large and unselected population sample in Germany.16 Among people aged 60 to 89, no ACB (ACB=0) was observed in 52.5 to 31.7% of men and 48.9 to 27.4% of women,16 with the prevalence of ACB=0 decreasing with age. The prevalence of clinically relevant ACB (ACB≥3) increased with age and was higher in women compared with men.16 However, it should be noted that in the sample of our study, the ACB score was significantly higher in younger participants.

The study by Grossi et al estimated the prevalence of anticholinergic drug use in England in 1991 and 2011 and found that the use of potent anticholinergic drugs in older adults almost doubled.1 We found only one study that examined the longitudinal development of anticholinergic scores over time, also in England and conducted by Mur et al.28 From 1990 to 2015, the mean yearly ACB increased fivefold according to the ACB.28 It was suggested that the increase in ACB was due to an increase in general multimedication, that is, the prescription of more anticholinergic drugs, rather than the prescription of stronger anticholinergics.28

Of all anticholinergics (including over-the-counter medications) prescribed to participants during the study period, the most commonly prescribed were metoprolol, metformin, furosemide, tramadol and digitoxin.

Beta-blockers such as metoprolol are prescribed for the treatment of various cardiovascular diseases such as heart failure, coronary heart disease and hypertension29 that often occur in old age. With an ACB score of 1 (weak anticholinergic effect),27 metoprolol was the most commonly prescribed anticholinergic medication in our sample (20.7 to 21.0% of participants). The antidiabetic drug metformin was the second most frequently prescribed drug with anticholinergic effects (ACB=1).27 We suggest that its prescription increased over time (13.7 to 14.9%), probably because diabetes prevalence increases with age.30 The same applies to the prescribing behaviour of the diuretic drug furosemide (ACB=1).27 Previous studies showed that the prevalence of diuretics also increases with age, as they are prescribed to treat common conditions such as heart failure and kidney disease.31 However, the prescription of furosemide remained stable in our study during the follow-up period (5.8 to 5.8%). Cardiac glycosides such as digitoxin (ACB=1)27 and digoxin are prescribed to treat patients with atrial fibrillation and heart failure.32 33 In line with changes in clinical guidelines, the prescribing rate of cardiac glycosides in the USA and Europe has decreased significantly in recent years32 33 due to potential risks of overdose and toxicity.34 In contrast, we observed no change in digitoxin-prescribing behaviour in our study sample (from 2.9% to 2.9%). The prescription of tramadol, which is used for the treatment of chronic pain, for example,35 fell slightly from 3.3% to 2.8%. According to the ACB score, tramadol has moderate anticholinergic effects (ACB=2).27

However, we found no significant increase in anticholinergic drug burden in the study sample over a 5 year period, as the ACB score remained stable. Compared with the study by Mur et al, our observation period may have been too short (5 vs 25 years) to detect significant changes in ACB scores.

Another aspect could be the effect of selective survival, as the average health status of the population deteriorated during the study period and a higher ACB is known to be associated with a higher mortality rate.13

The systematic review by Fox et al included 46 studies in which the effects of drugs with anticholinergic properties on relevant health outcomes were investigated.15 It was found that increasing ACB was associated with a significant decline in cognitive abilities and physical function.15 Antidepressants, for example, are one of the medication classes that contribute particularly to the cumulative ACB,16 so older adults with depression are at higher risk of having a clinically relevant ACB (ACB ≥3).36 Consistent with this, we found that patients with a higher number of depressive symptoms had higher ACB scores at baseline and during follow-up.

Discontinuation or at least dose reduction of drugs with anticholinergic properties in older adults has often been recommended7 17 but is difficult to implement in multimorbidity, especially when several clinical practice guidelines must be followed. Nevertheless, ACB should be taken into account when the patient’s health progressively deteriorates. While some anticholinergics may be beneficial for multimorbid patients (eg, metoprolol or furosemide), the use of other anticholinergics should be reconsidered by physicians (eg, tramadol or Digitoxin).

In our study, however, we observed an increase in ACB with increasing severity of illness. Our findings suggest that physicians prescribe even more drugs with anticholinergic properties rather than reducing them when their patients’ cognitive and physical functions deteriorate. The problem seems to be not only the prescription of strong anticholinergics (ACB=3) but rather the prescription of more drugs with weak anticholinergic properties (ACB=1), which cumulate to a high ACB.

We, therefore, recommend new practice guidelines for GPs to raise their awareness of the cumulative ACB and to enable them to discontinue or reduce the dose of some anticholinergic drugs where possible. We observed that, for example, the prescription of metformin (ACB=1) in older adults increased over time, although higher HbA1c (haemoglobin A1c) levels can be tolerated in old age. Even symptom-controlling therapy without HbA1c-dependent decisions is recommended if life expectancy is less than ten years.37 Another example indicating a lack of awareness is the unchanged prevalence of digitoxin in our study over time, despite the known adverse effects in older adults and the recommendations of clinical guidelines in recent years,32–34 as mentioned above. In older adults, at least the dose of digitoxin has to be reduced due to the lower skeletal muscle mass33 to avoid potential overdose and toxicity.

The strengths and limitations of the MultiCare Cohort Study have been described in detail elsewhere.5 22 In short, our exclusion criteria may have affected the generalisability of our results. For example, patients with dementia were excluded at baseline because they were unable to give informed consent. Since dementia and the use of anticholinergic drugs are linked and older adults with clinically significant cognitive impairment are often prescribed these substances,1 38 ACB scores may have been underestimated in our analyses. The disproportionate loss of people with dementia in later waves may also have contributed to the fact that no significant increase in ACB was found in the overall sample during the observation period. The same applies to patients living in nursing homes, as they are at a higher risk of receiving anticholinergic drugs39 but were also excluded. Furthermore, only patients from larger German cities were recruited, meaning that the current study does not cover rural areas. In addition, there were some restrictions on data collection. For example, cognitive function was only assessed with one instrument (LDST). Therefore, we were probably unable to capture the full spectrum of changes in cognitive performance.

Strengths of our study include regular training and monitoring of interviewers and data quality assurance, for example, detection of incomplete and inaccurately entered data, automatic plausibility and integrity checks and reporting of data errors. Our study is based on 5 years of observation with three follow-up observations, and our analyses also include participants with missing observations. Therefore, the effect of selective survival found in a sensitivity analysis including only patients participating in all waves of data collection could be reduced in the main analysis. We found no relevant bias due to patients with very high ACB scores. In addition, there is the high participation rate of over 46%. Other strengths include multivariable analyses adjusted for potential confounders, multilevel models allowing for cluster effects and an advanced treatment of missing values.

Our findings show that anticholinergic prescribing increases despite the deteriorating health status of older adults, which may lead to higher hospitalisation and mortality rates. New practice recommendations for GPs may be helpful in raising their awareness of cumulative ACB and enabling them to discontinue or reduce the dose of some anticholinergics where possible. However, further research is needed to assess the impact of our findings on prescribing behaviour in primary care.

Not applicable.

This study involves human participants. The study protocol was approved by the ethics committee of the Hamburg Medical Association in February 2008 and modified in November 2008 (approval no. 2881). Eligible patients gave written informed consent. We confirm that our study complies with the Declaration of Helsinki. Participants gave informed consent to participate in the study before taking part.