Dose response relationship of cumulative anticholinergic exposure with incident dementia: validation study of Korean anticholinergic burden scale

A longitudinal database was used for this case control study, the Korea National Health Insurance Service Senior Cohort database (2002–2013) that was developed and provided by the Korea National Health Insurance Service, a single and mandatory national insurance service. This database comprised 558,147 patients over 60 years, representing 10% of the Korean senior population. Its representativeness was supported by the results of a study comparing this database with population statistics based on resident registration, Statistics Korea’s data of December 2013 [13].

Older adults with no diagnosis of dementia during the prior years of 2002–2011 and alive as on January 1, 2013, were initially selected from the Korea National Health Insurance Service Senior Cohort database. Among them, patients who were diagnosed with dementia (either vascular or Alzheimer’s or both; ICD-10 codes: F00, F01, F02, F03, F051, G30, and G311) in 2012–2013 were selected. Index dates were defined as the earliest date of claims with the diagnostic code for dementia. Control cohorts were selected among the patients without dementia after propensity score matching. Details are described in statistical analysis.

Exposure to anticholinergic drugs was measured during the nine prior years starting 10 years before the index date and ending 1 year before the index date. Drug exposure in the recent one-year period was excluded to minimize the protopathic bias. Nine-year cumulative drug exposure was measured in all patients (Fig. 1). The anticholinergic burden for each patient was assessed using KABS and ACB (Additional file 1). Both scales assigned 1 for mild, 2 for moderate, and 3 for strong anticholinergics. First, the standard daily dose was calculated with the prescribed dose for each medication divided by the defined daily dose (DDD) of the World Health Organization. Then, we summed the standard daily dose for anticholinergic drugs after multiplying the assigned anticholinergic score during the period of drug exposure. This method was modified from that suggested by Gray et al. [8] with multiplying the anticholinergic score assigned by each anticholinergic burden scale. We finally calculated the average daily anticholinergic burden score during the drug exposure period. We categorized the anticholinergic burden according to daily average score as follows: < 0.25, minimal exposure; 0.25–1, low exposure; 1–2, intermediate exposure; and ≥ 2, high exposure. We further investigated the anticholinergic drugs most contributing to high exposure prior to the index date in patients with dementia according to KABS or ACB.

As confounders that might affect the onset of dementia, we selected baseline comorbid diseases such as hypertension, dyslipidemia, heart failure, atrial fibrillation, ischemic heart disease, diabetes mellitus, cerebrovascular disease, Parkinson’s disease, depression, anxiety, schizophrenia, bipolar disorder, insomnia, alcohol disease, obesity, substance abuse, and tobacco dependence and use.

These comorbid diseases were identified when there were two or more claims with the relevant diagnostic code (Additional file 2) before January 1, 2012. For covariates, the sedative load during the same period was calculated. We used the sedative load model for measuring the sedative load, which assigned 2 for the primary sedative and 1 for medications having sedation as a prominent side effect. We excluded the medications that were assigned 1 or a higher anticholinergic score by each scale for measuring the sedative load (Additional file 3).

To reduce the effect of confounding factors, control cohorts were selected using the propensity score matching method with the ratio of 1:2 using variables including age, sex, and baseline comorbid diseases that are known to cause of dementia including hypertension, dyslipidemia, heart failure, atrial fibrillation, ischemic heart disease, diabetes mellitus, cerebrovascular disease, Parkinson’s disease, depression, anxiety, insomnia, and alcohol disease. The matching was performed using a greedy algorithm with a caliper width equal to 0.2 standard deviations of the logit of the propensity score. The measured anticholinergic burden by both scales was described and compared between the case and control cohorts. We used Chi-square test for comparison of categorical variables for baseline characteristics. Multivariate logistic regression analysis was performed to evaluate the independent association between the anticholinergic burden and the incidence of dementia to adjust for the covariates including age, sex, the baseline co-morbid diseases, and sedative load. Adjusted odds ratios (aOR) with 95% confidence intervals (CI) are reported.

In the secondary analysis, we also investigated the impact of weak anticholinergic agents on each scale after excluding them. To minimize reverse causation bias, the association between incident dementia and the cumulative anticholinergic burden during the 5–10 years prior to the index date after excluding the most recent 4 years was analyzed as part of the sensitivity analysis. Data management and statistical analysis were performed using SAS version 9.3 (SAS Institute, Inc., Cary, NC, USA).

Among the 558,147 adults aged more than 60 as of 2002 included in the sample senior cohort, 104,087 patients and 137,680 patients were excluded because they were diagnosed with dementia before the year 2012 and because they had not survived until the year 2013, respectively. The total sampled population included 28,864 patients who were diagnosed with dementia between 2012 and 2013. After propensity score matching, 57,712 patients without dementia were identified as the control and in total 86,576 patients were included for the analysis (Fig. 2).

The median age at the time of dementia diagnosis was 79 (inter-quartile range (IQR), 75–83) and women accounted for 69.3%. As the comorbid condition, 79, 61, 49, 46, 45, 39, 35 and 30% of the patients had hypertension, dyslipidemia, diabetes, anxiety, cerebrovascular disease, insomnia, ischemic heart disease, and depression, respectively (Table 1).

During the 2–10 years before the index date, 46.2% of patients with dementia and 50.7% of patients without dementia were exposed minimally to anticholinergics (average daily KABS score <  0.25). Fewer patients were identified as having intermediate anticholinergic exposure using KABS compared to that using ACB in both cohorts with dementia (10.6 and 13.7%, p <  0.01, respectively) and without dementia (8.9 and 12.9%, p < 0.01). Proportions of patients with high exposure were 2.1% (for the KABS) and 2.8% (for the ACB) in the cohort without dementia and 3.2% (for the KABS) and 3.4% (for the ACB) in the cohort with dementia (Table 2).

The top 20 anticholinergic drugs that most contributed to high exposure prior to the index date in patients with dementia according to KABS and ACB are presented in Table 3. Cimetidine, dimenhydrinate, chlorpheniramine, furosemide, and diazepam contributed the most to the total KABS score, while dimenhydrinate, hydrochlorothiazide, isosorbide, chlorpheniramine, and nifedipine contributed most to the total ACB score in patients with dementia with high exposure. Drugs with a KABS score of 3, 2, and 1 contributed 47.4, 19.3, and 33.3%, respectively, to the cumulative amount of the KABS score, whereas those with an ACB score of 3, 2, and 1 contributed 41.4, 1.8, and 56.7%, respectively, to the cumulative amount of the ACB score in dementia patients with high exposure.

There were significant associations between incident dementia and nine-year cumulative dose adjusted anticholinergic burden of low, intermediate, and high exposure compared with that of minimal exposure assessed by KABS with corresponding aOR of 1.21 (95% CI, 1.17–1.25), 1.39 (95% CI, 1.31–1.46), and 1.71 (95% CI, 1.55–1.87), respectively. A similar association was observed when the anticholinergic burden was measured using ACB but corresponding aORs were lower than those measured using KABS. A high cumulative anticholinergic burden measured using the ACB was significantly associated with dementia incidence with an aOR of 1.22 (95% CI, 1.12–1.33) (Table 4).

When the anticholinergic burden was measured after excluding weak anticholinergics (score = 1), the proportion of patients with low to high anticholinergic exposure was significantly reduced, and this reduction was much greater using the ACB scale versus that using the KABS. The effect size of the association with incident dementia was similar when exposure was measured using KABS without regard to the inclusion of weak anticholinergics. The KABS scale with weak anticholinergics showed slightly greater association than that without weak anticholinergics regarding high exposure (aOR 1.71 [1.55–1.87] vs. 1.60 [1.38–1.86]). However, the odds ratios for low, intermediate, and high vs. minimal cumulative exposure increased from 1.06–1.22 to 1.24–1.41 when the anticholinergic burden was measured using ACB excluding weak anticholinergics. No substantial changes were found in the association of anticholinergic exposure with the incidence of dementia after measuring drug exposure 5–10 years prior to the index date. This led to small reductions in the associations.