Using the health care records of more than 1.4 million Ontario residents aged 67 y and older, we found that treatment with cholinesterase inhibitors was associated with a doubling in the risk of hospitalization for bradycardia. Importantly, although cholinesterase inhibitors are reversible precipitants of bradycardia, the drugs were resumed following discharge in greater than half the cases, presumably because the potential causative role of these drugs in the hospitalization was not appreciated. In many patients, cholinesterase inhibitors are associated with marginal improvement in cognition and global functioning 
. Consequently, recent guidelines suggest that cholinesterase inhibitors should not be standard of care for patients with dementia, but instead urge physicians to weigh each individual's expected risks and benefits before initiating therapy. Our large-scale population-based assessment of the risk of bradycardia with cholinesterase inhibitors in clinical practice should help inform the risk-benefit assessment for clinicians and patients. A recent study used an alternate cohort design to examine a primary outcome of syncope in patients receiving cholinesterase inhibitors. In a secondary analysis, these investigators also observed an elevated risk of bradycardia (adjusted hazard ratio 1.69; 95% CI 1.32–2.15) associated with cholinesterase inhibitor use, which complements the findings of our study 
The risk of bradycardia observed in our study may cause clinicians and patients to reconsider therapy with these drugs, particularly for patients in whom little or no cognitive improvement is observed early in therapy 
. At a minimum, our findings should alert clinicians to the potential role of cholinesterase inhibitors in patients with bradycardia, for whom resumption of treatment may be inadvisable.
Our study has several limitations that merit emphasis. Administrative databases contain information on hospitalizations, emergency department visits, diagnoses, procedures, physicians' claims, outpatient clinic visits, and drug dispensing data. However, they lack the clinical richness of a medical chart. Accordingly, we were not able to capture the severity of dementia, lifestyle habits such as smoking, diet, alcohol use, exercise, over-the-counter drug use, in-hospital drug use, laboratory values, and the results of diagnostic testing.
However, we relied on the unique crossover design feature that controls for fixed patient characteristics by allowing each patient to serve as his or her own control. Consequently, concerns regarding potential confounders, including disease severity, were largely overcome. The crossover design also relies on the temporal association between an exposure and outcome, and this feature enabled us to examine whether bradycardia-related hospitalization was likely secondary to cholinesterase inhibitors as opposed to underlying comorbidities. More specifically, the crossover design examines whether an exposure is more likely to occur immediately preceding an event or during another period when bradycardia did not occur. If the exposure is not associated with the outcome, then no temporal association would be expected. The presence of underlying comorbidities would not be expected to confound the cholinesterase inhibitor–bradycardia relationship unless the comorbidities were associated with both cholinesterase inhibitor use and bradycardia. To further control for confounding, we corrected for temporal changes in cholinesterase inhibitor use by matching each case with up to three control patients on the basis of an extensive disease risk index score.
We identified episodes of bradycardia resulting in emergency department visits or hospital admissions, but we were unable to identify individuals in whom bradycardia did not culminate in hospital care, including cases in which bradycardia led to death 
. Therefore, our analysis likely underestimates the true risk of cardiovascular harm associated with cholinesterase inhibitors. The coding for bradycardia has not been validated and it is possible that some cases of bradycardia were missed if the code for bradycardia has low sensitivity. However, we expect that the positive predictive value for the bradycardia code would be reasonable because the diagnosis for bradycardia is based upon a fairly straightforward medical assessment. The occurrence of random miscoding would only have attenuated our estimates and biased the results towards the null. It is likely that the cases of bradycardia captured in our study were clinically significant because they were severe enough to be documented in the patient's chart and coded in the CIHI database.
We were unable to assess the absolute risk of bradycardia due to cholinesterase inhibitor therapy, and we could not be certain that resumption of cholinesterase inhibitors following hospital discharge truly reflected a lack of appreciation for the potential negative chronotropic effects of therapy. Since most patients were taking donepezil, we were not able to contrast risks with individual cholinesterase inhibitors given the low prevalence of use with the other agents. Future studies are needed to address the relative harms of the individual drugs in this class.
Our post hoc examination of the 78 patients who resumed cholinesterase inhibitor after hospital discharge showed that only 4% were readmitted to hospital or visited the emergency department with a diagnosis of bradycardia in the 3 mo following resumption of therapy. It is hard to know how to interpret this post hoc analysis. We did not evaluate out-of-hospital death, and could not ascertain whether cholinesterase inhibitor therapy was restarted with a reduced dose or more gradual dose titration, whether changes to other medications were made, or whether closer outpatient follow up for bradycardia was undertaken.
Finally, because we relied on drug dispensing data as a proxy for drug adherence, there was a potential for exposure misclassification.
While bias and confounding can threaten any observational study, the likelihood of these was reduced substantially by the design of our study, in which every case served as his or her own control, thereby lessening between-patient variability, and by the use of a disease risk index, an advanced matching technique, which resulted in considerable similarity between cases and controls () 
In summary, we found that cholinesterase inhibitors are associated with a significantly increased risk of hospitalization for bradycardia among older outpatients, and that the risk is similar in patients with cardiovascular comorbidity and those receiving concurrent therapy with negative chronotropic drugs. Our findings highlight the importance of careful clinical evaluation prior to initiating cholinesterase inhibitor therapy, and vigilant monitoring thereafter in order to prevent adverse cardiac events. Cholinesterase inhibitors should be prescribed judiciously, and because these drugs carry a risk of serious adverse events, they should be continued only if the benefits outweigh the risks. Finally, it is important for clinicians to be aware of the potential negative chronotropic effects of cholinesterase inhibitors, and reassess the merits of continued therapy in patients who develop bradycardia while taking these drugs. The frequent resumption of cholinesterase inhibitors following discharge suggests that bradycardia may not be widely recognized as a potential adverse effect of this class of medications.