In this population-based sample of incident AD dementia, we observed that longer duration of exposure to all major classes of psychotropic medications (quantified by PI) was associated with a more rapid rate of decline in general cognition (MMSE), and several classes of medicines were associated with more rapid increase in dementia severity (CDR-Sum) and NPS (NPI). These results strengthen prior reported associations of poorer outcomes with exposure to antipsychotics and benzodiazepines and extend these observations to antidepressants, including SSRIs, and to both typical and atypical antipsychotics. The association of greater benzodiazepine PI with more rapid cognitive and functional decline adds to a substantial literature implicating benzodiazepines with adverse outcomes in AD and indeed in older persons in general. depicts a consistent result of our modeling that the rate of decline itself declines over time, that is, the rates of change in outcomes tend to decelerate over time.
All classes of psychotropic medication were widely used over the study duration (ranging from 15 to 48%) with close to half of the participants using antidepressants and one-quarter using antipsychotics and benzodiazepines over the course of the study. The duration of exposure differed by medication classes, with antipsychotics and benzodiazepines generally being taken less than half the time and antidepressants (mostly SSRIs) more than half the time. The data at enrollment provide strong evidence that psychotropic medications were being prescribed to participants who had greater symptom severity (Lopez et al., 2010
; Tschanz et al., 2011
). Greater baseline functional impairment (CDR-Sum) was associated with more use of all medication classes except typical antipsychotics. Antidepressants, including SSRIs, were used more by those with greater initial severity of NPS as evidenced by NPI-Total. Female gender and younger age also were associated with greater PI for all classes except atypical antipsychotics and benzodiazepines. Polypharmacy of psychotropic medications was common. The strength of the associations may diminish over long periods of time (10 years) as illustrated in , but it is important to note that these are models whose validity over long periods of time should be interpreted with caution.
These results support two plausible explanations: Psychotropic medications are prescribed to AD participants who were already at risk of poorer outcomes, and their use is associated with more rapid decline. These data cannot, however, determine whether the more rapid decline is causally related to the medications or whether an underlying feature of the pathobiology of a specific aspect of the disease causes more severe dementia, more NPS, and worse medical health and is, therefore, the cause of the more rapid decline. Psychotropic medication exposure was also associated with younger age, which has been reported to be a risk factor for more rapid cognitive and functional decline in AD (Lopez et al., 2010
These results do not support a long term benefit from the use of any class of psychotropic drug, but it is possible that decline would have been more rapid had these drugs not been used. A recent randomized controlled trial of antipsychotics for psychosis in AD lessens concern that participants’ use of the treatments was a reflection of their need for such medicines (e.g., because of greater severity of illness at enrollment), but it did not demonstrate long-term benefit either (Sultzer et al., 2008
). Taken together, these results support current practice of attempting to use non-pharmacologic treatments, first unless the risk of harm or distress is high, and of recommendations that trials off drug are indicated in most cases.
Strengths of this study include its examination of persons with incident AD who were ascertained from a population-based sample. Most prior reports have relied on clinical or convenience samples. Second, the diagnoses of AD were established by an interdisciplinary geropsychiatric team that had the benefit of multiple forms of information, including longitudinal observations in most instances and neuropathologic confirmation in a sample. Third, the mean duration of follow-up averaged 3.7 years, longer than any clinical trial and most observational studies. Fourth, ascertainment of medication use was by direct observation on home visits rather than self-report or pharmacy records, a method that is likely to maximize reliability of medicine use data.
The primary limitation of the study is its observational design that cannot exclude confounding by indication, that is, that the observed association is a reflection of some other shared attribute such as symptom or illness severity. Conversely, it is also possible that participants with improvement in NPI may have discontinued these medications, biasing the NPI outcomes toward a negative association with medication use. Further limitations include the following: (i) modifications to the wording of specific NPI domains may have served to bias NPI scores upward; (ii) the NPI total is the sum of 10 domains and may not have sufficient range or sensitivity to detect change in one domain from a single drug class; (iii) we lacked information about the duration of psychotropic medication use prior to baseline, which may have influenced the outcomes; (iv) the PI is a calculated variable that summarizes duration of medication exposure, but in summarizing does not incorporate the details of when medications were stopped and started. There are several types of associations that cannot be captured with such a summary variable, such as delayed or transient effects of medication exposure. Additionally, “confound by indication” can include the possibility that medication discontinuation was associated with improvement in symptoms (particularly NPI), which would tend to bias the associations in the directions we observed; and (v) although the associations present a relatively consistent pattern, certain associations (particularly associations of medication use with NPI trajectory) would not be significant after adjustment for multiple comparisons. Finally, the study would inevitably have benefited from a larger sample size, increased duration of follow-up, and use of a more elaborate summary measure of cognition than the MMSE.
We suggest that the most likely explanation of our findings is that psychotropic medications are preferentially prescribed to AD patients whose clinical characteristics are associated with poorer prognosis including neuropsychiatric symptoms and problem behaviors. We base this suggestion on the observation that virtually all classes of psychotropic medications were associated with poorer outcomes, suggesting that this is an effect of prescribing practices rather than specific effects of individual medications. However, given the prevalence of polypharmacy, this study has limited power to distinguish effects of individual medications. Given these caveats, we cannot rule out the possibility that psychotropic medications have a deleterious effect on the clinical course of AD. To some extent this sort of question can be resolved by larger observational data sets with more elaborate data collection routines and longer follow-up intervals. More likely, however, the ultimate answer to these questions is better left to randomized trials.