Our study suggests that there may be no acute protection afforded with respect to falls risk by selecting a newer versus older class of antidepressants or antipsychotics or a short acting versus long acting benzodiazepine. Previous studies have determined that the long term risk of falls and injurious falls was no different among users of tricyclic versus SSRI antidepressants
, or among users of typical versus atypical antipsychotics
. Similarly, use of both short and long acting benzodiazepines has been associated with an equally elevated longitudinal risk of falls in community dwellers
[17, 18]. Thus, there is no evidence to support selection of psychotropic agents based on risk of falls.
Our finding that there was a high risk of falls in the days immediately following a psychotropic drug change is consistent with other studies. Using a self-controlled, case-crossover approach, Sorock and Neutel both found that the fall risk in nursing home residents was maximally increased within 2–3 days of a change in a medication that affects the central nervous system
[7, 8]. Previous work by our group showed that changes in specifically non-SSRI antidepressant prescriptions increased the risk of falls within 2 days, with a subsequent decreased risk of falls over the next 5 days
. Zint et al. concluded that in community dwellers, hip fracture risk associated with benzodiazepines was highest in users within 14 days of initiation
Nonetheless, it is possible that the high rates of falls we observed are due to random variation in falls rates or because residents with psychotropic drug changes have more risk factors for falls, such as orthostatic hypotension, muscle weakness, or prior falls, predisposing them to fall. We did not have information on orthostatic hypotension or muscle weakness in order to adjust for these characteristics in our models. Although we had information on history of falls, this characteristic was not used in the final models due to statistical instability when we included both history of falls and an individual effect in the same model. When we included in our model a history of falls rather than a random patient effect, the associations between psychotropic drug changes and rate of falls did not change. We do not believe that any of these characteristics would fully account for the markedly elevated risk of falls we observed following a drug change. The rate of falls in the 7 days following an antidepressant, antipsychotic, or benzodiazepine drug change ranged from 13.7 to 18.4/1,000 person days. This is much greater than the annual rate of falls among residents that used psychotropic medications without change (4.1/1,000 person days) or the annual rate of falls among residents with a psychotropic drug change (7.1/1,000 person days).
While we observed that the rate of falls was highest one day following a psychotropic drug change, rates of falls were also elevated 3 to 4 days preceding the drug change. This suggests that the underlying medical condition for which the medication was prescribed also contributed to the increased risk of falls. In the nursing home, psychotropic medication changes are often made in an effort to manage psychiatric symptoms in persons with dementia, such as wandering, combativeness, and insomnia. Behavioral symptoms may begin days before pharmacologic treatment is selected resulting in an increased risk of falls before drug initiation.
Separating the effects of a drug from the effects of the underlying medical condition can be challenging. In our study, exposure to a psychotropic drug change was highly correlated with dementia with psychiatric symptoms. Attempts to adjust for psychiatric symptoms using a traditional model or propensity scores would have been of limited utility given the very high correlation between drug exposure and the psychiatric conditions
. Instead, we used a comparator medication strategy to determine if falls risk differed between classes of medications that are used to treat similar medical conditions. An alternative approach would have been a stratified or restrictive analysis, grouping residents with similar acute symptoms together
. One could test the hypothesis that it is the medical illness rather than the drug causing the observed risk by repeating the analysis with an alternative outcome that is known to have no association with the drug of interest
. We did not have an adequate sample size or precise information on other acute illnesses in order to use these strategies. Our study suggests that the rate of falls may be elevated in the days surrounding a psychotropic drug change. While it may be impractical to distinguish between the effects of the drug and the underlying medical condition, the implications of our findings remain: attention should be given to residents experiencing a psychotropic drug change in an effort to prevent falls.
Strengths of this study include a large, prospective cohort of long term care residents with complete information on medications and falls. Additionally our study addresses an important clinical question of whether certain psychotropic agents are less safe with respect to the acute risk of falls.
This study also has several potential limitations. First, we had small numbers of falls on individual days with respect to the drug change. While this may result in some variability of fall rates between days, overall trends were observed. Second, we were unable to consider the sizes of all new prescriptions and increased doses separately, and we did not consider concomitant medication changes given the limited number of falls in our study. For example, in our one-year study, 376 antidepressant changes occurred with only 36 falls within 7 days of the drug changes. Only a small fraction of antidepressant changes were associated with a concomitant benzodiazepine or antipsychotic drug change. Hanlon et al. found that risk of falling increases with the number and dosage of psychotropic medications
, and it is possible that acute fall rates were underestimated in our study among residents who received a higher dose or multiple psychotropic medications. We recommend that future studies with a longer period of follow-up consider the effects of dose and concomitant drug changes on the risk of falls.
Third, information was not available on decreases or discontinuations in dosage so that we were not able to examine the effect of psychotropic drug reductions on the acute risk of falls. Withdrawal syndromes are possible with discontinuation of psychotropic medications, and could potentially result in an acute increased risk of falls. It is unlikely that decreases are a common cause of falls as previous studies suggest withdrawal of psychotropic medications reduces the risk of falls
[24, 25]. Nonetheless, long term benefits of withdrawing psychotropic agents are not mutually exclusive with the possibility of an acute risk of falling in the days after a decrease or discontinuation. Fourth, we did not have resident specific information on days spent outside the facility. We expect this number would be low, and accounting for days outside the facility would likely increase the rates of falls only slightly.
Finally, our study was conducted in two facilities operated by a single organization. In comparison with other facilities in the U.S., the proportion of residents prescribed an antidepressant or benzodiazepine was greater than the national average, while the proportion of residents prescribed an antipsychotic was similar to the national average
[26, 27]. In other countries, the prevalence of psychotropic drug use may be lower. The absolute effect of psychotropic drug changes on falls may be more or less significant depending on the frequency of use.