Drugs and Falls: Rethinking the Approach to Medication Risk in Older Adults

In this issue of the Journal of the American Geriatrics Society, Ensrud et al.1 report on the risk of falling in more than 8,000 older women who took one or more central nervous system–active medications. The participants, women originally enrolled in the Study of Osteoporotic Fractures, were followed for an average of 1 year; more than one-quarter fell at least once, with frequent (two or more) falls occurring in 11% of the participants. Frequent falls occurred more commonly in those taking benzodiazepines (short- and long-acting forms), antidepressants, and anticonvulsants. Selective serotonin reuptake inhibitor (SSRI) antidepressant use was associated with a higher risk of falling than tricyclic antidepressant use, replicating findings from previous observational studies in long-term care patients.2, 3 The use of narcotics was not associated with a statistically significant increase in falls risk, as was also noted in a recent meta-analysis.4 The results of this study confirm a broad array of previous work that has linked central nervous system agents to the occurrence of falls in older adults. Although the study population consisted of exclusively white community-dwelling women from four cities, results are similar to those from other populations.2-5 The current study provides further compelling evidence for clinicians and public health professionals of the significant role that medications play in the occurrence of falls in older adults. Investigators hoping to elucidate the relationship between medication use and falls outcomes face several methodological challenges. Gathering complete and accurate medication histories in large cohorts of older persons remains a difficult task because of the panoply of prescription and over-the-counter medications currently marketed. Even the ascertainment of drug consumption is difficult, with substantial incongruence reported between true medication regimens and physician-elicited medication lists, for example.6 Questions also commonly arise concerning the frequency of medication changes in an older population over time and the establishment of temporal precedence between drug and outcome. The advent of computerized medical record systems should ameliorate some of these problems, although the lack of an integrated system connecting disparate provider, pharmacy, and health outcomes data will continue to thwart methodologically rigorous pharmacoepidemiological studies in many settings. The present study attempts to address several frequent criticisms of previous studies linking specific medications and falls. Confounding by indication, as the authors discuss, remains a particularly difficult challenge. Confounding by indication occurs when an older person at risk for an outcome (falls, in this case) takes a medication for a specific illness that itself can increase the risk of falls. Lack of adjustment for these clinical indications can lead to a falsely elevated association between risk of falling and use of a specific medication. Conversely, overestimation of a putative relationship between falls and a medication may occur when high-risk patients preferentially receive one drug over another because of the perceived superior side-effect profile of the chosen drug. Thus, without adequate adjustment, the seemingly safer medication may paradoxically appear to be associated with an increased risk of falls. SSRI antidepressants are a possible example of this type of confounding. In the case of confounding by indication or contraindication, careful consideration of baseline comorbidities and medical indications for the drugs is necessary but often is not possible in large population-based epidemiological studies in which in-depth clinical data may not be available. To their credit, Ensrud et al. adjusted for selected medical conditions (e.g., depression) that could be independent risk factors or confounders in the relationship between medication use and falls. Nevertheless, in the evaluation of benzodiazepines, adjustment for sleep difficulty and anxiety, two common indications, would also have been important to include in the multivariate modeling. Similarly, prophylactic use of anticonvulsant drugs for brain tumor, subarachnoid hemorrhage, or seizure history is common; all of these indications are risk factors for falling, albeit with differing potency of risk. Although the investigators laudably sought to address this serious methodological limitation of previously published studies, the importance of confounding by indication requires continued vigilance to ensure that the most rigorous analyses are undertaken to answer pertinent questions related to medication use and adverse outcomes such as falls. Another important methodological dilemma faced by investigators is the problem of medication classification. Most classification schemes, unfortunately, are based on various combinations of mechanistic, biochemical, and therapeutic indications criteria. The current study focuses on four main psychoactive medications: benzodiazepines (biochemical criterion), antidepressants (therapeutic indication criterion), anticonvulsants (therapeutic indication criterion), and narcotics (biochemical criterion). Aggregating diverse medications such as anticonvulsants, for example, into one class raises the question of the appropriateness of combining biochemically distinct subclasses of drugs: hydantoins (phenytoin), succinimides (ethosuximide), and other anticonvulsants (carbamazepine). Analyzing by subclasses of anticonvulsants, based upon the drugs' biochemical structures, may yield different relative risks for falls, if indeed the risk of falling is associated with the drug itself and not the indication for the drug. Although statistical considerations of power often influence such larger classification decisions, the ability to detect differences among subgroups is needed to answer clinically relevant questions about risk of medication use. The risk of medication use is important to investigate because of the frequency, and morbid consequences, of adverse drug events in older patients.7-10 Several of the most frequent manifestations of adverse drug effects—postural hypotension, cognitive changes, and dizziness—are risk factors themselves for falling. Falling therefore may represent a common final pathway of, and be a marker for, cumulative adverse drug effects. Nevertheless, most research has centered on identifying specific individual medications or categories of medications that are associated with an increase in risk of falling. But is this the issue faced in clinical decision making? How does the harried primary care physician, faced with an ever-growing list of drugs labeled problematic in older persons,11, 12 incorporate this knowledge into daily practice? How should public health professionals use the data to devise population-based interventions to improve prescribing? Certainly, determining the harms of individual medications is important, but the direct clinical relevance of these studies in the real-world practice of medicine is limited, given the multiple simultaneous medications individuals consume. Investigators have yet to articulate a clear and suitable framework to weigh decisions about medication use and management in older adults. Examining the totality of medications taken by a patient, rather than the effects of discrete drugs, may prove a more rationale approach to minimizing unwanted adverse events such as falls while maximizing health and functioning. Clinicians understand that a drug's action, although intended for a particular organ or organ system, is rarely limited to effects on the targeted endpoint. In fact, systemic side-effects are common and therefore important to consider in understanding how multidrug regimens affect outcomes such as dizziness, postural hypotension, and psychomotor performance—all components along the pathway to a final common adverse outcome such as falls. A model for examining total medication load, although implicitly acknowledging the beneficial role intended for medications, requires attention to the balance between treating discrete diseases and treating the whole patient. Investigators need to develop new ways of analyzing medication use, and clinicians need to consider medication effects from a broader perspective than merely the benefits and harms of individual medications for individual diseases.