©1997 by Eileen E. Ming. All rights reserved
Benzodiazepines, a widely used class of hypnotics-anxiolytics,
previously considered both safe and effective, have been associated with an
increased incidence of hip fracture among the elderly (Thapa et al 1995,
Ford et al 1990). This relationship illustrates how the interplay of
increased likelihood of drug use, increased sensitivity to medication effects,
and increased frailty enhances the risk of serious life-threatening injury. But
demonstrating this association, quantifying the risk and proving a cause-effect
relationship are difficult tasks (Campbell 1991). This report will discuss the
development of the evidence for an association between administration of
benzodiazepines and hip fracture in the elderly, an important modifiable cause
of injury.
Approximately 220 falls occur per 1000 person-years among those over age 65 (Monane and Avorn 1996), resulting in direct costs of $7.8 billion in 1985 (Ray 1992). In long term care settings, where 45% to 70% of residents fall each year (Thapa et al 1995), 1600 falls occur per 1000 person-years (Lauritzen et al 1993). One to 2% of falls result in hip fracture, and the risk of hip fracture increases almost 100-fold from age 60-64 to 80-84 (Cummings and Nevitt 1989). In the year following a fracture, there is a 23% mortality rate, compared to an expected 8%; 50% of the ambulatory lose the capacity to walk independently; one-third of the community-dwelling require long term nursing care; and many are incapacitated by the fear of falling again (Evans 1988, Ray et al 1990, Ray 1992).
Psychotropic drugs are used regularly by 45 to 55% of nursing home residents (Thapa et al 1995) and 10% of community-dwelling elderly (Sorock 1988). Sedative use, especially benzodiazepine (BZD) use, rises with age as the prevalence of insomnia rises with age (Ray et al 1990). Changes in pharmacokinetics and pharmacodynamics cause the effect of a drug to be more pronounced for older people for a given oral dose (Pomara et al 1985). BZDs have been found to impair basic psychomotor function and postural sway in normal volunteers, a side effect which lasts at least through 4 weeks of continuous use; impairment increases with dose, and with drug elimination half-life (usually dichotomized to long-acting, >24 hour half life, and short-acting, <24 hour half life) (Ray 1992). Sedatives slow reaction time and reduce coordination and alertness (Ray et al 1990): protective responses at the time of the fall may be too late to prevent a hip fracture (Cummings and Nevitt 1989). BZDs also may cause orthostatic hypotension, a fall in systolic blood pressure, and an increase in heart rate in the standing position in the elderly, thereby precipitating falls (Ford et al 1990).
As early as 1955, Droller et al reported an association between psychotropic drugs and falls among elderly people. Several population-based case-control studies in the early- and mid-1980's looked at the association between general categories of psychotropic medications and falls or hip fracture, with varying results. Prudham and Evans (1981), Wild et al (1981) and Tinetti et al (1986) found an increased risk, while Campbell et al (1981), Paganini-Hill et al (1981) and Rashiq and Logan (1986) did not.
Ray et al (1987) conducted the first study on psychotropic medications and hip fractures large enough to stratify by specific medications, including BZDs, by using Michigan's Medicaid computerized reimbursement records; 1021 incident hospitalizations for hip fracture and 5606 controls were identified among enrolled elderly in 1981-1982. Current use (prescription filled within 30 days before index date) of long-acting BZDs (LABZDs) significantly increased risk (flurazepam, OR=1.9, 95%CI=1.3-2.8; diazepam or chlordiazepoxide, OR=1.8, 95%CI=1.0-2.4), with 5.7% of the cases currently exposed. There was a significant dose-response (p<0.05 for trend). No association was found for former use (prescription filled between 90 days and 1 year before index date) of hypnotics-anxiolytics or for short-acting hypnotics-anxiolytics.
The evidence presented from this study is compelling. The obvious clinical presentation of a hip fracture, the use of population-based controls, and the ascertainment of drug exposure from examinations of pharmacy records of both cases and controls render bias unlikely as an alternate explanation for these significant results. The study sample was large, allowing stratification by drug type and detection of small increases in risk. Significant confounding by indication was unlikely: there was little difference in the odds ratios between community-dwelling subjects and nursing home residents despite greater prevalence of chronic and mental illness in nursing homes, and there was no increased risk with short half-life hypnotic-anxiolytic drugs, which are used in clinical circumstances similar to those for the long half-life drugs. In addition, there was an increased risk with increased dose, and a lack of increased risk among former users.
The limitations of this study are, as are its strengths, derived from using a large administrative database. Detailed information on possible confounders was not available, especially the indications for BZD use. There is high probability of misclassification of exposure: it is not known whether the drug was being used at the time of the fall. In addition, the findings may not be generalizable to non-Medicaid patients.
Since Ray and colleagues' first study implicating specific BZDs in the etiology of hip fractures, several studies have addressed the relationship of BZDs and the more general outcome of falls, the event leading to hip fracture. Cumming et al (1991) found an increased risk for LABZD and diazepam use, but not short acting BZD (SABZD) or BZD use, in an interview-based cross-sectional study among 1358 community-dwelling elderly in St. Louis in 1987-1988. Sorock and Shimkin (1988) found that, in a prospective cohort study of 169 community-dwelling elderly followed for a mean of 5.6 months, BZD use was not independently associated with falls, but only increased risk of falling synergistically with position sense loss (RR=3.00, 95%CI=1.87-4.80). Ruthazer and Lipsitz (1993) followed 635 nursing home residents for one month, and found no association with a prescription for BZD use. None of these three studies accounted for the possibility of confounding by indication. The Cumming et al (1991) study was very likely to have been subject to recall bias and reporting bias: up to two-thirds of persons with positive serum tests for psychotropic drugs will deny drug use (Ray 1992)
Ryynanen et al (1993) attempted to address the problem of recall bias in their population-based case-control study in Finland in 1987-88 by determining serum BZD concentrations at the time the patient sought treatment for a first injury due to a fall during the study. BZD was detected in the serum of 42% of the cases and 29% of the controls (p<0.001). Again, important potential confounders were not adjusted for, and exposure misclassification may have still occurred, since BZD use may have been affected by the event.
Thapa et al (1995) improved over past studies by following 282 nursing home residents for 250 person-years, more carefully classifying exposure and adjusting for confounders. The crude analysis revealed a nonsignificant increased risk for second fall during the study with regular (4 to 7 days per week) BZD use versus no use (in past 30 days); adjustment led to a significant increase in risk (RR=2.10, 95%CI=1.17-3.76). Residual confounding remained likely since age was dichotomized to <75 versus >75 years. These several studies of benzodiazepines and risk of falling left many methodological problems unresolved.
Two studies investigating benzodiazepines and hip fracture were studies of many risk factors, and were not designed specifically for valid ascertainment of drug use, and so are very likely to suffer from reporting bias. Grisso et al (1991) found no increased risk of self-report of recent use of LABZD in a hospital-based case-control study. Cummings et al (1995) found increased risk of hip fracture with self-reported LABZD use, but not with SABZD use, in a prospective open cohort of 9516 white women.
Cumming and Klineberg (1993) performed a case-control study in Australia in 1990 with 209 persons age 65+ who were hospitalized for a hip fracture in 1990-91 and 207 population-based controls. After controlling for several potential confounders, the risks associated with self-reported use of BZDs, LABZDs, and specific LABZDs were not significant. Use of the SABZD temazepam did increase risk significantly (OR=3.52, 95%CI=1.07-11.54). This was the first study to demonstrate an increased risk with a SABZD; it is likely that no statistically increased risk was found for LABZDs due to the small sample size and/or recall bias.
Lichtenstein et al (1994) compared 129 cases of hip fractures occurring in the hospital and 234 hospital-based controls with person time sampling. There was an increased risk of hip fracture within 48 hours of administration of BZD (OR=2.05, 95%CI=1.05-3.77). The method of choosing the comparison time for the controls (a random time during the hospitalizations) may have biased the results, and many potential confounders were not consistently mentioned in the hospital charts and were unavailable for adjustment.
Hip fracture is a rare event; therefore, prospectively observed cohort studies are not feasible. Several authors used large linked datasets to gather enough subjects and information to make inferences. Ray et al (1989) followed their pivotal 1987 study with a population-based nested case-control study using Saskatchewan health databases to compare the risk of hip fracture between LABZD and SABZD users during 1977-85. Among 4501 eligible hospitalizations for hip fracture and 24,041 age and sex matched controls, current users of LABZD (prescription filled within 30 days before index date) had an increased risk of hip fracture (RR=1.7, 95%CI=1.5-2.0) compared to subjects who filled no prescriptions for psychotropic medications in the past year; former users (prescription filled between 90 days and 1 year before index date) had no increased risk (RR=0.9, 95%CI=0.8-1.1). The relative risks were significant for the subgroups of current users receiving a dose of less than 2 times the standard daily dose (SDD), current users receiving 2 or more SDD, new users (first BZD prescription within 30 days of index date), and users of each of the three BZDs that accounted for more than 95% of the LABZD drug use (diazepam, flurazepam, and chlordiazepoxide). Current users of SABZDs had no increased risk of hip fracture (RR=1.1, 95%CI=0.9-1.3); no significant increases in risk were found for any of the subgroups. The three formulations that accounted for more than 95% of SABZD use were triazolam, oxazepam, and lorazepam.
Medical record review for a sample of 189 cases suggested that this finding was not due to confounding by dementia, ambulatory status, functional status, or body mass; for example, the prevalence of diagnosed dementia was no greater among LABZD users than non-LABZD users. In addition, there was no increased risk among former users and among users of SABZDs, which are prescribed in similar clinical circumstances. It is possible that the study not detect an increase in risk among users of SABZDs that was small, manifested only in certain patient subgroups or confined to the less frequently prescribed drugs.
Herings et al (1995), in a case control study in the Netherlands using a Dutch record linkage system, found that dosage is more important than elimination half-life in determining risk among 493 hospitalizations for a femur fracture from an accidental fall in 1986-92; three population-based controls per case were matched on age, sex, and pharmacy. Falls were significantly increased with current use of BZDs (OR=1.6, 95%CI=1.2-2.1), SABZDs (OR=1.5, 95%CI=1.1-2.0), sudden dose increases (OR=2.5, 95%CI=1.0-11.5), and concomitant use of several BZDs (OR=2.5, 95%CI=1.3 -4.9). A strong dose-response relationship (p<0.0001) among users of either SA or LABZDs suggests that these are explained primarily by dose.
These findings point to the importance of dosing BZDs and suggest that effect of pharmacological differences between particular subgroups of BZDs are superseded by the way BZDs are dosed and used in daily practice. Elimination half-life of BZDs may not be a good predictor of risk for falls because it is not a fixed value; rather it depends on inter-individual differences, and in the elderly, these differences probably are larger, and short-acting drugs are effective for longer periods than expected. Frequent dosing of SABZDs will result in sustained blood levels of BZDs. Further, half-life is based on elimination from the peripheral blood circulation, which is not necessarily related to concentrations of BZDs in the central nervous system. It is likely that dosage and elimination half life both contribute to the risk for hip fracture.
Confounding by indication remains a most important consideration in these studies. Dementia may lead to psychomotor impairment and increase risks of falls and injury (Monane and Avorn 1996); BZDs are often prescribed for the concomitant disruption of sleep (Ray 1992). Insomnia per se is unlikely to be a major confounder for fall risk; there is no evidence that milder forms of sleep problems are associated with performance deficits (Ray 1992). Therefore, the relationship between falls and medication-related falls in the elderly may be confounded by prevalence of dementia (Monane and Avorn 1996). Some studies have addressed this problem by assessing presence of dementia or cognitive impairment and adjusting for these in the analyses; the positive association remained. Room for confounding remains, though, especially in studies using administrative data bases, where symptoms of dementia may not be recorded.
Many studies have been designed to study multiple risk factors and have not focused on drugs per se, and thus often fail to avoid misclassification due to recall bias, to identify specific drug class in use, to determine drug use at the actual time of the fall, and to adequately measure administered and scheduled medications, and have inadequate sample size to detect a moderate increase in drug risk (Ray 1992, Monane and Avorn 1996). Some studies stratified by 10 year (or larger) age groups to avoid confounding by age; BZD use and incidence of falls are both so very tightly related to age, that 10-year age groupings are likely to allow for residual confounding.
Community based studies and nursing home based studies may find differing results because of the differences in the populations studied. One author suggested that two distinct groups comprised those elderly who broke their hips in falling: one consisted typically of frail, immobile and demented persons who fell indoors and did poorly after treatment, while in contrast there was a group of fit active old people who usually broke their femora out of home (Evans 1988). Differences in the relative proportions of these two types of cases in the study probably contributes to inconsistent estimates of effect.
There are a number of difficulties in establishing the relationship between drugs and falls in epidemiological studies, though. One method of substantiating an adverse drug effect is by demonstrating occurrence of effect on reintroduction of the drug. The sporadic nature of falls and the fact that the drug may contribute only to falls under particular circumstances usually means that re-challenging is not possible (Monane and Avorn 1996).
More studies are needed to clarify the remaining controversy. Studies should focus on drugs per se rather than considering drugs as a secondary design objective, and sample size should be adequate to study effects of specific drug classes. Drug use should be measured with unbiased techniques that account for changes in drug use over time; however, this may be difficult with interview-based methods, since nearly two-thirds of persons with positive serum tests for BZDs or other psychotropic drugs will deny drug use, and drawing blood samples is expensive (Ray 1992).
To assess the potential for confounding by indication, accurate measures of psychopathology are needed (Ray 1992). In prospective cohort studies, though, it may be prohibitively expensive to periodically assess mental health, and there is the ethical consideration of identification of persons with mental illness and then withholding intervention. In a case-control or retrospective cohort study, measurement of psychopathology after outcome may be questionable because the injury may make the subject unavailable or may alter the disease state (Ray 1992).
For younger and healthier elderly, environmental factors are important causes of falling, especially stairs and floor obstacles (Perry 1982). For older sicker elderly, host-related factors, such as drugs, are more important (Monane and Avorn 1996) and may represent a unique opportunity for injury prevention because medication use is a more readily modifiable risk factor than many of the intrinsic host defects associated with injury risk in the elderly (Ray 1992). If the magnitude of the role of BZDs in injury risk is as great as that suggested by these data, then these frequently used drugs cause substantial excess morbidity and mortality in the elderly, especially with higher doses and long half-life agents (Ray 1992, Monane and Avorn 1996).
Because of their controversial nature, though, the extensive data concerning the effects of BZDs on psychomotor function and injury risk have had limited influence on regulatory bodies, the pharmaceutical industry, public health agencies, and on clinical practice (Ray 1992). Many studies found that published recommendations for using smaller doses in the elderly and short duration of use were not followed by the prescribing physicians (Herings et al 1995).
Some suggestions for reducing the risk of falls from the use of psychotropic medications include: using the lowest effective dosage, urging supervision and use of walking aids, decreasing the chronic use of psychoactive medication via "drug holidays" with careful tapering-off and monitoring, and educating users about the risks of falling (Sorock 1988).
Many studies, with many different designs and in different populations, have found an increased risk of fall or hip fracture among elderly persons using benzodiazepines. The range of increased risk has been from 1.5 times (Ray et al 1987) to 5.80 times (Cumming et al 1991) compared to those not using psychoactive drugs. Those studies with no significant result were either not designed as drug studies (Cummings et al 1995, Grisso et al 1991), or did not adjust for potential confounders (e.g., Ruthazer and Lipsitz 1993). Additional large prospective epidemiological studies are necessary to determine definitively whether benzodiazepines cause increased risk for hip fracture, with careful design to account for confounding by indication.
Return to the DDIL Homepage
This
document has been provided through the Drugs and Devices Information Line,
http://www.hsph.harvard.edu/Organizations/DDIL/ddil.html