This cluster randomized trial of an intervention to increase bed alarm use in hospital nursing units showed that increased use had no statistically significant effect on the number or rate of falls, injurious falls, or patients restrained on intervention compared with control units. On the basis of the 95% CIs around our estimate in fall rate differences, our findings were statistically compatible with a decrease in falls by as much as 1 per 1000 patient-days and an increase in falls by as much as 2.47 per 1000 patient-days in the intervention units, so large, clinically significant benefits (or harms) are not probable.
To identify previous studies of alarms as a strategy for fall prevention, we searched PubMed and CINAHL databases from 1975 through April 2012 using the terms hospitals, accidental falls prevention
, and clinical alarms
. We identified 4 fall prevention studies in hospitals (13
) where alarm systems were the primary intervention. Each of these studies reported that falls were reduced by 20% to 60%; however, these results should be considered with caution because only the study by Tideiksaar and colleagues (24
) included a concurrent control group. In addition, 2 recently published cluster randomized studies included alarms as a part of multifactorial interventions to prevent falls in hospitals. One trial was effective (27
), but the other did not reduce falls (28
Our study was able to overcome many methodological weaknesses and informs the design of future interventions aimed at reducing falls in hospitals. Using a fall evaluator system in addition to hospital occurrence reports enhanced the accuracy and reduced potential for reporting bias. The study also used information from various data sources to develop unit-level covariates, including patient demographic characteristics, insurance, psychotropic medication use, and staffing. Finally, the long duration of the intervention permitted the novelty of the intervention to diminish, as evidenced by alarm use reaching an equilibrium state after approximately 6 months.
Our study has limitations. It was conducted at a single site; however, this assured fidelity of the intervention and facilitated standardizing our approach to end point and covariate ascertainment. The fall rates that we observed are typical of acute care hospitals in the United States (27
), and we found excellent separation of alarm use between intervention and control nursing units, suggesting that there was little evidence of contamination.
Furthermore, because of higher fall rates in the baseline period, our trial was ultimately underpowered to detect our primary end point, falls per 1000 patient-days, and was not designed to detect a difference in injurious falls. Therefore, our findings should be interpreted with caution. However, the lower 95% CIs exclude large benefits in fall rates with the intervention.
Another limitation is the inability to conduct our study in a blinded manner and to completely balance the exposure of the intervention and control nursing units to the study team. Although these might increase the risk for a Hawthorne effect, we do not believe that this played an important role in biasing our findings. Fall rates remained similar between baseline and study periods in both intervention and control units; furthermore, we found no difference in end points that would be less susceptible to reporting bias (for example, injurious falls and restraint use).
Although we were able to control for several demographic covariates, we could not completely control for fall risk at the unit level because it was captured in the electronic medical record beginning in the last month of the baseline period. However, as and the show, the unit-level fall risk in both intervention and control units was similar in both baseline and study periods.
Several plausible explanations are available for why the intervention did not reduce falls or related events despite a large increase in patient-days with bed alarms in place. Despite the support of the intervention team, false alarms are a common problem of bed alarm systems in the practice setting. In a field study of nursing home patients, Capezuti and colleagues (14
) found a high degree of both false-positive as well as false-negative events in a traditional alarm system like that used in this study. More advanced alarm systems, including infrared beam sensors, do exist and may produce more encouraging results (14
False alarms may also contribute to “alarm fatigue” (33
), in which staff no longer responds when an alarm appropriately sounds. Also, instruments to predict falls among hospitalized patients have limited specificity (34
), and the “wrong patients” may have been chosen for bed alarm monitoring. Finally, alarm signals may occur after patients had already fallen because they fell immediately on exiting the bed or chair. Although each of these factors was observed in the course of the intervention, we did not systematically quantify reasons for alarm failure.
In summary, although our intervention to increase bed alarm use increased use in intervention nursing units, there was little evidence of an effect on fall-related events or an effect on physical restraint use in intervention compared with control nursing units. Although the study was not designed to rigorously track alarm-related expenses, the costs of the study alarm system are substantial: The monitoring box, connection cables, and replacement cords cost approximately $350, and each disposable sensor pad costs $23. There are also facility expenses related to inventory control and maintenance. Thus, although bed alarms may yet prove useful as a part of a well-defined fall prevention program, hospitals should temper expectations that their use will provide a simple and cost-effective solution to the problem of falls.