Instruments for assessing the risk of falls in acute hospitalized patients: a systematic review and meta-analysis

For hospitalized patients, falls are a serious problem, reducing the duration and quality of life. Older people with injuries have higher mortality rates and stay longer in hospital due to comorbidity. Falls are the predominant cause of injury in older people (over 65 years), followed by traffic accidents, fires and burns, drowning and poisoning. It has been reported that in the European Union there are 13.3-164.5 deaths per 100,000 persons among those older than 65 years [1].

It is estimated that over 84% of all adverse events in hospitalized patients are related to falls [2]. Approximately 30% of the hospitalized patients who fall suffer injuries, of which 4-6% are severe, including fractures, subdural haematomas, bleeding and even death [3].

The importance of this issue is such that the Joint Commission International (JCI) includes it among its safety standards in the accreditation manual for hospitals “Reduce the risk of patient harm resulting from falls” [4]. Furthermore, falls ranked sixth on the list of JCI sentinel events in 2012, with 477 notifications [5], a position among the most common adverse events it has held for the past three years [5].

As well as the physical consequences, there are also psychological ones, constituting what is known as the “post-fall syndrome”, which include fear of another fall, and the loss of self-esteem and independence, compromising the patient’s lifestyle and impacting on family caregivers.

The costs arising from falls, particularly hip fractures, skull fractures and leg injuries, represent a large proportion of healthcare spending. It is estimated that 92% of the costs of health care for patients who have suffered a fall are attributable to this factor [6], although it is difficult to obtain an accurate figure because most studies only include the costs of patients admitted following an injury, and do not take into account those who fall within the hospital itself [7]. An estimate by the British National Health Service estimated that about £15 million a year are incurred in hospital costs as a result of falls (£92,000 per year for an 800-bed hospital) [8].

Various studies have investigated the risk factors for falls in hospitals [3, 9], identifying these as including advanced age, agitation, confusion or disorientation, generalized muscle and/or leg weakness, unstable gait, urinary incontinence, a history of previous falls, visual deficit or the use of certain medications (hypnotics, sedatives, vasodilators, diuretics, antidepressants, etc.) [3, 9, 10]. Moreover, the hospital environment itself can directly affect the incidence of falls. Extrinsic risk factors include the presence/absence of bed rails, the height and stability of any type of seat (including toilet) or obstacles in the form of clinical furniture and equipment [11]. The mere fact of hospitalisation represents a risk factor for falls. Older people especially can become more disoriented or agitated, or suffer diminished functionality during hospitalisation, and thus be at increased risk of falls [12].

Analysis of the circumstances in which falls occur among hospitalized acutely-ill patients and of the risk factors involved has led to the development of various instruments to assess the risk of falls, such as the Downton scale [13], the Morse Fall Scale (MFS) [14], the St. Thomas Risk Assessment Tool in falling elderly inpatients (STRATIFY) [15], the Tinetti test [16], the Conley scale [12], the Hendrich Fall Risk Model (HFRM) [17] and its later version HFRM II [18].

Some of these risk assessment tools have been tested in environments other than those for which they were developed [19‐23], with disparate results, including difficulties for widespread use, serious validity discrepancies between the original authors’ version and successive ones [24], and in the heterogeneity of diagnostic accuracy in terms of cutoff points [22, 25‐27]. However, a recent Cochrane review showed that multifactorial interventions in hospitals reduce the rate of falls (rate ratio 0.69, 95% CI: 0.49 - 0.96), although risk assessment is addressed as one of many interventions, and it is not easy to isolate its specific effect [28]. Hospitalized patients in the acute phase of their disease have specific characteristics. Changes in acuity of illness and medication will affect mobility, physical status and cognition [10], requiring a special assessment in this setting in order to prevent falls. Moreover, an unknown environment like the hospital can contribute to increase previous risk or generate new risk factors.

The methodological weaknesses that have been identified are subsequently reflected in the under or over-detection of patients at risk of falling, and the routine use of such methods may divert attention and resources toward patients who would least benefit from preventive measures, at the expense of those who do need them.

These considerations highlight controversies and inconsistencies regarding the widespread use of these tools in the field of acute hospital care. Therefore, in order to dispel some of these uncertainties and to determine which of the available instruments offers better diagnostic performance for fall prevention as part of a range of preventive interventions to minimise risk among hospitalized patients, we present an up-to-date, detailed analysis of the existing literature illustrating the scope of measures available. The results of this review will contribute to the implementation of best practices related to preventing falls in an acute care hospital setting.

The aim of this review is to determine the accuracy of instruments for detecting fall risk and predicting falls in acute hospitalized patients. The specific objectives were to analyse the diagnostic validity and psychometric properties of the various risk assessment tools for predicting falls in acute hospitalized patients, and to compare the effectiveness of risk assessment instruments for falls and its impact on the incidence of falls by acute hospitalized patients.

Following the Cochrane Manual for Diagnostic Test Accuracy [29], this systematic review focuses on establishing the accuracy of instruments, scales or questionnaires (index) developed for detecting or predicting falls (target condition) in acute hospitalized patients, aged 16 or over (patients).

Thus, the review determines what instruments are available for assessing the risk of falls by acute hospitalized patients, the differences among them in terms of diagnostic accuracy and/or psychometric properties and their potential impact on preventing falls when implemented in the clinical context.

A search within different databases and webs produced a total of 2,181 references (Table 1). After removing duplicates, there remained 2,006 articles, whose titles and abstracts were evaluated by blinded pairs of reviewers. After this first phase, 78 articles were selected as potentially eligible, and the full text was then read, again by blinded pairs, to assess its quality, extract data and determine its inclusion or otherwise in the meta-analysis. Finally, 14 studies [15, 20, 22, 23, 25‐27, 42‐48] were selected for this review (Figure 1). A summary of the references and the reasons for excluding the remaining 64 items is provided in Table 2.

A total of 14,663 patients were considered in the studies selected, although several of the latter, in addition to the diagnostic validation of the instrument or instruments in question, also conducted studies of intra-observer reliability [48] inter-observer reliability [26, 42, 48] or the procedure leading to the development of some risk assessment scales [15, 46]. A total of 13,284 patients were involved in the analysis of diagnostic accuracy. Although not all the studies provided the age and sex distribution of their populations, according to the published data these involved 5,504 men (41.43%) and 5,358 women (40.33%). All the studies were performed in hospitals for acutely-ill adults. Several of them focused on patients aged over 50 years [45], over 65 years [15, 20, 44], or on hospital departments that mainly treated the elderly [47]. In consequence, the overall mean age of the patients was 69.76 years (SD 9.56). In all cases, the diagnostic validation of the different risk assessment tools was performed prospectively. No experimental studies were conducted (Table 3).

Regarding the methodological quality of the studies, some shortcomings were identified, mainly related to two aspects: the lack of blinding in outcome assessment or lack of information in this respect, and doubt as to the representativeness of the study population, generally because the article failed to stipulate how the sample size was calculated (Table 4).

Although all the studies provided data enabling reproduction of the calculations of diagnostic validation, not all were included in the meta-analysis, because some scales did not contain sufficient studies for this (Figure 1). The Schmid study [46] was discarded because it described the development of an assessment tool for the risk of falls but this was not subsequently tested in any of the selected studies. We also excluded the Oliver study [15], which although it provided data on local validation and the remote validation of the STRATIFY scale, expressed the results in terms of ‘falls’ rather than ‘fallers’ as in all the selected studies. Therefore, and as done by this very author in a meta-analysis of the same scale [69], it was ruled out of our meta-analysis. With respect to the study by Milisen [23] we only considered the data for patients admitted to medical and surgical units (these data are available separately), but not to geriatric units, as the latter would not meet the criteria for inclusion in the present review, which is limited to acutely-ill patients.

Finally, the meta-analysis was performed with the Morse, STRATIFY and Hendrich II Fall Risk Model instruments. The results showed that the STRATIFY tool provided greater diagnostic validity, with a diagnostic odds ratio (DOR) value of 7.640 (95% CI: 4.862 - 12.007) versus 5.068 (95% CI: 3.747 - 6.857) for the MFS and 3.362 (95% CI: 2.107 - 5.364) for the HFRM II (Table 5). Figure 2 shows the forest plots with partial DOR of each study included into the meta-analysis, as well as the global DOR for each tool and the 95% confidence interval.

Sensitivity analyses performed for MFS showed that after removing the Kim EAN 2007 study [26], heterogeneity was markedly improved, possibly because it included younger patients (average age below 65 years), to whom lower cutoff points were applied (25 and 51). The same was true for the STRATIFY meta-analysis, where, as well as the Kim EAN 2007 study [26], the data referring to the original scale in the Barker 2011 study [42], were also removed. In both cases, the mean age of the study population was less than 65 years, although the cutoff points (2 and 3) were higher than those applied in the other study included in this meta-analysis with similar characteristics in terms of the age of the sample population, namely Milisen 2007 [23] (cutoff point, 1). The removal of these two studies significantly improved heterogeneity, although this remained high for specificity and for a positive likelihood ratio (LH+), without reaching statistical significance for the DOR. On the HFMR II tool, heterogeneity was over 50% in sensitivity and specificity, but was not significant for DOR, LH + or LH-.

A meta-regression was also performed, in order to assess the effect of average patient age over 65 years [20, 22, 25, 44, 45, 47, 48] and the performance or otherwise of risk reassessments during the patient’s stay, versus a single evaluation on admission [27, 45]. In the case of the MFS, the reassessment showed a significant reduction in the DOR on the tool (rDOR 0.75; 95% CI: 0.64 - 0.89, p = 0.0176).

Although other systematic reviews and meta-analyses of assessment tools for falls by hospitalized patients have been performed [9, 56, 57, 69, 106], ours is the first that includes only acute patients. This is particularly useful in clinical practice for identifying the behaviour of the instruments currently used exclusively in the hospital environment, where falls are among the most frequent adverse events [5], and thus are directly relevant to the development and implementation of safety policies in acute care hospitals.

Since V. Scott [106] and T.P. Haines [56] published their systematic reviews of fall risk assessment tools in 2007, there were no new updates focused on that instruments concerning acute hospitalized patients. The systematic review published by Oliver in 2009 focused only on the STRATIFY tool and was not limited to acute patients. In the present review, 9 [23, 25‐27, 42‐45, 48] of the 14 selected studies have been published since 2007, allowing an update of knowledge available on this topic. This is one of the strengths of this study. Another strength of this review is that contemplated studies assessing the psychometric properties of the fall risk assessment instruments.

This meta-analysis was carried out as a comparison of the Morse Fall Scale (MFS), the St. Thomas Risk Assessment Tool in Falling Elderly Inpatients (STRATIFY) and the Hendrich II Fall Risk Model (HFRM II). The results obtained showed the STRATIFY tool to be the best tool for assessing the risk of falls among hospitalized acutely ill adult patients, followed by MFS and finally HFRM II. STRATIFY produced the best values for sensitivity and had a specificity similar to that of MFS, and obtained the best values for DOR. In part, these results contradict those published in a recent meta-analysis focusing on MFS and STRATIFY [57], which found a higher sensitivity but lower specificity for the MFS with respect to STRATIFY. However, these results did not include the calculation of the DOR and some of the studies that were included were excluded from our analysis as not meeting the inclusion criteria. Moreover, we also considered another four studies published subsequently with data for these assessment tools [27, 42, 43, 48].

These three tools have been compared by their simultaneous application in a sample of hospitalized adult patients [26]. In this study, HFRM II was found to be the most suitable for identifying the patients at high risk of falls, with a sensitivity of 70% and a specificity of 61.8%. However, in the present meta-analysis, HFRM II proved to be the worst of the three instruments considered, due to its lower sensitivity (0.628), specificity (0.640) and DOR (3.362).

In another study in which four risk assessment instruments (STRATIFY, Tinetti, Downton and Tullamore) were tested simultaneously in an acute patient hospital environment [47], STRATIFY was completed most easily and in the least time; furthermore, it also presented the best predictive validity, although it was the least sensitive of the four. The short time required to administer this assessment scale and the fact that it is readily understandable for medical staff are very relevant factors in an acute hospital setting where work loads are high and periodic reassessments of patients are advisable.

In previous systematic reviews of this question, one of the inclusion criteria was that the selected studies should conduct a prospective validation of assessment tools for falls [9, 69, 106]. In the systematic review and meta-analysis carried out by Haines in 2007 [56] one of the practical implications described was that although retrospective evaluations are still valuable for generating initial results and identifying the tools and cutoff points that may be useful in clinical practice, less weight should be given to their results than to those obtained from prospective studies, with respect to selecting a detection tool for use in clinical practice. In coherence with this view, all the studies included in the present review conducted a prospective validation of the various instruments examined. Nine of the fourteen studies had been published since the completion of the above-mentioned systematic review. Moreover, the present analysis complies with one of the “gold standard” criteria described by Wyatt and Altman for such scales [107], although in none of the articles selected was a randomised controlled trial carried out, and this may be an area for improvement regarding the development of future research in this field.

Previous studies have argued that fall risk assessment performed only at the time of hospital admission does not identify changes in the patients’ clinical status during hospitalisation, although this is a common occurrence, especially among the elderly, who may become disoriented, agitated or lose functionality during hospitalisation, and thus be at greater risk of suffering a fall [12]. The acute phase of the disease and changes in medication can affect both mobility and the physical and cognitive status, and therefore hospitals need an instrument that can be used quickly and easily so that repeated assessments of these patients may be carried out [10]. In only four of the fourteen studies selected for this review was a reassessment conducted, whether on a weekly schedule [15, 27, 46] or following changes in the patient’s condition, after cognitive impairment, after significant changes in medication or after a fall [45]. Although the meta-regression analysis with respect to this criterion showed no effect for HFRM II, and could not be performed for STRATIFY due to the lack of studies in which a reassessment of patients was performed, in the case of MFS, the reassessment produced a significant reduction in the DOR. This may be explained, in part, as the MFS losing predictive capacity when the risk of patients’ suffering a fall decreases, as their condition improves. In the only study in which a reassessment was carried out with MFS [27] the mean age of the patients was below 65 years (62.6). The condition of these younger patients would presumably improve over time, and so their risk is more difficult to identify with this scale. However, as discussed above, in only one of the studies in which MFS was tested was a reassessment performed. Thus, further research is needed, including reassessment both with MFS and with the other instruments in order to achieve a more realistic analysis of this circumstance. It should be borne in mind that, in general, it is difficult to accurately predict the risk of falls among hospitalized adult patients who are subjected to external risk factors, specific to the hospital environment and which are not taken into account by any of the assessment instruments described.

This review and meta-analysis may also be affected by the limitations of the primary studies analysed. First, knowledge of the number of patients suffering a fall is always dependent on the voluntary reporting of this fact by the healthcare staff, and so falls may occur that are not reported, which would to some extent invalidate the results obtained in these studies. Second, the review may be affected by contamination related to the implementation of other actions taken to prevent falls in the different environments studied, and by a possible Hawthorne effect. Moreover, limitations arise from the questionable quality of some of the studies selected: some offered no data on the age and/or sex distribution of the study population [15, 43, 46], or were deficient regarding the representativeness of the sample [22, 25, 44, 46‐48] or regarding the blinding of the researchers [22, 27, 43‐45]. Another possible limitation concerns the search language: in the present review, the search languages used were limited to English, Spanish and Portuguese, and four studies were excluded for this reason [102‐105].

In short, despite the findings obtained, our analysis of the various studies clearly shows that the behaviour of these risk assessment instruments varies considerably depending on the population and the environment in which they are administered. In consequence, we cannot recommend the generalised adoption of any single method without its prior testing in the healthcare setting of the intended implementation. Moreover, it should be recalled that these instruments, or the actions taken including their use, will not be effective if healthcare personnel do not ensure patient safety procedures are followed, and this aspect remains to be investigated in the case of falls by hospitalized patients. A study of safety and security in Spanish hospitals reported that the majority of healthcare staff (77.8%) had not reported any event related to patient safety in the past year and that 95% had reported fewer than two such events [108]. This aspect, noted above as one of the limitations of our study, and the question of compliance by personnel with procedures established to prevent adverse events, are issues which must be addressed in order to achieve an effective culture of safety within hospitals.

The STRATIFY scale was found to be the best tool for assessing the risk of falls by hospitalized acutely-ill adults. With this scale, the DOR was higher than with the MFS and HFRM II. However, the behaviour of these instruments varies considerably depending on the population and the environment, and so their operation should be tested prior to implementation. Further studies are needed to investigate the effect of the reassessment of these instruments with respect to hospitalized adult patients, and to consider the real compliance by healthcare personnel with procedures related to patient safety, and in particular concerning the prevention of falls.