Falls risk-prediction tools for hospital inpatients. Time to put them to bed?

Accidental falls are the commonest safety incidents affecting hospital inpatients and care-home residents [1, 2]. The recent National Patient Safety Agency (NPSA) report, ‘Slips Trips and Falls in Hospital’ [2], identified over 200,000 reported falls incidents from acute, community and mental health trusts in England and Wales from 2004/2005 alone, (some 32% of all incidents in all age groups) though we know from this and other sources that such incidents are underreported [3]. Such falls are associated with a range of adverse outcomes including injury, impaired confidence and function, increased length of stay, institutionalisation anxiety and guilt for staff and relatives, complaint and litigation. They should, therefore, be a major risk management priority for hospitals and care homes (where around 50% of residents fall at least once a year) [4], and have recently been made a main focus for examining older patients’ care by the Healthcare Commission [5]. There is a growing body of evidence on interventions to prevent falls and falls-related injuries in hospital [1, 6, 7]. One component of many research interventions, and a common feature of ‘real-life’ falls policies in hospitals, is the use of falls risk-prediction tools. By this, I do not mean ‘checklists’ of common risk factors to prompt specific action by staff, which might, in turn, reduce falls. Such factors might include environmental and equipment safety, medication, hypotension, visual impairment, muscle weakness or postural instability, cognitive impairment, restlessness or agitation, all of which, amongst others, have been targeted in successful falls intervention programmes [8]. I have no argument with the use of such tools, which, in effect, prompt good comprehensive geriatric assessment and care-planning. My concern is over scoring tools, which purport simplistically to classify patients as having a ‘high’ or ‘low’ risk of falling so that interventions can be targeted to ‘high-risk’ patients. This approach can work well in other fields, for instance, with diagnostic screening tests such as troponin for acute coronary syndrome, or d-dimer for suspected pulmonary embolism (both of which are good ‘true negative’ tests with high specificity). It is also possible for clinical prediction tools with continuous scoring to be used to calculate overall probability of an adverse event. Examples of these include the Kings Fund ‘Patient at Risk of Readmission Scores’ (PARR), or Critical Care Early Warning Scores, both of which correlate tightly with the probability of the adverse event, (i.e. hospital admission or cardiac arrest/ICU admission) [9, 10]. Falls risk-assessment tools in hospital have rarely been used in this probabilistic way, rather, they place patients categorically as either at ‘high’ or ‘low’ risk. Though most work has been done in hospitals, my comments on the use of such tools in hospitals could apply equally to their use in long-term care or community settings. There have been a succession of systematic reviews on validation studies for hospital falls prediction tools in recent years [8, 11–13] and several more new validation studies are in press. In this issue of Age and Ageing, Vassallo et al. [14] report a prospective validation study comparing two tools (STRATIFY and Downton Score) with nurses’ clinical judgement (largely based on wandering behaviour) on a cohort of 200 geriatric patients. Meanwhile, Ashburn et al. [15] looked at 122 consecutive patients discharged from a stroke ward, following up on them to record further falls at home over 12 months. Sixty-three experienced one or more further falls. Before discharge, they collected a variety of structured clinical data, a score based on ‘near-falls’ in hospital, and poor upper limb function on retrospective fitting predicted falls with 70% specificity and 60% sensitivity. In Vassallo et al., the single item of ‘wandering’ identified by nurses conferred better predictive accuracy than either of the formal scores, but significantly lower sensitivity (though this might not be the case for populations where wandering is infrequent). These papers raise some interesting questions around the practical utility in predicting and preventing falls, the ‘trade-off’ between the various elements of predictive validity, their validity in other (quite different) populations and settings. A detailed exploration of these and allied issues is not possible here, but is explored in recent reviews. However, even as one of the authors of the most widely validated tool for use in hospital (STRATIFY) [16]—still used in many hospitals [1, 2]—I am happy to recant. I do not believe that STRATIFY, or any other tool, is good enough at its job. In