Can you predict who will fall in a hospital ward?
CLINICAL BOTTOM LINE:
The fall assessment tool STRATIFY performs little better than clinical judgement when used in an acute hospital setting. Neither STRATIFY nor clinical judgement are effective at predicting which patients will fall and both over-predict risk.
Your acute medical ward is looking at falls and how to prevent them. A meeting is set to brainstorm ideas about how to reduce the incidence of falls. You recall reading about a tool called STRATIFY that might be useful in predicting who will fall. You want to investigate before your ward starts choosing an assessment tool.
Is the STRATIFY tool effective at identifying which adults will fall in an acute hospital setting?
PubMed – STRATIFY and Falls.
Webster J, Courtney M, Marsh N, et al. The STRATIFY tool and clinical judgement were poor predictors of falling in an acute hospital setting. J Clin Epidemiol 2010; 63:109-13.
Patients who were admitted to internal medicine, surgery, orthopaedic, psychiatric, oncologic and geriatric rehabilitation services in a 982-bed tertiary teaching hospital in Australia between March and October 2007.
Inclusion criterion was patients being aged ≤ 65 years.
Patients were assessed within 48 hours of admission by research officers trained in the use of STRATIFY. At the same time the nurse caring for the patient was asked if they thought the patient was at risk of falling.
A fall was defined using the World Health Organization criteria and patients were followed up until discharge.
Two risk assessments. (1) STRATIFY tool – five-item instrument that assesses whether patient admitted after fall or has had fall since admission, if they are agitated, have visual impairment, require frequent toileting or have high transfer needs. Each question answered yes/no and yes scored as 1. Cut-off score of 2 suggested threshold for risk; (2) Clinical judgement of the nurse caring for the patient.
Reference standard. Reported falls identified from incident reports and on discharge, medical record assessed for evidence of a fall.
Follow-up over time was gold standard. Expected falls to be reported in incident database but also checked clinical record on discharge in case falls not reported.
Blind comparison? Falls assessment undertaken by research officers who could not know whether patient was actually going to fall or not (therefore blinded). Reporting of falls by ward staff but checking in medical records to ensure as many falls as possible recorded. Ward staff not blinded to falls so reporting could be influenced. However, falls prevalence of 9.2% suggests reporting and checking of records ensured accuracy as much as possible.
Appropriate spectrum of patients? A tertiary-level hospital with patients in variety of services aged 65 or more being assessed, so yes.
Reference standard applied regardless of results? The follow-up data was not available for 13 out of 801 patients screened.
Validated in a second independent group of patients? STRATIFY has been validated in previous studies in geriatric populations, but not in general hospital population.
Overall impression? A good-quality study with some limitations in terms of blinding, but unlikely to be able to blind ward staff to such an investigation.
Prospective cohort of 801 adults meeting inclusion criteria; 348 were medical patients, 306 were surgical, 55 oncologic, 37 were mental health patients, 46 were in geriatric rehabilitation and nine were in infectious diseases. Thirteen did not have data for follow up, so 788 included in analysis. Mean age of patients was 78 years and the mean length of stay was 27.7 days. 49% of the patients were male, 22% were agitated, 33% had experienced a previous fall, and 29% required frequent toileting. Seventy-two patients (9.2%) fell. Sensitivities, specificities and likelihood ratios presented in the table.
The mean length of stay high for an acute-care facility.
Both approaches will lead to high false positive rates in settings where falls prevalence is usually low, e.g. acute medical wards. Might be more acceptable in high-prevalence settings. Knowledge of own data will be useful in determining whether area is high or low prevalence (consider total number of falls/1000 patients admitted or total number of falls/1000 patient days to make such a determination).
No inter-rater reliability information was collected on agreement between the research officers using STRATIFY.
Reviewer: Dr Andrew Jull RN PhD – associate professor, School of Nursing, University of Auckland; nurse advisor, quality, Auckland District Health Board.