Predictive validity of the Hendrich fall risk model II in an acute ...

another geriatric unit of the Teaching Hospital at the end of
January and the research protocol could not be followed as
previously decided, the study stopped after 8 months, even
if a longer survey period had been planned.
The patients were screened with HFRM II at the
moment of, or at the most within 24 h upon, admission
at the geriatric unit by the trained nurses.
The patients were followed-up from their admission
until their first fall, discharge, death or transfer to another
unit, and falls were reported on the appropriate fall
incident report form. We defined a fall as ‘‘an event which
results in a person coming to rest inadvertently on the
ground or floor or other lower level’’ (Hauer et al., 2003).
The inter-rater reliability was evaluated in the first
period of the research. Two nurses applied consequently
and independently the Hendrich II Model to a convenience
sample of older patients admitted to the unit during the
first 2 weeks of the study.
The time of administration was evaluated asking all
nurses to estimate the mean time needed to fill in the risk
assessment tool after the routine assessment of the
patients at the admission.
2.5. Statistic analysis
Descriptive statistics were used for demographic data.
To measure the reproducibility of results by more than one
rater, the inter-rater agreement was calculated using
kappa index at 95% CI. The likelihood of older patients to
fall was examined creating a 2 2 table to record patients
with high and low risk who fell or did not fall and the
predictive validity was tested calculating sensitivity,
specificity, predictive positive (PPV) and negative values
(NPV). The sensitivity was calculated dividing the numbers
of patients with high risk scores who fell by the total
number of patients who fell, and specificity dividing the
numbers of patients with low risk scores who did not fall
by the total number of patients who did not experience a
fall. The PPV was calculated dividing the number of older
patients with high risk score by the total number of
patients with high risk scores, while the NPV dividing the
number of patients with low-risk scores who did not fall by
the number of patients with low risk score. Associated 95%
CIs were also reported.
A receiver operating characteristic (ROC) analysis was
also carried out to examine the relationship between
sensitivity and specificity of the tool for different cut points
in order to determine the optimal cutoff point. Moreover
we calculated the incidence rate ratio during the period of
study, and the fall index incidence ratio for days of hospital
stay; to evaluate the risk factors that mainly affected the
falls we calculated the odds ratio for each risk factor
considered in the tool.
The data was analyzed using SAS V9.0 for Windows
(SAS Inc., Cary, NC, USA) and P < 0.05 was considered
statistically significant.
2.6. Ethical considerations
We asked the authorization to perform the study to the
Teaching Hospital General Manager. The study protocol
D. Ivziku et al. / International Journal of Nursing Studies 48 (2011) 468–474 471
was reviewed and approved by the University Hospital
Review Board. At admission nurses gave patients, or their
relatives if the patients were unable to decide for
themselves, a brief description of the aim of the study,
its risk and benefits, assuring that, independently of
participation consent or tool scores, the best nursing care
was going to be provided during their hospital stay. As the
main aim of the study was a tool validation and posed only
minimal risk for patients only oral consent was requested,
as approved by review board. Anonymity and confidentiality
was guaranteed, all the patients’ forms were retrieved
in the unit daily by principal researcher and stored in a
closed box. During the data analysis each patient was
identified with a progressive number and only the main
researcher could go back to the identification of the
patients.
3. Results
3.1. Inter-rater reliability
During the first 2 weeks of the study 24 elderly patients
were admitted and the inter-rater reliability was estimated
on them. The kappa index was 0.87 (95% CI 0.71–
1.00).
Disagreements mainly concerned the classification of
patients as depressed, the presence of dizziness, and the
use of benzodiazepines (used at home, but not during the
first 24 h of hospital stay).
3.2. Time of administration
The average time needed to collect the data to fill in the
tool was about 1 min, as estimated by each nurse: in fact,
some information could be retrieved easily from the
clinical record or asking the patients or relatives during the
nursing assessment at the admission. The time needed to
perform the get up and go test was variable, depending on
the clinical general conditions of the patients and
associated diseases. Sometimes was not performed specifically,
because the nurse observed the patients during the
examination and how she/he moves in the room.
3.3. Population
A total of 179 patients met the inclusion criteria and
gave consent to participate during the 8 months of the
study. Of the 179 patients, 74 were male and 105 female.
The mean age was 79.47 years (SD 9.5), with female older
than male (81.7 vs 79.8).
3.4. Validity
A total of 14 falls were reported on 179 patients, with a
cumulative incidence of 7.8%, and a fall index of 7.5 per
1000 days of hospital stay.
The mean score of the HFRM II was 5.78 (SD 3.52, range
0–15). Among the risk factors evaluated by the tool, vertigo
(49%), incontinence (48%), depression (46%) and confusion
(32%) were the most prevalent risk factors screened on the
elderly (Table 1), consistently with the results of other