Search string terms and limiters applied to databases 3
Calculation of effect size estimates of primary outcomes 4
Table of risk ratio, confidence interval and standard error of studies used for meta-analysis 6
Forest plots of meta-analysis, subgroup meta-analysis and a priori meta-analysis 17
Characteristics of included studies 28
Characteristics of excluded studies 71
Search string terms and limiters applied to databases
(older adult OR older people OR older patient* OR aged OR elderly OR geriatric) AND (information OR counselling OR consultation OR advice OR discuss* OR education OR pamphlet* OR brochure* OR video OR media OR publication OR leaflet* OR internet) AND ( attitude* OR motivation OR intention OR participat* OR program OR prevent* OR adherence OR change OR action OR health beliefs OR awareness) AND (fall* OR fall* risk* OR accidental fall*) AND ( hospital* OR community OR discharge OR transition OR post hospitali*ation)
CINAHL limiters: English language, aged 65+years, human and research articles
Scopus limiters: English, aged, humans and journal. Inverted commas were used for phrasing health beliefs, fall* risk*, accidental fall and post hospitalisation.
Cochrane central register of control trials: No limiter
Calculation of effect size estimates of primary outcomes
For outcome i) The proportion of patients who became fallers and outcome iii) Proportion of patients who had an injurious fall (relative to all patients)
STATA (version 12, college station TX) was used to calculate a relative risk.
An integer of one was added to each cell in the 2x2 contingency table if there was a zero cell so as to obtain a finite odds ratio.
For studies that involved allocation of intervention and control conditions to hospital wards rather than individual patients (eg. cluster randomised trials, parallel control group studies), 95% confidence interval of the ratio was adjusted before pooling using the approach of White and Thomas (White and Thomas 2005)and the intra-cluster correlation coefficient reported by Cumming (2008)study.
Adjustment of 95% confidence interval for clustering used in cluster randomised trials
Step 1: To calculate design effect (DE) for the study:
DE= 1+ (n per cluster-1) x Intra cluster coefficient
Where n per cluster =
Intra cluster coefficient=0.014 from Cumming (2008)
Step 2: To calculate standard error (normal) for the study:
Standard error (normal)=
Step 3: To calculate standard error (corrected) for studies:
Standard error (corrected) =Standard error (normal) x
For outcome ii) Rate of falls, outcome iv) Rate of injurious falls, outcome v) Rate of hospital admission due to falls and outcome vi) Rate of emergency department presentations due to falls
if hazard ratios or incidence rate ratio were not provided, an estimate of the relative rate using the formula for calculating a relative risk (Altman and Deeks 2002)was calculated where the number of fallers was replaced with the number of falls in each group and the number of non-fallers with the number of patient days in each group.
Previous research has shown that this relative rate approach produced similar point estimates to survival analysis and negative binomial analysis approaches, however the relative rate approach produces 95% confidence intervals that are too narrow in range (Haines and Hill 2011). To account for this, an inflation factor was determined from two trials included in this review that had the same patient-level data that was calculated from a negative binomial regression (Haines 2004 and 2011). An inflation factor of 1.24 and 1.39 was required and therefore we decided to multiply the log natural standard error of relevant estimates by 1.3 in order for its 95% confidence intervals to be the same width as that from the negative binomial regression generated incidence rate ratio. This improved the estimate for inclusion in the meta-analysis (Haines and Hill 2011).These estimates were also inflated to account for the intra-cluster correlation as described above if warranted by the study design.
Table of risk ratio, confidence interval and standard error of studies used for meta-analysis
Table 6.1 Studies of hospitalized older adults (Targeted multifactorial fall prevention programme that consisted of educational component)