Allocation
We assessed risk of bias for random sequence generation as low
in eight studies (80%) and unclear in the remaining two studies
(20%). We assessed risk of bias for allocation concealment as low
for seven studies (70%) and unclear for the remaining three stu
dies
(30%) (
Figure 3
).
Blinding
With participant recall or active registration of falls by the p
ar-
ticipant themselves through a falls calendar or diary, we asse
ssed
the risk of bias for blinding as high in eight studies (80%) and
as
unclear in two studies (20%) (
Figure 3
). In the latter two studies
(
Sato 2005a
;
Sato 2011
), falls were recorded by nursing staff filling
out standardised fall protocols for institutionalised or ho
spitalised
people after stroke, but nurses recording falls would also be
dis-
pensing medication.
Incomplete outcome data
We scored the risk of bias for incomplete outcome data addressed
as low for eight studies (80%) and high for the remaining two
studies (20%) (
Figure 3
).
Selective reporting
We assessed reporting bias as low in eight studies (80%) and hi
gh
in the remaining two studies (20%) (
Figure 3
).
Other potential sources of bias
We also assessed whether the falls/fallers outcome was ascert
ained
reliably. For this item, we scored the risk of bias as low for se
ven
studies (70%), high for two studies (20%) and unclear for the
remaining study (10%) (
Figure 3
). The seven studies that scored
low all used a falls calendar that had to be returned monthly wi
th
follow-up telephone calls if necessary. The two studies that sco
red
high (
Dean 2010
;
Green 2002
) used retrospective recall of six
months and three months respectively. We assessed
Holmgren
2010
as being at unclear risk because it was not apparent whether
the falls calendar that they used for the six-month follow-up ha
d
to be returned monthly, three-monthly or after six months, an
d if
there were any follow-up telephone calls.
Effects of interventions
Exercises
Our searches identified five studies that evaluated the effect o
f
exercises on falls (
Dean 2010
;
Dean 2012
;
Green 2002
;
Lau
2012
;
Marigold 2005
). As we also identified one multiple trial
(
Holmgren 2010
) and one multifactorial trial (
Batchelor 2012)