Issues in the design of the randomised crossover trial
We welcome with interest the Christmas Special article by Rawlins et
al  on cardiopulmonary resuscitation training (CPR). Many snow-
shovellers suffer heart attack around this time, reputedly due to holding
their breath during the “shovel effort” so as to increase shovel-force via
a so called water-hammer effect. The importance of effective training in
CPR, especially at Xmas, therefore cannot be overemphasised and any
intervention that increases the effectiveness of training is worthwhile.
We feel the authors’ dismissal of Nellie The Elephant (NTE) as an
element of training may be premature. This is because there are some
issues of the design and interpretation of this crossover trial that may
have been missed. We are told of NTE’s success in delivering a
statistically significant greater proportion of trainees capable of 95 –
105 compressions/min (32% v only 12% and 9% with alternative
interventions). Music narrowed the distribution of compression rates (IQR
7 for TTW and 9 for NTE v 22 for no music) and NTE brought the centre of
the range closer to the desired 100/min (NTE 105 v 110 for TTW and 109 for
no music). However NTE v no music was then dismissed because too many
compressions were of inappropriate depth (86% NTE v 76% no music).
It appears music helps the rhythm but detracts from the depth. We
suggest that different (and possibly overlapping) elements of “music”
interventions may be involved:
(a) in narrowing the range of compressions/min; (b) in bringing median
compression rate close to 100; (c) in allowing correct depth of
compression. Rather than dismissing NTE, the positive elements of this
particular masterpiece should be built upon; for example if NTE was
delivered with a strong bass perhaps depth of compression would be
improved; only a small improvement is needed to equate with no music.
We disagree with the authors in their statement that blinding
investigators was impossible in this trial. Blinding of the researchers
can be done and should actually have been done. Moreover, if there is no
carryover, the trial should estimate the same treatment effect as parallel
group trials . This study did not report any parallel outcomes.
Academically speaking, crossover design should meet the following
- The “condition” should be chronic, incurable and stable. Clearly
the condition here is the inexperience in performing appropriate CPR.
Although this condition could be viewed as chronic, it is easily curable
and rapidly changes depending on training frequency. Obviously, it does
not fit well with this criterion. If the condition is rapidly changing and
treatments curative, cross-over trials may be infeasible.
- The issue of "order" effects. It is possible that the order in
which treatments are administered may affect the outcome. Previous
exposure to NTE in childhood may exert a greater influence if NTE is
- The “carryover” effect: For example, if a subject is changed from
intervention (a) to intervention (b), the observation under intervention
(b) will only be valid if there is no residual carryover from intervention
(a). There must be enough of a washout period to be sure none of the
intervention (a) or its effects actually remains. One minute washout (as
it is here), does not really seem long enough in this trial.
In conclusion, other hits by “Ralph Butler” and “Peter Hart” should
be investigated together with whatever “hit” tops the Xmas charts each
year as potentially the best current training intervention for that year.
Dr Yaser Adi(email@example.com)
Dr Martin Connock
Rawlins L ,Woollard M ,Williams J ,Hallam P. Effect of listening
to Nellie the Elephant during CPR training on performance of chest
compressions by lay people: randomised crossover trial.BMJ 2009;339:b4707,
doi: 10.1136/bmj.b4707 (Published 13 December 2009)
 Higgins JPT, Green S (editors). Cochrane Handbook for Systematic
Reviews of Interventions Version 5.0.1 [updated September 2008]. The
Cochrane Collaboration, 2008. Available from www.cochrane-handbook.org.
 Jadad A. Randomised Controlled Trial. BMJ Books, London, 2000.pp
 Gordis L. Epidemiology, 4th Edition. Philadelphia, W.B. Sounders,
2008 pp 140.
Competing interests: No competing interests