Available online at www.sciencedirect.com
Journal of Science and Medicine in Sport 14 (2011) 287–292
Original research
The impact of adherence on sports injury prevention effect estimates in
randomised controlled trials: Looking beyond the CONSORT statement
Evert A.L.M. Verhagen a,∗ , Maarten D.W. Hupperets a,b , Caroline F. Finch b ,
Willem van Mechelen a
a
Department of Public and Occupational Health, EMGO Institute for Health and Care Research, VU University Medical Center, Amsterdam, The Netherlands
b Monash University Accident Research Centre, Monash University, Melbourne, Vic, Australia
Received 28 September 2010; received in revised form 15 February 2011; accepted 18 February 2011
Abstract
Objective: To investigate estimated outcome effects of a sports injury prevention intervention when analysed by means of a per protocol (PP)
analysis approach. Design: Randomised controlled trial (RCT) involving 522 athletes who sustained a lateral ankle sprain allocated to either
an intervention (received a preventive programme in addition to usual care) or control group who were followed prospectively for one year.
Methods: Secondary analysis of data relating to registered ankle sprain recurrences, exposure and adherence to the allocated intervention using
a PP analysis approach. Results: Twenty-three percent of the RCT intervention group indicated to have fully adhered with the neuromuscular
training programme. A per protocol analysis only considering fully adherent athletes and control athletes, showed a Hazard Ratio of 0.18
(95% CI: 0.07–0.43). Significantly fewer recurrent ankle sprains were found in the fully adherent group compared to the group that was not
adherent (relative risk = 0.63; 95% CI: 0.43–0.99). Conclusions: A PP analysis on fully adherent athletes versus control group athletes showed
that the established intervention effect was over threefold higher compared to an earlier intention-to-treat based analysis approach. This shows
that outcomes of intervention studies are heavily biased by adherence to the allocated intervention.
© 2011 Sports Medicine Australia. Published by Elsevier Ltd. All rights reserved.
Keywords: CONSORT statement; RCT; Injury prevention; Adherence
1. Introduction
There has been increasing recognition of the need to formally evaluate the preventive capabilities of a range of sports
injury preventive interventions. The sequence of prevention1
has been widely used in sports injury prevention research.
More recently, the model has been extended to emphasise
the need for both efficacy and effectiveness studies as well as
an increased focus on actual behaviours within the context of
implementing preventive interventions.2 From a theoretical
and quality evidence perspective, the best way to evaluate
the effect of any preventive intervention is by performing a
randomised controlled trial (RCT).3
Although the RCT is considered the gold standard in
research,4 there are some important considerations and lim∗
Corresponding author.
E-mail address: e.verhagen@vumc.nl (E.A.L.M. Verhagen).
itations when the methodology is applied to sports injury
prevention interventions.5 Amongst others, most sports
injury prevention RCTs require athletes to adopt some form
of behaviour change for injury reduction benefits.6 The
long follow-up time that is required to demonstrate such
behaviour change can result in significant loss to followup.7–9 Another important consideration that determines the
outcome of an RCT is adherence to the intervention or the
required behaviour change. Adherence in sports injury prevention research is a term used to indicate the athlete’s correct
following of the prescribed intervention. Large cohorts are
harder to control and it can be difficult to directly influence
the behaviours of individuals. This can lead to compromised
adherence, as there is increased dependence on the athletes’
own motivation levels.
Sports medicine research can be used to exemplify the
importance of intervention adherence on RCT outcome. For
example, multiple studies have studied the effectiveness of
1440-2440/$ – see front matter © 2011 Sports Medicine Australia. Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.jsams.2011.02.007
288
E.A.L.M. Verhagen et al. / Journal of Science and Medicine in Sport 14 (2011) 287–292
exercise training programmes for the prevention of injuries.
Many such studies showed that these training programmes
can prevent injuries.10–15 In contrast, two recent studies have
failed to show a preventive effect, but did report suffering
from poor adherence with the prescribed intervention.16,17
The impact on outcome of ‘non-adherence’ with specific
interventions on injury reductions has also been highlighted
in other randomised controlled trials of lower limb injury
prevention interventions in soccer18–20 and for protective
equipment in Australian Football.21
With respect to analysis of RCT study results, the CONSORT statement advocates intention-to-treat (ITT).22 By
using an ITT approach, RCTs aim to minimise selection bias
and the likelihood that confounding factors influence results.
However, poor adherence to the intervention tends to dilute
treatment effects when analyses go by the ITT approach.23 A
different approach that has a stronger focus on the efficacy of
an intervention is analysis by treatment actually received, i.e.
per-protocol analysis (PP).24 By solely analysing adherers to
the intervention the maximal achievable effect of the intervention is shown. Recently, studies on protective equipment
in football17,25 and studies on injury prevention in soccer17,18
have adopted the PP approach.
What is lacking from the literature is a formal assessment
of the differences in injury prevention outcome effect sizes
derived from a PP approach compared to an ITT analysis,
although one rugby helmet study did present the results of
both an ITT and PP analysis.25 Therefore, the overall aim of
the present study was to investigate the estimated outcome
effects of a preventive sports injury intervention when analysed by means of a PP analysis and to compare this with
prior outcome effects estimated from an ITT approach. For
this purpose, we have used data from a published RCT with
522 participants on the prevention of ankle sprain recurrences
through an 8-week home based unsupervised neuromuscular
training programme.26 The main study design, intervention
and outcomes are described in detail elsewhere.26,27 Briefly,
a RCT was conducted in athletes (n = 522) who sprained
their ankle. Athletes all received treatment according to usual
care, after which they were randomised to a control group or
an intervention group. Athletes allocated to the intervention
group received an additional 8-week proprioceptive training
programme after treatment by usual care. This programme
consisted of 3 proprioceptive training sessions per week, with
a maximum duration of 30 min per session. The full programme is described elsewhere.27 Cox regression analysis
according to the ITT principle showed in this study that the
risk for incident ankle sprain recurrences was significantly
lower in the intervention group as compared to the control
group (HR = 0.63; 95% CI: 0.45–0.88).26
2. Methods
This paper reports a new PP analysis on these data, taking
adherence rates into account. As this adherence data has not
been previously reported, the methods used to collect it are
reported here.
During a one year follow-up, athletes reported all sudden inversions of the same ankle and details of their sports
participation for each training session through monthly questionnaires. The first and second monthly questionnaires of
the intervention participants also contained a Likert item on
adherence with the prescribed neuromuscular training programme: “The last four weeks I performed the exercises of the
programme as prescribed”. This item was answered on a fivepoint scale, ranging from complete disagreement (score = 1)
to complete agreement (score = 5). As no valid measurement
or consensus on adherence to training programmes exists, the
choice to measure adherence through this single Likert item
was arbitrarily. This choice was based on the general ease of
understanding of Likert items, and the general consistency of
participants’ responses on Likert items.
Participants rating the item with a four or a five on both
occasions (i.e. after one and two months) were considered
fully adherent. The term partially adherent was used for participants who scored a four or a five in only one out of the
two occasions. Participants who gave scores of three or lower
on both occasions were considered not to have adhered to the
prescribed programme. Although these three categories were
chosen arbitrarily, we believe that they rule out central tendency bias and differentiate between participants who did
and did not adhere to the prescribed programme. To rule
out spill over or contamination between intervention and
control group participants, control group participants were
asked monthly whether they had participated in neuromuscular training during the past month. These control participants
were excluded from the PP analysis.
Adherence numbers were presented as the absolute number of participants (and percentages) in each category.
Control group participants who performed some form of
neuromuscular training during the one-year follow-up were
presented as a total number and a percentage of the total
number of participants in that group. In the analyses, it was
assumed that 100% of the controls ‘adhered’ with their usual
treatment regimens, which were expected not to include the
exercises used by the intervention group, as these are not part
of usual treatment.
Clinical outcome was ankle sprain recurrence incidence
density and its 95% confidence interval (95% CI), calculated
as the number of incident ankle sprains recurrences per 1000 h
of exposure. Follow-up of athletes ceased after first injury
recurrence.
Missing exposure data were imputed through multiple
imputation by means of Multiple Imputation by the Chained
Equations procedure28 using R statistical software.29 This
procedure leads to a small spread between the imputed values, resulting in low uncertainty about the missing data.
Five multiple imputed datasets were generated and these
were then combined into 1 using Rubin’s rules.30 Participants with missing adherence data were excluded from the PP
analysis.
E.A.L.M. Verhagen et al. / Journal of Science and Medicine in Sport 14 (2011) 287–292
289
Table 1
Number of incident ankle sprain recurrences, mean hours of sports participation (SD), and ankle sprain recurrence incidence densities (95% CI), for the different
groups of interest.
Intervention group
Full adherence
Partial adherence
No adherence
Control group
Full adherence
No adherence
# Participants (% of total)
# Ankle sprain recurrences
Mean hours exposure (SD)
n/1000 h (95% CI)
256
58 (23%)
75 (29%)
89 (35%)
266
261 (98%)
5 (2%)
56
4
21
30
89
89
–
116.1 (93.5)
131.7 (821.8)
138.52 (93.4)
110.8 (84.2)
114.1 (113.5)
111.8 (112.9)
–
1.86 (1.37–2.35)
0.52 (0.01–1.04)
2.02 (1.16–2.89)
3.04 (1.95–4.13)
2.90 (2.30–3.50)
2.95 (2.34–3.57)
–
Cox regression analyses on overall intervention effects
were performed from a PP perspective. For this purpose, the
same modelling strategy was employed as for the original
RCT analysis, i.e. with adjustment for age, type of sport, and
level of sports.26 For the PP analysis, only the fully adherent
intervention participants were taken into account and compared to the control participants. Finally, a Cox regression
analysis was used to compare ankle sprain recurrence incidence density within the intervention group, using adherence
as an independent variable. Differences were considered statistically significant at P < 0.05.
3. Results
Fifty-eight (23%) intervention group athletes indicated to
have fully adhered to the eight-week proprioceptive training
programme; 75 (29%) intervention group athletes indicated
to have been partially adhered; 89 (35%) responded in such a
way that they were classified as not adherent (Table 1). Adherence to the training programme was unknown for 33 (13%)
athletes, since they did not complete the questionnaires. Five
out of 266 control group athletes (2%) reported to have performed some sort of neuromuscular training exercises during
the one-year follow-up. These athletes performed neuromuscular training exercises as part of medical treatment of an
ankle sprain recurrence and were not incorporated in the PP
analysis.
Cox regression analysis from a PP approach showed that
the risk for ankle sprain recurrences was significantly lower
in fully compliant intervention participants when compared
to control group athletes that did not follow any neuromuscular programme during follow-up (HR = 0.18; 95% CI:
0.07–0.43) (Fig. 1).
Ankle sprain recurrence incidence for fully adherent participants was 0.52 (95% CI: 0.01–1.04), whereas nonadherent
athletes showed an incidence of 3.14 (95% CI: 2.04–4.25)
(Table 1). Cox regression revealed that the risk for an ankle
sprain recurrence was significantly lower in the fully adherent group when compared to the partial adherent group
(HR = 0.24; 95% CI: 0.08–0.80) and nonadherent group
(HR = 0.17; 95% CI: 0.06–0.48) (Fig. 2). Ankle sprain recurrence risk for partial adherent participants did not differ
significantly from nonadherent participants (HR = 0.64; 95%
CI: 0.35–1.14).
Fig. 1. Survival plot derived from the cox regression analysis in the PP
analysis.
Fig. 2. Survival plot derived from the cox regression analysis on different
adherence levels within the intervention group.
290
E.A.L.M. Verhagen et al. / Journal of Science and Medicine in Sport 14 (2011) 287–292
4. Discussion
When compared to the ITT analysis outcomes of the RCT
reported by Hupperets et al.,26 these secondary analyses
based on a PP approach analyses showed that the established intervention effect was over threefold higher compared
to when an ITT approach was utilised. The ITT approach
resulted in a Hazard Ratio of 0.63 (95% CI: 0.45–0.88) in
favour of the intervention group, whereas the PP approach
resulted in a Hazard Ratio of 0.18 (95% CI: 0.07–0.43). This
means that the larger part of the ITT outcome is achieved
by only a small subsample of the intervention group that
was fully adherent to the prescribed intervention. Thus this
study clearly demonstrates that when adherence to the intervention is taken into account in the analysis of a sports
injury prevention study, then large significant effect differences can exist within an intervention group. Such important
information regarding effect of a studied sports injury prevention measure is missed in the commonly employed ITT
approach.
The benefit of the ITT principle is that it gives a pragmatic
estimate of the benefit of a change in treatment policy31 and
that it minimises selection bias.24 Furthermore, by including all study participants in the analysis, power of the trial
is improved. However, despite these benefits and despite the
ITT principle being widely used in injury preventive RCTs,
this method has some disadvantages. It tends to dilute the
treatment effect and leads to a loss of power compared with a
situation of full compliance of a total population. The problem is that the ITT analysis includes people who did not
apply the intervention at all. Conversely, perhaps the ITT
principle is a sensible approach, since it is best to be conservative when publishing effect sizes. Although ITT analysis
removes or adjusts for a placebo effect, for some injury interventions, such as wearing protective equipment, a placebo
effect cannot be possible. If people do not wear protective
equipment, they can never be prevented from sustaining significant impact, because there is no physical barrier to absorb
the energy exchange.
A PP analysis is justified if a preventive measure is shown
to be applicable in preventing injuries to further investigate which intervention component creates what effect. The
present study on the effectiveness on ankle sprain recurrences
of neuromuscular training is not the first to establish this
effect. Research has shown that such a training programme is
effective in preventing ankle sprain recurrences,11–16 therefore adherence to the programme is a key to establishing
proper results. A recent RCT on wearing padded headgear in rugby25 illustrated that another reason to adopt a
PP analysis is when an intervention is critical to injury
reduction. By not wearing headgear, there is no physical barrier between the rugby player and the ground or
another player during a collision and the transfer of energy
to the head cannot be prevented. This applies to all sports
injury prevention studies in which a ‘material’ intervention is
utilised.
However, caution is needed when utilising and interpreting
a PP analysis on data derived from a preventive RCT. In fact
the term ‘RCT’ is incorrect and misleading when applied to a
PP analysis as performed in the current study. Participants in
each of the ‘adherence level’ groups that form the basis of the
PP analysis are not randomly allocated to these groups. Moreover, group allocation is arguably to great extent influenced
by any number of factors that might each affect intervention
outcome. Importantly, this is a key reason why ITT is advocated as the optimal approach to analysis of RCTs – assuming
loss to follow-up (and exclusion of participants from analysis)
is minimal, it retains the original groups formed by random
allocation, which should reflect a fairly even distribution of
participants with different levels of ‘potentially confounding
variables’ across the randomly formed groups.
Adherence is the term most commonly used in clinical
RCTs, but in sports injury behavioural studies other terms
such as compliance, adoption,32 uptake,2 or sustainability33
are used as well. In essence, these terms are often used as
synonyms for adherence, since they cover the same construct.
An intervention, whether it is wearing protective equipment
or a training programme, is either ‘used’ by the athlete or
not. This is particularly true in protective equipment interventions, but as was shown in the present study and in other
exercise training programme interventions,17,18 there can be
degrees of usage.
It is possible that the adequacy of our measure of adherence could have contributed to the difference between the
two analyses, but this is unlikely. Though it should be noted
that adherence was measured through only one statement,
which made it impossible to rule out social desirability bias.
Nevertheless, athletes were encouraged to answer questions
honestly and they were made aware beforehand that answers
would never be related to an individual athlete. Despite this
shortcoming, a large effect on ankle sprain recurrence incidence density was shown for athletes who had reported
to adhere to the intervention. Occurrence of spill over or
contamination between intervention and control groups is
unlikely. Only five control group athletes indicated to have
performed neuromuscular training exercises as part of medical treatment of an ankle sprain recurrence that occurred
during the follow-up period. Reported neuromuscular training by these control participants did not affect results as
the outcome measure was ankle sprain recurrence incidence
density and the follow-up consequently ceased after injury
recurrence.
As shown by the present study, assessment of adherence
to an intervention programme and the subsequent reporting
of adherence is of importance for RCTs. Since adherence
is such a critical issue, there should be consensus on how
to measure adherence and how to report it. It is therefore
striking that the CONSORT guideline does not contain a section on adherence measurement and reporting. This has led
to differing approaches on how to measure, report and deal
with adherence.8,17,18 For all RCTs of injury interventions,
we strongly recommend that adherence is assessed at least
E.A.L.M. Verhagen et al. / Journal of Science and Medicine in Sport 14 (2011) 287–292
through multiple adherence questions and at different times
points throughout the study. This would help to ascertain the
accuracy of the information. One way to achieve this would be
by requiring participants complete an adherence log throughout the entire intervention period. A Likert scale containing
five-point Likert items as used in the present study could be
used as a means of scoring the above mentioned questions.
Furthermore, differentiation between full adherence, partial
adherence, and no adherence as presented in the present study
could be adopted in the reporting of compliance. However,
a more continuous adherence scale would be required to be
able to make statements about possible dose-response relationships between adherence and intervention effect.
Low adherence can make statistical power too low to
assess effects of the training programme in intervention
groups through a PP analysis. A recent study on injury prevention in soccer suffered from this shortcoming.17 Since
many cohorts are already large and maximum sample size
capacity is often reached, efforts should be made to minimise
non-adherence. Strategies to increase adherence significantly
need to be explored before the results of ITT studies can
effectively be applied in the real world.
In this study, we have considered the specific case when
the RCT comprised one intervention and one control group.
The issues relating to adherence with interventions become
even more complex when RCTs aim to compare more than
one intervention group, relative to controls. There is potential
for all data from such studies to be of poor quality due to
unanticipated low adherence rates, so that no conclusions
can be drawn about any of the interventions.
A major way to avoid this possibility is in the design phase
of the trial and to adopt a factorial study design.34 An example of the successful use of this design approach was a RCT
aimed at determining the effects of protective headgear and
custom-fitted mouth guards for preventing oro-facial injuries
in Australian football players.5 This study was designed as
a four arm factorial trial: headgear only, mouth guards only,
headgear + mouthguard, control (usual behaviour). Unfortunately, however, too few players wore the assigned headgear
rendering it impossible to assess its effectiveness in this trial.7
Because of the factorial design, it was possible (without loss
of power) to collapse the final study to a two arm factorial
trial and to still formally test the hypotheses relating to the
preventive potential of mouth guards.5 This example also
highlights the importance of only conducting RCTs of interventions which are likely to have good levels of acceptance,
and hence adherence, by the trial participants.
5. Conclusion
Although the CONSORT statement is widely used to guide
the reporting of RCTs in many major journals, in the specific
example of sports injury prevention research additional analysis strategies might also apply in addition to the favoured
ITT. From the present study it can be concluded that under
291
the influence of nonadherence to the allocated intervention,
differences in effect sizes can be found depending on the
type of analysis that is chosen. In determining effect size
of an intervention relative to control it is best to adopt the
ITT principle, since an intervention is eventually applied to
a general public, in which not every patient will adhere to
a preventive programme. Nevertheless, a PP analysis could
be presented alongside ITT analysis to provide more insight
into the effectiveness of the intervention in only those who
adhere to the intervention as per the protocol specifications.
A PP analysis is only possible if measurement of adherence
to the intervention programme is undertaken. Therefore, it
is highly recommended that RCTs incorporate proper adherence measurement strategies.
Practical implications
• Outcomes of intervention studies are heavily biased by
adherence to the allocated intervention.
• In randomised controlled trials a per-protocol analysis conducted alongside an intention-to-treat analyses provides
insight into the efficacy and true potential of the intervention.
• Assessment of adherence to an intervention programme
and the subsequent reporting of adherence is of critical
importance for RCTs.
Acknowledgements
This study is supported by a grant from the Netherlands
Organization for Health Research and Development
(ZonMw), grant number 750-20-002. Caroline Finch was
supported by a National Health and Medical Research Council (of Australia) Principal Research Fellowship (NHMRC
ID: 565900). No author or related institution has received
any financial benefit from research in this study.
References
1. Van Mechelen W, Hlobil H, Kemper HCG. Incidence, severity, aetiology and prevention of sports injuries. Sports Med 1992;14(2):82–99.
2. Finch C. A new framework for research leading to sports injury prevention. J Sci Med Sport 2006;9:3–9.
3. Altman DG. Better reporting of randomised controlled trials: the CONSORT statement. BMJ 1996;313:570–1.
4. Agabegi SS, Stern PJ. Bias in research. Am J Orthop 2008;37(5):242–8.
5. Emery CA. Research designs for evaluation studies. In: Sports injury
research oxford. Oxford University Press; 2010 [chapter 13].
6. Verhagen EALM, van Stralen MM, van Mechelen W. Behaviour, the
key factor for sports injury prevention. Sports Med 2010;40(12):1–8.
7. Verhagen EALM, Van der Beek AJ, Twisk JWR, et al. The effect
of a proprioceptive balance board training program for the prevention of ankle sprains: a prospective controlled trial. Am J Sports Med
2004;32(6):1385–93.
8. Lun V, Meeuwisse WH, Stergiou P, et al. Relation between running
injury and static lower limb alignment in recreational runners. Br J
Sports Med 2004;38:576–80.
292
E.A.L.M. Verhagen et al. / Journal of Science and Medicine in Sport 14 (2011) 287–292
9. Stevenson MR, Hamer P, Finch CF, et al. Sport, age, and sex specific incidence of sports injuries in Western Australia. Br J Sports Med
2000;34:188–94.
10. Bahr R, Bahr IA. Incidence of acute volleyball injuries: a prospective
cohort study of injury mechanisms and risk factors. Scand J Med Sci
Sports 1997;7:166–71.
11. Verhagen EALM, Van der Beek AJ, Bouter LM, et al. A one-season
prospective cohort study of volleyball injuries. Br J Sports Med
2004;38(4):477–81.
12. Stasinopoulos D. Comparison of three preventive methods in order to
reduce the incidence of ankle inversion sprains among female volleyball
players. Br J Sports Med 2004;38(2):182–5.
13. Petersen W, Hohmann G, Pufe T, et al. A controlled perspective
case control study of a prevention training program in female team
handball players: the German experience. Arch Orthop Trauma Surg
2005;125(9):614–21.
14. Wedderkopp N, Kaltoft M, Lundgaard B, et al. Prevention of injuries
in young female players in European team handball. A prospective
intervention study. Scand J Med Sci Sports 1999;9(1):41–7.
15. Tropp H, Askling C, Gillquist J. Prevention of ankle sprains. Am J
Sports Med 1985;13(4):259–62.
16. Engebretsen AH, Myklebust G, Holme I, et al. Prevention of injuries
among male soccer players: a prospective, randomized intervention
study targeting players with previous injuries or reduced function. Am
J Sports Med 2008;36:1052–60.
17. Steffen K, Myklebust G, Olsen OE, et al. Preventing injuries in female
youth football – a cluster-randomized controlled trial. Scand J Med Sci
Sports 2008;36(4):700–8.
18. Kilding A, Tunstall H, Kuzmic D. Suitability of FIFA’s “the 11” training
programme for young football players – impact on physical performance. Clin J Sports Med 2008;7:320–6.
19. Reider B. A popular cult. Am J Sports Med 2008;36:1467.
20. Croisier JL, Ganteaume S, Binet J, et al. Strength imbalances and prevention of hamstring injury in professional soccer players: a prospective
study. Am J Sports Med 2008;36:1469–75.
21. Finch C, Braham R, McIntosh A, et al. Should Australian football players wear custom-made mouthguards? Results from a group-randomised
controlled trial. Inj Prev 2005;11:242–6.
22. Schulz KF, Altman DG, Moher D. CONSORT 2010 statement:
updated guidelines for reporting parallel group randomised trials. BMJ
2010;340:c332.
23. Montori VM, Guyatt GH. Intention-to-treat principle. CMAJ
2001;165(10):1339–41.
24. Becque T, White IR. Regaining power lost by non-compliance via full
probability modeling. Statist Med 2008;27:5640–63.
25. McIntosh AS, McCrory P, Finch CF, et al. Does padded headgear
prevent head injury in rugby union football? Med Sci Sports Excerc
2009;41(2):306–13.
26. Hupperets MDW, Verhagen EALM, van Mechelen W. Unsupervised
home-based proprioceptive training reduces the incidence of ankle
sprain recurrences: report of a randomized controlled trial. BMJ
2009;339:b2684.
27. Hupperets MDW, Verhagen EALM, van Mechelen W. The 2BFit study:
is an unsupervised proprioceptive balance board training programme,
given in addition to usual care, effective in preventing ankle sprain
recurrences? Design of a randomized controlled trial. BMC Musc Dis
2008;9:71.
28. Van Buuren S, Oudshoorn K. Flexible multivariate imputation by
MICE. Technical Report. Leiden, Netherlands: TNO Quality of Life;
1999.
29. R Development Core Team. R: a language and environment for
statistical computing. Vienna, Austria: R Foundation for Statistical
Computing; 2008 [accessed 22.12.10] http://www.R-project.org.
30. Rubin DB. Multiple imputation for nonresponse in surveys. New York,
NY: John Wiley & Sons; 1987.
31. Hollis S, Campbell F. What is meant by intention to treat analysis? Survey of published randomised controlled trials. BMJ 1999;319:
670–4.
32. De Meij JS, Chinapaw MJ, Kremers SP, et al. Promoting physical activity in children: the stepwise development of the primary school-based
JUMP-in intervention applying the RE-AIM evaluation framework. Br
J Sports Med 2008;19 [epub ahead of print].
33. Eime R, Finch C, Wolfe R, et al. The effectiveness of a squash eyewear
promotion strategy. Br J Sports Med 2005;39:681–5.
34. Murray D. Design and analysis of group-randomized trials. New York:
Oxford University Press; 1998.