Letters

Effectiveness of hip protectors

Results may not be generalisable to the community

Effectiveness of studied hip protector was uncertain

Some clarifications would be useful

Authors' reply

Results may not be generalisable to the community

EDITORThe hip protector trial reported by Meyer et al is methodologically a notable improvement on previous hip protector trials.¹ However, some methodological shortcomings affect the study, and issues arise about its generalisability to people at high risk who are not resident in nursing homes.

An important problem, not noted by the authors, is that the study groups have differential loss to follow up. In the intervention group 64% of the participants completed the 18 month follow up compared with only 57% of the controls (P=0.04). This difference can introduce selection bias and could give a false estimate of effectiveness. In addition, there seems to be some evidence that the control group may have been frailer than the intervention group as the death rate was somewhat greater and this may have explained the higher incidence of falls in the control group. These differences could explain some of the apparent effectiveness of the hip protectors.

Interestingly, the compliance rate for the hip protectors was very low, at only 35%, not 68% as implied in the abstract. This low compliance rate is similar to that in our ongoing community study in which we simply post hip protectors to participants at high risk.

Finally, although these data may support the use of hip protectors among nursing home residents, evidence for their effectiveness among older people at high risk who are living in the community is still required. We are undertaking a large individually randomised trial among 4200 women at increased risk of hip fracture living in the community, the results of which will be reported this summer.

David Torgerson, director. 
York trials unit

Jill Porthouse, trial coordinator, primary care hip protector study. 
University of York, York YO10 5DD

Competing interests: None declared.

Effectiveness of studied hip protector was uncertain

EDITORThe trial reported by Meyer et al left the main issue of effectiveness of the studied hip protector unanswered.¹

Firstly, many important baseline characteristics or risk factors of hip fracture were not reported and included in the analysis. The hip protector group and the control group were not directly comparable since more controls could not be followed up to the end of the trial and had a (29%) higher incidence of falling than people in the protector group (both variables referring to more frailty in the controls).

Secondly, since the difference in the risk of hip fracture in the two groups was non-significant the preventive effect of the hip protector remained uncertain. This is also seen as an open end in the 95% confidence interval of the calculation of the number needed to treat. The authors' speculation about using a one sided hypothesis is not on firm ground: statisticians would not accept one sided hypothesis because at least two previous trials of the studied protector model have shown no effect. ^{2 3}

Thirdly, although a possible cluster effect was taken into account, analysis without cluster randomisation was not provided. It therefore remained unknown whether adjustment for this procedure had any effect on the relative risk of fracture and its P value. For falls, it had little effect.

Fourthly, since Meyer et al could not provide the fracture data for protected and unprotected falls in the two groups, the biomechanical efficacy of the studied protector in actual falls remained unclear. This information would also have been essential in interpreting the given relative risk of hip fracture, since the claimed risk reduction of 40% on an intention to treat basis in the protector group can be true only if the risk of falls was clearly reduced: about half of the falls in the protector group occurred without the protector.

Pekka A Kannus, chief physician. 
Accident and Trauma Research Center, FIN-33500 Tampere, Finland kipeka{at}uta.fi

Competing interests: None declared.

Some clarifications would be useful

EDITORMeyer et al examined factors influencing the use of hip protectors in nursing homes.¹ The analysis was adjusted for the effects of clustering, and the study provides stronger evidence for effectiveness of hip protectors than previously published studies.² We hope that the authors can provide further information to help others in applying the findings of the study.

What proportion of nursing home residents received (and understood) the educational intervention and what proportion were simply encouraged to wear hip protectors by the staff who had received the education are unclear. In Australia and other countries most residents of nursing homes have severe cognitive impairment, limiting their participation in decision making about the use of hip protectors. We would be interested to know what percentage of participants in this study had severe cognitive impairment.

The cost of hip protectors is clearly a disincentive to their use in countries such as Germany where they are not subsidised. Could Meyer et al speculate on how much of the effect of the intervention was due to the supply of free hip protectors and how much to the educational session? Were three pairs of hip protectors enough for the 15 months of the study? In our experience more than three pairs would be required if they are used regularly.

Reasons for non-adherence to using hip protectors should be elaborated as measures to address these might improve the limited adherence rates reported in this study. Some other methodological issues could be clarified, such as how hip fractures were ascertained.

Clarification of these issues would help clinicians working with older people at high risk of hip fractures. Several review articles recommend hip protectors for older people who have had multiple falls,³ and further evidence based information is necessary to ensure their appropriate use.

Susan E Kurrle, senior staff specialist geriatrician. 
Hornsby Ku-ring-gai Hospital, Hornsby, NSW 2077, Australia kurrle{at}bigpond.com

Ian D Cameron, chair, rehabilitation medicine. 
Rehabilitation Studies Unit, University of Sydney, PO Box 6, Ryde, NSW 1680, Australia

Competing interests: None declared.

Authors' reply

EDITORTorgerson and Porthouse and Kannus are wrong in assuming that the two groups were not comparable. Our optimal randomisation procedure ensured the comparability of groups as shown in similar baseline characteristics. There were no losses to follow up.

The trend to fewer falls and lower mortality associated with longer observation in the intervention group could be a result of the intervention. Analyses based on survival times taking into account different follow up times as well as cluster randomisation by using either (a) a Cox model with frailty or (b) a Cox model with robust score test to allow for correlation within clusters would yield the following estimated hazard ratios and two sided P values for the primary outcome of hip fracture: (a) hazard ratio 0.53, P=0.028; (b) hazard ratio 0.52, P=0.066. These results are similar to the reported results of the cluster adjusted ² test (relative risk 0.57, P=0.072); analysis not taking cluster randomisation into account would yield a lower P value (P=0.034).

When we planned our study there was already strong evidence from a randomised controlled trial that Safehip protectors prevent hip fractures.¹ Therefore, we chose a one sided hypothesis but we reported results for two sided 5% tests according to BMJ policy. We used the Safehip protector because it was the only protector available when we initiated the study. The studies cited by Kannus are small trials lacking the power to detect differences between groups, ^{2 3} and one is published in a supplement.³ In our study no hip fractures occurred with unambiguously documented use of the protector in either group.

Adherence was assessed by documenting hip protector use during a fall. We could not document how many of the residents who did not fall during the study were using the protector. Therefore, the proportions of residents who used the hip protector are worst scenario estimates based on the assumption that no resident without falls had used the hip protector. We included these figures on the request of the BMJ reviewers. Since these data are prone to misinterpretation we still think that they should not have been reported.

We have no separate data on the proportion of participants with severe cognitive impairment or on the proportion of residents who had participated in the programme in small groups. For further details on protected and unprotected falls, assessment of fractures (not blinded), and characteristics of the complex intervention programme we refer readers to the full paper on bmj.com. Space limitations prevent us discussing the reasons for non-adherence.

Gabriele Meyer, research fellow. 
Andrea Warnke, research fellow. 
Ingrid Mühlhauser, professor. 
Ingrid_Muehlhauser{at}uni-hamburg.de Unit of Health Sciences and Education, University of Hamburg, Martin-Luther-King-Platz 6, D-20146 Hamburg, Germany

Competing interests: AW was formerly an employee and is at present a consultant of Rölke Pharma, the German distributor of Safehip. AW and GM have received travel grants from Rölke Pharma.