Smoke detectors and house fires

Three themes recur in injury prevention: the need to implement fully what is already known, a preference for passive strategies over active ones, and pressure to evaluate new programmes formally. It is unusual for all to be reflected in a single paper, but each is evident in the report by DiGuiseppi et al in this issue (p 995).¹ It is also unusual for a report to be as flawless as this one seems to be. The scientific literature is plagued with overworked phrases such as “landmark” and “milestone,” yet this study describing the results of a cluster randomised trial of a distribution programme for smoke detectors fully deserves such accolades. To have evaluated a safety programme by using this immaculate design is a huge credit to the investigators and their funding bodies. It is also to the credit of this journal and its reviewers to publish a report whose findings are “negative.” Or are they?

For evangelists of injury prevention everywhere the greatest challenge is to implement what has been shown to be efficacious.² Smoke detectors must surely rank high on the list of measures that appear to work, alongside seat belts or bike helmets, although none has been subjected to the rigours of a randomised trial. Nevertheless, it is widely agreed that these devices are efficacious. What is not known is how effective they are under specific circumstances, such as when the target is a population at high risk. The central finding of this large and complex undertaking is that simply distributing smoke alarms to such a population does not protect it against fires if the devices are not operational. This finding should surprise no one. It could easily have been assumed by health authorities that the benefits of a distribution programme were so self evident that the programme need not be studied. The sobering finding demands further explanation.

One possibility is that the study groups were not large enough. When asked questions about sample size, pundits often suggest that you estimate the required number and then double or triple it—and, even when you do you are still likely to come up short. This study had enough power to detect an intervention effect as large as that found in a previous study on which it was modelled—that is, an 80% reduction.³ From a public health perspective, however, a much smaller reduction would be a great success, and ideally the sample should have been large enough to be able to detect it. But in light of the direction of the estimates, a much larger sample is unlikely to have yielded a different result.

A more likely explanation is that the effectiveness of any safety measure depends on whether it is used as intended. No one would declare a new drug to be useless if it were not taken in the proper dose or proper manner. The same applies to smoke detectors. The alarm prompts action that removes potential victims from danger.

One should remember that this remarkable report was not intended as a trial of the efficacy of smoke detectors. It aimed at examining a specific distribution programme, which, in spite of the earlier study, was well justified because it was not known if the same findings would apply to the low income, multiethnic communities in London.³ Moreover, DiGuiseppi and her colleagues noted a flaw in the design in the study by Mallonee. As this was an observational, not experimental, study the choice of groups made regression towards the mean a likely explanation for its apparent effectiveness.

What makes the conclusion by DiGuiseppi and her colleagues so frustrating and disappointing is that, in contrast, their study was exemplary in every conceivable respect. They took account of all the niceties of the cluster randomised trial—a situation where opportunities for inappropriate statistical assumptions abound. Specifically, they examined clustering effects and showed them to be negligible. They took steps to provide for the cultural differences between their target group and those in the American study. They followed the CONSORT guidelines to the letter.⁴ This is truly a model of how every safety programme should be evaluated—and evaluated they must be because not only are precious resources wasted on ineffective programmes but in some situations programmes may prove unexpectedly harmful.⁵

So what went wrong? For one reason or another, in spite of all efforts to improve adherence, too few of the distributed detectors were working properly when inspected. This reflects the need for more passive measures and less reliance on active—for example, educational—approaches. Like seat restraints, which must be properly fastened on each trip, smoke detectors require periodic checks to ensure that they are working properly.

Clearly, however, this population, for reasons that remain unclear, failed to adhere to the seemingly simple steps needed to install and maintain the detectors. How then can we prevent house fires among poor or elderly people, or homes with children? These results show that a simple give away is not the entire answer. For this population at least it may be necessary for public health officials to use more passive detectors, install and maintain them, or to require sprinklers in all new housing for people with low incomes. Disappointing as the results must be for this team, encouraging byproducts of their work are new directives from the local housing authority in the United Kingdom urging some such actions.⁶

Papers p 995