To the Editor: O’Grady and Noland, concerned about the consequences of privacy legislation,1 draw attention to “. . . the findings of an Australian survey in which 61% of adults believe that even their de-identified health information should not be used for research purposes without their consent”.

Unfortunately, they do not point out that the survey,2 commissioned by the Office of the Federal Privacy Commissioner, had a 20% response rate, making it effectively useless in determining what Australian adults really think about the use of their de-identified health information.

Alarmed at the possibility that “evidence” of this quality could be used to aid decision-making that had important implications for rigorous research, I turned to Google (<www.google.com>) for assistance. A Google search using the words “61% de-identified health information roy morgan” generated 21 hits, 15 of them unique, and only five related to the subject. One of the five was the letter by O’Grady and Noland, one was the report of the survey, and three specifically cited this survey result: Privacy Victoria,3 Privacy NSW,4 and the Office of the Federal Privacy Commissioner5 all used the result in formal submissions to reviews of privacy-related issues — without revealing the survey’s appalling response rate.

We have privacy commissioners who are powerful advocates of the principles of respect for privacy and autonomy. Perhaps the time has come for Australia to have “public interest commissioners” who can powerfully advocate for the public interest in high-quality health research.

In reply: In a world in which people are increasingly busy and mobile, increasingly concerned about invasions of privacy and increasingly approached to participate in surveys, survey response rates that could be readily achieved 25 years ago are now very much more difficult to attain. High response rates (ie, 60% and over) are still very desirable and can still be achieved. We have conducted surveys on sensitive issues, such as drug-taking, and achieved response rates that would probably satisfy even Arnolda. But this requires very intensive field activity that is not always justified by the nature of the project.

The survey in question was not a health survey. It set out to provide general background information, exploring comparative levels of concern about, and the relationships between, a very wide range of privacy issues on a scale adequate to allow relatively small groups within the population to be examined. It was not intended to yield precise and critical measurements. Measures of “concern” or “reluctance” are highly context-dependent, “soft” measures, subject to interpretation, both by the respondents and by end-users. External validity is therefore not the issue: a response rate of 80% would not have made the figures demonstrably more “accurate”. Given the imprecise nature of the measures obtained, overengineering the sample relative to other components of the survey design would have been a waste of (public) money, better devoted to further research. The survey was part of a wider-ranging project and was planned in close consultation with the Office of the Privacy Commissioner in the light of their needs and priorities.

It is appropriate that this issue arose out of a debate on privacy. As privacy constraints bite harder, whether imposed by statute or codes of practice, or arising through increasing resistance from subjects, medical researchers are going to have to come to terms with problems of non-response in the same way that social-survey researchers have.

How are researchers going to react when only 20% of potential subjects consent to have their information used? Will they use words like “appalling” and “effectively useless” to dismiss any studies based on such a subset? Or will they perhaps learn to use them, with due caution, for the valuable information they nevertheless contain? One does not have to look very far in the history of medicine or public health to find major advances in knowledge using less than perfect statistics.