Much epidemiological and health services research relies on surveys and Jousilahti et al’s paper¹ is a timely reminder to us about the biases caused by participants’ non-response. Although we all know that non-response is a socially patterned phenomenon, this knowledge is so familiar that it is easy to become complacent—how many of us have ticked the mental boxes of “>70%” and “information about response biases” as markers of a good quality survey when reading or reviewing a paper? Jousilahti et al’s paper is valuable in that it provides data from several national, large scale, well conducted Finnish surveys with high response rates, and it shows the differences in health outcome, in this case mortality, according to responder status. It is sobering to realise that even with an aggregate response rate of >80% findings may still be subject to substantial bias.

As if Jousilathi et al’s findings were not important enough, with their implications for the conduct and interpretation of population surveys, their study also has wider relevance. Unlike Finland, where studies such as this are possible because of the comprehensive records infrastructure (a national register, permissions for record linkage), other countries, such as the UK, Australia, Canada, and the USA, are moving towards a requirement of individual consent for the use of identifiable medical data in health research.^2–4 Hence in these countries investigators are experiencing increased difficulties in identifying unbiased samples for cross sectional surveys and cohort and case-control studies and in assembling accurate, comprehensive datasets for records based analysis. Published evidence has already shown the hugely detrimental effect of a requirement of individual consent.^5,6 However, the regulatory authorities seem unconvinced that bias in data is a legitimate concern and maintain that they are safeguarding the rights of the patient, for example:

“I would have to say that I’ve heard an argument made and I’ve heard a lot of anecdotal evidence that not having the full sample somehow badly skews epidemiological research. Not to say it is not out there, but I have not seen the persuasive evidence that this is in fact the case.”⁷“We have received a number of applications for Section 60 support to cover research studies where it has been argued that it would not be appropriate to seek consent from patients because failure of any patients to agree to participate in the study may bias the research findings. ... It is not sufficient for applicants to simply state that certain groups are less likely to provide consent; we seek evidence to justify such comments.”⁸

Perhaps this paper will help provide the evidence that bodies such as these seek. If individual consent for all secondary uses of health data becomes the established norm then most public health and health services population based research will be rendered impossible, to the detriment of the health evidence base. To prevent this, the public health community needs to convince those outside the research arena that existing (and recent) uses of identifiable health data are in the public interest.