The case for open data: the Duke clinical trials

A recent story in the Times Higher Educational Supplement, backed up by leader comment, provides a highly readable summary of a long and complex case of flawed clinical research and the difficulties encountered by those trying to expose the flaws. It also provides a strong argument for being open with data and code at an early stage, even where sensitive data is involved.

Since this research involved cancer chemotherapy, the lives of people and their quality of life whilst undergoing treatment potentially depended on the truth of the research findings. As the article shows, falsifying the findings would have been far easier and quicker had the original data, and the methods used to analyse it, been made available from the outset. Expensive clinical trials could have been avoided. Potentially, better treatments could have been brought to trial more quickly once the false promise of this particular intervention was clear.

It's often felt that whilst some subjects may be prime candidates for openness with data, those involving human subjects, and in particular clinical medicine, present too many ethical and regulatory challenges. Examples such as this show that such a position is wrong. Even if ethical and regulatory barriers exist, wider ethical issues - the avoidance of unnecessary human suffering being one - demand that we be as open as possible with clinical data. In this case, no identifying information needed to be released to allow others to validate or invalidate this work. Even when the inclusion of identifying information is inescapable, data can still be open in the sense that its existence is public and it is made available to anyone who can satisfy the necessary requirements of maintaining the confidentialiy of the individual subjects.

The key story here is not about the original flawed research; such flaws are to be expected since humans are themselves fallible. It is the difficulty others faced in demonstrating those flaws. Some of those difficulties weren't to do with open data, but with such things as the unwillingness of journals to print negative findings. But what's clear is that open data and code at the outset would have saved time, money and human experimentation, as well as protecting the reputation of the otherwise promising field of personalised medicine. There are lessons here for researchers and for the institutions which host them, both of whose reputations suffer from episodes such as this.