Privacy for research participants

Privacy for medical research participants is protected by several procedures such as informed consent, compliance with medical privacy laws, and transparency in how patient data is accumulated and analyzed.

People decide to participate in research for any number of different reasons, such as a personal interest, a desire to promote research which benefits their community, or for other reasons. Various guidelines for human subject research protect study participants who choose to participate in research, and the international consensus is that the rights of people who participate in studies are best protected when the study participant can trust that researchers will not connect the identities of study participants with their input into the study.

Many study participants have experienced problems when their privacy was not upheld after participating in research. Sometimes privacy is not kept because of insufficient study protection, but also sometimes it is because of unanticipated problems with the study design which inadvertently compromise privacy. The privacy of research participants is typically protected by the research organizer, but the institutional review board is a designated overseer which monitors the organizer to provide protection to study participants.

Information privacy

Researchers publish data that they get from participants. To preserve participants' privacy, the data goes through a process to de-identify it. The goal of such a process would be to remove protected health information which could be used to connect a study participant to their contribution to a research project so that the participants will not suffer from data re-identification.

Privacy attacks

A privacy attack is the exploitation of an opportunity for someone to identify a study participant based on public research data. The way that this might work is that researchers collect data, including confidential identifying data, from study participants. This produces an identified dataset. Before the data is sent for research processing, it is "de-identified", which means that personally identifying data is removed from the dataset. Ideally, this means that the dataset alone could not be used to identify a participant.

In some cases, the researchers simply misjudge the information in a de-identified dataset and actually it is identifying, or perhaps the advent of new technology makes the data identifying. In other cases, the published de-identified data can be cross-referenced with other data sets, and by finding matches between an identified dataset and the de-identified data set, participants in the de-identified set may be revealed. This is particularly the case with medical research data because traditional data anonymization techniques designed for numerical data are not as effective for the nonnumerical data contained in medical data, such as rare diagnoses and personalized treatments. Thus, in cases like medical research data that contain unique nonnumerical data, only removing identifying numerical features, such as age and social security number, may not be enough to mitigate privacy attacks.

Risk mitigation

The ideal situation from the research perspective is the free sharing of data. Since privacy for research participants is a priority, though, various proposals for protecting participants have been made for different purposes. Replacing the real data with synthetic data allows the researchers to show data which gives a conclusion equivalent to the one drawn by the researchers, but the data may have problems such as being unfit for repurposing for other research. Other strategies include "noise addition" by making random value changes or "data swapping" by exchanging values across entries. Still another approach is to separate the identifiable variables in the data from the rest, aggregate the identifiable variables and reattach them with the rest of the data. This principle has been used successfully in creating maps of diabetes in Australia and the United Kingdom using confidential General Practice clinic data.