To protect research subjects through prompt remedial actions
This applies to all kinds of research, whether or not risky interventions are undertaken. ‘Adversity’ may also be interpreted in a broader sense to include both medical and non-medical events (such as social or financial harm).
The fact that ‘something bad’ has occurred to someone when they are a subject in your study means that there is a possibility, however tiny, that they could have been in a better position had they not participated in your study, or if you had conducted your study in a different manner.
For example, although a ruptured appendix is very unlikely to have been caused by answering a 15-minute-long questionnaire, a subject might have received medical treatment fifteen minutes earlier if (s)he had not agreed to spend that time answering your questionnaire; or if you had not been so eager to complete the questionnaire, so much so that you failed to realise that (s)he was unwell.