Comparison communities in a cluster randomised trial innovate in response to 'being controlled'

We conducted qualitative interviews among primary health care teams and community agencies in eight communities in Victoria, Australia which had (1) agreed to be part of a universal primary care and community development intervention to reduce post natal depression and promote maternal health; and (2) were randomised to the comparison arm. The purpose was to document their experience with and interpretation of the trial. Although 'control' in a controlled trial refers to the control of confounding of the trial result by factors other than allocation to the intervention, participants interpreted 'control' to mean restrictions on what they were allowed to do during the trial period. They had agreed not to use the Edinburgh Post Natal Depression Scale or the SF 36 in clinical practice and not to implement any of the elements of the intervention. We found that no elements of the intervention were implemented. However, the extension of the trial from three to five years made the trial agreement a strain. The imposition of trial conditions also encouraged a degree of lateral thinking and innovation in service delivery (quality improvement). This may have potentially contributed to the null trial results. The observations invite interrogation of intervention theory and consequent rethinking of the way contamination in a cluster trial is defined.