A randomised fractional factorial screening experiment to predict effective features of audit and feedback

Background

Audit and feedback aims to improve patient care by comparing healthcare performance against explicit standards. It is used to monitor and improve patient care, including through National Clinical Audit (NCA) programmes in the UK. Variability in effectiveness of audit and feedback is attributed to intervention design; separate randomised trials to address multiple questions about how to optimise effectiveness would be inefficient. We evaluated different feedback modifications to identify leading candidates for further “real-world” evaluation.

Methods

Using an online fractional factorial screening experiment, we randomised recipients of feedback from five UK NCAs to different combinations of six feedback modifications applied within an audit report excerpt: use effective comparators, provide multimodal feedback, recommend specific actions, provide optional detail, incorporate the patient voice, and minimise cognitive load. Outcomes, assessed immediately after exposure to the online modifications, included intention to enact audit standards (primary outcome, ranked on a scale of −3 to +3, tailored to the NCA), comprehension, user experience, and engagement.

Results

We randomised 1241 participants (clinicians, managers, and audit staff) between April and October 2019. Inappropriate repeated participant completion occurred; we conservatively excluded participant entries during the relevant period, leaving a primary analysis population of 638 (51.4%) participants.

None of the six feedback modifications had an independent effect on intention across the five NCAs. We observed both synergistic and antagonistic effects across outcomes when modifications were combined; the specific NCA and whether recipients had a clinical role had dominant influences on outcome, and there was an antagonistic interaction between multimodal feedback and optional detail. Among clinical participants, predicted intention ranged from 1.22 (95% confidence interval 0.72, 1.72) for the least effective combination in which multimodal feedback, optional detail, and reduced cognitive load were applied within the audit report, up to 2.40 (95% CI 1.88, 2.93) for the most effective combination including multimodal feedback, specific actions, patient voice, and reduced cognitive load.

Conclusion

Potentially important synergistic and antagonistic effects were identified across combinations of feedback modifications, audit programmes, and recipients, suggesting that feedback designers must explicitly consider how different features of feedback may interact to achieve (or undermine) the desired effects.