University of Canberra researchers ran Evidano three times to cross-check a modified Delphi analysis

Researchers at the University of Canberra and Queensland University of Technology wanted to establish practical guidance for training the foot’s spring mechanism during landing. This mechanism allows the medial longitudinal arch to absorb, store, and return energy as the foot deforms under load and recoils. The study was published in the Journal of Science and Medicine in Sport.

The team used a modified Delphi design. Twenty-eight experts, including clinicians, biomechanists, an exercise physiologist, a sport scientist, and an athletic trainer, completed all three rounds. Participants answered open-ended questions about optimal foot function, proposed exercises and prescription parameters, and rated candidate recommendations on five-point agreement scales.

The panel brought substantial experience to the study: participants averaged 16.6 years of relevant experience, 21 had authored relevant peer-reviewed work, and 19 were registered health practitioners.

The primary investigator first analysed the open-ended responses using Braun and Clarke’s thematic-analysis approach. Evidano was then used to check the themes generated from round-one responses about optimal foot function.

The researchers did not rely on a single AI output. They analysed the data three times, randomizing the order of participant responses on each run to reduce the possibility that input order affected the result. The human- and AI-generated analyses were then compared to reduce the likelihood of researcher bias.

The paper reports that the human analysis produced four key themes and Evidano produced five. The resulting thematic structure covered arch compression and recoil, active and passive foot structures, coordination and alignment, energy recycling, and proprioception and balance.

Across three rounds, 23 of 46 exercises reached consensus. The accepted exercises tended to be multi-joint, functionally integrated movements, including plyometric exercises that resemble the way the foot absorbs and returns energy during landing.

The researchers generated 54 exercise-prescription statements from the open-ended responses. Twenty-one reached consensus of agreement and one reached consensus of disagreement. The accepted prescriptions broadly aligned with American College of Sports Medicine guidance and with parameters used in plyometric-training research.

The paper’s practical contribution is a consensus-informed starting point for clinicians and exercise professionals who want to train foot-spring function. Its methodological contribution is equally useful: it shows how an AI-assisted second pass can be stress-tested through repeated runs and then compared with a human analysis before the findings are used.

This is a particularly transparent example of AI-assisted triangulation. The researchers disclosed the tool, described exactly where it entered the workflow, reran the analysis to test order sensitivity, and kept both thematic interpretation and Delphi consensus with human experts.

It also extends Evidano’s published-use record into sport science and modified Delphi research, a method where open-ended expert input must be transformed carefully into statements that a panel can evaluate.

The authors note that qualitative analysis necessarily involves researcher interpretation. Evidano was used to reduce the risk that one researcher’s framing dominated the themes, not to eliminate interpretation. The study also cautions that most experts worked in Western, developed-country settings, especially Australia, so the consensus may not transfer unchanged to every clinical or training context.