CSIRO and Queensland Children’s Hospital combined Evidano with implementation-science frameworks

Researchers from CSIRO’s Australian e-Health Research Centre and Queensland Children’s Hospital co-designed a test-result review app for antimicrobial stewardship in a paediatric emergency department.

Microbiology results are often still pending when a patient leaves an emergency department. When those results later show that a prescribed antibiotic may not match the organism, clinicians need to find the case, review it, document action, and contact the patient or treating team. The prototype was designed to prioritize potentially actionable results by cross-referencing microbiology findings, susceptibility information, and antibiotics recorded at discharge.

The qualitative study mapped the existing workflow, gathered feedback on the prototype, and identified factors likely to shape implementation. Five sessions involved five nurse practitioners and three senior paediatric medical officers. Each session lasted about an hour and used simulated cases so participants could inspect or discuss the prototype in context.

Evidano, cited in the preprint under its former name AILYZE, supported several distinct stages. Audio recordings were transcribed in Evidano, and a researcher checked every transcript against the recording and corrected errors before analysis.

Evidano was then asked to identify themes related to the existing workflow, technology-design feedback, and implementation suggestions. The text segments and themes it surfaced guided manual coding of workflow and app-design evidence.

A human-coded transcript was used as a calibration case for CFIR analysis: researchers compared Evidano’s output with their manual CFIR coding before applying the workflow to the remaining material. AI-generated themes and subthemes were then manually checked, modified, regrouped, and validated.

This is not a one-click analysis. It is a layered workflow in which AI output is repeatedly compared with audio, transcripts, manual codes, and established implementation-science frameworks.

The workflow analysis identified activities including checking and assessing results, escalation and consultation, antibiotic review, documentation, sign-off, and patient follow-up. Participants described high result volumes, weak prioritization, incomplete documentation, navigation across multiple systems, and the workload of following up patients after discharge.

Clinicians responded positively to the prototype, especially its visual flagging of possible drug-organism mismatches. Their feedback emphasized integration with existing systems, clear visual prioritization, usability, and the need to fit the tool into real clinical responsibilities rather than create a parallel process.

The researchers identified ten CFIR-based implementation factors spanning suitability, adaptability, design, relative advantage, complexity, technology infrastructure, work infrastructure, local conditions, communication, and the perceived need for change. The resulting ERIC recommendations included tailoring the tool to context, strengthening stakeholder relationships, and using iterative evaluation before wider deployment.

This study demonstrates a richer role for Evidano than theme generation alone. The platform supported transcription, evidence organization, a manually calibrated codebook workflow, and structured translation from qualitative material to implementation strategy.

It is also a strong example of traceable human oversight in a sensitive clinical setting. Researchers checked transcripts against audio, used two-person review, tested AI-to-human alignment on a calibration transcript, and retained clinical and methodological judgment throughout.

This source is a preprint posted on 28 August 2026 and had not completed journal peer review when this case study was prepared. The participant group was small and drawn from one paediatric emergency department. The findings are therefore most useful as a detailed co-design and pre-implementation example, not as proof of clinical effectiveness.

The underlying interview data is not public because of privacy and ethics constraints.