Diagnostic stewardship in emergency departments using the UNTIE framework: a stepped-wedge cluster randomised study protocol Authors: Pramod Chandru, Naren Gunja, Simon Poon, James L Mallows, Khanh Nguyen, Richard McNulty, Armando Teixeira-Pinto, Clement Loy Published: 2026 ABSTRACT AbstractIntroductionPathology testing accounts for over half of non-labour costs in Australian Emergency Departments (EDs), yet up to one-fifth of tests are unnecessary. Despite national guidance from the Australasian College for Emergency Medicine (ACEM) and the Royal College of Pathologists of Australasia (RCPA), variation in test ordering persists across departments and across clinical presentations. The Unnecessary Tests in Emergency (UNTIE) study aims to evaluate whether audit and feedback using a novel indicator can reduce unnecessary pathology testing.Methods and analysisUNTIE is a multicentre, stepped-wedge cluster randomised trial conducted across five metropolitan EDs in New South Wales, Australia. All adult ED presentations that fall within the 29 clinical conditions covered by the 2023 ACEM–RCPA guideline will be included. Presentations were identified using routinely collected ED triage diagnoses. Because the ACEM–RCPA pathology guideline specifies clinical presentations rather than diagnostic codes, a mapping process was developed to align triage terms used in the electronic medical record with the guideline categories. Approximately 100 common triage diagnoses were reviewed by a panel of emergency physicians and mapped to the 29 eligible presentations. This mapping dictionary was used to automatically classify all ED encounters during data extraction. Ambiguous or unmapped triage terms were reviewed and assigned by consensus.The intervention comprises a multifaceted audit-feedback programme—combining local champions, education sessions, visual prompts and electronic dashboards. Two automated indicators will be generated for every eligible ED encounter: UNTIE-U (Unnecessary Testing Index) representing the proportion of tests performed but not recommended and UNTIE-N (Necessary Testing Index) representing the proportion of guideline-recommended tests performed. The primary outcome is change in UNTIE-U before and after intervention; the secondary outcome is change in UNTIE-N.Ethics and disseminationEthics approval has been granted by the Western Sydney Local Health District Human Research Ethics Committee (HREC/17/WMEAD/274, 2022/STE03249). All analyses will use de-identified data. Findings will be disseminated through peer-reviewed publication, conference presentation and feedback to participating EDs.Trial registration numberAustralian New Zealand Clinical Trials Registry: ACTRN12623001130651, UTN: U1111-1297-0386 STRENGTHS AND LIMITATIONS OF THIS STUDY Large multicentre stepped-wedge cluster randomised design allows causal inference while ensuring all sites receive the intervention. Large multicentre stepped-wedge cluster randomised design allows causal inference while ensuring all sites receive the intervention. Automated indicators enable near-real-time objective assessment of test-ordering behaviour using routinely collected data. Automated indicators enable near-real-time objective assessment of test-ordering behaviour using routinely collected data. Multifaceted audit-feedback strategy grounded in implementation science principles and delivered at a whole-of-department level. Multifaceted audit-feedback strategy grounded in implementation science principles and delivered at a whole-of-department level. Case identification is based on triage diagnoses, and residual non-differential misclassification may occur, likely biasing effect estimates towards the null rather than generating spurious effects. Case identification is based on triage diagnoses, and residual non-differential misclassification may occur, likely biasing effect estimates towards the null rather than generating spurious effects. Limited to metropolitan New South Wales (NSW) emergency departments; generalisability to regional or differently structured health systems may be reduced. Limited to metropolitan New South Wales (NSW) emergency departments; generalisability to regional or differently structured health systems may be reduced. Background and introduction Diagnostic testing plays a critical role in emergency departments (EDs). Thirteen per cent of Australians attend an ED each year, accounting for 3.59 million presentations in New South Wales (NSW) alone.1 2 Approximately half of all ED visits involve pathology testing. Unnecessary testing not only represents waste but can also lead to adverse patient and system outcomes, including prolonged length of stay and resource diversion.2 3 It is estimated that up to 20% of pathology tests are unnecessary, with studies reporting ranges between 12% and 44%. Efforts to systematically reduce unnecessary testing have been hampered by two key barriers: first, a lack of consensus on what constitutes ‘appropriate’ testing, and second, the absence of a readily measurable indicator. Several strategies have been shown to reduce low-value testing in specific contexts, including request restrictions,4 education programmes and audit-feedback interventions.5 A recent scoping review confirmed that targeted interventions within EDs can achieve meaningful reductions in test ordering.5 The joint Australasian College for Emergency Medicine (ACEM) and Royal College of Pathologists of Australasia (RCPA) guideline provides a timely opportunity to overcome previous barriers by establishing consensus on appropriate testing for clinical presentations and offering a foundation for developing measurable performance indicators in regard to clinical test ordering behaviour. Study rationale and objectives The ACEM–RCPA Pathology Testing in the Emergency Department guideline uses a ‘traffic-light’ classification system, categorising tests for each of 29 common ED presentations as ‘Perform’ (green), ‘Consider or Ask Supervisor’ (yellow) or ‘Not Generally Indicated’ (red). Building on this framework, we developed two automated indicators to quantify test-ordering behaviour: UNTIE-N (Unnecessary Tests in Emergency–Necessary): The percentage of tests listed as ‘Perform’ for a specific presentation that were actually performed during the ED visit. A score<100% indicates potential underutilisation of necessary tests. UNTIE-N (Unnecessary Tests in Emergency–Necessary): The percentage of tests listed as ‘Perform’ for a specific presentation that were actually performed during the ED visit. A score<100% indicates potential underutilisation of necessary tests. UNTIE-U (Unnecessary Tests in Emergency–Unnecessary): The percentage of tests performed that were not listed as ‘Perform’ or ‘Consider/Ask Supervisor’ for that presentation, divided by the number of ‘Perform’ tests. A score>0% indicates potential overutilisation. UNTIE-U (Unnecessary Tests in Emergency–Unnecessary): The percentage of tests performed that were not listed as ‘Perform’ or ‘Consider/Ask Supervisor’ for that presentation, divided by the number of ‘Perform’ tests. A score>0% indicates potential overutilisation. These indicators will be generated automatically on a fortnightly basis using routinely collected data and displayed on an interactive dashboard to provide real-time feedback on departmental performance across the 29 clinical presentations. This study aims to implement a multifaceted audit-feedback programme—co-designed with local departmental champions—to deliver education and performance feedback to ED clinicians and evaluate the effectiveness of audit and feedback compared with usual practice in improving appropriate test ordering using a stepped-wedge randomised design. Methods Study design and setting This study is a 50-week stepped-wedge cluster randomised controlled trial conducted across EDs in Western Sydney and Nepean Local Health Districts, New South Wales, Australia. Trial commencement is from February 2025 to February 2026. EDs at five hospitals will be participating in the trial: Auburn, Blacktown, Mount Druitt, Nepean and Westmead Hospitals. EDs at Blacktown and Mount Druitt Hospitals have a high degree of shared staffing and are therefore combined into one site to minimise ‘contamination’ of intervention, resulting in a total of 4 trial sites. A participating site is randomised to commence intervention at 10-week intervals. This protocol is reported in accordance with the Consolidated Standards of Reporting Trials (CONSORT) extension for stepped-wedge cluster randomised trials.6 7 Eligibility criteria Inclusion criteria Consecutive ED presentations that fall within one of the 29 clinical presentations defined by the joint ACEM and RCPA Pathology Testing in the Emergency Department guideline (2023). Triage diagnoses were matched to these presentations: Abdominal pain—severe (upper/epigastric). Abdominal pain—severe (upper/epigastric). Abdominal pain—severe (lower). Abdominal pain—severe (lower). Back pain—atraumatic. Back pain—atraumatic. Cellulitis (requiring admission). Cellulitis (requiring admission). Cerebrovascular accident. Cerebrovascular accident. Chest pain—suspected ischaemic heart disease. Chest pain—suspected ischaemic heart disease. Chest pain—suspected pulmonary embolism. Chest pain—suspected pulmonary embolism. Confusion/syncope. Confusion/syncope. Diabetic ketoacidosis. Diabetic ketoacidosis. Fever for investigation/fever in returned traveller. Fever for investigation/fever in returned traveller. Fracture—neck of femur or other major long bone. Fracture—neck of femur or other major long bone. Fracture—requiring theatre. Fracture—requiring theatre. Gastrointestinal bleeding. Gastrointestinal bleeding. Jaundice for investigation. Jaundice for investigation. Oncology patients (febrile neutropenia). Oncology patients (febrile neutropenia). Overdose. Overdose. Per vaginal bleeding in first trimester. Per vaginal bleeding in first trimester. Pneumonia (requiring admission). Pneumonia (requiring admission). Pyelonephritis (not simple cystitis). Pyelonephritis (not simple cystitis). Renal colic (first episode). Renal colic (first episode). Seizures (first episode). Seizures (first episode). Seizures (recurrent). Seizures (recurrent). Sepsis. Sepsis. Septic joint. Septic joint. Shortness of breath—asthma. Shortness of breath—asthma. Shortness of breath—suspected acute pulmonary oedema. Shortness of breath—suspected acute pulmonary oedema. Shortness of breath—suspected chronic obstructive airways disease. Shortness of breath—suspected chronic obstructive airways disease. Snake bite trauma. Snake bite trauma. Exclusion criteria There were no exclusion criteria. Case identification and mapping As ED triage terminology does not align directly with the ACEM–RCPA guideline categories, approximately 100 commonly used triage diagnoses were reviewed by a multidisciplinary panel and mapped to the 29 eligible presentations. This mapping was updated in September 2024, prior to commencement of the trial, to reflect the updated guideline. This mapping dictionary was applied automatically during data extraction to classify all encounters. Ambiguous terms were adjudicated by consensus (PC, CL). The mapping process is included in an online supplemental appendix to enable replication. Intervention The intervention comprises a multifaceted audit-feedback programme delivered to participating EDs. UNTIE scores—UNTIE-U (Unnecessary Testing Index) and UNTIE-N (Necessary Testing Index)—will be calculated for all eligible ED visits throughout the study period, including both control and intervention phases, to provide baseline and ongoing measurement. UNTIE-U: percentage of tests performed that are not classified as ‘Perform’ or ‘Consider/Ask Supervisor’ in the guideline, divided by the number of Perform tests for that presentation. A score>0% indicates potential overutilisation. UNTIE-U: percentage of tests performed that are not classified as ‘Perform’ or ‘Consider/Ask Supervisor’ in the guideline, divided by the number of Perform tests for that presentation. A score>0% indicates potential overutilisation. UNTIE-N: percentage of Perform tests for a presentation that were actually ordered. A score<100% indicates potential underutilisation. UNTIE-N: percentage of Perform tests for a presentation that were actually ordered. A score<100% indicates potential underutilisation. The denominator (number of Perform tests) was selected to anchor unnecessary testing relative to the expected core investigations for each presentation, thereby normalising for variation in baseline recommended testing intensity across clinical contexts. Alternative formulations (eg, proportion of all tests ordered) were considered but were less sensitive to guideline-aligned overuse. Scores are generated fortnightly using automated data extraction and displayed on a performance dashboard. Each site will receive a structured audit-feedback intervention once randomised. Feedback will be delivered fortnightly via the following: In-person education sessions and presentations. In-person education sessions and presentations. Visual prompts such as posters in clinical areas. Visual prompts such as posters in clinical areas. Regular e-mail bulletins summarising performance. Regular e-mail bulletins summarising performance. Recorded sessions accessible to staff. Recorded sessions accessible to staff. Local clinical champions and research staff will lead delivery, ensuring contextual alignment with site workflows. Data from all sessions will be stored in a secure study database. Audit-feedback dashboards are generated centrally and accessible to the research team. Aggregated findings are disseminated to clinical staff, including nursing and medical personnel, through multidisciplinary meetings, educational sessions, written summaries and other local communication channels. The intervention is designed at a whole-of-department level and aims to influence multidisciplinary test-ordering behaviour rather than individual clinicians. Comparator (control arm) The comparator is usual practice at each site within the trial period, prior to researcher intervention (see schedule). Assignment of intervention An independent statistician generated the randomisation sequence using a computer-based random number generator in R. Each participating hospital was assigned a random number, and the hospitals were ordered according to these values to determine the sequence in which clusters crossed from control to intervention. The intervention leads are only informed of the randomisation order 1 week prior to intervention at each block to minimise anticipatory behaviour. Schedule of intervention rollout SitesBlock 1Block 2Block 3Block 4Block 5Site 1Usual practiceInterventionInterventionInterventionInterventionSite 2Usual practiceUsual practiceInterventioninterventionInterventionSite 3Usual practiceUsual practiceUsual practiceInterventionInterventionSite 4Usual practiceUsual practiceUsual practiceUsual practiceIntervention Protocol adherence and modification Participants are included if their triage diagnosis maps to one of the 29 presentations. Those outside these categories are excluded from analysis. Adherence to the intervention is promoted through multimodal engagement (education, visuals and e-mail prompts). Clinician judgement may over-ride guideline recommendations when clinically indicated; such cases will be discussed during feedback sessions and incorporated into ongoing refinement. Nurse-initiated pathology protocols Nurse-initiated pathology ordering (eg, Emergency care assessment and treatment (ECAT) protocols) is included in the calculation of UNTIE-U and UNTIE-N. The ordering clinician role (nurse vs medical officer) is identifiable within the dataset. As ECAT protocols were implemented across all participating sites 6 months prior to recruitment and remained stable throughout the trial period, nurse-initiated testing is considered a time-invariant cluster-level characteristic. The intervention is directed at both nursing and medical staff, and audit-feedback materials are accessible to all clinical personnel. Nurses are explicit targets of the audit-feedback intervention, and dashboard feedback and educational sessions are accessible to both nursing and medical staff. The primary analysis will therefore include all ED-initiated pathology testing regardless of ordering role, consistent with the pragmatic, whole-of-department design of the intervention. Outcomes Primary outcome The primary outcome is UNTIE-U, which is generated blinded to each ED’s intervention status. Individual ED encounters are the unit of analysis, with UNTIE-U measured at the level of each eligible clinical presentation. The primary estimand is the mean change in UNTIE-U between control and intervention periods, estimated using a linear mixed effects model with identity link, including fixed effects for intervention status and baseline covariates, and a random intercept for cluster (site). Complementary descriptive measures, including total number of tests per presentation and proportion of patients receiving at least one unnecessary test, will also be reported to aid interpretation. Secondary outcome UNTIE-N represents the proportion of guideline-recommended tests performed. This indicator will also be generated blinded to intervention status using the same data sources. Full details of the statistical model, including specification of covariates, variance estimation and df, are provided in the statistical analysis section below. Risk of bias Randomisation: A stepped-wedge cluster randomised design was used because individual clinician randomisation was neither feasible nor appropriate; clinicians frequently rotate across teams and patients, making contamination unavoidable, whereas cluster-level randomisation preserves the integrity of the intervention rollout across EDs. Randomisation: A stepped-wedge cluster randomised design was used because individual clinician randomisation was neither feasible nor appropriate; clinicians frequently rotate across teams and patients, making contamination unavoidable, whereas cluster-level randomisation preserves the integrity of the intervention rollout across EDs. Blinding: Blinding of participating clinicians to the intervention was not possible due to the overt nature of audit-feedback activities (education sessions, posters, emailed dashboards); however, UNTIE-U and UNTIE-N indicators are generated automatically and blinded to site allocation during data processing to minimise detection bias. Blinding: Blinding of participating clinicians to the intervention was not possible due to the overt nature of audit-feedback activities (education sessions, posters, emailed dashboards); however, UNTIE-U and UNTIE-N indicators are generated automatically and blinded to site allocation during data processing to minimise detection bias. Sample size calculation: A formal sample size calculation was performed using the Hemming and Girling method for stepped-wedge trials. Sample size calculation: A formal sample size calculation was performed using the Hemming and Girling method for stepped-wedge trials. Exclusion of data: Data were excluded only when triage diagnoses did not map to one of the 29 ACEM–RCPA guideline presentations; this exclusion avoids misclassification and ensures the indicators reflect guideline-defined clinical scenarios rather than heterogeneous or non-applicable presentations. Exclusion of data: Data were excluded only when triage diagnoses did not map to one of the 29 ACEM–RCPA guideline presentations; this exclusion avoids misclassification and ensures the indicators reflect guideline-defined clinical scenarios rather than heterogeneous or non-applicable presentations. Missing data: As all outcomes are derived from routinely collected, complete pathology and administrative datasets, missing outcome data are expected to be minimal. Missing data: As all outcomes are derived from routinely collected, complete pathology and administrative datasets, missing outcome data are expected to be minimal. Statistical analysis We hypothesise that the intervention will reduce UNTIE-U by 0.0625. This estimate assumes that approximately 20% of ED tests are unnecessary, corresponding to a baseline UNTIE-U of 0.25 (20/80) and a target post-intervention UNTIE-U of 0.1875 (15/80). Sample size estimation was performed using the Hemming and Girling method for stepped-wedge cluster randomised trials.6 Individual ED encounters are the unit of analysis, with UNTIE-U measured at the level of each eligible clinical presentation. The primary estimand is the mean change in UNTIE-U between control and intervention periods, estimated using a linear mixed effects model with identity link, including fixed effects for intervention status and baseline covariates, and a random intercept for cluster (site). Pre-specified baseline covariates include patient age, sex, triage category and presentation type, included to improve precision and address potential imbalance between clusters. Further analyses will explore lag effects by adding time period (block) and an interaction between time and intervention to the model,8 9 consistent with methodological guidance highlighting exposure-time-varying treatment effects as a key concern in stepped-wedge designs.1012 Robust (sandwich) variance estimators will be used to account for potential model misspecification.9 Given the small number of clusters (I=4), statistical inference will be based on a t-distribution with df equal to I − 2 (ie, 2 degrees of freedom), rather than normal approximation, in accordance with recommendations for stepped-wedge cluster randomised trials with few clusters.13 All analyses will follow the intention-to-treat principle, with observations classified according to the randomised step irrespective of implementation delays. Statistical analyses will be conducted using Stata/SE (V.18, StataCorp, TX, USA). Sensitivity analysis As a sensitivity analysis, a cluster fixed-effects model with robust variance estimation will be implemented. This approach adjusts for all time-invariant cluster-level covariates and mitigates potential bias arising from chance imbalance in small-cluster stepped-wedge trials.14 Additional sensitivity analyses will explore lag effects and exposure time-varying treatment effects by incorporating time period (block) and an interaction between time and intervention status into the primary model, consistent with methodological guidance for stepped-wedge designs.1012 Sensitivity analyses related to nurse-initiated testing Ordering clinician role (nurse vs medical officer) is identifiable within the underlying dataset but is not retained in the automated UNTIE score output. As the intervention operates at a whole-of-department level and ECAT protocols were implemented across all sites 6 months prior to recruitment and remained stable throughout the study period, nurse-initiated testing is considered a time-invariant cluster-level characteristic. As a sensitivity analysis, the proportion of pathology tests initiated by nursing staff will be calculated for each site-period and included as a fixed-effect covariate in secondary models. Temporal stability in nurse-initiated testing across pre-intervention and post-intervention periods will also be assessed. Power calculation Sample-size estimation followed the Hemming and Girling method for stepped-wedge designs, implemented in Stata (6). Power calculations were based on a two-sided test of difference in proportions using normal approximation under an exchangeable correlation structure. Input included hypothesised UNTIE-U figures above (0.25 and 0.1875), four steps, one cluster randomised at each step, average cluster size of 5498 (average number of presentations for this group of hospitals over each time block) and significance level of 0.05. The intra-cluster correlation for UNTIE-U is unknown, and we varied it from 0.01 to 0.99. The total projected sample size is 109 960 ED presentations across all clusters and time periods. Power exceeded 90% across plausible Interclass Correlation Coefficient values (0.01–0.10) Additional power calculations under a nested exchangeable (Hooper–Girling) structure were conducted and yielded comparable power (>0.95 across plausible parameter values); therefore, the original calculation is retained. Patient and public involvement Patients or the public were not involved in the design, or conduct, or reporting, or dissemination plans of our research. Ethics and dissemination Ethics approval has been granted by the Western Sydney Local Health District Human Research Ethics Committee (HREC/17/WMEAD/274, 2022/STE03249). All analyses will use de-identified data. Findings will be disseminated through peer-reviewed publication, conference presentation and feedback to participating EDs. Participants were not required to give informed consent for the collection of this data. Clinical staff at each participating site were provided with a Staff Information Sheet prior to commencement of the intervention, outlining the nature and purpose of the study and the voluntary nature of participation (see online supplemental material). Data monitoring Test usage is ascertained through routinely collected administrative data with no data monitoring required. Data availability Individual participant-level data from this trial are derived from routinely collected ED and pathology datasets held by NSW Health and cannot be made publicly available due to data governance and privacy restrictions. De-identified aggregate data and the full data dictionary may be made available on reasonable request to the corresponding author and subject to approval by the Western Sydney Local Health District Human Research Ethics Committee. Statistical code used for data processing and analysis will be shared via a public repository (eg, GitHub or Figshare) on publication. No identifiable data will be shared. Trial protocol and statistical analysis plan The full trial protocol is available via the Australian New Zealand Clinical Trials Registry (ANZCTR) under the registered trial number ACTRN12623001130651. Any protocol amendments will be updated on the ANZCTR record. The statistical analysis plan will be uploaded to the same registry prior to completion of data collection and will also be available from the corresponding author on reasonable request. Harms/auditing While the intervention is low risk and does not involve direct patient contact, potential unintended consequences include under-testing, delayed investigations or workflow disruption. These will be monitored through UNTIE-N scores (necessary testing), routine clinical safety indicators and local governance processes at participating sites. Data access Restricted to study investigators and data custodians. Dissemination Findings will be reported to participating EDs, presented at scientific meetings and published in peer-reviewed journals to inform wider implementation of audit-feedback interventions. Supplementary material