Chest pain is the leading cause of emergency department (ED) presentations worldwide, contributing to ED overcrowding and treatment delays that are associated with adverse outcomes [1]. Overcrowding is further aggravated by a trend to reduce bed capacities in hospitals [2]. The introduction of high sensitivity troponin (hs-cTn) assays has not only enabled to detect smaller myocardial infarcts but also earlier detection of acute myocardial injury [3]. Consequently, clinical guidelines now recommend the use of hs-cTn assays with fast diagnostic protocols that incorporate shorter time intervals for repeat testing [4, 5]. Clinical decision pathways (CDPs) are being promoted to help to categorize patients into low-, intermediate- and high-risk strata and hence facilitate disposition and subsequent diagnostic evaluation [5]. The ESC 0/1, ESC 0/2 and ESC 0/3 h algorithms have convincingly demonstrated high diagnostic performance, effectiveness and safety [6]. However, their adoption in clinical routine lags behind expectations [7].

In a recent update of the CARMAGUE survey [7] covering the period from 2019 to 2022 and data from 664 laboratories across 76 countries, a utilization rate of the ESC 0/1 h and ESC 0/2 h algorithm of only 39.8% was demonstrated. A delta for hs-cTn was reported in only 62.5%, and only 44.3% reported an absolute delta, which is an essential component of fast protocols. In addition, the diagnostic algorithms require assay-specific cutoffs for ESC 0/1, ESC 0/2 and ESC 0/3 h algorithms and mandate a strict adherence to time to repeat sampling with a tolerance interval of ± 10 min [4]. Guidelines motivate to obtain blood samples for hs-cTn at 0 h and 1 h, irrespective of other clinical details and pending results. This strategy will likely result in unnecessary cardiac troponin measurements in perhaps 10–15% of patients with very low 0 h concentrations and chest pain onset > 3 h but is believed to substantially facilitate the implementation of ESC 0/1 h algorithm [8].

A real-world implementation study from the UK [9] on 5496 patients in 2020 and 7363 patients in 2021 patients with suspected acute coronary syndrome (ACS) reported multi-faceted, practical limitations of achieving protocol adherence to the ESC 0/1 h algorithm that persisted after the familiarization period [9]. We hypothesize that a well-designed clinical decision support system (CDSS), based on clinical guidelines and embedded into an electronic healthcare system, can improve implementation, protocol adherence, cost-effectiveness and patient safety of clinical decision pathways.

Methods and CDSS designDefinition and components of the clinical decision support system

CDSSs are platforms that integrate multiple clinical data inputs to generate outputs comprising a diagnosis, a clinical recommendation or risk stratification. These systematic outputs can help clinicians guide the disposition and management of patients. CDSS tools are classified based on the timing of support (before, during or after the clinical decision is made) and the degree of activity (active alerts versus passive response to user input or patient data) (10 Berner). Moreover, they are distinguished as either knowledge-based systems (using explicit rules, such as if–then statements) or non-knowledge-based systems (employing methods like machine learning algorithms) [10]. A typical CDSS consists of three main components, the knowledge-base, the inference/reasoning engine and a user interface (UI) [10, 11]. Specifically, the knowledge base consists of compiled information, often in the form of structured clinical rules. For example, IF hs-cTn is below the limit of detection (LoD) AND onset of symptoms is more than 3 h, THEN myocardial infarction can be ruled out with the 0-h algorithm. The reasoning engine contains the formulas for combining the rules or associations in the knowledge base with actual patient data, and finally, the UI transmits the system’s output (e.g. triage category, alert) to the user.

Data for the CDSS must either be inserted manually or, ideally, incorporated directly into the electronic health record (EHR) system. In the latter case, patient data is automatically retrieved from the computer-based record, laboratory, or other integrated hospital system (Fig. 1) [10].

Fig. 1The alternative text for this image may have been generated using AI.

System architecture and data integration. The schematic illustrates the CDSS’ hybrid input model, designed to integrate seamlessly with clinical workflows. Data acquisition (Left): Critical temporal variables (specifically time of symptom onset) are entered manually by the clinician to ensure accuracy for early presenters, while quantitative hs-cTn concentrations are retrieved automatically from the Laboratory Information System (LIS) to prevent transcription errors. Processing (Center): These inputs are aggregated within the hospital’s electronic health record (EHR) and accessed by the CDSS Engine via a secure Application programming interface (API). The engine applies the encoded 2023 ESC Guideline rules (Knowledge Base) to calculate the patient’s risk status. Output (Right): The user interface delivers a time adjusted triage recommendation (rule-in, rule-out or observe) alongside the specific algorithmic rationale (e.g. “0 h hs-cTn > ULN”), providing a “White Box” approach to decision support. API, application programming interface; hs-cTn, high sensitivity cardiac troponin; ULN, upper limit of normal

Addressed gaps and challenges in fast protocols addressed by the CDSS

The CDSS was specifically engineered to overcome known barriers in the implementation of fast diagnostic protocols (Table 1). While the ESC 0/1 h and ESC 0/2 h algorithms are the preferred standard of care, their complexity often leads to protocol violations in routine practice. Specific challenges addressed by the tool include the rigorous requirement for assay-specific delta calculations, the management of sex-specific 99th percentile upper limits of normal (ULN) and the strict adherence to serial sampling intervals. By automating these complex calculations, the CDSS aims to mitigate common errors such as the underestimation of risk in early presenters, the overuse of troponin testing beyond diagnostic necessity and the associated prolongation of ED length of stay (LOS) (Table 2).

Table 1 Gaps and challenges in fast protocolsTable 2 Potential advantages of the CDSS in the EDDesign of the CDSS using triage algorithms and temporal logic

The system requires configuration of the locally used hs-cTn assay, including its validated baseline values, delta change criteria for the ESC 0/1 h, ESC 0/2 h and ESC 0/3 h algorithms, and the option to use sex-specific 99th percentile upper limit of normal (ULN) cutoffs (Fig. 2). The novel knowledge-based CDSS was designed to correctly triage patients into “rule-out”, “rule-in” or “observe” categories based on the diagnostic triage algorithm proposed by the 2023 ESC guidelines on ACS [4]. ESC guidelines recommend the ESC 0/1 h and ESC 0/2 h algorithm as the preferred CDP options, with the ESC 0/3 h algorithm as an alternative when the faster protocols are not routinely available. A threshold of 3 h or less for the time from onset of symptoms (last severe episode leading to admission) to admission is used to define “early presentation”. In the latter scenario, triage cannot be based on a single hs-cTn below LoD but requires consecutive troponin testing after 60 to 120 min. To consider actual elapsed time intervals between baseline and subsequent testing, a tolerance time of ± 30 min for each CDP was used to cover time intervals continuously without overlap or gaps. If triage could not be unequivocally determined after two blood draws, a third blood draw at 3 h or more is recommended to resolve the “observe zone” into rule-out or rule-in. The criteria were based on the recommendations of a consensus document of the ESC biomarker group [12] and a recent validation of the ESC 0/3-h protocol [13, 14]. To ensure patient safety, the CDSS incorporates specific fail-safe mechanisms. If critical input variables (e.g. time of symptom onset) are missing or entered in an illogical format (e.g. future timestamp), the system prevents the generation of a triage recommendation and instead issues a warning prompting data correction. The development process for this knowledge-based CDS system considered the GUIDES checklist, a recognized standard developed by experts to facilitate the adoption and evaluation of such tools in clinical settings [15].

Fig. 2The alternative text for this image may have been generated using AI.

Flowchart of the CDSS decision logic and triage algorithm procedure. This flowchart delineates the step-by-step logic utilized by the CDSS for NSTE-ACS triage. The procedure begins with regional algorithm selection (US vs. non-US application) and the determination of whether to apply sex-specific ULN cutoffs. The system then processes clinical inputs—including symptom onset time and serial hs-cTn measurements—to navigate through 0 h, 1 h, 2 h or 3 h diagnostic pathways. If initial or secondary results fall into an “observation zone”, the logic directs the clinician toward a third blood draw to reach a final rule-in or rule-out classification based on the ESC 0/3 h algorithm. CDS/CDSS, clinical decision support/clinical decision support system; hs-cTn, high-sensitivity cardiac troponin; ULN, upper limit of normal; NSTE-ACS, non-ST-elevation acute coronary syndrome

For convenience, the CDSS requires only a limited set of variables. After patient presentation, the system requires manual input of the time of symptom onset, as this parameter is typically not available through the electronic health records (EHR). For convenience, a manual imputation of time of symptom onset is only requested if the initial hs-cTn is below the decision threshold for a single sample strategy, e.g. below LoD (< 5 ng/L). Otherwise, a serial sampling is the default strategy if initial hs-cTn exceeds this particular decision threshold. The following variables can typically be retrieved via EHR interfaces: sex, time and exact troponin concentration (rounded to 1 decimal place) for the first, second and (if required) third blood test result. Data integration is achieved via a secure application programming interface (API) utilizing standard interoperability formats (HL7 FHIR) to minimize manual entry and ensure real-time data synchronization. The system provides a “White Box” approach to decision support. Alongside the triage category, the UI displays the specific rationale for the recommendation (e.g. citing the specific guideline rule triggered), thereby maintaining transparency and allowing the clinician to verify the algorithmic logic against their clinical judgment (Fig. 3). The output consists of a time adjusted triage category with reference to the current guideline (e.g. “Recommendation: Rule-in”) or a notification prompting a follow-up measurement at a designated time. The tool also displays specific instructions based on the relevant ESC algorithm (Table 3).

Fig. 3The alternative text for this image may have been generated using AI.

User interface and algorithmic output of the guideline-based cDSS. This figure illustrates the interactive components of the CDSS designed for NSTE-ACS triage. Section (a) displays the manual and automatic data entry fields, including time of symptom onset and serial high-sensitivity cardiac troponin (hs-cTn) concentrations. Section (b) demonstrates the system’s “White Box” output, providing an unequivocal triage recommendation (e.g. “Rule-in”) based on the ESC 0/1 h algorithm. The tool automatically adjusts for the truly elapsed time between blood collections, maintaining transparency by citing the specific guideline source used for the recommendation. CDS/CDSS, clinical decision support/clinical decision support system; hs-cTn, high-sensitivity cardiac troponin; ULN, upper limit of normal; NSTE-ACS, non-ST-elevation acute coronary syndrome

Table 3 Data input requirements for the clinical decision support systemPlanned prospective validation

In a sequential manner over a recruitment period of 24 weeks, all-comers presenting with suspected NSTE-ACS will be assigned to either standard-of-care or CDS tool. Assignment to either arm will alternate every 2 weeks (Fig. 4). The primary endpoint was defined as the proportion of protocol adherence. In this context, protocol deviations are defined as an inappropriate number of troponin tests, i.e. less than required or more than required including inappropriate testing leaving patients in the observe zone. In addition, protocol violations include time intervals that exceed the tolerance interval of the default CDP, e.g. ESC 0/1 h or ESC 0/2 h, with or without a re-classification to a different triage category. Adherence to sex-specific cutoffs for the 99th percentile ULN represents an optional component of the primary endpoint. Secondary endpoints include the accuracy of triage classification (defined as the correct assignment to either immediate rule-in or rule-out using the ESC 0 h algorithm or correct assignment to triage categories using serial measurements and the proportion of re-classifications), overall length of ED stay, length of ED stay stratified by appropriate triage, discharge and admission rates, overall 30-day mortality rates, and mortality rates split by appropriate triage. The study also assesses the effect of crowding on triage performance. The specific in- and exclusion criteria are provided in the Supplementary Material. At the planned sample size (N = 1200), simulated power was 85.7% for the pre-specified minimum clinically meaningful improvement of protocol adherence of + 10 percentage points (see Supplementary Materials for Details on power calculation).

Fig. 4The alternative text for this image may have been generated using AI.

Schematic of the prospective validation study design. The study utilizes a sequential, pseudo-randomized roll-out design over a 24-week enrollment period with alternating 2-week blocks. The ED workflow alternates between the standard of care (control) phase and the CDSS implementation (intervention) phase (12 blocks total, 6 per arm). The primary endpoint (protocol adherence) is compared between these distinct time blocks CDSS, clinical decision support system

Primary endpoint—protocol adherence

Protocol adherence is defined as a binary, patient-level outcome. A patient is classified as “adherent” if all of the following criteria are met: (a) the correct clinical decision pathway (CDP) is applied based on ESC 0/1 h or ESC 0/2 h as the default pathway; ESC 0/3 h only when faster protocols are not routinely available; (b) the number of hs-cTn measurements matches the protocol requirement (no overuse, defined as additional tests beyond protocol necessity; no underuse, defined as fewer tests than required, leaving the patient in an unresolved observe zone); and (c) time intervals between serial blood draws fall within the protocol-specified tolerance of ± 30 min for the default CDP. Protocol non-adherence is classified into the following violation types: timing violation, underuse, overuse, pathway selection error or cutoff application error.

Potential sources of bias and mitigation strategies

The alternating-block design, while pragmatic for a single-centre ED-wide workflow intervention, is susceptible to several sources of bias that are addressed through design choices and prespecified analytic strategies. Calendar time (period number) is included as a fixed covariate in the primary GEE model to adjust for secular trends in adherence that may arise from seasonal ED volume fluctuations, staffing changes or institutional process evolution. A daily ED crowding metric will be recorded and included as a time-varying covariate in the analysis model. Crowding-stratified results (above vs. below median daily census) are reported as an exploratory analysis to assess effect modification. Clinicians exposed to CDSS guidance during intervention blocks may retain improved adherence behaviours during subsequent control blocks (contamination). As pre-specified sensitivity analyses, we will (a) exclude the first 3 days of each control block following a CDSS block (washout) and (b) test for a monotonic time trend in adherence across control blocks to detect a learning effect. A significant positive trend would suggest contamination, which would bias results toward the null (conservative). Weekday versus weekend presentation status is recorded and included as a covariate to account for differences in staffing patterns and clinical workflow.