Case Study: Reducing the Review Burden of Unactionable Ambulatory ECG Signals

Background 

Ambulatory ECGs are fundamental tools in the diagnosis and monitoring of cardiac conditions, providing critical insights into cardiac function over extended periods. Traditionally, these ECGs are used in settings where continuous monitoring is necessary to capture transient or infrequent cardiac events that might not be detected during a standard 12-lead ECG. The presence of noise and artefacts in ECG recordings complicates the diagnostic process, requiring healthcare professionals to spend considerable time distinguishing between poor-quality, non-actionable data and clinically relevant signals. This differentiation is critical because erroneous interpretations due to noise can lead to misdiagnoses or unnecessary further testing, both of which have implications for patient safety and healthcare costs.  

The Challenge 

The high prevalence of noise in ambulatory ECGs poses a substantial challenge, with healthcare professionals often needing to manually review large volumes of ECG data, a process that is both time-consuming and prone to human error. Although signal quality assessment algorithms have been hailed as solutions to reduce the review burden of unactionable ECG, the integration of such features within standard clinical practice has been slow, primarily due to a lack of evidence-based research highlighting the potential benefits they can endow. 

Study Aims 

The primary aim of this study was to evaluate the impact of HeartKey Rhythm’s Signal Quality algorithm on the clinical workflow of an Independent Diagnostic Testing Facility (IDTF). By integrating automated signal quality indicators, we sought to reduce the time healthcare professionals spend reviewing unactionable ECG data, thereby enhancing workflow efficiency. 

Study Design

We conducted a two-phase study on patients who were referred to an IDTF (TZ Medical, Tualatin) for continuous cardiac monitoring, utilizing a three-channel cardiac monitor (Trident®). The HeartKey Signal Quality algorithm was employed to evaluate each ECG lead, segmenting the data into two-second intervals classified as either high or low quality. The overall event quality was determined based on the proportion of low quality segments, with validations against manual annotations. As signal quality is not often consistent across all recorded leads, we employed an aggregate metric derived from the highest quality lead per event to represent the best-case scenario. 

Phase 1 

A retrospective analysis of 19,392 ECG arrhythmia events from 291 patients. This phase focused on defining an optimal threshold of low quality data to be automatically discarded prior to ECG triage, while also maintaining minimal risk of discarding actional data. At a threshold where >90% of an event had been classified by HeartKey as low quality, the IDTF were able to reduce the number of unactionable events that required manual review by 32.4%, with only a negligible 0.3% chance of discarding actionable data.  

Phase 2 

A prospective, two-arm analysis of 33,249 events over a one-week period. Events ranged from 60-180-seconds (mean: 72 seconds) and totalled 669 hours of ECG data. This phase involved further optimisation of the thresholds identified in Phase 1, which resulted in a change of focus from percentage of unactionable data to duration of unactionable data. Healthcare professionals determined that <8-seconds of combined high quality data across the aggregate lead would be deemed unactionable, translating to a proportional threshold, as defined in Phase 1, ranging from 87.7% to 95.6%. Utilising this new threshold of <8-seconds of high quality data enabled a reduction in all ECG events requiring manual review by 17.1%, with a reduction in artefact events by a staggering 40.6%, while still maintaining a specificity of 99.8%. 

An overview of the study design (*results compared to outputs from ECG Tech Review during live triage of events)

Implications for Clinical Workflow 

The integration of HeartKey Signal Quality indicators into the ECG review process ensured that healthcare professionals spent their time reviewing only the most clinically relevant data by significantly reducing the volume of data requiring manual review. For a healthcare professional reviewing 50 events per day at an average two-minute review time per event, this equates to a daily time saving of >17 minutes, enabling the review of an additional 8.6 events per day. 

Conclusions 

The implementation of automated signal quality indicators within the ECG review workflow presents a substantial advancement in the management of ambulatory ECG data. By significantly reducing the time clinicians spend on reviewing unactionable data, HeartKey enables more efficient patient monitoring and potentially quicker diagnostic turnaround. This case study underscores B-Secur’s commitment to enhancing the utility and efficiency of cardiac monitoring technologies in clinical settings.