False Arrhythmia Research Articles

A large, real-world cohort analysis of arrhythmia detections with insertable cardiac monitors

Abstract Background Insertable cardiac monitors (ICMs) provide long-term continuous monitoring for arrhythmia diagnosis and management for various clinical indications. However, little data exists on the real-world diagnostic yield of ICMs across indicated patient populations, including both expected and incidental arrhythmic findings. Validated artificial intelligence (AI) algorithms have been developed to identify true arrhythmia episodes while significantly reducing false positives, enabling the adjudication of ICM episodes for larger cohorts of patients. Purpose To characterize comprehensive true arrhythmia detection rates in a large real-world population with long-term ICM monitoring. Methods De-identified patients ≥18 years receiving a Reveal LINQ ICM between Oct 1, 2016 – Sept 30, 2020 with no prior cardiac implantable electronic devices and ≥1 year of follow-up in the CareLink monitoring database were identified (N=12,013). Patients were stratified by reason for ICM placement: N=9,841 who received ICM for diagnosis of suspected arrhythmias (40.8% unexplained syncope, 38.3% cryptogenic stroke, 9.7% palpitations, 9.7% suspected atrial fibrillation (AF), 1.6% ventricular tachycardia), and N=2,172 who received ICM for management of known AF. Device-detected ECGs during long-term follow-up were identified and processed through previously validated arrhythmia-specific AI algorithms designed to reduce false positive arrhythmia detections (each trained using ICM data from ≥25,000 adjudicated ECG episodes and validated on an independent set of ≥5,000 adjudicated ECGs). Results Mean (SD) and median follow-up were 24.6(12.7) and 24.5 months, respectively. In patients receiving ICM for Suspected Arrhythmia, 5,492 patients (55.8%) had ≥1 AI-adjudicated arrhythmia detected during follow-up, with 2,472 (45.0%) of these patients having ≥2 arrhythmia types detected (Table). Mean (SD) time to 1st arrhythmia detection was 7.7(9.1) months; time to 2nd arrhythmia type detected was 12.3(10.4) months. Among the 6,958 patients with longer-term ICM follow-up ≥2 years, 2,931 (42.1%) had no arrhythmias detected (i.e., arrhythmia rule-out occurred). In patients receiving ICM for AF Management, 1,737 patients (80.0%) had ≥1 arrhythmia detected, and notably 1,074 (61.8%) of these patients had ≥1 arrhythmia detected during follow-up that was not AF. Mean (SD) time to 1st arrhythmia detection was 5.3 (7.7) months; time to 2nd arrhythmia type detected was 10.4(9.9) months. Conclusions In a large real-world ICM population, >50% of patients had AI-adjudicated arrhythmia detections during long-term follow-up, with a significant proportion receiving additional arrhythmia diagnoses beyond the initial type diagnosed or suspected. These results suggest that in indicated patients, long-term continuous monitoring with ICMs can enable diagnosis of a variety of cardiac arrhythmias that may have otherwise remained undetected.

PO-03-046 ACCELERATING IMPROVEMENTS TO ARRYTHMIA CLASSIFIERS WITH AI-CURATED DATASETS

AccuRhythm AI is a suite of cloud-based AI algorithms that filter and suppress false arrhythmia detections from insertable cardiac monitors (ICMs). Its initial release represented a large reduction to remote monitoring burden, suppressing 74.1% of false AF alerts. Deploying performance updates is easy with cloud-based AI, but the adjudication effort to train and test updates becomes increasingly burdensome as the AI performance improves. Ever larger datasets are required to obtain enough challenging episodes to differentiate performance between AI iterations. Reduce manual adjudication requirements for AI retraining and validation by leveraging the AI to curate datasets of challenging episodes. An independent dataset of 739,188 unadjudicated AF episodes was created from a 7-month subset of an ICM database. Challenging episodes were iteratively identified from this subset as those with uncertain or discordant outputs between AI iterations. Random samples of these episodes were labeled and incorporated into the original training dataset over several retraining iterations. The final retrained AI was compared to the original AI by applying both to a 2nd independent dataset of 5,030 episodes and manually adjudicating those with discordant AI outputs. A 3rd independent dataset of 955 AF alerts (1,856 episodes, 955 ICMs) was randomly drawn from 3-month follow-ups of 3,031 consecutively implanted ICMs to test generalization. Each episode was manually adjudicated by 2-3 reviewers. A total of 10,790 challenging episodes were incorporated into the 115,395-episode training dataset over 10 retraining iterations. There were 521 discordant AI classifications from 431 ICMs within the 5,030-episode comparison dataset. The retrained AI demonstrated superior sensitivity and specificity, increasing the number of true positives by 4 and reducing false negatives by 420. This performance generalized to the 1,856-episode validation dataset with relative alert sensitivity and specificity of 99.0% (494/499) and 88.2% (402/456), respectively, representing a 54% reduction in false alerts relative to the original AI. Targeted adjudication of challenging episodes based on AI output increased the efficiency of training and validating an AI update by focusing manual efforts on the most relevant examples. This accelerates the creation of highly generalizable AI by incorporating more rare episodes than would otherwise occur with representative sampling.

Insertion procedure assessment, sensing and detection performance of a novel miniaturized implantable cardiac monitor with a long sensing vector

Abstract Funding Acknowledgements Type of funding sources: Private grant(s) and/or Sponsorship. Main funding source(s): unrestricted grant by Biotronik Background The Biotronik BioMonitor III (BM III) is a novel implantable cardiac monitor (ICM) with a miniaturized profile, long sensing vector and simplified implantation technique. Objective We assessed insertion procedure, reliability of sensing quality and detection performance of the BM III. Methods The ICM was implanted in the left parasternal region between the suprasternal notch and the nipple, preferentially at a 45° angle to the sternum. R-wave amplitude was recorded immediately post implantation and at day 1. Follow-up was after 3 months or an event. All episodes, R-wave amplitude and noise burden were documented. The anatomical ICM position was determined at day 1 and after 3 months. A patient questionnaire on pain after implantation, prolonged paresthesia, restriction of daily activity, handling of the HomeMonitoring system and overall satisfaction was conducted after 3 months. Results In 30 patients (age 67 years, 40% male) a BM III was successfully inserted, without complications. Indications were: unexplained syncope (80%), suspected AF (3%) and cryptogenic stroke (17%). Median implantation time was 6 min. Mean R-wave amplitude was 0.73 mV at insertion, 0.77 mV at day 1 (p=0.59) and 0.81 mV after 107 days (p=0.39). At day 1 and at follow-up visit, an anatomically stable device position was demonstrated. Vertical distance from fossa jugularis to the lowest part of the ICM was 14.5 cm at implantation and 14.5 cm at follow-up (p=0.94). The horizontal cranial sternal-to-device distance was 4.4 cm and 4.3 cm at follow-up (p=0.87). The horizontal caudal sternal-to-device distance was 6.0 cm and 5.7 cm at follow-up (p=0.7). Median time to first event was 12 days. In 13 (43%) patients, true episodes were detected. False arrhythmia alerts were detected in 12 (40%) patients, in 7/12 false AF alerts due to recurrent premature atrial contractions (PAC) were recorded. Total number of false detections was low. Patient-satisfaction rate was high, with very few limitations in daily activity. Conclusion Insertion of the novel Biotronik BioMonitor III ICM is fast, uncomplicated and keeps an anatomically stable position of the device over time. Initial sensing values are good and remain stable over time. False arrhythmia alert, mainly due to PACs, occurred in 40% of patients, with a low total number of false detections.

A contrastive learning approach for\xa0ICU false arrhythmia\xa0alarm reduction

The high rate of false arrhythmia alarms in Intensive Care Units (ICUs) can lead to disruption of care, negatively impacting patients’ health through noise disturbances, and slow staff response time due to alarm fatigue. Prior false-alarm reduction approaches are often rule-based and require hand-crafted features from physiological waveforms as inputs to machine learning classifiers. Despite considerable prior efforts to address the problem, false alarms are a continuing problem in the ICUs. In this work, we present a deep learning framework to automatically learn feature representations of physiological waveforms using convolutional neural networks (CNNs) to discriminate between true vs. false arrhythmia alarms. We use Contrastive Learning to simultaneously minimize a binary cross entropy classification loss and a proposed similarity loss from pair-wise comparisons of waveform segments over time as a discriminative constraint. Furthermore, we augment our deep models with learned embeddings from a rule-based method to leverage prior domain knowledge for each alarm type. We evaluate our method using the dataset from the 2015 PhysioNet Computing in Cardiology Challenge. Ablation analysis demonstrates that Contrastive Learning significantly improves the performance of a combined deep learning and rule-based-embedding approach. Our results indicate that the final proposed deep learning framework achieves superior performance in comparison to the winning entries of the Challenge.

The BIOMONITOR III Injectable Cardiac Monitor: Clinical Experience with a Novel Injectable Cardiac Monitor.

Background: Injectable cardiac monitors (ICMs) are leadless subcutaneous devices for long-term monitoring of arrhythmias. The BIOTRONIK BIOMONITOR III is a novel ICM with a miniaturized profile, long sensing vector, and simplified implantation technique. Methods: R-wave amplitude was recorded immediately after implantation, the day after implantation, and after 3 months. Follow-up was scheduled after 3 months or after an event. All data from the ICM were retrieved. The anatomical position of the ICM was determined post-implantation and after 3 months. A patient questionnaire was conducted after 3 months. Results: In 36 patients (mean age 67 ± 13 years; 40% male) an ICM was inserted. Six patients were not included in the final analysis. The median time from skin cut to wound closure was 6 [IQR 5–7] minutes. Mean R-wave amplitude increased over time (0.73 ± 32 mV vs. 0.78 ± 0.38 mV vs. 0.81 ± 0.39 mV; p = ns). Three months after implantation, the ICM was in an anatomically stable position. In 14 (47%) patients, true episodes were detected. False arrhythmia alerts were detected in 13 (43%) patients. The total number of false detections was low, and the patient satisfaction rate was high. Conclusion: Implantation of the novel BIOMONITOR III is fast and uncomplicated; its sensing characteristics are excellent and improve over time, and patient satisfaction is high.

Computer Assisted Patient Monitoring: Associated Patient, Clinical and ECG Characteristics and Strategy to Minimize False Alarms

This chapter is a review of studies that have examined false arrhythmia alarms during in-hospital electrocardiographic (ECG) monitoring in the intensive care unit. In addition, we describe an annotation effort being conducted at the UCSF School of Nursing, Center for Physiologic Research designed to improve algorithms for lethal arrhythmias (i.e., asystole, ventricular fibrillation, and ventricular tachycardia). Background: Alarm fatigue is a serious patient safety hazard among hospitalized patients. Data from the past five years, showed that alarm fatigue was responsible for over 650 deaths, which is likely lower than the actual number due to under-reporting. Arrhythmia alarms are a common source of false alarms and 90% are false. While clinical scientists have implemented a number of interventions to reduce these types of alarms (e.g., customized alarm settings; daily skin electrode changes; disposable vs. non-disposable lead wires; and education), only minor improvements have been made. This is likely as these interventions do not address the primary problem of false arrhythmia alarms, namely deficient and outdated arrhythmia algorithms. In this chapter we will describe a number of ECG features associated with false arrhythmia alarms. In addition, we briefly discuss an annotation effort our group has undertaken to improve lethal arrhythmia algorithms.

P1362Diagnostic accuracy of R-wave detection by insertable cardiac monitors

Abstract OnBehalf 1.Department of Cardiology, Ghent University Hospital, Ghent, Belgium Introduction Insertable cardiac monitors (ICM) allow automatic arrhythmia detection but the diagnostic performance can be hampered by poor R-wave sensing leading to false positive recordings. Purpose This study assesses the prevalence and predictors of R-wave under- and oversensing among different ICM types. Methods Patients implanted with an ICM at the University Hospital of Ghent between January 2017 and August 2018 were included. All ICM tracings recorded at ICM interrogation or transmitted by remote monitoring were reviewed for inadequate R-wave sensing leading to false arrhythmia alerts. Inadequate R wave sensing was defined as undersensing due to sudden reduction in R wave amplitude or oversensing due to the occurrence of artefacts. Patient and implant characteristics were retrieved from the medical record and implant reports. Results The study screened 135 patients (age 59 ± 19 years, 44% female) implanted with different ICM types: Reveal XT and LINQ n = 92 (68%), Confirm and Confirm Rx n = 35 (26%), BioMonitor n = 8 (6%). ICM analysis was conducted in 112 patients (83%). Overall, false arrhythmia alerts due to inadequate R wave sensing occurred in 22 (20%) of the patients. Among these 22 patients, false diagnosis of bradycardia or pauses were documented in 64%, false high ventricular rates in 14% and false atrial fibrillation alerts in 22% of the patients. Of interest, in an additional 7 (6%) patients significant changes in R wave sensing occurred, although not causing false arrhythmic alerts. Occurrence of R-wave changes occurred in 19% of Reveal, 17% of Confirm and 50% of BioMonitor ICMs (p = not significant). Occurrence of R-wave changes among the different types of ICMs were not related to age, sex, BMI, time of the day, heart frequency or implant R-wave sensing. Conclusion Poor R wave sensing, leading to false arrhythmia detection, occurred in a significant number of ICM, and seems to occur independent of ICM type, patient or implant characteristics. Abstract Figure.

Diagnostic accuracy of R-wave detection by insertable cardiac monitors.

Insertable cardiac monitors (ICM) allow prolonged rhythm monitoring, but the diagnostic performance can be hampered by false positive arrhythmia alerts related to inadequate R-wave sensing. This study assesses the prevalence and predictors of inadequate R-wave sensing (both over- and undersensing) among different ICM types. Patients implanted with an ICM at Ghent University Hospital between January 2017 and August 2018 were included. ICM tracings recorded at interrogation or transmitted by remote monitoring were reviewed for inadequate R-wave sensing leading to false arrhythmia alerts. Patient and implant characteristics were retrieved from the medical records and implant reports. The study screened 135 patients (age 59 ± 19 years, 44% female) implanted with different ICM types: Reveal LINQ™ and XT (Medtronic): n=92 (68%), Confirm and Confirm Rx (Abbott): n=35 (26%), and BioMonitor 2 (Biotronik): n=8 (6%). ICM tracings were analyzed in 112 patients (83%). False arrhythmia alerts occurred in 22 (20%) patients, most frequently related to undersensing (77%). False diagnosis of bradycardia or pause was documented in 64%, false high ventricular rates in 14%, and false atrial fibrillation alerts in 22%. Occurrence of R-wave changes was not related to patient characteristics or implant R-wave sensing. A trend toward higher number of inadequate R-wave sensing seems to occur with nonparasternal implant sites (P=.074). False arrhythmia alerts due to inadequate R-wave sensing occurred in 20% of ICM patients independent of implant features and patient characteristics.

Double Trouble: Patients With Both True and False Arrhythmia Alarms

Patients with both true and false arrhythmia alarms pose a challenge because true alarms might be buried among a large number of false alarms, leading to missed true events. To determine (1) the frequency of patients with both true and false arrhythmia alarms; (2) patient, clinical, and electrocardiographic characteristics associated with both true and false alarms; and (3) the frequency and types of true and false arrhythmia alarms. This was a secondary analysis using data from an alarm study conducted at a tertiary academic medical center. Of 461 intensive care unit patients, 211 (46%) had no arrhythmia alarms, 12 (3%) had only true alarms, 167 (36%) had only false alarms, and 71 (15%) had both true and false alarms. Ventricular pacemaker, altered mental status, mechanical ventilation, and cardiac intensive care unit admission were present more often in patients with both true and false alarms than among other patients (P < .001). Intensive care unit stays were longer in patients with only false alarms (mean [SD], 106 [162] hours) and those with both true and false alarms (mean [SD], 208 [333] hours) than in other patients. Accelerated ventricular rhythm was the most common alarm type (37%). An awareness of factors associated with arrhythmia alarms might aid in developing solutions to decrease alarm fatigue. To improve detection of true alarms, further research is needed to build and test electrocardiographic algorithms that adjust for clinical and electrocardiographic characteristics associated with false alarms.

Blind Source Separation for the Aggregation of Machine Learning Algorithms: An Arrhythmia Classification Case

In this work, we present an application of the blind source separation (BSS) algorithm to reduce false arrhythmia alarms and to improve the classification accuracy of artificial neural networks (ANNs). The research was focused on a new approach for model aggregation to deal with arrhythmia types that are difficult to predict. The data for analysis consisted of five-minute-long physiological signals (ECG, BP, and PLETH) registered for patients with cardiac arrhythmias. For each patient, the arrhythmia alarm occurred at the end of the signal. The data present a classification problem of whether the alarm is a true one—requiring attention or is false—should not have been generated. It was confirmed that BSS ANNs are able to detect four arrhythmias—asystole, ventricular tachycardia, ventricular fibrillation, and tachycardia—with higher classification accuracy than the benchmarking models, including the ANN, random forest, and recursive partitioning and regression trees. The overall challenge scores were between 63.2 and 90.7.

Single-modal and multi-modal false arrhythmia alarm reduction using attention-based convolutional and recurrent neural networks.

This study proposes a deep learning model that effectively suppresses the false alarms in the intensive care units (ICUs) without ignoring the true alarms using single- and multi- modal biosignals. Most of the current work in the literature are either rule-based methods, requiring prior knowledge of arrhythmia analysis to build rules, or classical machine learning approaches, depending on hand-engineered features. In this work, we apply convolutional neural networks to automatically extract time-invariant features, an attention mechanism to put more emphasis on the important regions of the segmented input signal(s) that are more likely to contribute to an alarm, and long short-term memory units to capture the temporal information presented in the signal segments. We trained our method efficiently using a two-step training algorithm (i.e., pre-training and fine-tuning the proposed network) on the dataset provided by the PhysioNet computing in cardiology challenge 2015. The evaluation results demonstrate that the proposed method obtains better results compared to other existing algorithms for the false alarm reduction task in ICUs. The proposed method achieves a sensitivity of 93.88% and a specificity of 92.05% for the alarm classification, considering three different signals. In addition, our experiments for 5 separate alarm types leads significant results, where we just consider a single-lead ECG (e.g., a sensitivity of 90.71%, a specificity of 88.30%, an AUC of 89.51 for alarm type of Ventricular Tachycardia arrhythmia).

Reduction of false alarms in the intensive care unit using an optimized machine learning based approach

This work attempts to reduce the number of false alarms generated by bedside monitors in the intensive care unit (ICU), as a majority of current alarms are false. In this study, we applied methods that can be categorized into three stages: signal processing, feature extraction, and optimized machine learning. At the stage of signal processing, we ensured that the heartbeats were properly annotated. During feature extraction, besides extracting features that are relevant to the arrhythmic alarms, we also extracted a set of signal quality indices (SQIs), which we used to distinguish noise/artifact from normal physiological signals. When applying a machine learning algorithm (Random Forest), we performed feature selection in order to reduce the complexity of the models and improve the efficiency of the algorithm. The dataset used is from Reducing False Arrhythmia Alarms in the ICU: the PhysioNet/Computing in Cardiology Challenge 2015. Using the performance metric “score” from the Challenge, we achieved a score of 83.08 in the real-time category on the hidden test set, which is the highest in all published work.

Reducing False Arrhythmia Alarms Using Different Methods of Probability and Class Assignment in Random Forest Learning Methods.

The literature indicates that 90% of clinical alarms in intensive care units might be false. This high percentage negatively impacts both patients and clinical staff. In patients, false alarms significantly increase stress levels, which is especially dangerous for cardiac patients. In clinical staff, alarm overload might lead to desensitization and could result in true alarms being ignored. In this work, we applied the random forest method to reduce false arrhythmia alarms and specifically explored different methods of probability and class assignment, as these affect the classification accuracy of the ensemble classifiers. Due to the complex nature of the problem, i.e., five types of arrhythmia and several methods to determine probability and the alarm class, a synthetic measure based on the ranks was proposed. The novelty of this contribution is the design of a synthetic measure that helps to leverage classification results in an ensemble model that indicates a decision path leading to the best result in terms of the area under the curve (AUC) measure or the global accuracy (score). The results of the research are promising. The best performance in terms of the AUC was 100% accuracy for extreme tachycardia, whereas the poorest results were for ventricular tachycardia at 87%. Similarly, in terms of the accuracy, the best results were observed for extreme tachycardia (91%), whereas ventricular tachycardia alarms were the most difficult to detect, with an accuracy of only 51%.

Relationship between ECG beat detector performance over short time intervals and correct detection of arrhythmic events

Introduction Performance of ECG beat detectors are traditionally assessed using gross sensitivity (Se) and positive predictive value (PPV) on human-annotated datasets over large time windows (e.g.: 30 min). Exceptional performance on gross scores (> 99%) can be achieved by same number of incorrect beat detections occurring nearby or faraway over time. However, missing a true arrhythmic event or detecting an arrhythmic event when one is not actually present can be caused by a limited number of incorrect beat detections occurring nearby over a short period. We compared the relationship between performance of ECG beat detectors measured over short time intervals or by traditional approaches with the correct detection of arrhythmic events. Methods We assessed performance (Se and PPV) of 5 beat detector algorithms over short time intervals (10 s epochs with 5 s overlap) on MIT-BIH arrhythmia database signals. Histograms of Se and PPV were calculated with bin size 1%. Using the histogram, estimated cumulative distribution function (ecdf) for “performance over short interval” was constructed and the area under the ecdf (AUecdf) was calculated. ECG signals where reviewer beat annotations confirmed a true/false alarm in the last 20 s were selected from PhysioNet/Computing in Cardiology Challenge (CinC) 2015 training set. Asystole, bradycardia, and tachycardia alarms were considered. A missed alarm was declared if detected beats did not meet the alarm condition during last 20 s epoch on a record annotated as true alarm. A false alarm was declared if the alarm condition was met on a record annotated as false alarm. For each beat detector total number of incorrect events was the sum of missed and false alarms. The monotonicity of association between performance metrics and number of incorrect events was tested using Spearman's rank correlation coefficient (ρ). Results Table shows performance results for each detector. The number of incorrect events on CinC had no significant correlation with MIT-BIH performance reported as Gross Se (ρ = − 0.60, p = 0.28) or PPV (ρ = − 0.40, p = 0.50) or AUecdf PPV (ρ = − 0.10, p = 0.87). However, AUecdf for Se had the strongest relationship (ρ = 0.90, p = 0.04). Detector Beat detector performance on MIT-BIH arrhythmia database Incorrect arrhythmia detections on CinC 2015 training set Gross score AUecdf Missed alarms False alarms Incorrect arrhythmia events Se PPV Se PPV A B T A B T 1 99.15 99.55 1.98 1.46 1 0 6 17 6 3 33 2 98.59 99.67 2.13 1.28 0 1 18 7 4 0 30 3 99.78 98.82 1.64 2.26 1 1 0 5 2 6 15 4 99.72 99.71 1.57 1.54 0 0 3 1 3 1 8 5 98.86 99.31 2.38 1.85 0 2 14 13 6 4 39 A: Asystole; B: Bradycardia; T: Tachycardia. Discussion A significant relationship between detector performance reported as short time interval metrics with the number of incorrect arrhythmia events was found for Se but not for PPV. Considering performance over short intervals may provide insights on real world performance of beat detectors as automated arrhythmia detectors.

Patient characteristics associated with false arrhythmia alarms in intensive care.

IntroductionA high rate of false arrhythmia alarms in the intensive care unit (ICU) leads to alarm fatigue, the condition of desensitization and potentially inappropriate silencing of alarms due to frequent invalid and nonactionable alarms, often referred to as false alarms.ObjectiveThe aim of this study was to identify patient characteristics, such as gender, age, body mass index, and diagnosis associated with frequent false arrhythmia alarms in the ICU.MethodsThis descriptive, observational study prospectively enrolled patients who were consecutively admitted to one of five adult ICUs (77 beds) at an urban medical center over a period of 31 days in 2013. All monitor alarms and continuous waveforms were stored on a secure server. Nurse scientists with expertise in cardiac monitoring used a standardized protocol to annotate six clinically important types of arrhythmia alarms (asystole, pause, ventricular fibrillation, ventricular tachycardia, accelerated ventricular rhythm, and ventricular bradycardia) as true or false. Total monitoring time for each patient was measured, and the number of false alarms per hour was calculated for these six alarm types. Medical records were examined to acquire data on patient characteristics.ResultsA total of 461 unique patients (mean age =60±17 years) were enrolled, generating a total of 2,558,760 alarms, including all levels of arrhythmia, parameter, and technical alarms. There were 48,404 hours of patient monitoring time, and an average overall alarm rate of 52 alarms/hour. Investigators annotated 12,671 arrhythmia alarms; 11,345 (89.5%) were determined to be false. Two hundred and fifty patients (54%) generated at least one of the six annotated alarm types. Two patients generated 6,940 arrhythmia alarms (55%). The number of false alarms per monitored hour for patients’ annotated arrhythmia alarms ranged from 0.0 to 7.7, and the duration of these false alarms per hour ranged from 0.0 to 158.8 seconds. Patient characteristics were compared in relation to 1) the number and 2) the duration of false arrhythmia alarms per 24-hour period, using nonparametric statistics to minimize the influence of outliers. Among the significant associations were the following: age ≥60 years (P=0.013; P=0.034), confused mental status (P<0.001 for both comparisons), cardiovascular diagnoses (P<0.001 for both comparisons), electrocardiographic (ECG) features, such as wide ECG waveforms that correspond to ventricular depolarization known as QRS complex due to bundle branch block (BBB) (P=0.003; P=0.004) or ventricular paced rhythm (P=0.002 for both comparisons), respiratory diagnoses (P=0.004 for both comparisons), and support with mechanical ventilation, including those with primary diagnoses other than respiratory ones (P<0.001 for both comparisons).ConclusionPatients likely to trigger a higher number of false arrhythmia alarms may be those with older age, confusion, cardiovascular diagnoses, and ECG features that indicate BBB or ventricular pacing, respiratory diagnoses, and mechanical ventilatory support. Algorithm improvements could focus on better noise reduction (eg, motion artifact with confused state) and distinguishing BBB and paced rhythms from ventricular arrhythmias. Increasing awareness of patient conditions that apparently trigger a higher rate of false arrhythmia alarms may be useful for reducing unnecessary noise and improving alarm management.

Reducing false arrhythmia alarm rates using robust heart rate estimation and cost-sensitive support vector machines

To lessen the rate of false critical arrhythmia alarms, we used robust heart rate estimation and cost-sensitive support vector machines. The PhysioNet MIMIC II database and the 2015 PhysioNet/CinC Challenge public database were used as the training dataset; the 2015 Challenge hidden dataset was for testing. Each record had an alarm labeled with asystole, extreme bradycardia, extreme tachycardia, ventricular tachycardia or ventricular flutter/fibrillation. Before alarm onsets, 300 s multimodal data was provided, including electrocardiogram, arterial blood pressure and/or photoplethysmogram. A signal quality modified Kalman filter achieved robust heart rate estimation. Based on this, we extracted heart rate variability features and statistical ECG features. Next, we applied a genetic algorithm (GA) to select the optimal feature combination. Finally, considering the high cost of classifying a true arrhythmia as false, we selected cost-sensitive support vector machines (CSSVMs) to classify alarms. Evaluation on the test dataset showed the overall true positive rate was 95%, and the true negative rate was 85%.

Patient characteristics associated with false arrhythmia alarms in intensive care

Patient characteristics associated with false arrhythmia alarms in intensive care

A novel algorithm for reducing false arrhythmia alarms in intensive care units.

Alarm fatigue in intensive care units (ICU) is one of the top healthcare issues in the US. False alarms in ICU will decrease the quality of care and staff response time over the alarms. Normally, false alarm will cause desensitization of the clinical staff which leads to warnings and misleading, if the triggered alarm is true. In this study, we have proposed a multi-model ensemble approach to reduce the false alarm rate in monitoring systems. We have used 750 patient records from PhysioNet database. At First arrhythmia based features from electrocardiogram (ECG), arterial blood pressure (ABP) and photoplethysmogram (PPG) features were extracted from the records. Next, the dataset has been separated into two subsets on the basis of available features information. The first dataset (DS1) is the combination of ECG physiological, ABP and PPG features. Their correlation coefficient and p-values criteria have been applied for relevant alarm-wise feature-set selection, and random forest classifier was used for model development and validation. The threshold based approach was used on second dataset (DS2) which is the combination of arrhythmia, ABP and PPG features. The developed ensemble model is able to achieve sensitivity 83.33-100 % (average 95.56 %) being true alarms and suppress false alarms rate 66.67-89% (average 77.25%). The predictability of classifier shows the advantage to deal with unbalanced set of information, therefore overall model performance has reached to 83.96% accuracy.

A practical algorithm to reduce false critical ECG alarms using arterial blood pressure and/or photoplethysmogram waveforms

There has been a high rate of false alarms for the critical electrocardiogram (ECG) arrhythmia events in intensive care units (ICUs), from which the ‘crying-wolf’ syndrome may be resulted and patient safety may be jeopardized. This article presents an algorithm to reduce false critical arrhythmia alarms using arterial blood pressure (ABP) and/or photoplethysmogram (PPG) waveform features. We established long duration reference alarm datasets which consist of 573 ICU waveform-alarm records (283 for development set and 290 for test set) with total length of 551 patent days. Each record has continuous recordings of ECGs, ABP and/or PPG signals and contains one or multiple critical ECG alarms. The average length of a record is 23 h. There are totally 2408 critical ECG alarms (1414 in the development set and 994 in the test set), each of which was manually annotated by experts. The algorithm extracts ABP/PPG pulse features on a beat-by-beat basis. For each pulse, five event feature indicators (EFIs), which correspond to the five critical ECG alarms, are generated. At the time of a critical ECG alarm, the corresponding EFI values of those ABP/PPG pulses around the alarm time are checked for adjudicating (accept/reject) this alarm. The algorithm retains all (100%) the true alarms and significantly reduces the false alarms. Our results suggest that the algorithm is effective and practical on account of its real-time dynamic processing mechanism and computational efficiency.

Detection of false arrhythmia alarms with emphasis on ventricular tachycardia

Our approach to detecting false arrhythmia alarms in the intensive care unit breaks down into several tasks. It involves beat detection on different signals: electrocardiogram, photoplethysmogram and arterial blood pressure. The quality of each channel has to be estimated in order to evaluate the reliability of obtained beat detections. The information about the heart rate from the different channels must be integrated in order to find a final conclusion. Some alarm types require particular detectors as is the case of ventricular fibrillation. To identify false ventricular tachycardia alarms we needed to classify heart beats as normal/ventricular. For that purpose we introduce a new feature, QRS polarity type. This feature was important in order to reduce misclassification of ventricular beats: there was an improvement in the ventricular tachycardia alarm true positive rate from 69% to 81%. However, the true negative rate was reduced from 95% to 69% and our global challenge score (real-time event) dropped from 79.02 to 74.28.Our challenge algorithm achieved the third best score in the 2015 PhysioNet/CinC challenge event 1 (real time).