Table of contents

High false alarm rates in the ICU decrease quality of care by slowing staff response times while increasing patient delirium through noise pollution. The 2015 PhysioNet/Computing in Cardiology Challenge provides a set of 1250 multi-parameter ICU data segments associated with critical arrhythmia alarms, and challenges the general research community to address the issue of false alarm suppression using all available signals. Each data segment was 5 minutes long (for real time analysis), ending at the time of the alarm. For retrospective analysis, we provided a further 30 seconds of data after the alarm was triggered.

A total of 750 data segments were made available for training and 500 were held back for testing. Each alarm was reviewed by expert annotators, at least two of whom agreed that the alarm was either true or false. Challenge participants were invited to submit a complete, working algorithm to distinguish true from false alarms, and received a score based on their program's performance on the hidden test set. This score was based on the percentage of alarms correct, but with a penalty that weights the suppression of true alarms five times more heavily than acceptance of false alarms.

We provided three example entries based on well-known, open source signal processing algorithms, to serve as a basis for comparison and as a starting point for participants to develop their own code. A total of 38 teams submitted a total of 215 entries in this year's Challenge.

This editorial reviews the background issues for this challenge, the design of the challenge itself, the key achievements, and the follow-up research generated as a result of the Challenge, published in the concurrent special issue of Physiological Measurement. Additionally we make some recommendations for future changes in the field of patient monitoring as a result of the Challenge.

This paper presents a novel approach for false alarm suppression using machine learning tools. It proposes a multi-modal detection algorithm to find the true beats using the information from all the available waveforms. This method uses a variety of beat detection algorithms, some of which are developed by the authors. The outputs of the beat detection algorithms are combined using a machine learning approach. For the ventricular tachycardia and ventricular fibrillation alarms, separate classification models are trained to distinguish between the normal and abnormal beats. This information, along with alarm-specific criteria, is used to decide if the alarm is false. The results indicate that the presented method was effective in suppressing false alarms when it was tested on a hidden validation dataset.

In this paper, we propose an algorithm that classifies whether a generated cardiac arrhythmia alarm is true or false. The large number of false alarms in intensive care is a severe issue. The noise peaks caused by alarms can be high and in a noisy environment nurses can experience stress and fatigue. In addition, patient safety is compromised because reaction time of the caregivers to true alarms is reduced.

The data for the algorithm development consisted of records of electrocardiogram (ECG), arterial blood pressure, and photoplethysmogram signals in which an alarm for either asystole, extreme bradycardia, extreme tachycardia, ventricular fibrillation or flutter, or ventricular tachycardia occurs. First, heart beats are extracted from every signal. Next, the algorithm selects the most reliable signal pair from the available signals by comparing how well the detected beats match between different signals based on ${{\text{F}}_{1}}$-score and selecting the best match. From the selected signal pair, arrhythmia specific features, such as heart rate features and signal purity index are computed for the alarm classification. The classification is performed with five separate Random Forest models. In addition, information on the local noise level of the selected ECG lead is added to the classification. The algorithm was trained and evaluated with the PhysioNet/Computing in Cardiology Challenge 2015 data set. In the test set the overall true positive rates were 93 and 95% and true negative rates 80 and 83%, respectively for events with no information and events with information after the alarm. The overall challenge scores were 77.39 and 81.58.

The purpose of this study was to develop algorithms to lower the incidence of false arrhythmia alarms in the ICU using information from independent sources, namely electrocardiogram (ECG), arterial blood pressure (ABP) and photoplethysmogram (PPG). Our approach relies on robust adaptive signal processing techniques in order to extract accurate heart rate (HR) values from the different waveforms. Based on the quality of available signals, heart rate was either estimated from pulsatile waveforms using an adaptive frequency tracking algorithm or computed from ECGs using an adaptive mathematical morphology approach. Furthermore, we developed a supplementary measure based on the spectral purity of the ECGs to determine whether a ventricular tachycardia or flutter/fibrillation arrhythmia has taken place. Finally, alarm veracity was determined based on a set of decision rules on HR and spectral purity values. The proposed method was evaluated on the PhysioNet/CinC Challenge 2015 database, which is composed of 1250 life-threatening alarm recordings, each categorized into either bradycardia, tachycardia, asystole, ventricular tachycardia or ventricular flutter/fibrillation arrhythmia. This resulted in overall true positive rates of 95%/99% and overall true negative rates of 76%/80% on the real-time and retrospective subsets of the test dataset, respectively.

False arrhythmia alarms pose a major threat to the quality of care in today's ICU. Thus, the PhysioNet/Computing in Cardiology Challenge 2015 aimed at reducing false alarms by exploiting multimodal cardiac signals recorded by a patient monitor. False alarms for asystole, extreme bradycardia, extreme tachycardia, ventricular flutter/fibrillation as well as ventricular tachycardia were to be reduced using two electrocardiogram channels, up to two cardiac signals of mechanical origin as well as a respiratory signal.

In this paper, an approach combining multimodal rhythmicity estimation and machine learning is presented. Using standard short-time autocorrelation and robust beat-to-beat interval estimation, the signal's self-similarity is analyzed. In particular, beat intervals as well as quality measures are derived which are further quantified using basic mathematical operations (min, mean, max, etc). Moreover, methods from the realm of image processing, 2D Fourier transformation combined with principal component analysis, are employed for dimensionality reduction. Several machine learning approaches are evaluated including linear discriminant analysis and random forest.

Using an alarm-independent reduction strategy, an overall false alarm reduction with a score of 65.52 in terms of the real-time scoring system of the challenge is achieved on a hidden dataset. Employing an alarm-specific strategy, an overall real-time score of 78.20 at a true positive rate of 95% and a true negative rate of 78% is achieved. While the results for some categories still need improvement, false alarms for extreme tachycardia are suppressed with 100% sensitivity and specificity.

In this paper, as a contribution to the Physionet/Computing in Cardiology 2015 Challenge, we present individual algorithms to accurately classify five different life threatening arrhythmias with the goal of suppressing false alarm generation in intensive care units. Information obtained by analysing electrocardiogram, photoplethysmogram and arterial blood pressure signals was utilized to develop the classification models. Prior to classification, the signals were subject to a signal pre-processing stage for quality analysis. Classification was performed using a combination of support vector machine based machine learning approach and logical analysis techniques. The predicted result for a certain arrhythmia classification model was verified by logical analysis to aid in reduction of false alarms. Separate feature vectors were formed for predicting the presence or absence of each arrhythmia, using both spectral and time-domain information. The training and test data were obtained from the Physionet/CinC Challenge 2015 database. Classification algorithms were written for two different categories of data, namely real-time and retrospective, whose data lengths were 10 s and an additional 30 s, respectively. For the real-time test dataset, sensitivity of 94% and specificity of 82% were obtained. Similarly, for the retrospective test dataset, sensitivity of 94% and specificity of 86% were obtained.

False intensive care unit (ICU) alarms induce stress in both patients and clinical staff and decrease the quality of care, thus significantly increasing both the hospital recovery time and rehospitalization rates. In the PhysioNet/CinC Challenge 2015 for reducing false arrhythmia alarms in ICU bedside monitor data, this paper validates the application of a real-time arrhythmia detection library (ADLib, Schiller AG) for the robust detection of five types of life-threatening arrhythmia alarms. The strength of the application is to give immediate feedback on the arrhythmia event within a scan interval of 3 s–7.5 s, and to increase the noise immunity of electrocardiogram (ECG) arrhythmia analysis by fusing its decision with supplementary ECG quality interpretation and real-time pulse wave monitoring (quality and hemodynamics) using arterial blood pressure or photoplethysmographic signals. We achieved the third-ranked real-time score (79.41) in the challenge (Event 1), however, the rank was not officially recognized due to the 'closed-source' entry. This study shows the optimization of the alarm decision module, using tunable parameters such as the scan interval, lead quality threshold, and pulse wave features, with a follow-up improvement of the real-time score (80.07). The performance (true positive rate, true negative rate) is reported in the blinded challenge test set for different arrhythmias: asystole (83%, 96%), extreme bradycardia (100%, 90%), extreme tachycardia (98%, 80%), ventricular tachycardia (84%, 82%), and ventricular fibrillation (78%, 84%). Another part of this study considers the validation of ADLib with four reference ECG databases (AHA, EDB, SVDB, MIT-BIH) according to the international recommendations for performance reports in ECG monitors (ANSI/AAMI EC57). The sensitivity (Se) and positive predictivity (+P) are: QRS detector QRS (Se, +P)  >  99.7%, ventricular ectopic beat (VEB) classifier VEB (Se, +P)  =  95%, and ventricular fibrillation detector VFIB (P  +  = 94.8%)  >  VFIB (Se  =  86.4%), adjusted to the clinical setting requirements, giving preference to low false positive alarms.

False alarm (FA) rates as high as 86% have been reported in intensive care unit monitors. High FA rates decrease quality of care by slowing staff response times while increasing patient burdens and stresses. In this study, we proposed a rule-based and multi-channel information fusion method for accurately classifying the true or false alarms for five life-threatening arrhythmias: asystole (ASY), extreme bradycardia (EBR), extreme tachycardia (ETC), ventricular tachycardia (VTA) and ventricular flutter/fibrillation (VFB). The proposed method consisted of five steps: (1) signal pre-processing, (2) feature detection and validation, (3) true/false alarm determination for each channel, (4) 'real-time' true/false alarm determination and (5) 'retrospective' true/false alarm determination (if needed). Up to four signal channels, that is, two electrocardiogram signals, one arterial blood pressure and/or one photoplethysmogram signal were included in the analysis. Two events were set for the method validation: event 1 for 'real-time' and event 2 for 'retrospective' alarm classification. The results showed that 100% true positive ratio (i.e. sensitivity) on the training set were obtained for ASY, EBR, ETC and VFB types, and 94% for VTA type, accompanied by the corresponding true negative ratio (i.e. specificity) results of 93%, 81%, 78%, 85% and 50% respectively, resulting in the score values of 96.50, 90.70, 88.89, 92.31 and 64.90, as well as with a final score of 80.57 for event 1 and 79.12 for event 2. For the test set, the proposed method obtained the score of 88.73 for ASY, 77.78 for EBR, 89.92 for ETC, 67.74 for VFB and 61.04 for VTA types, with the final score of 71.68 for event 1 and 75.91 for event 2.

False alarms in intensive care units represent a serious threat to patients. We propose a method for detection of five live-threatening arrhythmias. It is designed to work with multimodal data containing electrocardiograph and arterial blood pressure or photoplethysmograph signals. The presented method is based on descriptive statistics and Fourier and Hilbert transforms. It was trained using 750 records. The method was validated during the follow-up phase of the CinC/Physionet Challenge 2015 on a hidden dataset with 500 records, achieving a sensitivity of 93% (95%) and a specificity of 87% (88%) for real-time (retrospective) files. The given sensitivity and specificity resulted in score of 81.62 (84.96) for real-time (retrospective) records. The presented method is an improved version of the original algorithm awarded the first and the second prize in CinC/Physionet Challenge 2015.

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).

This study developed algorithms to decrease the arrhythmia false alarms in the ICU by processing multimodal signals of photoplethysmography (PPG), arterial blood pressure (ABP), and two ECG signals. The goal was to detect the five critical arrhythmias comprising asystole (ASY), extreme bradycardia (EBR), extreme tachycardia (ETC), ventricular tachycardia (VTA), and ventricular flutter or fibrillation (VFB). The different characteristics of the arrhythmias suggested the application of individual signal processing for each alarm and the combination of the algorithms to enhance false alarm detection. Thus, different features and signal processing techniques were used for each arrhythmia type. The ECG signals were first processed to reduce the signal interference. Then, a Hilbert-transform based QRS detector algorithm was utilized to identify the QRS complexes, which were then processed to determine the instantaneous heart rate. The pulsatile signals (PPG and ABP) were processed to discover the pulse onset of beats which were then employed to measure the heart rate. The signal quality index (SQI) of the signals was implemented to verify the integrity of the heart rate information. The overall score obtained by our algorithms in the 2015 Computing in Cardiology Challenge was a score of 74.03% for retrospective and 69.92% for real-time analysis.

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.

False and non-actionable alarms in critical care can be reduced by developing algorithms which assess the trueness of an arrhythmia alarm from a bedside monitor. Computational approaches that automatically identify artefacts in ECG signals are an important branch of physiological signal processing which tries to address this issue. Signal quality indices (SQIs) derived considering differences between artefacts which occur in ECG signals and normal QRS morphology have the potential to discriminate pathologically different arrhythmic ECG segments as artefacts. Using ECG signals from the PhysioNet/Computing in Cardiology Challenge 2015 training set, we studied previously reported ECG SQIs in the scientific literature to differentiate ECG segments with artefacts from arrhythmic ECG segments. We found that the ability of SQIs to discriminate between ECG artefacts and arrhythmic ECG varies based on arrhythmia type since the pathology of each arrhythmic ECG waveform is different. Therefore, to reduce the risk of SQIs classifying arrhythmic events as noise it is important to validate and test SQIs with databases that include arrhythmias. Arrhythmia specific SQIs may also minimize the risk of misclassifying arrhythmic events as noise.

Electrocardiograms (ECGs) recorded from patients in intensive care were investigated to quantify any relationship between ECG signal quality and false monitoring alarms. False alarms are a considerable problem for nursing and medical staff as they distract from clinical care, and are also a problem for patients as they disturb rest, which is important for clinical recovery. ECG and alarm data were obtained for 750 patient alarms from the PhysioNet database. The final 8 s period before the alarm was triggered was investigated. All but one ECG channel in 38 ECG recordings with out-of-range data were associated with false positive alarms (p  <  0.0001). The frequency contributions for baseline (BL) instability, electromyogram (EMG) muscle noise, and high frequency (HF) noise were calculated. For all three frequency bands, the contributions associated with false positive alarms were very significantly greater than for true positive alarms (p  <  0.0001). The greatest difference was for BL with a mean level for false positive alarms 4.0 times greater than for true positive alarms, followed by EMG and HF at 1.6 times and 1.4 times respectively. These results confirm that attention needs to be taken to improve ECG signal quality to reduce the frequency of clinical false alarms, and hence improve conditions for clinical staff and patients.