Dataset Information

Using Supervised Machine Learning to Classify Real Alerts and Artifact in Online Multisignal Vital Sign Monitoring Data.

ABSTRACT: The use of machine-learning algorithms to classify alerts as real or artifacts in online noninvasive vital sign data streams to reduce alarm fatigue and missed true instability.Observational cohort study.Twenty-four-bed trauma step-down unit.Two thousand one hundred fifty-three patients.Noninvasive vital sign monitoring data (heart rate, respiratory rate, peripheral oximetry) recorded on all admissions at 1/20 Hz, and noninvasive blood pressure less frequently, and partitioned data into training/validation (294 admissions; 22,980 monitoring hours) and test sets (2,057 admissions; 156,177 monitoring hours). Alerts were vital sign deviations beyond stability thresholds. A four-member expert committee annotated a subset of alerts (576 in training/validation set, 397 in test set) as real or artifact selected by active learning, upon which we trained machine-learning algorithms. The best model was evaluated on test set alerts to enact online alert classification over time.The Random Forest model discriminated between real and artifact as the alerts evolved online in the test set with area under the curve performance of 0.79 (95% CI, 0.67-0.93) for peripheral oximetry at the instant the vital sign first crossed threshold and increased to 0.87 (95% CI, 0.71-0.95) at 3 minutes into the alerting period. Blood pressure area under the curve started at 0.77 (95% CI, 0.64-0.95) and increased to 0.87 (95% CI, 0.71-0.98), whereas respiratory rate area under the curve started at 0.85 (95% CI, 0.77-0.95) and increased to 0.97 (95% CI, 0.94-1.00). Heart rate alerts were too few for model development.Machine-learning models can discern clinically relevant peripheral oximetry, blood pressure, and respiratory rate alerts from artifacts in an online monitoring dataset (area under the curve > 0.87).

SUBMITTER: Chen L

PROVIDER: S-EPMC4911247 | biostudies-literature | 2016 Jul

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Using Supervised Machine Learning to Classify Real Alerts and Artifact in Online Multisignal Vital Sign Monitoring Data.

Chen Lujie L Dubrawski Artur A Wang Donghan D Fiterau Madalina M Guillame-Bert Mathieu M Bose Eliezer E Kaynar Ata M AM Wallace David J DJ Guttendorf Jane J Clermont Gilles G Pinsky Michael R MR Hravnak Marilyn M

Critical care medicine 20160701 7

<h4>Objective</h4>The use of machine-learning algorithms to classify alerts as real or artifacts in online noninvasive vital sign data streams to reduce alarm fatigue and missed true instability.<h4>Design</h4>Observational cohort study.<h4>Setting</h4>Twenty-four-bed trauma step-down unit.<h4>Patients</h4>Two thousand one hundred fifty-three patients.<h4>Intervention</h4>Noninvasive vital sign monitoring data (heart rate, respiratory rate, peripheral oximetry) recorded on all admissions at 1/20 ...[more]

PMID: 26992068

Similar Datasets

Project description:BackgroundTimely recognition of patient deterioration remains challenging. Ambulatory monitoring systems (AMSs) may provide support to current monitoring practices; however, they need to be thoroughly tested before implementation in the clinical environment for early detection of deterioration.ObjectiveThe objective of this study was to assess the wearability of a selection of commercially available AMSs to inform a future prospective study of ambulatory vital sign monitors in an acute hospital ward.MethodsFive pulse oximeters (4 with finger probes and 1 wrist-worn only, collecting pulse rates and oxygen saturation) and 2 chest patches (collecting heart rates and respiratory rates) were selected to be part of this study: The 2 chest-worn patches were VitalPatch (VitalConnect) and Peerbridge Cor (Peerbridge); the 4 wrist-worn devices with finger probe were Nonin WristOx2 3150 (Nonin), Checkme O2+ (Viatom Technology), PC-68B, and AP-20 (both from Creative Medical); and the 1 solely wrist-worn device was Wavelet (Wavelet Health). Adult participants wore each device for up to 72 hours while performing usual "activities of daily living" and were asked to score the perceived exertion and perception of pain or discomfort by using the Borg CR-10 scale; thoughts and feelings caused by the AMS using the Comfort Rating Scale (CRS); and to provide general free text feedback. Median and IQRs were reported and nonparametric tests were used to assess differences between the devices' CRS scores.ResultsQuantitative scores and feedback were collected in 70 completed questionnaires from 20 healthy volunteers, with each device tested approximately 10 times. The Wavelet seemed to be the most wearable device (P<.001) with an overall median (IQR) CRS score of 1.00 (0.88). There were no statistically significant differences in wearability between the chest patches in the CRS total score; however, the VitalPatch was superior in the Attachment section (P=.04) with a median (IQR) score of 3.00 (1.00). General pain and discomfort scores and total percentage of time worn are also reflective of this.ConclusionsOur results suggest that adult participants prefer to wear wrist-worn pulse oximeters without a probe compressing the fingertip and they prefer to wear a smaller chest patch. A compromise between wearability, reliability, and accuracy should be made for successful and practical integration of AMSs within the hospital environment.

Project description:BackgroundThe standard of care in general wards includes periodic manual measurements, with the data entered into track-and-trigger charts, either on paper or electronically. Wearable devices may support health care staff, improve patient safety, and promote early deterioration detection in the interval between periodic measurements. However, regulatory standards for ambulatory cardiac monitors estimating heart rate (HR) and respiratory rate (RR) do not specify performance criteria during patient movement or clinical conditions in which the patient's oxygen saturation varies. Therefore, further validation is required before clinical implementation and deployment of any wearable system that provides continuous vital sign measurements.ObjectiveThe objective of this study is to determine the agreement between a chest-worn patch (VitalPatch) and a gold standard reference device for HR and RR measurements during movement and gradual desaturation (modeling a hypoxic episode) in a controlled environment.MethodsAfter the VitalPatch and gold standard devices (Philips MX450) were applied, participants performed different movements in seven consecutive stages: at rest, sit-to-stand, tapping, rubbing, drinking, turning pages, and using a tablet. Hypoxia was then induced, and the participants' oxygen saturation gradually reduced to 80% in a controlled environment. The primary outcome measure was accuracy, defined as the mean absolute error (MAE) of the VitalPatch estimates when compared with HR and RR gold standards (3-lead electrocardiography and capnography, respectively). We defined these as clinically acceptable if the rates were within 5 beats per minute for HR and 3 respirations per minute (rpm) for RR.ResultsComplete data sets were acquired for 29 participants. In the movement phase, the HR estimates were within prespecified limits for all movements. For RR, estimates were also within the acceptable range, with the exception of the sit-to-stand and turning page movements, showing an MAE of 3.05 (95% CI 2.48-3.58) rpm and 3.45 (95% CI 2.71-4.11) rpm, respectively. For the hypoxia phase, both HR and RR estimates were within limits, with an overall MAE of 0.72 (95% CI 0.66-0.78) beats per minute and 1.89 (95% CI 1.75-2.03) rpm, respectively. There were no significant differences in the accuracy of HR and RR estimations between normoxia (≥90%), mild (89.9%-85%), and severe hypoxia (<85%).ConclusionsThe VitalPatch was highly accurate throughout both the movement and hypoxia phases of the study, except for RR estimation during the two types of movements. This study demonstrated that VitalPatch can be safely tested in clinical environments to support earlier detection of cardiorespiratory deterioration.Trial registrationISRCTN Registry ISRCTN61535692; https://www.isrctn.com/ISRCTN61535692.

Project description:BackgroundContinuous monitoring of patient vital signs may improve patient outcomes. Head-worn displays (HWDs) can provide hands-free access to continuous vital sign information of patients in critical and acute care contexts and thus may reduce instances of unrecognized patient deterioration.ObjectiveThe purpose of the study is to conduct a systematic review of the literature to evaluate clinical, surrogate, and process outcomes when clinicians use HWDs for continuous patient vital sign monitoring.MethodsThe review was registered with PROSPERO (CRD42019119875) and followed the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-analyses) guidelines. A literature search was conducted for articles published between January 1995 and June 2020 using the following databases: PubMed, Embase, CINAHL, PsycINFO, and Web of Science. Overall, 2 reviewers independently screened titles and abstracts and then assessed the full text of the articles. Original research articles that evaluated the clinical, surrogate, or process outcomes of head-mounted displays for continuous vital sign monitoring in critical care or acute care contexts were included.ResultsOf the 214 records obtained, 15 (7%) articles met the predefined criteria and were included in this review. Of the 15 studies, 7 (47%) took place in a clinical context, whereas the remainder took place in a simulation environment. In 100% (7/7) of the studies that evaluated gaze behavior, changes were found in gaze direction with HWDs. Change detection improvements were found in 67% (2/3) of the studies evaluating changes in the participants' ability to detect changes in vital signs. Of the 10 studies assessing the ease of use of the HWD, most participants of 7 (70%) studies reported that the HWD was easy to use. In all 6 studies in which participants were asked if they would consider using the HWD in their practice, most participants responded positively, but they often suggested improvements on the HWD hardware or display design. Of the 7 studies conducted in clinical contexts, none reported any clinical outcomes.ConclusionsAlthough there is limited and sometimes conflicting evidence about the benefits of HWDs from certain surrogate and process outcomes, evidence for clinical outcomes is lacking. Recommendations are to employ user-centered design when developing HWDs, perform longitudinal studies, and seek clinical outcomes.Trial registrationPROSPERO International Prospective Register of Systematic Reviews CRD42019119875; https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=119875.

Dataset Information

Using Supervised Machine Learning to Classify Real Alerts and Artifact in Online Multisignal Vital Sign Monitoring Data.

Publications

Using Supervised Machine Learning to Classify Real Alerts and Artifact in Online Multisignal Vital Sign Monitoring Data.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure