Robust parameter extraction for decision support using multimodal intensive care data

A broadly applicable approach to enrich electronic-health-record cohorts by identifying patients with complete data: a multisite evaluation

OBJECTIVE

Patients who receive most care within a single healthcare system (colloquially called a "loyalty cohort" since they typically return to the same providers) have mostly complete data within that organization's electronic health record (EHR). Loyalty cohorts have low data missingness, which can unintentionally bias research results. Using proxies of routine care and healthcare utilization metrics, we compute a per-patient score that identifies a loyalty cohort.

MATERIALS AND METHODS

We implemented a computable program for the widely adopted i2b2 platform that identifies loyalty cohorts in EHRs based on a machine-learning model, which was previously validated using linked claims data. We developed a novel validation approach, which tests, using only EHR data, whether patients returned to the same healthcare system after the training period. We evaluated these tools at 3 institutions using data from 2017 to 2019.

RESULTS

Loyalty cohort calculations to identify patients who returned during a 1-year follow-up yielded a mean area under the receiver operating characteristic curve of 0.77 using the original model and 0.80 after calibrating the model at individual sites. Factors such as multiple medications or visits contributed significantly at all sites. Screening tests' contributions (eg, colonoscopy) varied across sites, likely due to coding and population differences.

DISCUSSION

This open-source implementation of a "loyalty score" algorithm had good predictive power. Enriching research cohorts by utilizing these low-missingness patients is a way to obtain the data completeness necessary for accurate causal analysis.

CONCLUSION

i2b2 sites can use this approach to select cohorts with mostly complete EHR data.

Statistical shape modeling-based algorithm for replacing missing beats in blood pressure signals

In clinical environments, such as the Intensive Care Unit (ICU), continuous and uninterrupted monitoring of vital signs is critical for the early detection of patient deterioration and prompt intervention. Since data collected in these settings are often corrupted by noise, artifacts, or recording gaps, it is important to estimate missing data for a more accurate signal assessment.In this study, we propose an automatic algorithm for reconstructing of arterial blood pressure signal waveforms. The methodological core of the algorithm is based on the idea of statistical shape modeling, which basically estimates the shape variation of beat waveforms in order to reconstruct them in noisy segments. The waveform reconstruction is achieved by combining the average beat template from a 90-second segment of clean signal preceding the gap with the main shape variations of the estimated waveform.The algorithm was validated using arterial blood pressure recordings from 9 subjects admitted in the ICU and collected in the MIMIC-III Waveform Database, each lasting 1 hour and sampled at 125 Hz. For each record, ten fictitious gaps were created, and the reconstructed segments were compared to the original signals with the metrics proposed within the PhysioNet / Computing in Cardiology Challenge 2010. Results demonstrate the excellent performance of the proposed algorithm, with overall averages of both Q1 and Q2 metrics greater than 0.85.

Missing data matter: an empirical evaluation of the impacts of missing EHR data in comparative effectiveness research

OBJECTIVES

The impacts of missing data in comparative effectiveness research (CER) using electronic health records (EHRs) may vary depending on the type and pattern of missing data. In this study, we aimed to quantify these impacts and compare the performance of different imputation methods.

MATERIALS AND METHODS

We conducted an empirical (simulation) study to quantify the bias and power loss in estimating treatment effects in CER using EHR data. We considered various missing scenarios and used the propensity scores to control for confounding. We compared the performance of the multiple imputation and spline smoothing methods to handle missing data.

RESULTS

When missing data depended on the stochastic progression of disease and medical practice patterns, the spline smoothing method produced results that were close to those obtained when there were no missing data. Compared to multiple imputation, the spline smoothing generally performed similarly or better, with smaller estimation bias and less power loss. The multiple imputation can still reduce study bias and power loss in some restrictive scenarios, eg, when missing data did not depend on the stochastic process of disease progression.

DISCUSSION AND CONCLUSION

Missing data in EHRs could lead to biased estimates of treatment effects and false negative findings in CER even after missing data were imputed. It is important to leverage the temporal information of disease trajectory to impute missing values when using EHRs as a data resource for CER and to consider the missing rate and the effect size when choosing an imputation method.

Integrating Physiological Data Artifacts Detection With Clinical Decision Support Systems: Observational Study

BACKGROUND

Clinical decision support systems (CDSS) have the potential to lower the patient mortality and morbidity rates. However, signal artifacts present in physiological data affect the reliability and accuracy of the CDSS. Moreover, patient monitors and other medical devices generate false alarms while processing physiological data, further leading to alarm fatigue because of increased noise levels, staff disruption, and staff desensitization in busy critical care environments. This adversely affects the quality of care at the patient bedside. Hence, artifact detection (AD) algorithms play a crucial role in assessing the quality of physiological data and mitigating the impact of these artifacts.

OBJECTIVE

The aim of this study is to evaluate a novel AD framework for integrating AD algorithms with CDSS. We designed the framework with features that support real-time implementation within critical care. In this study, we evaluated the framework and its features in a false alarm reduction study. We developed static framework component models, followed by dynamic framework compositions to formulate four CDSS. We evaluated these formulations using neonatal patient data and validated the six framework features: flexibility, reusability, signal quality indicator standardization, scalability, customizability, and real-time implementation support.

METHODS

We developed four exemplar static AD components with standardized requirements and provisions interfaces that facilitate the interoperability of framework components. These AD components were mixed and matched into four different AD compositions to mitigate the artifacts' effects. We developed a novel static clinical event detection component that is integrated with each AD composition to formulate and evaluate a dynamic CDSS for peripheral oxygen saturation (SpO2) alarm generation. This study collected data from 11 patients with diverse pathologies in the neonatal intensive care unit. Collected data streams and corresponding alarms include pulse rate and SpO2 measured from a pulse oximeter (Masimo SET SmartPod) integrated with an Infinity Delta monitor and the heart rate derived from electrocardiography leads attached to a second Infinity Delta monitor.

RESULTS

A total of 119 SpO2 alarms were evaluated. The lowest achievable SpO2 false alarm rate was 39%, with a sensitivity of 80%. This demonstrates the framework's utility in identifying the best possible dynamic composition to serve the clinical need for false SpO2 alarm reduction and subsequent alarm fatigue, given the limitations of a small sample size.

CONCLUSIONS

The framework features, including reusability, signal quality indicator standardization, scalability, and customizability, allow the evaluation and comparison of novel CDSS formulations. The optimal solution for a CDSS can then be hard-coded and integrated within clinical workflows for real-time implementation. The flexibility to serve different clinical needs and standardized component interoperability of the framework supports the potential for a real-time clinical implementation of AD.

Data-driven curation process for describing the blood glucose management in the intensive care unit

Analysis of real-world glucose and insulin clinical data recorded in electronic medical records can provide insights into tailored approaches to clinical care, yet presents many analytic challenges. This work makes publicly available a dataset that contains the curated entries of blood glucose readings and administered insulin on a per-patient basis during ICU admissions in the Medical Information Mart for Intensive Care (MIMIC-III) database version 1.4. Also, the present study details the data curation process used to extract and match glucose values to insulin therapy. The curation process includes the creation of glucose-insulin pairing rules according to clinical expert-defined physiologic and pharmacologic parameters. Through this approach, it was possible to align nearly 76% of insulin events to a preceding blood glucose reading for nearly 9,600 critically ill patients. This work has the potential to reveal trends in real-world practice for the management of blood glucose. This data extraction and processing serve as a framework for future studies of glucose and insulin in the intensive care unit.

Uncertainty in heart rate complexity metrics caused by R-peak perturbations

Heart rate complexity (HRC) is a proven metric for gaining insight into human stress and physiological deterioration. To calculate HRC, the detection of the exact instance of when the heart beats, the R-peak, is necessary. Electrocardiogram (ECG) signals can often be corrupted by environmental noise (e.g., from electromagnetic interference, movement artifacts), which can potentially alter the HRC measurement, producing erroneous inputs which feed into decision support models. Current literature has only investigated how HRC is affected by noise when R-peak detection errors occur (false positives and false negatives). However, the numerical methods used to calculate HRC are also sensitive to the specific location of the fiducial point of the R-peak. This raises many questions regarding how this fiducial point is altered by noise, the resulting impact on the measured HRC, and how we can account for noisy HRC measures as inputs into our decision models. This work uses Monte Carlo simulations to systematically add white and pink noise at different permutations of signal-to-noise ratios (SNRs), time segments, sampling rates, and HRC measurements to characterize the influence of noise on the HRC measure by altering the fiducial point of the R-peak. Using the generated information from these simulations provides improved decision processes for system design which address key concerns such as permutation entropy being a more precise, reliable, less biased, and more sensitive measurement for HRC than sample and approximate entropy.

"Usability of data integration and visualization software for multidisciplinary pediatric intensive care: a human factors approach to assessing technology"

Background

Intensive care clinicians use several sources of data in order to inform decision-making. We set out to evaluate a new interactive data integration platform called T3™ made available for pediatric intensive care. Three primary functions are supported: tracking of physiologic signals, displaying trajectory, and triggering decisions, by highlighting data or estimating risk of patient instability. We designed a human factors study to identify interface usability issues, to measure ease of use, and to describe interface features that may enable or hinder clinical tasks.

Methods

Twenty-two participants, consisting of bedside intensive care physicians, nurses, and respiratory therapists, tested the T3™ interface in a simulation laboratory setting. Twenty tasks were performed with a true-to-setting, fully functional, prototype, populated with physiological and therapeutic intervention patient data. Primary data visualization was time series and secondary visualizations were: 1) shading out-of-target values, 2) mini-trends with exaggerated maxima and minima (sparklines), and 3) bar graph of a 16-parameter indicator. Task completion was video recorded and assessed using a use error rating scale. Usability issues were classified in the context of task and type of clinician. A severity rating scale was used to rate potential clinical impact of usability issues.

Results

Time series supported tracking a single parameter but partially supported determining patient trajectory using multiple parameters. Visual pattern overload was observed with multiple parameter data streams. Automated data processing using shading and sparklines was often ignored but the 16-parameter data reduction algorithm, displayed as a persistent bar graph, was visually intuitive. However, by selecting or automatically processing data, triggering aids distorted the raw data that clinicians use regularly. Consequently, clinicians could not rely on new data representations because they did not know how they were established or derived.

Conclusions

Usability issues, observed through contextual use, provided directions for tangible design improvements of data integration software that may lessen use errors and promote safe use. Data-driven decision making can benefit from iterative interface redesign involving clinician-users in simulated environments. This study is a first step in understanding how software can support clinicians’ decision making with integrated continuous monitoring data. Importantly, testing of similar platforms by all the different disciplines who may become clinician users is a fundamental step necessary to understand the impact on clinical outcomes of decision aids.

Electronic supplementary material

The online version of this article (doi:10.1186/s12911-017-0520-7) contains supplementary material, which is available to authorized users.

A Review of Wearable Sensor Systems for Monitoring Body Movements of Neonates

Characteristics of physical movements are indicative of infants' neuro-motor development and brain dysfunction. For instance, infant seizure, a clinical signal of brain dysfunction, could be identified and predicted by monitoring its physical movements. With the advance of wearable sensor technology, including the miniaturization of sensors, and the increasing broad application of micro- and nanotechnology, and smart fabrics in wearable sensor systems, it is now possible to collect, store, and process multimodal signal data of infant movements in a more efficient, more comfortable, and non-intrusive way. This review aims to depict the state-of-the-art of wearable sensor systems for infant movement monitoring. We also discuss its clinical significance and the aspect of system design.

-Omic and Electronic Health Record Big Data Analytics for Precision Medicine

OBJECTIVE

Rapid advances of high-throughput technologies and wide adoption of electronic health records (EHRs) have led to fast accumulation of -omic and EHR data. These voluminous complex data contain abundant information for precision medicine, and big data analytics can extract such knowledge to improve the quality of healthcare.

METHODS

In this paper, we present -omic and EHR data characteristics, associated challenges, and data analytics including data preprocessing, mining, and modeling.

RESULTS

To demonstrate how big data analytics enables precision medicine, we provide two case studies, including identifying disease biomarkers from multi-omic data and incorporating -omic information into EHR.

CONCLUSION

Big data analytics is able to address -omic and EHR data challenges for paradigm shift toward precision medicine.

SIGNIFICANCE

Big data analytics makes sense of -omic and EHR data to improve healthcare outcome. It has long lasting societal impact.

A review of methods for the signal quality assessment to improve reliability of heart rate and blood pressures derived parameters

The assessment of signal quality has been a research topic since the late 1970s, as it is mainly related to the problem of false alarms in bedside monitors in the intensive care unit (ICU), the incidence of which can be as high as 90 %, leading to alarm fatigue and a drop in the overall level of nurses and clinicians attention. The development of efficient algorithms for the quality control of long diagnostic electrocardiographic (ECG) recordings, both single- and multi-lead, and of the arterial blood pressure (ABP) signal is therefore essential for the enhancement of care quality. The ECG signal is often corrupted by noise, which can be within the frequency band of interest and can manifest similar morphologies as the ECG itself. Similarly to ECG, also the ABP signal is often corrupted by non-Gaussian, nonlinear and non-stationary noise and artifacts, especially in ICU recordings. Moreover, the reliability of several important parameters derived from ABP such as systolic blood pressure or pulse pressure is strongly affected by the quality of the ABP waveform. In this work, several up-to-date algorithms for the quality scoring of a single- or multi-lead ECG recording, based on time-domain approaches, frequency-domain approaches or a combination of the two will be reviewed, as well as methods for the quality assessment of ABP. Additionally, algorithms exploiting the relationship between ECG and pulsatile signals, such as ABP and photoplethysmographic recordings, for the reduction in the false alarm rate will be presented. Finally, some considerations will be drawn taking into account the large heterogeneity of clinical settings, applications and goals that the reviewed algorithms have to deal with.

Electrocardiogram ST-Segment Morphology Delineation Method Using Orthogonal Transformations

Differentiation between ischaemic and non-ischaemic transient ST segment events of long term ambulatory electrocardiograms is a persisting weakness in present ischaemia detection systems. Traditional ST segment level measuring is not a sufficiently precise technique due to the single point of measurement and severe noise which is often present. We developed a robust noise resistant orthogonal-transformation based delineation method, which allows tracing the shape of transient ST segment morphology changes from the entire ST segment in terms of diagnostic and morphologic feature-vector time series, and also allows further analysis. For these purposes, we developed a new Legendre Polynomials based Transformation (LPT) of ST segment. Its basis functions have similar shapes to typical transient changes of ST segment morphology categories during myocardial ischaemia (level, slope and scooping), thus providing direct insight into the types of time domain morphology changes through the LPT feature-vector space. We also generated new Karhunen and Lo ève Transformation (KLT) ST segment basis functions using a robust covariance matrix constructed from the ST segment pattern vectors derived from the Long Term ST Database (LTST DB). As for the delineation of significant transient ischaemic and non-ischaemic ST segment episodes, we present a study on the representation of transient ST segment morphology categories, and an evaluation study on the classification power of the KLT- and LPT-based feature vectors to classify between ischaemic and non-ischaemic ST segment episodes of the LTST DB. Classification accuracy using the KLT and LPT feature vectors was 90% and 82%, respectively, when using the k-Nearest Neighbors (k = 3) classifier and 10-fold cross-validation. New sets of feature-vector time series for both transformations were derived for the records of the LTST DB which is freely available on the PhysioNet website and were contributed to the LTST DB. The KLT and LPT present new possibilities for human-expert diagnostics, and for automated ischaemia detection.

Machine Learning and Decision Support in Critical Care

Clinical data management systems typically provide caregiver teams with useful information, derived from large, sometimes highly heterogeneous, data sources that are often changing dynamically. Over the last decade there has been a significant surge in interest in using these data sources, from simply re-using the standard clinical databases for event prediction or decision support, to including dynamic and patient-specific information into clinical monitoring and prediction problems. However, in most cases, commercial clinical databases have been designed to document clinical activity for reporting, liability and billing reasons, rather than for developing new algorithms. With increasing excitement surrounding "secondary use of medical records" and "Big Data" analytics, it is important to understand the limitations of current databases and what needs to change in order to enter an era of "precision medicine." This review article covers many of the issues involved in the collection and preprocessing of critical care data. The three challenges in critical care are considered: compartmentalization, corruption, and complexity. A range of applications addressing these issues are covered, including the modernization of static acuity scoring; on-line patient tracking; personalized prediction and risk assessment; artifact detection; state estimation; and incorporation of multimodal data sources such as genomic and free text data.

Integrating Real Time Data to Improve Outcomes in Acute Kidney Injury

Critically ill patients with acute kidney injury requiring renal replacement therapy have a poor prognosis. Despite well-known factors, which contribute to outcomes, including dose delivery, patients frequently miss the target dose and volume removal. One major barrier to effective care of these patients is the traditional dissociation of dialysis device data from other clinical information systems, notably the electronic health record (EHR). This lack of integration and the resulting manual documentation leads to errors and biases in documentation and missed opportunities to intervene in a timely fashion. This review summarizes the technological advancements facilitating direct connection of dialysis devices to EHRs. This connection facilitates automated data capture of many variables - including delivered dose, ultrafiltration rate and pressure measurements - which in turn can be leveraged for data mining, quality improvement and real-time targeted therapy adjustments. These interventions hold the promise to significantly improve outcomes for this patient population.

Signal quality measures on pulse oximetry and blood pressure signals acquired from self-measurement in a home environment

Recently, decision support system (DSSs) have become more widely accepted as a support tool for use with telehealth systems, helping clinicians to summarize and digest what would otherwise be an unmanageable volume of data. One of the pillars of a home telehealth system is the performance of unsupervised physiological self-measurement by patients in their own homes. Such measurements are prone to error and noise artifact, often due to poor measurement technique and ignorance of the measurement and transduction principles at work. These errors can degrade the quality of the recorded signals and ultimately degrade the performance of the DSS system, which is aiding the clinician in their management of the patient. Developed algorithms for automated quality assessment for pulse oximetry and blood pressure (BP) signals were tested retrospectively with data acquired from a trial that recorded signals in a home environment. The trial involved four aged subjects who performed pulse oximetry and BP measurements by themselves at their home for ten days, three times per day. This trial was set up to mimic the unsupervised physiological self-measurement as in a telehealth system. A manually annotated "gold standard" (GS) was used as the reference against which the developed algorithms were evaluated after analyzing the recordings. The assessment of pulse oximetry signals shows 95% of good sections and 67% of noisy sections were correctly detected by the developed algorithm, and a Cohen's Kappa coefficient (κ) of 0.58 was obtained in 120 pooled signals. The BP measurement evaluation demonstrates that 75% of the actual noisy sections were correctly classified in 120 pooled signals, with 97% and 91% of the signals correctly identified as worthy of attempting systolic and/or diastolic pressure estimation, respectively, with a mean error and standard deviation of 2.53±4.20 mmHg and 1.46±5.29 mmHg when compared to a manually annotated GS. These results demonstrate the feasibility, and highlight the potential benefit, of incorporating automated signal quality assessment algorithms for pulse oximetry and BP recording within a DSS for telehealth patient management.

The Design of Safe Networked Supervisory Medical Systems Using Organ-Centric Hierarchical Control Architecture

There are growing demands to leverage network connectivity and interoperability of medical devices in order to improve patient safety and the effectiveness of medical services. However, if not properly designed, the integration of medical devices through networking could significantly increase the complexity of the system and make the system more vulnerable to potential errors, jeopardizing patient safety. The system must be designed and verified to guarantee the safety of patients and the effectiveness of medical services in the face of potential problems such as network failures. In this paper, we propose organ-centric hierarchical control architecture as a viable solution that reduces the complexity in system design and verification. In our approach, medical devices are grouped into clusters according to organ-specific human physiology. Each cluster captures common patterns arising out of medical device interactions and becomes a survivable semiautonomous unit during network failures. Further, safety verification and runtime enforcement can be modularized along organ-centric hierarchical control structure. We show the feasibility of the proposed approach under Simulink's model-based development framework. A simplified scenario for airway laser surgery is used as a case study.

Gaussian processes for personalized e-health monitoring with wearable sensors

Advances in wearable sensing and communications infrastructure have allowed the widespread development of prototype medical devices for patient monitoring. However, such devices have not penetrated into clinical practice, primarily due to a lack of research into "intelligent" analysis methods that are sufficiently robust to support large-scale deployment. Existing systems are typically plagued by large false-alarm rates, and an inability to cope with sensor artifact in a principled manner. This paper has two aims: 1) proposal of a novel, patient-personalized system for analysis and inference in the presence of data uncertainty, typically caused by sensor artifact and data incompleteness; 2) demonstration of the method using a large-scale clinical study in which 200 patients have been monitored using the proposed system. This latter provides much-needed evidence that personalized e-health monitoring is feasible within an actual clinical environment, at scale, and that the method is capable of improving patient outcomes via personalized healthcare.

Developing predictive models using electronic medical records: challenges and pitfalls

While Electronic Medical Records (EMR) contain detailed records of the patient-clinician encounter - vital signs, laboratory tests, symptoms, caregivers' notes, interventions prescribed and outcomes - developing predictive models from this data is not straightforward. These data contain systematic biases that violate assumptions made by off-the-shelf machine learning algorithms, commonly used in the literature to train predictive models. In this paper, we discuss key issues and subtle pitfalls specific to building predictive models from EMR. We highlight the importance of carefully considering both the special characteristics of EMR as well as the intended clinical use of the predictive model and show that failure to do so could lead to developing models that are less useful in practice. Finally, we describe approaches for training and evaluating models on EMR using early prediction of septic shock as our example application.

Disease-based modeling to predict fluid response in intensive care units

OBJECTIVE

To compare general and disease-based modeling for fluid resuscitation and vasopressor use in intensive care units.

METHODS

Retrospective cohort study involving 2944 adult medical and surgical intensive care unit (ICU) patients receiving fluid resuscitation. Within this cohort there were two disease-based groups, 802 patients with a diagnosis of pneumonia, and 143 patients with a diagnosis of pancreatitis. Fluid resuscitation either progressing to subsequent vasopressor administration or not was used as the primary outcome variable to compare general and disease-based modeling.

RESULTS

Patients with pancreatitis, pneumonia and the general group all shared three common predictive features as core variables, arterial base excess, lactic acid and platelets. Patients with pneumonia also had non-invasive systolic blood pressure and white blood cells added to the core model, and pancreatitis patients additionally had temperature. Disease-based models had significantly higher values of AUC (p < 0.05) than the general group (0.82 ± 0.02 for pneumonia and 0.83 ± 0.03 for pancreatitis vs. 0.79 ± 0.02 for general patients).

CONCLUSIONS

Disease-based predictive modeling reveals a different set of predictive variables compared to general modeling and improved performance. Our findings add support to the growing body of evidence advantaging disease specific predictive modeling.

Gaussian process regression in vital-sign early warning systems

The current standard of clinical practice for patient monitoring in most developed nations is connection of patients to vital-sign monitors, combined with frequent manual observation. In some nations, such as the UK, manual early warning score (EWS) systems have been mandated for use, in which scores are assigned to the manual observations, and care escalated if the scores exceed some pre-defined threshold. We argue that this manual system is far from ideal, and can be improved using machine learning techniques. We propose a system based on Gaussian process regression for improving the efficacy of existing EWS systems, and then demonstrate the method using manual observation of vital signs from a large-scale clinical study.

Design of a Decision Support System for a Home Telehealth Application

A decision support system (DSS) that has been designed to manage patients using a home telehealth system is presented. The DSS has been developed to assist home telehealth clinical support staff with their workload, and to provide more effective communication between multiple home telehealth users. The three-tier system architecture that consists of a data layer; a business logic layer; and a front-end layer employs business processes and uses a rule engine for its logic and knowledge base. This paper discusses the design considerations involved in the construction of a DSS for the purpose of home telehealth, and illustrates how it may be developed using entirely open source software.

Implementation of artifact detection in critical care: a methodological review

Artifact detection (AD) techniques minimize the impact of artifacts on physiologic data acquired in critical care units (CCU) by assessing quality of data prior to clinical event detection (CED) and parameter derivation (PD). This methodological review introduces unique taxonomies to synthesize over 80 AD algorithms based on these six themes: 1) CCU; 2) physiologic data source; 3) harvested data; 4) data analysis; 5) clinical evaluation; and 6) clinical implementation. Review results show that most published algorithms: a) are designed for one specific type of CCU; b) are validated on data harvested only from one OEM monitor; c) generate signal quality indicators (SQI) that are not yet formalized for useful integration in clinical workflows; d) operate either in standalone mode or coupled with CED or PD applications; e) are rarely evaluated in real-time; and f) are not implemented in clinical practice. In conclusion, it is recommended that AD algorithms conform to generic input and output interfaces with commonly defined data: 1) type; 2) frequency; 3) length; and 4) SQIs. This shall promote: a) reusability of algorithms across different CCU domains; b) evaluation on different OEM monitor data; c) fair comparison through formalized SQIs; d) meaningful integration with other AD, CED and PD algorithms; and e) real-time implementation in clinical workflows.

Robust classification of neonatal apnoea-related desaturations

Respiratory signals monitored in the neonatal intensive care units are usually ignored due to the high prevalence of noise and false alarms (FA). Apneic events are generally therefore indicated by a pulse oximeter alarm reacting to the subsequent desaturation. However, the high FA rate in the photoplethysmogram may desensitize staff, reducing the reaction speed. The main reason for the high FA rates of critical care monitors is the unimodal analysis behaviour. In this work, we propose a multimodal analysis framework to reduce the FA rate in neonatal apnoea monitoring. Information about oxygen saturation, heart rate, respiratory rate and signal quality was extracted from electrocardiogram, impedance pneumogram and photoplethysmographic signals for a total of 20 features in the 5 min interval before a desaturation event. 1616 desaturation events from 27 neonatal admissions were annotated by two independent reviewers as true (physiologically relevant) or false (noise-related). Patients were divided into two independent groups for training and validation, and a support vector machine was trained to classify the events as true or false. The best classification performance was achieved on a combination of 13 features with sensitivity, specificity and accuracy of 100% in the training set, and a sensitivity of 86%, a specificity of 91% and an accuracy of 90% in the validation set.

Monitoring in cardiovascular disease patients by nonlinear biomedical signal processing

Due to recent advances in technology extensive cardiovascular monitoring is widely introduced today. An essential component of cardiovascular monitoring is the analysis of several biosignals as electrocardiogram, blood pressure and other vital signs. This manuscript provides an overview about several application fields of cardiovascular monitoring with the main focus on nonlinear dynamics analysis.

Relationship between mortality and first-day events index from routinely gathered physiological variables in ICU patients

OBJECTIVE

To test the hypothesis that the degree and duration of alterations in physiological variables routinely gathered by intensive care unit (ICU) monitoring systems during the first day of admission to the ICU, together with a few additional routinely recorded data, yield information similar to that obtained by traditional mortality prediction systems.

DESIGN

A prospective observational multicenter study (EURICUS II) was carried out.

SETTING

Fifty-five European ICUs.

PATIENTS

A total of 17,598 consecutive patients admitted to the ICU over a 10-month period.

INTERVENTIONS

None.

MAIN VARIABLES OF INTEREST

Hourly data were manually gathered on alterations or "events" in systolic blood pressure, heart rate and oxygen saturation throughout ICU stay to construct an events index and mortality prediction models.

RESULTS

The mean first-day events index was 6.37±10.47 points, and was significantly associated to mortality (p<0.001), with a discrimination capacity for hospital mortality of 0.666 (area under the ROC curve). A second index included this first-day events index, age, pre-admission location, and the Glasgow coma score. A model constructed with this second index plus diagnosis upon admission was validated by using the Jackknife method (Hosmer-Lemeshow, H=13.8554, insignificant); the area under ROC curve was 0.818.

CONCLUSIONS

A prognostic index with performance very similar to that of habitual systems can be constructed from routine ICU data with only a few patient characteristics. These results may serve as a guide for the possible automated construction of ICU prognostic indexes.

Predictive monitoring for early detection of subacute potentially catastrophic illnesses in critical care

We wish to save lives of patients admitted to ICUs. Their mortality is high enough based simply on the severity of the original injury or illness, but is further raised by events during their stay. We target those events that are subacute but potentially catastrophic, such as infection. Sepsis, for example, is a bacterial infection of the bloodstream, that is common in ICU patients and has a >25% risk of death. Logically, early detection and treatment with antibiotics should improve outcomes. Our fundamental precepts are (1) some potentially catastrophic medical and surgical illnesses have subclinical phases during which early diagnosis and treatment might have life-saving effects, (2) these phases are characterized by changes in the normal highly complex but highly adaptive regulation and interaction of the nervous system and other organs such as the heart and lungs, (3) teams of clinicians and quantitative scientists can work together to identify clinically important abnormalities of monitoring data, to develop algorithms that match the clinicians' eye in detecting abnormalities, and to undertake the clinical trials to test their impact on outcomes.

An embedded reconfigurable architecture for patient-specific multi-paramater medical monitoring

A robust medical monitoring device should be able to provide intelligent diagnosis based on accurate analysis of physiological parameters in real-time. At the same time, such device must be able to adapt to the characteristics of a specific patient and desired diagnostic needs, and continue to operate even in presence of unexpected artifacts and accidental errors. A reconfigurable architecture is proposed for real-time assessment of individual's health status based on development of a patient-specific health index and online analysis and fusion of multi-parameter physiological signals. This is achieved by static configuration of processing elements and communication blocks in the architecture based on the patient's diagnostic needs. The proposed architecture is prototyped as a single integrated device on an FPGA platform and is evaluated using multi-parameter data from intensive care units (ICUs). Three representative test cases of concurrently analyzing Blood Pressure, Heart Rate, and Electrocardiogram (ECG) data from MIMIC database are presented. The results show the effectiveness of the proposed technique in eliminating false alarms caused by patient movements, monitor noise, or imperfections in the detection schemes.

University of Queensland vital signs dataset: development of an accessible repository of anesthesia patient monitoring data for research

BACKGROUND

Data recorded from the devices used to monitor a patient's vital signs are often used in the development of displays, alarms, and information systems, but high-resolution, multiple-parameter datasets of anesthesia monitoring data from patients during anesthesia are often difficult to obtain. Existing databases have typically been collected from patients in intensive care units. However, the physical state of intensive care patients is dissimilar to those undergoing surgery, more frequent and marked changes to cardiovascular and respiratory variables are seen in operating room patients, and additional and highly relevant information to anesthesia (e.g., end-tidal agent monitoring, etc.) is omitted from these intensive care databases. We collected a set of high-quality, high-resolution, multiple-parameter monitoring data suitable for anesthesia monitoring research.

METHODS

Vital signs data were recorded from patients undergoing anesthesia at the Royal Adelaide Hospital. Software was developed to capture, time synchronize, and interpolate vital signs data from Philips IntelliVue MP70 and MP30 patient monitors and Datex-Ohmeda Aestiva/5 anesthesia machines into 10 millisecond resolution samples. The recorded data were saved in a variety of accessible file formats.

RESULTS

Monitoring data were recorded from 32 cases (25 general anesthetics, 3 spinal anesthetics, 4 sedations) ranging in duration from 13 minutes to 5 hours (median 105 min). Most cases included data from the electrocardiograph, pulse oximeter, capnograph, noninvasive arterial blood pressure monitor, airway flow, and pressure monitor and, in a few cases, the Y-piece spirometer, electroencephalogram monitor, and arterial blood pressure monitor. Recorded data were processed and saved into 4 file formats: (1) comma-separated values text files with full numerical and waveform data, (2) numerical parameters recorded in comma-separated values files at 1-second intervals, (3) graphical plots of all waveform data in a range of resolutions as Portable Network Graphics image files, and (4) graphical overview plots of numerical data for entire cases as Portable Network Graphics and Scalable Vector Graphics files. The complete dataset is freely available online via doi:102.100.100/6914 and has been listed in the Australian National Data Service Collections Registry.

DISCUSSION

The present dataset provides clinical anesthesia monitoring data from entire surgical cases where patients underwent anesthesia, includes a wide range of vital signs variables that are commonly monitored during surgery, and is published in accessible, user-friendly file formats. The text and image file formats let researchers without engineering or computer science backgrounds easily access the data using standard spreadsheet and image browsing software. In future work, monitoring data should be collected from a wider range and larger number of cases, and software tools are needed to support searching and navigating the database.

Breath-by-breath analysis of cardiorespiratory interaction for quantifying developmental maturity in premature infants

In healthy neonates, connections between the heart and lungs through brain stem chemosensory pathways and the autonomic nervous system result in cardiorespiratory synchronization. This interdependence between cardiac and respiratory dynamics can be difficult to measure because of intermittent signal quality in intensive care settings and variability of heart and breathing rates. We employed a phase-based measure suggested by Schäfer and coworkers (Schäfer C, Rosenblum MG, Kurths J, Abel HH. Nature 392: 239-240, 1998) to obtain a breath-by-breath analysis of cardiorespiratory interaction. This measure of cardiorespiratory interaction does not distinguish between cardiac control of respiration associated with cardioventilatory coupling and respiratory influences on the heart rate associated with respiratory sinus arrhythmia. We calculated, in sliding 4-min windows, the probability density of heartbeats as a function of the concurrent phase of the respiratory cycle. Probability density functions whose Shannon entropy had a <0.1% chance of occurring from random numbers were classified as exhibiting interaction. In this way, we analyzed 18 infant-years of data from 1,202 patients in the Neonatal Intensive Care Unit at University of Virginia. We found evidence of interaction in 3.3 patient-years of data (18%). Cardiorespiratory interaction increased several-fold with postnatal development, but, surprisingly, the rate of increase was not affected by gestational age at birth. We find evidence for moderate correspondence between this measure of cardiorespiratory interaction and cardioventilatory coupling and no evidence for respiratory sinus arrhythmia, leading to the need for further investigation of the underlying mechanism. Such continuous measures of physiological interaction may serve to gauge developmental maturity in neonatal intensive care patients and prove useful in decisions about incipient illness and about hospital discharge.

A Hypotensive Episode Predictor for Intensive Care based on Heart Rate and Blood Pressure Time Series

In the intensive care unit (ICU), prompt therapeutic intervention to hypotensive episodes (HEs) is a critical task. Advance alerts that can prospectively identify patients at risk of developing an HE in the next few hours would be of considerable clinical value. In this study, we developed an automated, artificial neural network HE predictor based on heart rate and blood pressure time series from the MIMIC II database. The gap between prediction time and the onset of the 30-minute target window was varied from 1 to 4 hours. A 30-minute observation window preceding the prediction time provided input information to the predictor. While individual gap sizes were evaluated independently, weighted posterior probabilities based on different gap sizes were also investigated. The results showed that prediction performance degraded as gap size increased and the weighting scheme induced negligible performance improvement. Despite low positive predictive values, the best mean area under ROC curve was 0.934.

An investigation of patterns in hemodynamic data indicative of impending hypotension in intensive care

BACKGROUND

In the intensive care unit (ICU), clinical staff must stay vigilant to promptly detect and treat hypotensive episodes (HEs). Given the stressful context of busy ICUs, an automated hypotensive risk stratifier can help ICU clinicians focus care and resources by prospectively identifying patients at increased risk of impending HEs. The objective of this study was to investigate the possible existence of discriminatory patterns in hemodynamic data that can be indicative of future hypotensive risk.

METHODS

Given the complexity and heterogeneity of ICU data, a machine learning approach was used in this study. Time series of minute-by-minute measures of mean arterial blood pressure, heart rate, pulse pressure, and relative cardiac output from 1,311 records from the MIMIC II Database were used. An HE was defined as a 30-minute period during which the mean arterial pressure was below 60 mmHg for at least 90% of the time. Features extracted from the hemodynamic data during an observation period of either 30 or 60 minutes were analyzed to predict the occurrence of HEs 1 or 2 hours into the future. Artificial neural networks (ANNs) were trained for binary classification (normotensive vs. hypotensive) and regression (estimation of future mean blood pressure).

RESULTS

The ANNs were successfully trained to discriminate patterns in the multidimensional hemodynamic data that were predictive of future HEs. The best overall binary classification performance resulted in a mean area under ROC curve of 0.918, a sensitivity of 0.826, and a specificity of 0.859. Predicting further into the future resulted in poorer performance, whereas observation duration minimally affected performance. The low prevalence of HEs led to poor positive predictive values. In regression, the best mean absolute error was 9.67%.

CONCLUSIONS

The promising pattern recognition performance demonstrates the existence of discriminatory patterns in hemodynamic data that can indicate impending hypotension. The poor PPVs discourage a direct HE predictor, but a hypotensive risk stratifier based on the pattern recognition algorithms of this study would be of significant clinical value in busy ICU environments.

Data Fusion for Improved Respiration Rate Estimation

We present an application of a modified Kalman-Filter (KF) framework for data fusion to the estimation of respiratory rate from multiple physiological sources which is robust to background noise. A novel index of the underlying signal quality of respiratory signals is presented and then used to modify the noise covariance matrix of the KF which discounts the effect of noisy data. The signal quality index, together with the KF innovation sequence, is also used to weight multiple independent estimates of the respiratory rate from independent KFs. The approach is evaluated on both a realistic artificial ECG model (with real additive noise), and on real data taken from 30 subjects with overnight polysomnograms, containing ECG, respiration and peripheral tonometry waveforms from which respiration rates were estimated. Results indicate that our automated voting system can out-perform any individual respiration rate estimation technique at all levels of noise and respiration rates exhibited in our data. We also demonstrate that even the addition of a noisier extra signal leads to an improved estimate using our framework. Moreover, our simulations demonstrate that different ECG respiration extraction techniques have different error profiles with respect to the respiration rate, and therefore a respiration rate-related modification of any fusion algorithm may be appropriate.

Artificial arterial blood pressure artifact models and an evaluation of a robust blood pressure and heart rate estimator

Background

Within the intensive care unit (ICU), arterial blood pressure (ABP) is typically recorded at different (and sometimes uneven) sampling frequencies, and from different sensors, and is often corrupted by different artifacts and noise which are often non-Gaussian, nonlinear and nonstationary. Extracting robust parameters from such signals, and providing confidences in the estimates is therefore difficult and requires an adaptive filtering approach which accounts for artifact types.

Methods

Using a large ICU database, and over 6000 hours of simultaneously acquired electrocardiogram (ECG) and ABP waveforms sampled at 125 Hz from a 437 patient subset, we documented six general types of ABP artifact. We describe a new ABP signal quality index (SQI), based upon the combination of two previously reported signal quality measures weighted together. One index measures morphological normality, and the other degradation due to noise. After extracting a 6084-hour subset of clean data using our SQI, we evaluated a new robust tracking algorithm for estimating blood pressure and heart rate (HR) based upon a Kalman Filter (KF) with an update sequence modified by the KF innovation sequence and the value of the SQI. In order to do this, we have created six novel models of different categories of artifacts that we have identified in our ABP waveform data. These artifact models were then injected into clean ABP waveforms in a controlled manner. Clinical blood pressure (systolic, mean and diastolic) estimates were then made from the ABP waveforms for both clean and corrupted data. The mean absolute error for systolic, mean and diastolic blood pressure was then calculated for different levels of artifact pollution to provide estimates of expected errors given a single value of the SQI.

Results

Our artifact models demonstrate that artifact types have differing effects on systolic, diastolic and mean ABP estimates. We show that, for most artifact types, diastolic ABP estimates are less noise-sensitive than mean ABP estimates, which in turn are more robust than systolic ABP estimates. We also show that our SQI can provide error bounds for both HR and ABP estimates.

Conclusion

The KF/SQI-fusion method described in this article was shown to provide an accurate estimate of blood pressure and HR derived from the ABP waveform even in the presence of high levels of persistent noise and artifact, and during extreme bradycardia and tachycardia. Differences in error between artifact types, measurement sensors and the quality of the source signal can be factored into physiological estimation using an unbiased adaptive filter, signal innovation and signal quality measures.