Search

Heterogeneous patient populations, complex pharmacology and low recruitment rates in the Intensive Care Unit (ICU) have led to the failure of many clinical trials. Recently, machine learning (ML) emerged as a new technology to process and identify big data relationships, enabling a new era in clinical trial design. In this study, we designed a ML model for predictively stratifying acute respiratory distress syndrome (ARDS) patients, ultimately reducing the required number of patients by increasing statistical power through cohort homogeneity. From the Philips eICU Research Institute (eRI) database, no less than 51,555 ARDS patients were extracted. We defined three subpopulations by outcome: (1) rapid death, (2) spontaneous recovery, and (3) long-stay patients. A retrospective univariate analysis identified highly predictive variables for each outcome. All 220 variables were used to determine the most accurate and generalizable model to predict long-stay patients. Multiclass gradient boosting was identified as the best-performing ML model. Whereas alterations in pH, bicarbonate or lactate proved to be strong predictors for rapid death in the univariate analysis, only the multivariate ML model was able to reliably differentiate the disease course of the long-stay outcome population (AUC of 0.77). We demonstrate the feasibility of prospective patient stratification using ML algorithms in the by far largest ARDS cohort reported to date. Our algorithm can identify patients with sufficiently long ARDS episodes to allow time for patients to respond to therapy, increasing statistical power. Further, early enrollment alerts may increase recruitment rate.

Infectious threats, like the COVID-19 pandemic, hinder maintenance of a productive and healthy workforce. If subtle physiological changes precede overt illness, then proactive isolation and testing can reduce labor force impacts. This study hypothesized that an early infection warning service based on wearable physiological monitoring and real-time machine learning could be developed and deployed. We developed a prototype tool, first deployed June 23, 2020, that delivered continuously updated scores of infection risk for SARS-CoV-2. Data were acquired from 9381 United States Department of Defense (US DoD) personnel wearing Garmin and Oura devices, totaling 599,174 user-days of service and 201 million hours of data. There were 491 COVID-19 positive cases. A predictive algorithm identified infection before diagnostic testing with an AUC of 0.82. Barriers to implementation included adequate data capture (at least 48% data was needed) and delays in data transmission. We observe increased risk scores as early as 6 days prior to diagnostic testing (2.3 days average). This study showed feasibility of a real-time risk prediction score to minimize workforce impacts of infection.

Background: Infection prevention surveillance for cross transmission is often performed by manual review of microbiologic culture results to identify geotemporally related clusters. However, the sensitivity and specificity of this approach remains uncertain. Whole-genome sequencing (WGS) analysis can help provide a gold-standard for identifying cross-transmission events. Objective: We employed a published WGS program, the Philips IntelliSpace Epidemiology platform, to compare accuracy of two surveillance methods: (i.) a virtual infection practitioner (VIP) with perfect recall and automated analysis of antibiotic susceptibility testing (AST), sample collection timing, and patient location data and (ii) a novel clinical matching (CM) algorithm that provides cluster suggestions based on a nuanced weighted analysis of AST data, timing of sample collection, and shared location stays between patients. Methods: WGS was performed routinely on inpatient and emergency department isolates of Enterobacter cloacae, Enterococcus faecium, Klebsiella pneumoniae, and Pseudomonas aeruginosa at an academic medical center. Single-nucleotide variants (SNVs) were compared within core genome regions on a per-species basis to determine cross-transmission clusters. Moreover, one unique strain per patient was included within each analysis, and duplicates were excluded from the final results. Results: Between May 2018 and April 2019, clinical data from 121 patients were paired with WGS data from 28 E. cloacae, 21 E. faecium, 61 K. pneumoniae, and 46 P. aeruginosa isolates. Previously published SNV relatedness thresholds were applied to define genomically related isolates. Mapping of genomic relatedness defined clusters as follows: 4 patients in 2 E. faecium clusters and 2 patients in 1 P. aeruginosa cluster. The VIP method identified 12 potential clusters involving 28 patients, all of which were “pseudoclusters.” Importantly, the CM method identified 7 clusters consisting of 27 patients, which included 1 true E. faecium cluster of 2 patients with genomically related isolates. Conclusions: In light of the WGS data, all of the potential clusters identified by the VIP were pseudoclusters, lacking sufficient genomic relatedness. In contrast, the CM method showed increased sensitivity and specificity: it decreased the percentage of pseudoclusters by 14% and it identified a related genomic cluster of E. faecium. These findings suggest that integrating clinical data analytics and WGS is likely to benefit institutions in limiting expenditure of resources on pseudoclusters. Therefore, WGS combined with more sophisticated surveillance approaches, over standard methods as modeled by the VIP, are needed to better identify and address true cross-transmission events.Funding: This study was supported by Philips Healthcare.Disclosures: None

Background: Infection prevention and control (IPC) workflows are often retrospective and manual. New tools, however, have entered the field to facilitate rapid prospective monitoring of infections in hospitals. Although artificial intelligence (AI)–enabled platforms facilitate timely, on-demand integration of clinical data feeds with pathogen whole-genome sequencing (WGS), a standardized workflow to fully harness the power of such tools is lacking. We report a novel, evidence-based workflow that promotes quicker infection surveillance via AI-assisted clinical and WGS data analysis. The algorithm suggests clusters based on a combination of similar minimum inhibitory concentration (MIC) data, timing of sample collection, and shared location stays between patients. It helps to proactively guide IPC professionals during investigation of infectious outbreaks and surveillance of multidrug-resistant organisms and healthcare-acquired infections. Methods: Our team established a 1-year workgroup comprised of IPC practitioners, clinical experts, and scientists in the field. We held weekly roundtables to study lessons learned in an ongoing surveillance effort at a tertiary care hospital—utilizing Philips IntelliSpace Epidemiology (ISEpi), an AI-powered system—to understand how such a tool can enhance practice. Based on real-time case discussions and evidence from the literature, a workflow guidance tool and checklist were codified. Results: In our workflow, data-informed clusters posed by ISEpi underwent triage and expert follow-up analysis to assess: (1) likelihood of transmission(s); (2) potential vector(s) identity; (3) need to request WGS; and (4) intervention(s) to be pursued, if warranted. In a representative sample (spanning October 17, 2019, to November 7, 2019) of 67 total isolates suggested for inclusion in 19 unique cluster investigations, we determined that 9 investigations merited follow-up. Collectively, these 9 investigations involved 21 patients and required 115 minutes to review in ISEpi and an additional 70 minutes of review outside of ISEpi. After review, 6 investigations were deemed unlikely to represent a transmission; the other 3 had potential to represent transmission for which interventions would be performed. Conclusions: This study offers an important framework for adaptation of existing infection control workflow strategies to leverage the utility of rapidly integrated clinical and WGS data. This workflow can also facilitate time-sensitive decisions regarding sequencing of specific pathogens given the preponderance of available clinical data supporting investigations. In this regard, our work sets a new standard of practice: precision infection prevention (PIP). Ongoing effort is aimed at development of AI-powered capabilities for enterprise-level quality and safety improvement initiatives.Funding: Philips Healthcare provided support for this study.Disclosures: Alan Doty and Juan Jose Carmona report salary from Philips Healthcare.

Whole-genome sequencing is increasingly adopted in clinical settings to identify pathogen transmissions, though largely as a retrospective tool. Prospective monitoring, in which samples are continuously added and compared to previous samples, can generate more actionable information. To enable prospective pathogen comparison, genomic relatedness metrics based on single-nucleotide differences must be consistent across time, efficient to compute and reliable for a large variety of samples. The choice of genomic regions to compare, i.e., the core genome, is critical to obtain a good metric. We propose a novel core genome method that selects conserved sequences in the reference genome by comparing its k-mer content to that of publicly available genome assemblies. The conserved-sequence genome is sample set-independent, which enables prospective pathogen monitoring. Based on clinical data sets of 3436 S. aureus, 1362 K. pneumoniae and 348 E. faecium samples, ROC curves demonstrate that the conserved-sequence genome disambiguates same-patient samples better than a core genome consisting of conserved genes. The conserved-sequence genome confirms outbreak samples with high sensitivity: in a set of 2335 S. aureus samples, it correctly identifies 44 out of 44 known outbreak samples, whereas the conserved-gene method confirms 38 known outbreak samples.

Background:Determining infectious cross-transmission events in healthcare settings involves manual surveillance of case clusters by infection control personnel, followed by strain typing of clinical/environmental isolates suspected in said clusters. Recent advances in genomic sequencing and cloud computing now allow for the rapid molecular typing of infecting isolates.Objective:To facilitate rapid recognition of transmission clusters, we aimed to assess infection control surveillance using whole-genome sequencing (WGS) of microbial pathogens to identify cross-transmission events for epidemiologic review.Methods:Clinical isolates ofStaphylococcus aureus,Enterococcus faecium,Pseudomonas aeruginosa, andKlebsiella pneumoniaewere obtained prospectively at an academic medical center, from September 1, 2016, to September 30, 2017. Isolate genomes were sequenced, followed by single-nucleotide variant analysis; a cloud-computing platform was used for whole-genome sequence analysis and cluster identification.Results:Most strains of the 4 studied pathogens were unrelated, and 34 potential transmission clusters were present. The characteristics of the potential clusters were complex and likely not identifiable by traditional surveillance alone. Notably, only 1 cluster had been suspected by routine manual surveillance.Conclusions:Our work supports the assertion that integration of genomic and clinical epidemiologic data can augment infection control surveillance for both the identification of cross-transmission events and the inclusion of missed and exclusion of misidentified outbreaks (ie, false alarms). The integration of clinical data is essential to prioritize suspect clusters for investigation, and for existing infections, a timely review of both the clinical and WGS results can hold promise to reduce HAIs. A richer understanding of cross-transmission events within healthcare settings will require the expansion of current surveillance approaches.

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.

Whole-genome sequencing is increasingly adopted in clinical settings to identify pathogen transmissions. Currently, such studies are performed largely retrospectively, but to be actionable they need to be carried out prospectively, in which samples are continuously added and compared to previous samples. To enable prospective pathogen comparison, genomic relatedness metrics based on single nucleotide differences must be consistent across time, efficient to compute and reliable for a large variety of samples. The choice of genomic regions to compare, i.e., the core genome, is critical to obtain a good metric.We propose a novel core genome method that selects conserved sequences in the reference genome by comparing its k-mer content to that of publicly available genome assemblies. The conserved-sequence genome is sample set-independent, which enables prospective pathogen monitoring. Based on clinical data sets of 3436 S. aureus, 1362 K. pneumoniae and 348 E. faecium samples, we show that the conserved-sequence genome disambiguates same-patient samples better than a core genome consisting of conserved genes. The conserved-sequence genome confirms outbreak samples with high accuracy: in a set of 2335 S. aureus samples, it correctly identifies 44 out of 45 outbreak samples, whereas the conserved gene method confirms 38 out of 45 outbreak samples.

ObjectivesClinicians in the intensive care unit (ICU) are presented with a large number of physiological data consisting of periodic and frequently sampled measurements, such as heart rate and blood pressure, as well as aperiodic measurements, such as noninvasive blood pressure and laboratory studies. Because this data can be overwhelming, there is considerable interest in designing algorithms that help integrate and interpret this data and assist ICU clinicians in detecting or predicting in advance patients who may be deteriorating. In order to decide whether to deploy such algorithms in a clinical trial, it is important to evaluate these algorithms using retrospective data. However, the fact that these algorithms will be running continuously, i.e., repeatedly sampling incoming patient data, presents some novel challenges for algorithm evaluation. Commonly used measures of performance such as sensitivity and positive predictive value (PPV) are easily applied to static “snapshots” of patient data, but can be very misleading when applied to indicators or alerting algorithms that are running on continuous data. Our objective is to create a method for evaluating algorithm performance on retrospective data with the algorithm running continuously throughout the patient’s stay as it would in a real ICU.MethodsWe introduce our evaluation methodology in the context of evaluating an algorithm, a Hemodynamic Instability Indicator (HII), for assisting bedside ICU clinicians with the early detection of hemodynamic instability before the onset of acute hypotension. Each patient’s ICU stay is divided into segments that are labelled as hemodynamically stable or unstable based on clinician interventions typically aimed at treating hemodynamic instability. These segments can be of varying length with varying degrees of exposure to potential alerts, whether true positive or false positive. Furthermore, to simulate how clinicians might interact with the alerting algorithm, we use a dynamic alert supervision mechanism which suppresses subsequent alerts unless the indicator has significantly deteriorated since the prior alert. Under these conditions determining what counts as a positive or negative instance, and calculations of sensitivity, specificity, and positive predictive value can be problematic. We introduce a methodology for consistently counting positive and negative instances. The methodology distinguishes between counts based on alerting events and counts based on sub-segments, and show how they can be applied in calculating measures of performance such as sensitivity, specificity, positive predictive value.ResultsThe introduced methodology is applied to retrospective evaluation of two algorithms, HII and an alerting algorithm based on systolic blood pressure. We use a database, consisting of data from 41,707 patients from 25 US hospitals, to evaluate the algorithms. Both algorithms are evaluated running continuously throughout each patient’s stay as they would in a real ICU setting. We show how the introduced performance measures differ for different algorithms and for different assumptions.DiscussionThe standard measures of diagnostic tests in terms of true positives, false positives, etc. are based on certain assumptions which may not apply when used in the context of measuring the performance on an algorithm running continuously, and thus repeatedly sampling from the same patient. When such measures are being reported it is important that the underlying assumptions be made explicit; otherwise, the results can be very misleading.ConclusionWe introduce a methodology for evaluating how an alerting algorithm or indicator will perform running continuously throughout every patient’s ICU stay, not just for a subset of patients for selected episodes.

Background Typical Infection Prevention to detect pathogen transmission in hospitals has relied on observation of (1) uncommon pathogen phenotypes or (2) greater than expected number of pathogen phenotypes in a given timeframe and/or location. Genome sequencing of targeted organisms in conjunction with routine patient geo-temporal information and antibiotic susceptibility data holds promise in identifying transmissions with greater sensitivity and specificity, saving time and effort in reviewing for transmission events. Methods In an on-going genomic sequencing surveillance effort in a tertiary care hospital, drug-resistant clinical isolates from the “ESKAPE” pathogens were routinely sequenced in 2017. In parallel, potential clusters were identified for 2017 through conventional Infection Prevention approaches. Groups identified by their genetic distances along with visualizations on antimicrobial susceptibilities, and patient location histories and dates were displayed in an interactive interface, Philips IntelliSpace Epidemiology (PIE), and reviewed by Infection Prevention. Results Among 656 patients, 1,239 drug-resistant ESKAPE samples were sequenced. Thirty-eight genetically related groups involving 196 patients were identified. Groups ranged in size from two to 44 patients, primarily consisting of VRE and MRSA. Notably, a review of the 38 groups identified 20 groups where the information at hand suggested a concern for transmission. 16 of the 20 were not previously identified by Infection Prevention. Using PIE to review all 38 groups identified from 1 year’s worth of data required 3 hours of time by an Infection Prevention professional, averaging less than 5 minutes per cluster, less than 1 minute per patient, and 11 minutes of review time per actionable opportunity. By conventional means, approximately 23 hours would have been required to review the genomic groups without the aid of the PIE tool. Conclusion The use of PIE’s genomic-defined groups, along with the integrated clinical data platform, allows for a greater ability, certainty, and speed to detect clusters of organisms representing transmission in the hospital setting. Applied prospectively, PIE can detect transmissions sooner than by conventional means for potential patient safety gains and cost savings. Disclosures D. Chen, Philips: Scientific Advisor, Consulting fee. M. Fortunato-Habib, Philips Healthcare: Collaborator and Employee, Salary. A. Hoss, Philips: Employee, Salary. R. Kolde, Philips: Employee, Salary. A. Dhand, Merck: Speaker’s Bureau, Speaker honorarium. Astellas: Scientific Advisor, Consulting fee. R. Sussner, Philips: Scientific Advisor, Consulting fee. J. Carmona, Philips Healthcare: Employee, Salary. B. Gross, Philips Healthcare: Employee, Investigator, Research Contractor, Scientific Advisor and Shareholder, Salary. J. Fallon, Philips Healthcare: Investigator, Research support.

Background Each year, nearly 2 million patients contract and are affected by healthcare-associated infections (HAIs) in the United States alone, resulting in nearly 100K deaths. According to the Centers for Disease Control and Prevention (CDC), more patients die from HAIs in the United States per year than all breast and prostate cancer cases combined (National Vital Statistics Report, 2016). In addition to the mortality burden, the financial impact of HAIs within the hospital ecosystem is estimated to total between $28–45 billion. However, no economic model has demonstrated how early effective identification and mitigation of infection clusters can result in cost savings for hospitals until now. Methods As there is no publicly available data for infection cluster rates, we based our analysis on anonymized real-world retrospective data spanning 18 months (November 2016 to June 2018) from two US-based academic tertiary hospitals with a combined total of about 1,700 beds, then normalized to 800 beds. A cloud-computing platform (Philips IntelliSpace Epidemiology) was used for whole-genome sequence analysis and cluster identification. We determined that an average 800-bed facility would have an occurrence of 46 genetically related infectious clusters involving 2 or more patients (mean of 7.9, median of 3), affecting 180 patients in total. Results Given the average HAI treatment cost of $24,512 (average costs rescaled from literature to 2019 USD using PPI data), this represents a total cost of $4,412,160. If these clusters could have been limited to 2 patients, an additional 96 infections might have been prevented, representing a potentially avoidable economic burden of $2,353,152 for this 800-bed institution. Our data show that a 20% reduction in transmissions would drive a 3% overall reduction in HAIs, but results in savings of over $450,000. Conclusion Active, genomic-based surveillance can inform timely and precise preventative steps to help lower the size of infectious clusters. This health economic modeling shows that such measures can result in significant cost savings. As such, it recommends that prompt, dynamic detection of infectious clusters via genomics and active surveillance offers a relevant and timely strategy for cost savings within the healthcare ecosystem. Disclosures All authors: No reported disclosures.

Background Whole-genome sequencing (WGS) has shown promise in identifying transmissions of healthcare-associated infections (HAIs), but it may be costly to sequence all potential HAIs. By automatically identifying samples likely to be HAIs, WGS can be focused on specific samples. We describe an algorithm that quickly identifies potential HAI clusters by analyzing patient geotemporal and pathogen microbiology data. This approach systematically triages potential HAI investigations to aid infection control professionals (ICPs) in their workflow. Methods This novel algorithm within Philips IntelliSpace Epidemiology scores the potential of transmission for pairs of infections. Inputs include microbiology (MB) data (genus- or species-level identification and antimicrobial susceptibility test results) and geotemporal (GT) data (timing of sample collection and shared location stays). From the resulting pairwise scores, clusters of potential HAIs are identified. Leveraging 9 months (June, 2018 – March, 2019) of data from a 900-bed US hospital (i.e., 2825 samples, 1814 patients and 13 organisms—of which a subset of 404 samples had WGS performed concomitantly with MB studies), we evaluated the extent to which this algorithm captures genetically similar sample pairs. Results Pairwise scores enrich for genetically similar samples when considering MB data only (odds ratio: 17.3), GT only (odds ratio: 6.1) and a combination of both (odds ratio: 19.8), with highly significant P-values for all (P < 10−16). Considering MB only, 91% of samples group together in potential transmission clusters. With MB and GT data, this fraction drops to 24.6% (694 samples) forming 178 possible clusters, 173 of which contain fewer than ten samples each. The 5 larger clusters contain 40–64 samples each and span multiple units in the hospital. Conclusion The proposed system automatically suggests potential HAI clusters. By combining MB and GT data, the number of samples to review is reduced, enabling ICPs to focus their attention and sequencing efforts. By focusing on a targeted group of higher probability clusters, ICPs may be able to increase their efficiency and effectiveness in controlling the spread of HAIs—thus boosting potential for patient safety and amelioration of cost of care. Disclosures All authors: No reported disclosures.

Background Whole-genome sequencing (WGS) of bacteria is becoming a routine tool within microbiology, yet its utility to help guide infection control (IC) practice longitudinally is underexplored. As with any technology adopted in the hospital, the integration of WGS into IC practice must be carefully managed and considered. We qualitatively report an evidence-based implementation workflow that considers WGS to help proactively guide IC professionals during investigation of infectious outbreaks. Methods We built upon lessons learned in an ongoing surveillance effort at a tertiary care hospital—utilizing retrospective WGS data within the Philips IntelliSpace Epidemiology system—to understand facilitators and barriers to the use of bacterial WGS longitudinally to inform IC workflow. Our team established a 9-month workgroup to study the practical aspects of implementing WGS in routine IC practice. From expert opinion collected via the workgroup, in addition to evidence from the literature, a workflow guidance document and checklist were codified. New ideas included incorporating education to promote the establishment of an IC triage process. Results Facilitators to implementation included ability to display genomic relatedness alongside relevant patient data to enable clinical actionability, ability to pivot time and resources rapidly when infections are a pseudo outbreak (false positive) or missed outbreak (false negative), opportunities for nuanced staff education, and willingness to be a first-of-kind adopter. Barriers were communication of genomic concepts to IC professionals and relevant institutional stakeholders, maintaining sharable notes of active investigations to promote data-sharing practices, and timing and review of relevant interventions into the facility workflow. Strategies to address these issues are considered. Conclusion This study provides a novel framework for adaptation of existing IC workflow strategies to leverage the utility of bacterial WGS, and it presents a schema to effectively engage relevant stakeholders, based on an analysis of the unique challenges inherent within IC practice. It also offers an innovative model for the development and implementation of IC workflows to account for, and adapt to, site-specific conditions. Disclosures All authors: No reported disclosures.

It has been known to the public that high frequency of false and/or unnecessary alarms from patient monitoring devices causes "alarm fatigue" in critical care. But little is known about the impact to care on the less acute patients located outside the critical care areas, such as the traditional medical/surgical (med/surg) floor.As part of a larger population management study, we initiated continuous physiological monitoring to 79 beds of floor patients in a community hospital. In order to qualify the patient monitoring alarm load for subacute medical and surgical floor patients, we assessed alarm data from April 2009 to January 2010. A standard critical care monitoring system (Philips IntelliVue MP-5 and Telemetry) was installed and set to the default alarm limits. All waveform data available for the patient (typically ECG, RESP, PPG at 125hz 8 bit), all alarm conditions declared by the monitoring system, and 1 minute parameter trend data were saved to disk every 8 hours for all patients. A monitoring care protocol was created to determine whether the patient was monitored via the hardwired bedside or wirelessly via telemetry. Alarms were not announced on the care unit but instead notifications were the responsibility of remote telehealth center personnel. We retrospectively evaluated the frequency of alarms over specific physiologic thresholds (n= 4104 patients) and conducted adjudication of all alarms based on a smaller sampling (n=30 patients).For all patients, the average hours of monitoring per patient were 16.5 hours with a standard deviation (s) of 8.3 hours and a median of 22 hours. The average number of alarms (all severities) per patient was 69.7 (s =90.3, median =28) alarms. When this is adjusted to the duration of monitoring, the average per patient, per day rate was 95.6 (s =124.2, median =34.2) alarms. The adjudicated sample (n=30 patients) resulted in 34% of critical alarms (lethal arrhythmias, extreme high or low heart rate [HR], extreme desaturation, apnea) being true and 63% of the high priority alarms (high or low HR, high or low RR, Low SpO(2), pause, Missed Beat, Pair PVCs, Pacer Not Pace, Non Sustain VT, Irregular HR, Multiform) being true. Analysis of alarm history resulted in the ability to reduce the HR alarm load by more than 50% with a simple limit adjustment of high HR from 120 to 130 bpm and a 36% or 65% reduction in SpO(2) alarm load by reducing the SpO(2) limit from 90% to 85% or 80% respectively.1) Standard critical care alarm limits appear be too sensitive for subacute care areas of the hospital. 2) For most patients these alarm limits do not create a significant alarm load; however, for a small number of patients they cause a significant alarm load. 3) Alarm loads can be controlled with alarm limit settings appropriate to the population. 4) Current technology for HR and SpO(2) appear suitable for continuous monitoring of this population.

Current algorithms identifying hemodynamically unstable intensive care unit patients typically are limited to detecting existing dangerous conditions and suffer from high false alert rates. Our objective was to predict hemodynamic instability at least two hours before patient deterioration while maintaining a low false alert rate, using minute-by-minute heart rate (HR) and blood pressure (BP) data. We identified 66 stable and 104 unstable patients meeting our stability-instability criteria from the MIMIC II database, and developed multi-parameter measures using HR and BP. An instability index combining measures of BP, shock index, rate pressure product, and HR variation was developed from a multivariate regression model to predict hemodynamic instability (ROC of 0.82±0.03, sensitivity of 0.57±0.07 when the specificity was targeted at 0.90; the alert rate ratio of unstable to stable patients was 7.62). We conclude that these algorithms could form the basis for reliable predictive clinical alerts which identify patients likely to become hemodynamically unstable within the next few hours so that the clinicians can proactively manage these patients and provide necessary care.

Mechanical ventilation is an important life support tool for patients in intensive care units (ICU). For various research purposes related to patient hemodynamic and cardiopulmonary monitoring, it is important to know when a patient is on a ventilator. Unfortunately, the widely used MIMIC-II database contains results from user charted data, where the user did not always store ventilation on and off times explicitly and accurately. The resulting ventilation-related data are subject to error. Therefore, there are no simple rules to define ventilation times retrospectively for this dataset. Hence, we designed a simple set of rules to determine the ventilation times using multiple sources of mechanical ventilator-related settings and physiological measurements by expert heuristics. The rules worked well in comparison with nursing notes regarding ventilation events. We conclude that our rule sets for determining ventilation times may be useful in assisting with MIMIC-II database analysis.

Acute lung injury (ALI) and acute respiratory distress syndrome (ARDS) contribute to the morbidity and mortality of intensive care patients worldwide, and have large associated human and financial costs. We identified a reference data set of 624 mechanically-ventilated patients in the MIMIC-II intensive care database with and without low PaO 2 /FiO 2 ratios (termed respiratory instability), and developed prediction algorithms for distinguishing these patients prior to the critical event. In the end, we had four rule sets using mean airway pressure, plateau pressure, total respiratory rate and oxygen saturation (SpO 2 ), where the specificity/sensitivity rates were either 80%/60% or 90%/50%.

Background: Identifying hemodynamically unstable patients in a timely fashion in intensive care units (ICUs) is crucial because it can lead to earlier interventions and thus to potentially better patient outcomes. Current alert algorithms are typically limited to detecting dangerous conditions only after they have occurred and suffer from high false alert rates. Our objective was to predict hemodynamic instability at least two hours before a major clinical intervention (e.g., vasopressor administration), while maintaining a low false alert rate. Study population: From the MIMIC II database, containing ICU minute-by-minute heart rate (HR) and invasive arterial blood pressure (BP) monitoring trend data collected between 2001 and 2005, we identified 132 stable and 104 unstable patients that met our stability-instability criteria and had sufficient data points. Method: We first derived additional physiological parameters of shock index, rate pressure product, heart rate variability, and two measures of trending based on HR and BP. Then we developed 220 statistical features and systematically selected a small set to use for classification. We applied multi-variable logistic regression modeling to do classification and implemented validation via bootstrapping. Results: Area under receiver-operating curve (ROC) 0.83±0.03, sensitivity 0.75±0.06, and specificity 0.80±0.07; if the specificity is targeted at 0.90, then the sensitivity is 0.57±0.07. Based on our preliminary results, we conclude that the algorithms we developed using HR and BP trend data may provide a promising perspective toward reliable predictive alerts for hemodynamically unstable patients.

An intraluminal membrane oxygenator (IMO) prototype was surgically inserted in the ileum and evaluated as a method of supporting systemic oxygenation in an acutely hypoxemic porcine model. Animals were assigned randomly to the test (n = 12) or the control (n = 8) groups, which underwent identical protocols with the exception of the O2 flow in the IMO device, which was shut off in the control group. In each case, hypoxia was induced by a reduction in the inspired oxygen fraction (Fio2) to 0.14. A highly significant improvement (p < 0.005) in arterial and venous O2 content and lower arteriovenous O2 difference (p < 0.05), cardiac output, and hemoglobin (p < 0.005) were found in the test group during hypoxia. The results show that it is possible to meet a physiologically significant portion of the body's O2 demands via the intestine during respiratory hypoxia and suggests that similar devices may be of significant potential value as a supplemental oxygenation device in cases of respiratory distress.