Search

Objective To propose an algorithm that utilizes only timestamps of longitudinal electronic health record data to classify clinical deterioration events. Materials and methods This retrospective study explores the efficacy of machine learning algorithms in classifying clinical deterioration events among patients in intensive care units using sequences of timestamps of vital sign measurements, flowsheets comments, order entries, and nursing notes. We design a data pipeline to partition events into discrete, regular time bins that we refer to as timesteps. Logistic regressions, random forest classifiers, and recurrent neural networks are trained on datasets of different length of timesteps, respectively, against a composite outcome of death, cardiac arrest, and Rapid Response Team calls. Then these models are validated on a holdout dataset. Results A total of 6720 intensive care unit encounters meet the criteria and the final dataset includes 830 578 timestamps. The gated recurrent unit model utilizes timestamps of vital signs, order entries, flowsheet comments, and nursing notes to achieve the best performance on the time-to-outcome dataset, with an area under the precision-recall curve of 0.101 (0.06, 0.137), a sensitivity of 0.443, and a positive predictive value of 0. 092 at the threshold of 0.6. Discussion and Conclusion This study demonstrates that our recurrent neural network models using only timestamps of longitudinal electronic health record data that reflect healthcare processes achieve well-performing discriminative power.

Mechanical ventilation is an essential tool in the management of Acute Respiratory Distress Syndrome (ARDS), but it exposes patients to the risk of ventilator-induced lung injury (VILI). The human lung-ventilator system (LVS) involves the interaction of complex anatomy with a mechanical apparatus, which limits the achievable flexibility and fidelity needed to provide individualized clinical support by modeling lung processes. This work proposes a hypothesis-driven strategy for LVS modeling, in which robust personalization is achieved using a pre-defined parameter basis in a non-physiological model. Model inversion, here via windowed data assimilation, forges observed waveforms into interpretable parameter values that characterize the data rather than quantifying physiological processes. Inference experiments performed on human pressure waveform data indicate the flexible model accurately estimates parameters for a variety of breath types, including breaths of markedly dyssynchronous LVSs. Parameter estimates generate static characterizations of the data that are 50–70% more accurate than breath-wise single-compartment model estimates. They also retain sufficient information to distinguish between the types of breath they represent. However, the fidelity and interpetability of model characterizations are tied to parameter definitions and model resolution. These additional factors must be considered in conjunction with the objectives of specific applications, such as identifying and tracking the development of human VILI.

Background It would be useful to be able to assess the utility of predictive models of continuous values before clinical trials are performed. Objective The aim of the study is to compare metrics to assess the potential clinical utility of models that produce continuous value forecasts. Methods We ran a set of data assimilation forecast algorithms on time series of glucose measurements from neurological intensive care unit patients. We evaluated the forecasts using four sets of metrics: glucose root mean square (RMS) error, a set of metrics on a transformed glucose value, the estimated effect on clinical care based on an insulin guideline, and a glucose measurement error grid (Parkes grid). We assessed correlation among the metrics and created a set of factor models. Results The metrics generally correlated with each other, but those that estimated the effect on clinical care correlated with others the least and were generally associated with their own independent factors. The other metrics appeared to separate into those that emphasized errors in low glucose versus errors in high glucose. The Parkes grid was well correlated with the transformed glucose but not the estimation of clinical care. Discussion Our results indicate that we need to be careful before we assume that commonly used metrics like RMS error in raw glucose or even metrics like the Parkes grid that are designed to measure importance of differences will correlate well with actual effect on clinical care processes. A combination of metrics appeared to explain the most variance between cases. As prediction algorithms move into practice, it will be important to measure actual effects.

Diabetes is caused by dysfunction of electrically coupled heterogeneous β-cells within the pancreatic islet. Functional networks have been used to represent cellular synchronization and study β-cells subpopulations, which play an important role in driving dynamics. The mechanism by which highly synchronized β-cell subpopulations drive islet function is not clear. We used experimental and computational techniques to investigate the relationship between functional networks, structural (gap-junction) networks, and underlying β-cell dynamics. Highly synchronized subpopulations in the functional network were differentiated by metabolic dynamics rather than structural coupling. Consistent with this, metabolic similarities were more predictive of edges in the islet functional network. Finally, removal of gap junctions, as occurs in diabetes, caused decreases in the efficiency and clustering of the functional network. These results indicate that metabolism rather than structure drives connections in the function network, deepening our interpretation of functional networks and the formation of functional sub-populations in dynamic tissues such as the islet.

Objectives Nursing documentation behavior within electronic health records may reflect a nurse’s concern about a patient and can be used to predict patient deterioration. Our study objectives were to quantify variations in nursing documentation patterns, confirm those patterns and variations with clinicians, and identify which patterns indicate patient deterioration and recovery from clinical deterioration events in the critical and acute care settings. Methods We collected patient data from electronic health records and conducted a regression analysis to identify different nursing documentation patterns associated with patient outcomes resulting from clinical deterioration events in the intensive care unit (ICU) and acute care unit (ACU). The primary outcome measures were whether patients were discharged alive from the hospital or expired during their hospital encounter. Secondary outcome measures were clinical deterioration events. Results In the ICU, the increased documentation of heart rate, body temperature, and withheld medication administrations were significantly associated with inpatient mortality. In the ACU, the documentation of blood pressure, respiratory rate with comments, singular vital signs, and withheld medications were significantly related to inpatient mortality. In contrast, the documentation of heart rate and “as needed” medication administrations were significantly associated with patient survival to discharge in the ACU. Conclusion We successfully identified and confirmed the clinical relevancy of the nursing documentation patterns indicative of patient deterioration and recovery from clinical deterioration events in both the ICU and ACU.

In this paper, we describe two case studies of research projects that attempt to scale up HCI research beyond traditional small evaluation studies. The first of these projects focused on evaluating an interactive web application for promoting problem-solving in self-management of type 2 diabetes mellitus (T2DM) in a randomized clinical trial; the second one included deployment in the wild of a smartphone app that provided individuals with T2DM with personalized predictions for changes in blood glucose levels in response to meals. We highlight lessons learned during these two projects and describe four different design considerations important for large scale studies. These include designing for longevity, diversity, adoption, and abandonment. We then discuss implications for future research that targets large scale deployment studies.

Self-tracking can help personalize self-management interventions for chronic conditions like type 2 diabetes (T2D), but reflecting on personal data requires motivation and literacy. Machine learning (ML) methods can identify patterns, but a key challenge is making actionable suggestions based on personal health data. We introduce GlucoGoalie, which combines ML with an expert system to translate ML output into personalized nutrition goal suggestions for individuals with T2D. In a controlled experiment, participants with T2D found that goal suggestions were understandable and actionable. A 4-week in-the-wild deployment study showed that receiving goal suggestions augmented participants’ self-discovery, choosing goals highlighted the multifaceted nature of personal preferences, and the experience of following goals demonstrated the importance of feedback and context. However, we identified tensions between abstract goals and concrete eating experiences and found static text too ambiguous for complex concepts. We discuss implications for ML-based interventions and the need for systems that offer more interactivity, feedback, and negotiation.

ObjectiveStudies have independently shown associations of lower hemoglobin levels with larger admission intracerebral hemorrhage (ICH) volumes and worse outcomes. We investigated whether lower admission hemoglobin levels are associated with more hematoma expansion (HE) after ICH and whether this mediates lower hemoglobin levels' association with worse outcomes.MethodsConsecutive patients enrolled between 2009 and 2016 to a single-center prospective ICH cohort study with admission hemoglobin and neuroimaging data to calculate HE (>33% or >6 mL) were evaluated. The association of admission hemoglobin levels with HE and poor clinical outcomes using modified Rankin Scale (mRS 4–6) were assessed using separate multivariable logistic regression models. Mediation analysis investigated causal associations among hemoglobin, HE, and outcome.ResultsOf 256 patients with ICH meeting inclusion criteria, 63 (25%) had HE. Lower hemoglobin levels were associated with increased odds of HE (odds ratio [OR] 0.80 per 1.0 g/dL change of hemoglobin; 95% confidence interval [CI] 0.67–0.97) after adjusting for previously identified covariates of HE (admission hematoma volume, antithrombotic medication use, symptom onset to admission CT time) and hemoglobin (age, sex). Lower hemoglobin was also associated with worse 3-month outcomes (OR 0.76 per 1.0 g/dL change of hemoglobin; 95% CI 0.62–0.94) after adjusting for ICH score. Mediation analysis revealed that associations of lower hemoglobin with poor outcomes were mediated by HE (p = 0.01).ConclusionsFurther work is required to replicate the associations of lower admission hemoglobin levels with increased odds of HE mediating worse outcomes after ICH. If confirmed, an investigation into whether hemoglobin levels can be a modifiable target of treatment to improve ICH outcomes may be warranted.

Background The clinical mitigation of intracranial hypertension due to traumatic brain injury requires timely knowledge of intracranial pressure to avoid secondary injury or death. Noninvasive intracranial pressure (nICP) estimation that operates sufficiently fast at multihour timescales and requires only common patient measurements is a desirable tool for clinical decision support and improving traumatic brain injury patient outcomes. However, existing model-based nICP estimation methods may be too slow or require data that are not easily obtained. Objective This work considers short- and real-time nICP estimation at multihour timescales based on arterial blood pressure (ABP) to better inform the ongoing development of practical models with commonly available data. Methods We assess and analyze the effects of two distinct pathways of model development, either by increasing physiological integration using a simple pressure estimation model, or by increasing physiological fidelity using a more complex model. Comparison of the model approaches is performed using a set of quantitative model validation criteria over hour-scale times applied to model nICP estimates in relation to observed ICP. Results The simple fully coupled estimation scheme based on windowed regression outperforms a more complex nICP model with prescribed intracranial inflow when pulsatile ABP inflow conditions are provided. We also show that the simple estimation data requirements can be reduced to 1-minute averaged ABP summary data under generic waveform representation. Conclusions Stronger performance of the simple bidirectional model indicates that feedback between the systemic vascular network and nICP estimation scheme is crucial for modeling over long intervals. However, simple model reduction to ABP-only dependence limits its utility in cases involving other brain injuries such as ischemic stroke and subarachnoid hemorrhage. Additional methodologies and considerations needed to overcome these limitations are illustrated and discussed.

OBJECTIVE: Targets for treatment of raised intracranial pressure (ICP) or decreased cerebral perfusion pressure (CPP) in pediatric neurocritical care are not well-defined. Current pediatric guidelines, based on traumatic brain injury (TBI), suggest an ICP target of 15 mmHg, male sex, and TBI status were independently associated with in-hospital mortality (OR 14.23[5.55-36.46], 2.77[1.04-7.39], and 2.57[1.03-6.38], respectively; all p

Medical practice in the intensive care unit is based on the supposition that physiological systems such as the human glucose-insulin system are predictable. We demonstrate that delay within the glucose-insulin system can induce sustained temporal chaos, rendering the system unpredictable. Specifically, we exhibit such chaos for the Ultradian glucose-insulin model. This well-validated, finite-dimensional model represents feedback delay as a three-stage filter. Using the theory of rank one maps from smooth dynamical systems, we precisely explain the nature of the resulting delay-induced uncertainty (DIU). We develop a recipe one may use to diagnose DIU in a general oscillatory dynamical system. For infinite-dimensional delay systems, no analog of the theory of rank one maps exists. Nevertheless, we show that the geometric principles encoded in our DIU recipe apply to such systems by exhibiting sustained temporal chaos for a linear shear flow. Our results are potentially broadly applicable because delay is ubiquitous throughout mathematical physiology., 19 pages; 9 figures

BACKGROUND Background: It would be useful to be able to assess the utility of predictive models of continuous values before clinical trials are carried out. OBJECTIVE Objective: To compare metrics to assess the potential clinical utility of models that produce continuous value forecasts. METHODS Methods: We ran a set of data assimilation forecast algorithms on time series of glucose measurements from intensive care unit patients. We evaluated the forecasts using four sets of metrics: glucose root mean square error, a set of metrics on a transformed glucose value, the estimated effect on clinical care based on an insulin guideline, and a glucose measurement error grid (Parkes grid). We assessed correlation among the metrics and created a set of factor models. RESULTS Results: The metrics generally correlated with each other, but those that estimated the effect on clinical care correlated with the others the least and were generally associated with their own independent factors. The other metrics appeared to separate into those that emphasized errors in low glucose versus errors in high glucose. The Parkes grid was well correlated with the transformed glucose but not the estimation of clinical care. CONCLUSIONS Discussion: Our results indicate that we need to be careful before we assume that commonly used metrics like RMS error in raw glucose or even metrics like the Parkes grid that are designed to measure importance of differences will correlate well with actual effect on clinical care processes. A combination of metrics appeared to explain the most variance between cases. As prediction algorithms move into practice, it will be important to measure actual effects.

Decision-making related to health is complex. Machine learning (ML) and patient generated data can identify patterns and insights at the individual level, where human cognition falls short, but not all ML-generated information is of equal utility for making health-related decisions. We develop and apply attributable components analysis (ACA), a method inspired by optimal transport theory, to type 2 diabetes self-monitoring data to identify patterns of association between nutrition and blood glucose control. In comparison with linear regression, we found that ACA offers a number of characteristics that make it promising for use in decision support applications. For example, ACA was able to identify non-linear relationships, was more robust to outliers, and offered broader and more expressive uncertainty estimates. In addition, our results highlight a tradeoff between model accuracy and interpretability, and we discuss implications for ML-driven decision support systems.

Background Semisupervised and unsupervised anomaly detection methods have been widely used in various applications to detect anomalous objects from a given data set. Specifically, these methods are popular in the medical domain because of their suitability for applications where there is a lack of a sufficient data set for the other classes. Infection incidence often brings prolonged hyperglycemia and frequent insulin injections in people with type 1 diabetes, which are significant anomalies. Despite these potentials, there have been very few studies that focused on detecting infection incidences in individuals with type 1 diabetes using a dedicated personalized health model. Objective This study aims to develop a personalized health model that can automatically detect the incidence of infection in people with type 1 diabetes using blood glucose levels and insulin-to-carbohydrate ratio as input variables. The model is expected to detect deviations from the norm because of infection incidences considering elevated blood glucose levels coupled with unusual changes in the insulin-to-carbohydrate ratio. Methods Three groups of one-class classifiers were trained on target data sets (regular days) and tested on a data set containing both the target and the nontarget (infection days). For comparison, two unsupervised models were also tested. The data set consists of high-precision self-recorded data collected from three real subjects with type 1 diabetes incorporating blood glucose, insulin, diet, and events of infection. The models were evaluated on two groups of data: raw and filtered data and compared based on their performance, computational time, and number of samples required. Results The one-class classifiers achieved excellent performance. In comparison, the unsupervised models suffered from performance degradation mainly because of the atypical nature of the data. Among the one-class classifiers, the boundary and domain-based method produced a better description of the data. Regarding the computational time, nearest neighbor, support vector data description, and self-organizing map took considerable training time, which typically increased as the sample size increased, and only local outlier factor and connectivity-based outlier factor took considerable testing time. Conclusions We demonstrated the applicability of one-class classifiers and unsupervised models for the detection of infection incidence in people with type 1 diabetes. In this patient group, detecting infection can provide an opportunity to devise tailored services and also to detect potential public health threats. The proposed approaches achieved excellent performance; in particular, the boundary and domain-based method performed better. Among the respective groups, particular models such as one-class support vector machine, K-nearest neighbor, and K-means achieved excellent performance in all the sample sizes and infection cases. Overall, we foresee that the results could encourage researchers to examine beyond the presented features into other additional features of the self-recorded data, for example, continuous glucose monitoring features and physical activity data, on a large scale.

Medical practice in the intensive care unit is based on the supposition that physiological systems such as the human glucose-insulin system are reliabile. Reliability of dynamical systems refers to response to perturbation: A dynamical system is reliable if it behaves predictably following a perturbation. Here, we demonstrate that reliability fails for an archetypal physiological model, the Ultradian glucose-insulin model. Reliability failure arises because of the presence of delay. Using the theory of rank one maps from smooth dynamical systems, we precisely explain the nature of the resulting delay-induced uncertainty (DIU). We develop a recipe one may use to diagnose DIU in a general dynamical system. Guided by this recipe, we analyze DIU emergence first in a classical linear shear flow model and then in the Ultradian model. Our results potentially apply to a broad class of physiological systems that involve delay.

Background Type 1 diabetes is a chronic condition of blood glucose metabolic disorder caused by a lack of insulin secretion from pancreas cells. In people with type 1 diabetes, hyperglycemia often occurs upon infection incidences. Despite the fact that patients increasingly gather data about themselves, there are no solid findings that uncover the effect of infection incidences on key parameters of blood glucose dynamics to support the effort toward developing a digital infectious disease detection system. Objective The study aims to retrospectively analyze the effect of infection incidence and pinpoint optimal parameters that can effectively be used as input variables for developing an infection detection algorithm and to provide a general framework regarding how a digital infectious disease detection system can be designed and developed using self-recorded data from people with type 1 diabetes as a secondary source of information. Methods We retrospectively analyzed high precision self-recorded data of 10 patient-years captured within the longitudinal records of three people with type 1 diabetes. Obtaining such a rich and large data set from a large number of participants is extremely expensive and difficult to acquire, if not impossible. The data set incorporates blood glucose, insulin, carbohydrate, and self-reported events of infections. We investigated the temporal evolution and probability distribution of the key blood glucose parameters within a specified timeframe (weekly, daily, and hourly). Results Our analysis demonstrated that upon infection incidence, there is a dramatic shift in the operating point of the individual blood glucose dynamics in all the timeframes (weekly, daily, and hourly), which clearly violates the usual norm of blood glucose dynamics. During regular or normal situations, higher insulin and reduced carbohydrate intake usually results in lower blood glucose levels. However, in all infection cases as opposed to the regular or normal days, blood glucose levels were elevated for a prolonged period despite higher insulin and reduced carbohydrates intake. For instance, compared with the preinfection and postinfection weeks, on average, blood glucose levels were elevated by 6.1% and 16%, insulin (bolus) was increased by 42% and 39.3%, and carbohydrate consumption was reduced by 19% and 28.1%, respectively. Conclusions We presented the effect of infection incidence on key parameters of blood glucose dynamics along with the necessary framework to exploit the information for realizing a digital infectious disease detection system. The results demonstrated that compared with regular or normal days, infection incidence substantially alters the norm of blood glucose dynamics, which are quite significant changes that could possibly be detected through personalized modeling, for example, prediction models and anomaly detection algorithms. Generally, we foresee that these findings can benefit the efforts toward building next generation digital infectious disease detection systems and provoke further thoughts in this challenging field.

We studied how lagged linear regression can be used to detect the physiologic effects of drugs from data in the electronic health record (EHR). We systematically examined the effect of methodological variations ((i) time series construction, (ii) temporal parameterization, (iii) intra-subject normalization, (iv) differencing (lagged rates of change achieved by taking differences between consecutive measurements), (v) explanatory variables, and (vi) regression models) on performance of lagged linear methods in this context. We generated two gold standards (one knowledge-base derived, one expert-curated) for expected pairwise relationships between 7 drugs and 4 labs, and evaluated how the 64 unique combinations of methodological perturbations reproduce the gold standards. Our 28 cohorts included patients in the Columbia University Medical Center/NewYork-Presbyterian Hospital clinical database, and ranged from 2,820 to 79,514 patients with between 8 and 209 average time points per patient. The most accurate methods achieved AUROC of 0.794 for knowledge-base derived gold standard (95%CI [0.741, 0.847]) and 0.705 for expert-curated gold standard (95% CI [0.629, 0.781]). We observed a mean AUROC of 0.633 (95%CI [0.610, 0.657], expert-curated gold standard) across all methods that re-parameterize time according to sequence and use either a joint autoregressive model with time-series differencing or an independent lag model without differencing. The complement of this set of methods achieved a mean AUROC close to 0.5, indicating the importance of these choices. We conclude that time-series analysis of EHR data will likely rely on some of the beneficial pre-processing and modeling methodologies identified, and will certainly benefit from continued careful analysis of methodological perturbations. This study found that methodological variations, such as pre-processing and representations, have a large effect on results, exposing the importance of thoroughly evaluating these components when comparing machine-learning methods.

We introduce data assimilation as a computational method that uses machine learning to combine data with human knowledge in the form of mechanistic models in order to forecast future states, to impute missing data from the past by smoothing, and to infer measurable and unmeasurable quantities that represent clinically and scientifically important phenotypes. We demonstrate the advantages it affords in the context of type 2 diabetes by showing how data assimilation can be used to forecast future glucose values, to impute previously missing glucose values, and to infer type 2 diabetes phenotypes. At the heart of data assimilation is the mechanistic model, here an endocrine model. Such models can vary in complexity, contain testable hypotheses about important mechanics that govern the system (eg, nutrition’s effect on glucose), and, as such, constrain the model space, allowing for accurate estimation using very little data.

ObjectiveTo develop and test a visual analytics tool to help clinicians identify systematic and clinically meaningful patterns in patient-generated data (PGD) while decreasing perceived information overload.MethodsParticipatory design was used to develop Glucolyzer, an interactive tool featuring hierarchical clustering and a heatmap visualization to help registered dietitians (RDs) identify associative patterns between blood glucose levels and per-meal macronutrient composition for individuals with type 2 diabetes (T2DM). Ten RDs participated in a within-subjects experiment to compare Glucolyzer to a static logbook format. For each representation, participants had 25 minutes to examine 1 month of diabetes self-monitoring data captured by an individual with T2DM and identify clinically meaningful patterns. We compared the quality and accuracy of the observations generated using each representation.ResultsParticipants generated 50% more observations when using Glucolyzer (98) than when using the logbook format (64) without any loss in accuracy (69% accuracy vs 62%, respectively, p = .17). Participants identified more observations that included ingredients other than carbohydrates using Glucolyzer (36% vs 16%, p = .027). Fewer RDs reported feelings of information overload using Glucolyzer compared to the logbook format. Study participants displayed variable acceptance of hierarchical clustering.ConclusionsVisual analytics have the potential to mitigate provider concerns about the volume of self-monitoring data. Glucolyzer helped dietitians identify meaningful patterns in self-monitoring data without incurring perceived information overload. Future studies should assess whether similar tools can support clinicians in personalizing behavioral interventions that improve patient outcomes.

We study the question of how to represent or summarize raw laboratory data taken from an electronic health record (EHR) using parametric model selection to reduce or cope with biases induced through clinical care. It has been previously demonstrated that the health care process (Hripcsak and Albers, 2012, 2013), as defined by measurement context (Hripcsak and Albers, 2013; Albers et al., 2012) and measurement patterns (Albers and Hripcsak, 2010, 2012), can influence how EHR data are distributed statistically (Kohane and Weber, 2013; Pivovarov et al., 2014). We construct an algorithm, PopKLD, which is based on information criterion model selection (Burnham and Anderson, 2002; Claeskens and Hjort, 2008), is intended to reduce and cope with health care process biases and to produce an intuitively understandable continuous summary. The PopKLD algorithm can be automated and is designed to be applicable in high-throughput settings; for example, the output of the PopKLD algorithm can be used as input for phenotyping algorithms. Moreover, we develop the PopKLD-CAT algorithm that transforms the continuous PopKLD summary into a categorical summary useful for applications that require categorical data such as topic modeling. We evaluate our methodology in two ways. First, we apply the method to laboratory data collected in two different health care contexts, primary versus intensive care. We show that the PopKLD preserves known physiologic features in the data that are lost when summarizing the data using more common laboratory data summaries such as mean and standard deviation. Second, for three disease-laboratory measurement pairs, we perform a phenotyping task: we use the PopKLD and PopKLD-CAT algorithms to define high and low values of the laboratory variable that are used for defining a disease state. We then compare the relationship between the PopKLD-CAT summary disease predictions and the same predictions using empirically estimated mean and standard deviation to a gold standard generated by clinical review of patient records. We find that the PopKLD laboratory data summary is substantially better at predicting disease state. The PopKLD or PopKLD-CAT algorithms are not meant to be used as phenotyping algorithms, but we use the phenotyping task to show what information can be gained when using a more informative laboratory data summary. In the process of evaluation our method we show that the different clinical contexts and laboratory measurements necessitate different statistical summaries. Similarly, leveraging the principle of maximum entropy we argue that while some laboratory data only have sufficient information to estimate a mean and standard deviation, other laboratory data captured in an EHR contain substantially more information than can be captured in higher-parameter models.

Ventilator dyssynchrony is potentially harmful to patients with or at risk for the acute respiratory distress syndrome. Automated detection of ventilator dyssynchrony from ventilator waveforms has been difficult. It is unclear if certain types of ventilator dyssynchrony deliver large tidal volumes and whether levels of sedation alter the frequency of ventilator dyssynchrony.A prospective observational study.A university medical ICU.Patients with or at risk for acute respiratory distress syndrome.Continuous pressure-time, flow-time, and volume-time data were directly obtained from the ventilator. The level of sedation and the use of neuromuscular blockade was extracted from the medical record. Machine learning algorithms that incorporate clinical insight were developed and trained to detect four previously described and clinically relevant forms of ventilator dyssynchrony. The association between normalized tidal volume and ventilator dyssynchrony and the association between sedation and the frequency of ventilator dyssynchrony were determined.A total of 4.26 million breaths were recorded from 62 ventilated patients. Our algorithm detected three types of ventilator dyssynchrony with an area under the receiver operator curve of greater than 0.89. Ventilator dyssynchrony occurred in 34.4% (95% CI, 34.41-34.49%) of breaths. When compared with synchronous breaths, double-triggered and flow-limited breaths were more likely to deliver tidal volumes greater than 10 mL/kg (40% and 11% compared with 0.2%; p0.001 for both comparisons). Deep sedation reduced but did not eliminate the frequency of all ventilator dyssynchrony breaths (p0.05). Ventilator dyssynchrony was eliminated with neuromuscular blockade (p0.001).We developed a computerized algorithm that accurately detects three types of ventilator dyssynchrony. Double-triggered and flow-limited breaths are associated with the frequent delivery of tidal volumes of greater than 10 mL/kg. Although ventilator dyssynchrony is reduced by deep sedation, potentially deleterious tidal volumes may still be delivered. However, neuromuscular blockade effectively eliminates ventilator dyssynchrony.

Electronic health record phenotyping is the use of raw electronic health record data to assert characterizations about patients. Researchers have been doing it since the beginning of biomedical informatics, under different names. Phenotyping will benefit from an increasing focus on fidelity, both in the sense of increasing richness, such as measured levels, degree or severity, timing, probability, or conceptual relationships, and in the sense of reducing bias. Research agendas should shift from merely improving binary assignment to studying and improving richer representations. The field is actively researching new temporal directions and abstract representations, including deep learning. The field would benefit from research in nonlinear dynamics, in combining mechanistic models with empirical data, including data assimilation, and in topology. The health care process produces substantial bias, and studying that bias explicitly rather than treating it as merely another source of noise would facilitate addressing it.

Cohort identification for clinical studies tends to be laborious, time-consuming, and expensive. Developing automated or semi-automated methods for cohort identification is one of the “holy grails” in the field of biomedical informatics. We propose a high-throughput similarity-based cohort identification algorithm by applying numerical abstractions on Electronic Health Records (EHR) data. We implement this algorithm using the Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM), which enables sites using this standardized EHR data representation to avail this algorithm with minimum effort for local implementation. We validate its performance for a retrospective cohort identification task on six clinical trials conducted at the Columbia University Medical Center. Our algorithm achieves an average Area Under the Curve (AUC) of 0.966 and an average Precision at 5 of 0.983. This interoperable method promises to achieve efficient cohort identification in EHR databases. We discuss suitable applications of our method and its limitations and propose warranted future work.

Subarachnoid hemorrhage (SAH) is a devastating form of stroke. Approximately one in four patients develop progressive neurological deterioration and silent infarction referred to as delayed cerebral ischemia (DCI). DCI is a complex, multifactorial secondary brain injury pattern and its pathogenesis is not fully understood. We aimed to study the relationship between cerebral blood flow (CBF) and neuronal activity at both the cortex and in scalp using electroencephalography (EEG) in poor-grade SAH patients undergoing multimodality intracranial neuromonitoring. Twenty patients were included, of whom half had DCI median 4.7 days (interquartile range (IQR): 4.0–5.6) from SAH bleed. The rate of decline in regional cerebral blood flow (rCBF) was significant in both those with and without DCI and occurred between days 4 and 7 post-SAH. The scalp EEG alpha-delta ratio declined early in those with DCI. In the group without DCI, CBF and cortical EEG alpha-delta ratio were correlated (r = 0.53; p

Objectives This review aims to: 1) evaluate the quality of model reporting, 2) provide an overview of methodology for developing and validating Early Warning Score Systems (EWSs) for adult patients in acute care settings, and 3) highlight the strengths and limitations of the methodologies, as well as identify future directions for EWS derivation and validation studies. Methodology A systematic search was conducted in PubMed, Cochrane Library, and CINAHL. Only peer reviewed articles and clinical guidelines regarding developing and validating EWSs for adult patients in acute care settings were included. 615 articles were extracted and reviewed by five of the authors. Selected studies were evaluated based on the Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) checklist. The studies were analyzed according to their study design, predictor selection, outcome measurement, methodology of modeling, and validation strategy. Results A total of 29 articles were included in the final analysis. Twenty-six articles reported on the development and validation of a new EWS, while three reported on validation and model modification. Only eight studies met more than 75% of the items in the TRIPOD checklist. Three major techniques were utilized among the studies to inform their predictive algorithms: 1) clinical-consensus models (n = 6), 2) regression models (n = 15), and 3) tree models (n = 5). The number of predictors included in the EWSs varied from 3 to 72 with a median of seven. Twenty-eight models included vital signs, while 11 included lab data. Pulse oximetry, mental status, and other variables extracted from electronic health records (EHRs) were among other frequently used predictors. In-hospital mortality, unplanned transfer to the intensive care unit (ICU), and cardiac arrest were commonly used clinical outcomes. Twenty-eight studies conducted a form of model validation either within the study or against other widely-used EWSs. Only three studies validated their model using an external database separate from the derived database. Conclusion This literature review demonstrates that the characteristics of the cohort, predictors, and outcome selection, as well as the metrics for model validation, vary greatly across EWS studies. There is no consensus on the optimal strategy for developing such algorithms since data-driven models with acceptable predictive accuracy are often site-specific. A standardized checklist for clinical prediction model reporting exists, but few studies have included reporting aligned with it in their publications. Data-driven models are subjected to biases in the use of EHR data, thus it is particularly important to provide detailed study protocols and acknowledge, leverage, or reduce potential biases of the data used for EWS development to improve transparency and generalizability.

Accepted manuscript version, licensed CC BY-NC-ND 4.0. Background: Diabetes mellitus (DM) is a metabolic disorder that causes abnormal blood glucose (BG) regulation that might result in short and long-term health complications and even death if not properly managed. Currently, there is no cure for diabetes. However, self-management of the disease, especially keeping BG in the recommended range, is central to the treatment. This includes actively tracking BG levels and managing physical activity, diet, and insulin intake. The recent advancements in diabetes technologies and self-management applications have made it easier for patients to have more access to relevant data. In this regard, the development of an artificial pancreas (a closed-loop system), personalized decision systems, and BG event alarms are becoming more apparent than ever. Techniques such as predicting BG (modeling of a personalized profile), and modeling BG dynamics are central to the development of these diabetes management technologies. The increased availability of sufficient patient historical data has paved the way for the introduction of machine learning and its application for intelligent and improved systems for diabetes management. The capability of machine learning to solve complex tasks with dynamic environment and knowledge has contributed to its success in diabetes research. Motivation: Recently, machine learning and data mining have become popular, with their expanding application in diabetes research and within BG prediction services in particular. Despite the increasing and expanding popularity of machine learning applications in BG prediction services, updated reviews that map and materialize the current trends in modeling options and strategies are lacking within the context of BG prediction (modeling of personalized profile) in type 1 diabetes. Objective: The objective of this review is to develop a compact guide regarding modeling options and strategies of machine learning and a hybrid system focusing on the prediction of BG dynamics in type 1 diabetes. The review covers machine learning approaches pertinent to the controller of an artificial pancreas (closed-loop systems), modeling of personalized profiles, personalized decision support systems, and BG alarm event applications. Generally, the review will identify, assess, analyze, and discuss the current trends of machine learning applications within these contexts. Method: A rigorous literature review was conducted between August 2017 and February 2018 through various online databases, including Google Scholar, PubMed, ScienceDirect, and others. Additionally, peer-reviewed journals and articles were considered. Relevant studies were first identified by reviewing the title, keywords, and abstracts as preliminary filters with our selection criteria, and then we reviewed the full texts of the articles that were found relevant. Information from the selected literature was extracted based on predefined categories, which were based on previous research and further elaborated through brainstorming among the authors. Results: The initial search was done by analyzing the title, abstract, and keywords. A total of 624 papers were retrieved from DBLP Computer Science (25), Diabetes Technology and Therapeutics (31), Google Scholar (193), IEEE (267), Journal of Diabetes Science and Technology (31), PubMed/Medline (27), and ScienceDirect (50). After removing duplicates from the list, 417 records remained. Then, we independently assessed and screened the articles based on the inclusion and exclusion criteria, which eliminated another 204 papers, leaving 213 relevant papers. After a full-text assessment, 55 articles were left, which were critically analyzed. The inter-rater agreement was measured using a Cohen Kappa test, and disagreements were resolved through discussion. Conclusion: Due to the complexity of BG dynamics, it remains difficult to achieve a universal model that produces an accurate prediction in every circumstance (i.e., hypo/eu/hyperglycemia events). Recently, machine learning techniques have received wider attention and increased popularity in diabetes research in general and BG prediction in particular, coupled with the ever-growing availability of a self-collected health data. The stateof-the-art demonstrates that various machine learning techniques have been tested to predict BG, such as recurrent neural networks, feed-forward neural networks, support vector machines, self-organizing maps, the Gaussian process, genetic algorithm and programs, deep neural networks, and others, using various group of input parameters and training algorithms. The main limitation of the current approaches is the lack of a welldefined approach to estimate carbohydrate intake, which is mainly done manually by individual users and is prone to an error that can severely affect the predictive performance. Moreover, a universal approach has not been established to estimate and quantify the approximate effect of physical activities, stress, and infections on the BG level. No researchers have assessed model predictive performance during stress and infection incidences in a free-living condition, which should be considered in future studies. Furthermore, a little has been done regarding model portability that can capture inter- and intra-variability among patients. It seems that the effect of time lags between the CGM readings and the actual BG levels is not well covered. However, in general, we foresee that these developments might foster the advancement of next-generation BG prediction algorithms, which will make a great contribution in the effort to develop the long–awaited, so-called artificial pancreas (a closed-loop system).

The increasing availability of health data and knowledge about computationally modeling human physiology opens new opportunities for personalized predictions in health. Yet little is known about how individuals interact and reason with personalized predictions. To explore these questions, we developed a smartphone app, GlucOracle, that uses self-tracking data of individuals with type 2 diabetes to generate personalized forecasts for post-meal blood glucose levels. We pilot-tested GlucOracle with two populations: members of an online diabetes community, knowledgeable about diabetes and technologically savvy; and individuals from a low socio-economic status community, characterized by high prevalence of diabetes, low literacy and limited experience with mobile apps. Individuals in both communities engaged with personal glucose forecasts and found them useful for adjusting immediate meal options, and planning future meals. However, the study raised new questions as to appropriate time, form, and focus of forecasts and suggested new research directions for personalized predictions in health.

Background - Diabetes mellitus is a chronic metabolic disorder that results in abnormal blood glucose (BG) regulations. The BG level is preferably maintained close to normality through self-management practices, which involves actively tracking BG levels and taking proper actions including adjusting diet and insulin medications. BG anomalies could be defined as any undesirable reading because of either a precisely known reason (normal cause variation) or an unknown reason (special cause variation) to the patient. Recently, machine-learning applications have been widely introduced within diabetes research in general and BG anomaly detection in particular. However, irrespective of their expanding and increasing popularity, there is a lack of up-to-date reviews that materialize the current trends in modeling options and strategies for BG anomaly classification and detection in people with diabetes. Objective - This review aimed to identify, assess, and analyze the state-of-the-art machine-learning strategies and their hybrid systems focusing on BG anomaly classification and detection including glycemic variability (GV), hyperglycemia, and hypoglycemia in type 1 diabetes within the context of personalized decision support systems and BG alarm events applications, which are important constituents for optimal diabetes self-management. Methods - A rigorous literature search was conducted between September 1 and October 1, 2017, and October 15 and November 5, 2018, through various Web-based databases. Peer-reviewed journals and articles were considered. Information from the selected literature was extracted based on predefined categories, which were based on previous research and further elaborated through brainstorming. Results - The initial results were vetted using the title, abstract, and keywords and retrieved 496 papers. After a thorough assessment and screening, 47 articles remained, which were critically analyzed. The interrater agreement was measured using a Cohen kappa test, and disagreements were resolved through discussion. The state-of-the-art classes of machine learning have been developed and tested up to the task and achieved promising performance including artificial neural network, support vector machine, decision tree, genetic algorithm, Gaussian process regression, Bayesian neural network, deep belief network, and others. Conclusions - Despite the complexity of BG dynamics, there are many attempts to capture hypoglycemia and hyperglycemia incidences and the extent of an individual’s GV using different approaches. Recently, the advancement of diabetes technologies and continuous accumulation of self-collected health data have paved the way for popularity of machine learning in these tasks. According to the review, most of the identified studies used a theoretical threshold, which suffers from inter- and intrapatient variation. Therefore, future studies should consider the difference among patients and also track its temporal change over time. Moreover, studies should also give more emphasis on the types of inputs used and their associated time lag. Generally, we foresee that these developments might encourage researchers to further develop and test these systems on a large-scale basis.

One way to interject knowledge into clinically impactful forecasting is to use data assimilation, a nonlinear regression that projects data onto a mechanistic physiologic model, instead of a set of functions, such as neural networks. Such regressions have an advantage of being useful with particularly sparse, non-stationary clinical data. However, physiological models are often nonlinear and can have many parameters, leading to potential problems with parameter identifiability, or the ability to find a unique set of parameters that minimize forecasting error. The identifiability problems can be minimized or eliminated by reducing the number of parameters estimated, but reducing the number of estimated parameters also reduces the flexibility of the model and hence increases forecasting error. We propose a method, the parameter Houlihan, that combines traditional machine learning techniques with data assimilation, to select the right set of model parameters to minimize forecasting error while reducing identifiability problems. The method worked well: the data assimilation-based glucose forecasts and estimates for our cohort using the Houlihan-selected parameter sets generally also minimize forecasting errors compared to other parameter selection methods such as by-hand parameter selection. Nevertheless, the forecast with the lowest forecast error does not always accurately represent physiology, but further advancements of the algorithm provide a path for improving physiologic fidelity as well. Our hope is that this methodology represents a first step toward combining machine learning with data assimilation and provides a lower-threshold entry point for using data assimilation with clinical data by helping select the right parameters to estimate.

Ensemble Kalman methods constitute an increasingly important tool in both state and parameter estimation problems. Their popularity stems from the derivative-free nature of the methodology which may be readily applied when computer code is available for the underlying state-space dynamics (for state estimation) or for the parameter-to-observable map (for parameter estimation). There are many applications in which it is desirable to enforce prior information in the form of equality or inequality constraints on the state or parameter. This paper establishes a general framework for doing so, describing a widely applicable methodology, a theory which justifies the methodology, and a set of numerical experiments exemplifying it.

Objective To investigate how individuals with diabetes and diabetes educators reason about data collected through self-monitoring and to draw implications for the design of data-driven self-management technologies. Materials and Methods Ten individuals with diabetes (six type 1 and four type 2) and 2 experienced diabetes educators were presented with a set of self-monitoring data captured by an individual with type 2 diabetes. The set included digital images of meals and their textual descriptions, and blood glucose (BG) readings captured before and after these meals. The participants were asked to review a set of meals and associated BG readings, explain differences in postprandial BG levels for these meals, and predict postprandial BG levels for the same individual for a different set of meals. Researchers compared conclusions and predictions reached by the participants with those arrived at by quantitative analysis of the collected data. Results The participants used both macronutrient composition of meals, most notably the inclusion of carbohydrates, and names of dishes and ingredients to reason about changes in postprandial BG levels. Both individuals with diabetes and diabetes educators reported difficulties in generating predictions of postprandial BG; their predictions varied in their correlations with the actual captured readings from r = 0.008 to r = 0.75. Conclusion Overall, the study showed that identifying trends in the data collected with self-monitoring is a complex process, and that conclusions reached by both individuals with diabetes and diabetes educators are not always reliable. This suggests the need for new ways to facilitate individuals’ reasoning with informatics interventions.

Introduction Self-monitoring technologies produce patient-generated data that could be leveraged to personalize nutritional goal setting to improve population health; however, most computational approaches are limited when applied to individual-level personalization with sparse and irregular self-monitoring data. We applied informatics methods from expert suggestion systems to a challenging clinical problem: generating personalized nutrition goals from patient-generated diet and blood glucose data. Materials and methods We applied qualitative process coding and decision tree modeling to understand how registered dietitians translate patient-generated data into recommendations for dietary self-management of diabetes (i.e., knowledge model). We encoded this process in a set of functions that take diet and blood glucose data as an input and output diet recommendations (i.e., inference engine). Dietitians assessed face validity. Using four patient datasets, we compared our inference engine’s output to clinical narratives and gold standards developed by expert clinicians. Results To dietitians, the knowledge model represented how recommendations from patient data are made. Inference engine recommendations were 63 % consistent with the gold standard (range = 42 %–75 %) and 74 % consistent with narrative clinical observations (range = 63 %–83 %). Discussion Qualitative modeling and automating how dietitians reason over patient data resulted in a knowledge model representing clinical knowledge. However, our knowledge model was less consistent with gold standard than narrative clinical recommendations, raising questions about how best to evaluate approaches that integrate patient-generated data with expert knowledge. Conclusion New informatics approaches that integrate data-driven methods with expert decision making for personalized goal setting, such as the knowledge base and inference engine presented here, demonstrate the potential to extend the reach of patient-generated data by synthesizing it with clinical knowledge. However, important questions remain about the strengths and weaknesses of computer algorithms developed to discern signal from patient-generated data compared to human experts.

Mortality associated with the acute respiratory distress syndrome remains unacceptably high due in part to ventilator-induced lung injury (VILI). Ventilator dyssynchrony is defined as the inappropriate timing and delivery of a mechanical breath in response to patient effort and may cause VILI. Such deleterious patient–ventilator interactions have recently been termed patient self-inflicted lung injury. This narrative review outlines the detection and frequency of several different types of ventilator dyssynchrony, delineates the different mechanisms by which ventilator dyssynchrony may propagate VILI, and reviews the potential clinical impact of ventilator dyssynchrony. Until recently, identifying ventilator dyssynchrony required the manual interpretation of ventilator pressure and flow waveforms. However, computerized interpretation of ventilator waive forms can detect ventilator dyssynchrony with an area under the receiver operating curve of >0.80. Using such algorithms, ventilator dyssynchrony occurs in 3%–34% of all breaths, depending on the patient population. Moreover, two types of ventilator dyssynchrony, double-triggered and flow-limited breaths, are associated with the more frequent delivery of large tidal volumes >10 mL/kg when compared with synchronous breaths (54% [95% confidence interval (CI), 47%–61%] and 11% [95% CI, 7%–15%]) compared with 0.9% (95% CI, 0.0%–1.9%), suggesting a role in propagating VILI. Finally, a recent study associated frequent dyssynchrony-defined as >10% of all breaths-with an increase in hospital mortality (67 vs. 23%, P = 0.04). However, the clinical significance of ventilator dyssynchrony remains an area of active investigation and more research is needed to guide optimal ventilator dyssynchrony management.

Type 2 Diabetes Mellitus (T2DM) is a common chronic condition that requires management of one's lifestyle, including nutrition. Critically, patients often lack a clear understanding of how everyday meals impact their blood glucose. New predictive analytics approaches can provide personalized mealtime blood glucose forecasts. While communicating forecasts can be challenging, effective strategies for doing so remain little explored. In this study, we conducted focus groups with 13 participants to identify approaches to visualizing personalized blood glucose forecasts that can promote diabetes self-management and understand key styles and visual features that resonate with individuals with diabetes. Focus groups demonstrated that individuals rely on simple heuristics and tend to take a reactive approach to their health and nutrition management. Further, the study highlighted the need for simple and explicit, yet information-rich design. Effective visualizations were found to utilize common metaphors alongside words, numbers, and colors to convey a sense of authority and encourage action and learning.