SOFA score performs worse than age for predicting mortality in patients with COVID-19

The use of the Sequential Organ Failure Assessment (SOFA) score, originally developed to describe disease morbidity, is commonly used to predict in-hospital mortality. During the COVID-19 pandemic, many protocols for crisis standards of care used the SOFA score to select patients to be deprioritized due to a low likelihood of survival. A prior study found that age outperformed the SOFA score for mortality prediction in patients with COVID-19, but was limited to a small cohort of intensive care unit (ICU) patients and did not address whether their findings were unique to patients with COVID-19. Moreover, it is not known how well these measures perform across races. In this retrospective study, we compare the performance of age and SOFA score in predicting in-hospital mortality across two cohorts: a cohort of 2,648 consecutive adult patients diagnosed with COVID-19 who were admitted to a large academic health system in the northeastern United States over a 4-month period in 2020 and a cohort of 75,601 patients admitted to one of 335 ICUs in the eICU database between 2014 and 2015. We used age and the maximum SOFA score as predictor variables in separate univariate logistic regression models for in-hospital mortality and calculated area under the receiver operator characteristic curves (AU-ROCs) and area under precision-recall curves (AU-PRCs) for each predictor in both cohorts. Among the COVID-19 cohort, age (AU-ROC 0.795, 95% CI 0.762, 0.828) had a significantly better discrimination than SOFA score (AU-ROC 0.679, 95% CI 0.638, 0.721) for mortality prediction. Conversely, age (AU-ROC 0.628 95% CI 0.608, 0.628) underperformed compared to SOFA score (AU-ROC 0.735, 95% CI 0.726, 0.745) in non-COVID-19 ICU patients in the eICU database. There was no difference between Black and White COVID-19 patients in performance of either age or SOFA Score. Our findings bring into question the utility of SOFA score-based resource allocation in COVID-19 crisis standards of care.

Biometric contrastive learning for data-efficient deep learning from electrocardiographic images

OBJECTIVE

Artificial intelligence (AI) detects heart disease from images of electrocardiograms (ECGs). However, traditional supervised learning is limited by the need for large amounts of labeled data. We report the development of Biometric Contrastive Learning (BCL), a self-supervised pretraining approach for label-efficient deep learning on ECG images.

MATERIALS AND METHODS

Using pairs of ECGs from 78 288 individuals from Yale (2000-2015), we trained a convolutional neural network to identify temporally separated ECG pairs that varied in layouts from the same patient. We fine-tuned BCL-pretrained models to detect atrial fibrillation (AF), gender, and LVEF < 40%, using ECGs from 2015 to 2021. We externally tested the models in cohorts from Germany and the United States. We compared BCL with ImageNet initialization and general-purpose self-supervised contrastive learning for images (simCLR).

RESULTS

While with 100% labeled training data, BCL performed similarly to other approaches for detecting AF/Gender/LVEF < 40% with an AUROC of 0.98/0.90/0.90 in the held-out test sets, it consistently outperformed other methods with smaller proportions of labeled data, reaching equivalent performance at 50% of data. With 0.1% data, BCL achieved AUROC of 0.88/0.79/0.75, compared with 0.51/0.52/0.60 (ImageNet) and 0.61/0.53/0.49 (simCLR). In external validation, BCL outperformed other methods even at 100% labeled training data, with an AUROC of 0.88/0.88 for Gender and LVEF < 40% compared with 0.83/0.83 (ImageNet) and 0.84/0.83 (simCLR).

DISCUSSION AND CONCLUSION

A pretraining strategy that leverages biometric signatures of different ECGs from the same patient enhances the efficiency of developing AI models for ECG images. This represents a major advance in detecting disorders from ECG images with limited labeled data.

Predicting Major Adverse Events in Patients Undergoing Transcatheter Left Atrial Appendage Occlusion

BACKGROUND

The National Cardiovascular Data Registry Left Atrial Appendage Occlusion Registry (LAAO) includes the vast majority of transcatheter LAAO procedures performed in the United States. The objective of this study was to develop a model predicting adverse events among patients undergoing LAAO with Watchman FLX.

METHODS

Data from 41 001 LAAO procedures with Watchman FLX from July 2020 to September 2021 were used to develop and validate a model predicting in-hospital major adverse events. Randomly selected development (70%, n=28 530) and validation (30%, n=12 471) cohorts were analyzed with 1000 bootstrapped samples, using forward stepwise logistic regression to create the final model. A simplified bedside risk score was also developed using this model.

RESULTS

Increased age, female sex, low preprocedure hemoglobin, no prior attempt at atrial fibrillation termination, and increased fall risk most strongly predicted in-hospital major adverse events and were included in the final model along with other clinically relevant variables. The median in-hospital risk-standardized adverse event rate was 1.50% (range, 1.03%-2.84%; interquartile range, 1.42%-1.64%). The model demonstrated moderate discrimination (development C-index, 0.67 [95% CI, 0.65-0.70] and validation C-index, 0.66 [95% CI, 0.62-0.70]) with good calibration. The simplified risk score was well calibrated with risk of in-hospital major adverse events ranging from 0.26% to 3.90% for a score of 0 to 8, respectively.

CONCLUSIONS

A transcatheter LAAO risk model using National Cardiovascular Data Registry and LAAO Registry data can predict in-hospital major adverse events, demonstrated consistency across hospitals and can be used for quality improvement efforts. A simple bedside risk score was similarly predictive and may inform shared decision-making.

Automated Diagnostic Reports from Images of Electrocardiograms at the Point-of-Care

Timely and accurate assessment of electrocardiograms (ECGs) is crucial for diagnosing, triaging, and clinically managing patients. Current workflows rely on a computerized ECG interpretation using rule-based tools built into the ECG signal acquisition systems with limited accuracy and flexibility. In low-resource settings, specialists must review every single ECG for such decisions, as these computerized interpretations are not available. Additionally, high-quality interpretations are even more essential in such low-resource settings as there is a higher burden of accuracy for automated reads when access to experts is limited. Artificial Intelligence (AI)-based systems have the prospect of greater accuracy yet are frequently limited to a narrow range of conditions and do not replicate the full diagnostic range. Moreover, these models often require raw signal data, which are unavailable to physicians and necessitate costly technical integrations that are currently limited. To overcome these challenges, we developed and validated a format-independent vision encoder-decoder model - ECG-GPT - that can generate free-text, expert-level diagnosis statements directly from ECG images. The model shows robust performance, validated on 2.6 million ECGs across 6 geographically distinct health settings: (1) 2 large and diverse US health systems- Yale-New Haven and Mount Sinai Health Systems, (2) a consecutive ECG dataset from a central ECG repository from Minas Gerais, Brazil, (3) the prospective cohort study, UK Biobank, (4) a Germany-based, publicly available repository, PTB-XL, and (5) a community hospital in Missouri. The model demonstrated consistently high performance (AUROC≥0.81) across a wide range of rhythm and conduction disorders. This can be easily accessed via a web-based application capable of receiving ECG images and represents a scalable and accessible strategy for generating accurate, expert-level reports from images of ECGs, enabling accurate triage of patients globally, especially in low-resource settings.

Severe aortic stenosis detection by deep learning applied to echocardiography

BACKGROUND AND AIMS

Early diagnosis of aortic stenosis (AS) is critical to prevent morbidity and mortality but requires skilled examination with Doppler imaging. This study reports the development and validation of a novel deep learning model that relies on two-dimensional (2D) parasternal long axis videos from transthoracic echocardiography without Doppler imaging to identify severe AS, suitable for point-of-care ultrasonography.

METHODS AND RESULTS

In a training set of 5257 studies (17 570 videos) from 2016 to 2020 [Yale-New Haven Hospital (YNHH), Connecticut], an ensemble of three-dimensional convolutional neural networks was developed to detect severe AS, leveraging self-supervised contrastive pretraining for label-efficient model development. This deep learning model was validated in a temporally distinct set of 2040 consecutive studies from 2021 from YNHH as well as two geographically distinct cohorts of 4226 and 3072 studies, from California and other hospitals in New England, respectively. The deep learning model achieved an area under the receiver operating characteristic curve (AUROC) of 0.978 (95% CI: 0.966, 0.988) for detecting severe AS in the temporally distinct test set, maintaining its diagnostic performance in geographically distinct cohorts [0.952 AUROC (95% CI: 0.941, 0.963) in California and 0.942 AUROC (95% CI: 0.909, 0.966) in New England]. The model was interpretable with saliency maps identifying the aortic valve, mitral annulus, and left atrium as the predictive regions. Among non-severe AS cases, predicted probabilities were associated with worse quantitative metrics of AS suggesting an association with various stages of AS severity.

CONCLUSION

This study developed and externally validated an automated approach for severe AS detection using single-view 2D echocardiography, with potential utility for point-of-care screening.

Clinical Phenotyping with an Outcomes-driven Mixture of Experts for Patient Matching and Risk Estimation

Observational medical data present unique opportunities for analysis of medical outcomes and treatment decision making. However, because these datasets do not contain the strict pairing of randomized control trials, matching techniques are to draw comparisons among patients. A key limitation to such techniques is verification that the variables used to model treatment decision making are also relevant in identifying the risk of major adverse events. This article explores a deep mixture of experts approach to jointly learn how to match patients and model the risk of major adverse events in patients. Although trained with information regarding treatment and outcomes, after training, the proposed model is decomposable into a network that clusters patients into phenotypes from information available before treatment. This model is validated on a dataset of patients with acute myocardial infarction complicated by cardiogenic shock. The mixture of experts approach can predict the outcome of mortality with an area under the receiver operating characteristic curve of 0.85 ± 0.01 while jointly discovering five potential phenotypes of interest. The technique and interpretation allow for identifying clinically relevant phenotypes that may be used both for outcomes modeling as well as potentially evaluating individualized treatment effects.

Variational Autoencoders for Biomedical Signal Morphology Clustering and Noise Detection

Accurate estimation of physiological biomarkers using raw waveform data from non-invasive wearable devices requires extensive data preprocessing. An automatic noise detection method in time-series data would offer significant utility for various domains. As data labeling is onerous, having a minimally supervised abnormality detection method for input data, as well as an estimation of the severity of the signal corruptness, is essential. We propose a model-free, time-series biomedical waveform noise detection framework using a Variational Autoencoder coupled with Gaussian Mixture Models, which can detect a range of waveform abnormalities without annotation, providing a confidence metric for each segment. Our technique operates on biomedical signals that exhibit periodicity of heart activities. This framework can be applied to any machine learning or deep learning model as an initial signal validator component. Moreover, the confidence score generated by the proposed framework can be incorporated into different models' optimization to construct confidence-aware modeling. We conduct experiments using dynamic time warping (DTW) distance of segments to validated cardiac cycle morphology. The result confirms that our approach removes noisy cardiac cycles and the remaining signals, classified as clean, exhibit a 59.92% reduction in the standard deviation of DTW distances. Using a dataset of bio-impedance data of 97885 cardiac cycles, we further demonstrate a significant improvement in the downstream task of cuffless blood pressure estimation, with an average reduction of 2.67 mmHg root mean square error (RMSE) of Diastolic Blood pressure and 2.13 mmHg RMSE of systolic blood pressure, with increases of average Pearson correlation of 0.28 and 0.08, with a statistically significant improvement of signal-to-noise ratio respectively in the presence of different synthetic noise sources. This enables burden-free validation of wearable sensor data for downstream biomedical applications.

Biometric Contrastive Learning for Data-Efficient Deep Learning from Electrocardiographic Images

Objective

Artificial intelligence (AI) detects heart disease from images of electrocardiograms (ECGs), however traditional supervised learning is limited by the need for large amounts of labeled data. We report the development of Biometric Contrastive Learning (BCL), a self-supervised pretraining approach for label-efficient deep learning on ECG images.

Materials and Methods

Using pairs of ECGs from 78,288 individuals from Yale (2000-2015), we trained a convolutional neural network to identify temporally-separated ECG pairs that varied in layouts from the same patient. We fine-tuned BCL-pretrained models to detect atrial fibrillation (AF), gender, and LVEF<40%, using ECGs from 2015-2021. We externally tested the models in cohorts from Germany and the US. We compared BCL with random initialization and general-purpose self-supervised contrastive learning for images (simCLR).

Results

While with 100% labeled training data, BCL performed similarly to other approaches for detecting AF/Gender/LVEF<40% with AUROC of 0.98/0.90/0.90 in the held-out test sets, it consistently outperformed other methods with smaller proportions of labeled data, reaching equivalent performance at 50% of data. With 0.1% data, BCL achieved AUROC of 0.88/0.79/0.75, compared with 0.51/0.52/0.60 (random) and 0.61/0.53/0.49 (simCLR). In external validation, BCL outperformed other methods even at 100% labeled training data, with AUROC of 0.88/0.88 for Gender and LVEF<40% compared with 0.83/0.83 (random) and 0.84/0.83 (simCLR).

Discussion and Conclusion

A pretraining strategy that leverages biometric signatures of different ECGs from the same patient enhances the efficiency of developing AI models for ECG images. This represents a major advance in detecting disorders from ECG images with limited labeled data.

Hypothesis Scoring for Confidence-Aware Blood Pressure Estimation With Particle Filters

We propose our Confidence-Aware Particle Filter (CAPF) framework that analyzes a series of estimated changes in blood pressure (BP) to provide several true state hypotheses for a given instance. Particularly, our novel confidence-awareness mechanism assigns likelihood scores to each hypothesis in an effort to discard potentially erroneous measurements - based on the agreement amongst a series of estimated changes and the physiological plausibility when considering DBP/SBP pairs. The particle filter formulation (or sequential Monte Carlo method) can jointly consider the hypotheses and their probabilities over time to provide a stable trend of estimated BP measurements. In this study, we evaluate BP trend estimation from an emerging bio-impedance (Bio-Z) prototype wearable modality although it is applicable to all types of physiological modalities. Each subject in the evaluation cohort underwent a hand-gripper exercise, a cold pressor test, and a recovery state to increase the variation to the captured BP ranges. Experiments show that CAPF yields superior continuous pulse pressure (PP), diastolic blood pressure (DBP), and systolic blood pressure (SBP) estimation performance compared to ten baseline approaches. Furthermore, CAPF performs on track to comply with AAMI and BHS standards for achieving a performance classification of Grade A, with mean error accuracies of -0.16 ± 3.75 mmHg for PP (r = 0.81), 0.42 ± 4.39 mmHg for DBP (r = 0.92), and -0.09 ± 6.51 mmHg for SBP (r = 0.92) from more than test 3500 data points.

Detection of Left Ventricular Systolic Dysfunction From Electrocardiographic Images

BACKGROUND

Left ventricular (LV) systolic dysfunction is associated with a >8-fold increased risk of heart failure and a 2-fold risk of premature death. The use of ECG signals in screening for LV systolic dysfunction is limited by their availability to clinicians. We developed a novel deep learning-based approach that can use ECG images for the screening of LV systolic dysfunction.

METHODS

Using 12-lead ECGs plotted in multiple different formats, and corresponding echocardiographic data recorded within 15 days from the Yale New Haven Hospital between 2015 and 2021, we developed a convolutional neural network algorithm to detect an LV ejection fraction <40%. The model was validated within clinical settings at Yale New Haven Hospital and externally on ECG images from Cedars Sinai Medical Center in Los Angeles, CA; Lake Regional Hospital in Osage Beach, MO; Memorial Hermann Southeast Hospital in Houston, TX; and Methodist Cardiology Clinic of San Antonio, TX. In addition, it was validated in the prospective Brazilian Longitudinal Study of Adult Health. Gradient-weighted class activation mapping was used to localize class-discriminating signals on ECG images.

RESULTS

Overall, 385 601 ECGs with paired echocardiograms were used for model development. The model demonstrated high discrimination across various ECG image formats and calibrations in internal validation (area under receiving operation characteristics [AUROCs], 0.91; area under precision-recall curve [AUPRC], 0.55); and external sets of ECG images from Cedars Sinai (AUROC, 0.90 and AUPRC, 0.53), outpatient Yale New Haven Hospital clinics (AUROC, 0.94 and AUPRC, 0.77), Lake Regional Hospital (AUROC, 0.90 and AUPRC, 0.88), Memorial Hermann Southeast Hospital (AUROC, 0.91 and AUPRC 0.88), Methodist Cardiology Clinic (AUROC, 0.90 and AUPRC, 0.74), and Brazilian Longitudinal Study of Adult Health cohort (AUROC, 0.95 and AUPRC, 0.45). An ECG suggestive of LV systolic dysfunction portended >27-fold higher odds of LV systolic dysfunction on transthoracic echocardiogram (odds ratio, 27.5 [95% CI, 22.3-33.9] in the held-out set). Class-discriminative patterns localized to the anterior and anteroseptal leads (V2 and V3), corresponding to the left ventricle regardless of the ECG layout. A positive ECG screen in individuals with an LV ejection fraction ≥40% at the time of initial assessment was associated with a 3.9-fold increased risk of developing incident LV systolic dysfunction in the future (hazard ratio, 3.9 [95% CI, 3.3-4.7]; median follow-up, 3.2 years).

CONCLUSIONS

We developed and externally validated a deep learning model that identifies LV systolic dysfunction from ECG images. This approach represents an automated and accessible screening strategy for LV systolic dysfunction, particularly in low-resource settings.

Detection of left ventricular systolic dysfunction from single-lead electrocardiography adapted for portable and wearable devices

Artificial intelligence (AI) can detect left ventricular systolic dysfunction (LVSD) from electrocardiograms (ECGs). Wearable devices could allow for broad AI-based screening but frequently obtain noisy ECGs. We report a novel strategy that automates the detection of hidden cardiovascular diseases, such as LVSD, adapted for noisy single-lead ECGs obtained on wearable and portable devices. We use 385,601 ECGs for development of a standard and noise-adapted model. For the noise-adapted model, ECGs are augmented during training with random gaussian noise within four distinct frequency ranges, each emulating real-world noise sources. Both models perform comparably on standard ECGs with an AUROC of 0.90. The noise-adapted model performs significantly better on the same test set augmented with four distinct real-world noise recordings at multiple signal-to-noise ratios (SNRs), including noise isolated from a portable device ECG. The standard and noise-adapted models have an AUROC of 0.72 and 0.87, respectively, when evaluated on ECGs augmented with portable ECG device noise at an SNR of 0.5. This approach represents a novel strategy for the development of wearable-adapted tools from clinical ECG repositories.

Clinical Risk Prediction Models with Meta-Learning Prototypes of Patient Heterogeneity

Hospitalized patients sometimes have complex health conditions, such as multiple diseases, underlying diseases, and complications. The heterogeneous patient conditions may have various representations. A generalized model ignores the differences among heterogeneous patients, and personalized models, even with transfer learning, are still limited to the small amount of training data and the repeated training process. Meta-learning provides a solution for training similar patients based on few-shot learning; however, cannot address common cross-domain patients. Inspired by prototypical networks [1], we proposed a meta-prototype for Electronic Health Records (EHR), a meta-learning-based model with flexible prototypes representing the heterogeneity in patients. We apply this technique to cardiovascular diseases in MIMIC-III and compare it against a set of benchmark models, and demonstrate its ability to address heterogeneous patient health conditions and improve the model performances from 1.2% to 11.9% on different metrics and prediction tasks.Clinical relevance- Developing an adaptive EHR risk prediction model for outcomes-driven phenotyping of heterogeneous patient health conditions.

Classification of blood pressure during sleep impacts designation of nocturnal nondipping

The identification of nocturnal nondipping blood pressure (< 10% drop in mean systolic blood pressure from awake to sleep periods), as captured by ambulatory blood pressure monitoring, is a valuable element of risk prediction for cardiovascular disease, independent of daytime or clinic blood pressure measurements. However, capturing measurements, including determination of wake/sleep periods, is challenging. Accordingly, we sought to evaluate the impact of different definitions and algorithms for defining sleep onset on the classification of nocturnal nondipping. Using approaches based upon participant self-reports, applied definition of a common sleep period (12 am -6 am), manual actigraphy, and automated actigraphy we identified changes to the classification of nocturnal nondipping, and conducted a secondary analysis on the potential impact of an ambulatory blood pressure monitor on sleep. Among 61 participants in the Eastern Caribbean Health Outcomes Research Network hypertension study with complete ambulatory blood pressure monitor and sleep data, the concordance for nocturnal nondipping across methods was 0.54 by Fleiss' Kappa (depending on the method, 36 to 51 participants classified as having nocturnal nondipping). Sleep quality for participants with dipping versus nondipping was significantly different for total sleep length when wearing the ambulatory blood pressure monitor (shorter sleep duration) versus not (longer sleep duration), although there were no differences in sleep efficiency or disturbances. These findings indicate that consideration of sleep time measurements is critical for interpreting ambulatory blood pressure. As technology advances to detect blood pressure and sleep patterns, further investigation is needed to determine which method should be used for diagnosis, treatment, and future cardiovascular risk.

Automated Diet Capture Using Voice Alerts and Speech Recognition on Smartphones: Pilot Usability and Acceptability Study

BACKGROUND

Effective monitoring of dietary habits is critical for promoting healthy lifestyles and preventing or delaying the onset and progression of diet-related diseases, such as type 2 diabetes. Recent advances in speech recognition technologies and natural language processing present new possibilities for automated diet capture; however, further exploration is necessary to assess the usability and acceptability of such technologies for diet logging.

OBJECTIVE

This study explores the usability and acceptability of speech recognition technologies and natural language processing for automated diet logging.

METHODS

We designed and developed base2Diet-an iOS smartphone application that prompts users to log their food intake using voice or text. To compare the effectiveness of the 2 diet logging modes, we conducted a 28-day pilot study with 2 arms and 2 phases. A total of 18 participants were included in the study, with 9 participants in each arm (text: n=9, voice: n=9). During phase I of the study, all 18 participants received reminders for breakfast, lunch, and dinner at preselected times. At the beginning of phase II, all participants were given the option to choose 3 times during the day to receive 3 times daily reminders to log their food intake for the remainder of the phase, with the ability to modify the selected times at any point before the end of the study.

RESULTS

The total number of distinct diet logging events per participant was 1.7 times higher in the voice arm than in the text arm (P=.03, unpaired t test). Similarly, the total number of active days per participant was 1.5 times higher in the voice arm than in the text arm (P=.04, unpaired t test). Furthermore, the text arm had a higher attrition rate than the voice arm, with only 1 participant dropping out of the study in the voice arm, while 5 participants dropped out in the text arm.

CONCLUSIONS

The results of this pilot study demonstrate the potential of voice technologies in automated diet capturing using smartphones. Our findings suggest that voice-based diet logging is more effective and better received by users compared to traditional text-based methods, underscoring the need for further research in this area. These insights carry significant implications for the development of more effective and accessible tools for monitoring dietary habits and promoting healthy lifestyle choices.

Nonexercise machine learning models for maximal oxygen uptake prediction in national population surveys

OBJECTIVE

Nonexercise algorithms are cost-effective methods to estimate cardiorespiratory fitness (CRF), but the existing models have limitations in generalizability and predictive power. This study aims to improve the nonexercise algorithms using machine learning (ML) methods and data from US national population surveys.

MATERIALS AND METHODS

We used the 1999-2004 data from the National Health and Nutrition Examination Survey (NHANES). Maximal oxygen uptake (VO2 max), measured through a submaximal exercise test, served as the gold standard measure for CRF in this study. We applied multiple ML algorithms to build 2 models: a parsimonious model using commonly available interview and examination data, and an extended model additionally incorporating variables from Dual-Energy X-ray Absorptiometry (DEXA) and standard laboratory tests in clinical practice. Key predictors were identified using Shapley additive explanation (SHAP).

RESULTS

Among the 5668 NHANES participants in the study population, 49.9% were women and the mean (SD) age was 32.5 years (10.0). The light gradient boosting machine (LightGBM) had the best performance across multiple types of supervised ML algorithms. Compared with the best existing nonexercise algorithms that could be applied to the NHANES, the parsimonious LightGBM model (RMSE: 8.51 ml/kg/min [95% CI: 7.73-9.33]) and the extended LightGBM model (RMSE: 8.26 ml/kg/min [95% CI: 7.44-9.09]) significantly reduced the error by 15% and 12% (P < .001 for both), respectively.

DISCUSSION

The integration of ML and national data source presents a novel approach for estimating cardiovascular fitness. This method provides valuable insights for cardiovascular disease risk classification and clinical decision-making, ultimately leading to improved health outcomes.

CONCLUSION

Our nonexercise models provide improved accuracy in estimating VO2 max within NHANES data as compared to existing nonexercise algorithms.

Quantifying Blood Pressure Visit-to-Visit Variability in the Real-World Setting: A Retrospective Cohort Study

BACKGROUND

Visit-to-visit variability (VVV) in blood pressure values has been reported in clinical studies. However, little is known about VVV in clinical practice and whether it is associated with patient characteristics in real-world setting.

METHODS

We conducted a retrospective cohort study to quantify VVV in systolic blood pressure (SBP) values in a real-world setting. We included adults (age ≥18 years) with at least 2 outpatient visits between January 1, 2014 and October 31, 2018 from Yale New Haven Health System. Patient-level measures of VVV included SD and coefficient of variation of a given patient's SBP across visits. We calculated patient-level VVV overall and by patient subgroups. We further developed a multilevel regression model to assess the extent to which VVV in SBP was explained by patient characteristics.

RESULTS

The study population included 537 218 adults, with a total of 7 721 864 SBP measurements. The mean age was 53.4 (SD 19.0) years, 60.4% were women, 69.4% were non-Hispanic White, and 18.1% were on antihypertensive medications. Patients had a mean body mass index of 28.4 (5.9) kg/m2 and 22.6%, 8.0%, 9.7%, and 5.6% had a history of hypertension, diabetes, hyperlipidemia, and coronary artery disease, respectively. The mean number of visits per patient was 13.3, over an average period of 2.4 years. The mean (SD) intraindividual SD and coefficient of variation of SBP across visits were 10.6 (5.1) mm Hg and 0.08 (0.04). These measures of blood pressure variation were consistent across patient subgroups defined by demographic characteristics and medical history. In the multivariable linear regression model, only 4% of the variance in absolute standardized difference was attributable to patient characteristics.

CONCLUSIONS

The VVV in real-world practice poses challenges for management of patients with hypertension based on blood pressure readings in outpatient settings and suggest the need to go beyond episodic clinic evaluation.

Timing of Blood Draws Among Patients Hospitalized in a Large Academic Medical Center

This study describes the degree to which blood draws occurred among hospitalized patients during traditional sleep hours and investigates trends over time.

A Review of Digital Innovations for Diet Monitoring and Precision Nutrition

This article provides an up-to-date review of technological advances in 3 key areas related to diet monitoring and precision nutrition. First, we review developments in mobile applications, with a focus on food photography and artificial intelligence to facilitate the process of diet monitoring. Second, we review advances in 2 types of wearable and handheld sensors that can potentially be used to fully automate certain aspects of diet logging: physical sensors to detect moments of dietary intake, and chemical sensors to estimate the composition of diets and meals. Finally, we review new programs that can generate personalized/precision nutrition recommendations based on measurements of gut microbiota and continuous glucose monitors with artificial intelligence. The article concludes with a discussion of potential pitfalls of some of these technologies.

Data-Driven Guided Attention for Analysis of Physiological Waveforms with Deep Learning

Estimating physiological parameters - such as blood pressure (BP) - from raw sensor data captured by noninvasive, wearable devices rely on either burdensome manual feature extraction designed by domain experts to identify key waveform characteristics and phases, or deep learning (DL) models that require extensive data collection. We propose the Data-Driven Guided Attention (DDGA) framework to optimize DL models to learn features supported by the underlying physiology and physics of the captured waveforms, with minimal expert annotation. With only a single template waveform cardiac cycle and its labelled fiducial points, we leverage dynamic time warping (DTW) to annotate all other training samples. DL models are trained to first identify them before estimating BP to inform them which regions of the input represent key phases of the cardiac cycle, yet we still grant the flexibility for DL to determine the optimal feature set from them. In this study, we evaluate DDGA's improvements to a BP estimation task for three prominent DL-based architectures with two datasets: 1) the MIMIC-III waveform dataset with ample training data and 2) a bio-impedance (Bio-Z) dataset with less than abundant training data. Experiments show that DDGA improves personalized BP estimation models by an average 8.14% in root mean square error (RMSE) when there is an imbalanced distribution of target values in a training set and improves model generalizability by an average 4.92% in RMSE when testing estimation of BP value ranges not previously seen in training.

Abstract 242: Assessing Variation In Data Captured By Clinical Registries And The Electronic Health Record: Clinical Phenotypes Of Patients Undergoing Cardiac Catheterization

Background: Cardiovascular registries are built through expert abstraction of summary clinical information often represented as binary flags used to define diagnosis status. We considered underlying variation in clinical data obtained from the electronic health record (EHR) with respect to binary flags denoting presence or absence of cardiometabolic disease in a clinical data registry.

Methods: We linked data abstracted for percutaneous coronary intervention (PCI) visits registered in NCDR CathPCI® with EHR data from Yale New Haven Hospital. We identified registry flags indicating diagnoses of hyperlipidemia (HLD), hypertension (HTN) and diabetes mellitus (DM) and evaluated EHR data with respect to registry diagnosis status.

Results: Among 3,473 patients, mean age was 67 ± 12 years and 960 (28%) patients were female. HLD, HTN and DM registry diagnoses were present (+) in 2694 (78%), 2833 (82%) and 1358 (39%) patients, and absent (-) in 779 (22%), 640 (18%) and 2115 (61%) patients. EHR data revealed mean baseline total cholesterol of 168 ± 51 mg/dL (+HLD) vs 182 ± 44 mg/dL (-HLD), presenting systolic blood pressure of 130 ± 19 mmHg (+HTN) vs 120 ± 17 mmHg (-HTN), and A1c of 7.9% ± 1.8% (+DM) vs 5.8% ± 0.7% (-DM). Prescription histories of patients with HLD, HTN or DM diagnoses included 1 or more relevant drug class in 2409 (89%), 2620 (92%) and 1222 (90%) patients, and 2 or more drug classes in 450 (17%), 2020 (71%) and 820 (60%) patients. Patients without these diagnoses had been prescribed 1 or more drug class in 268 (34%), 223 (35%) and 55 (3%) patients, and 2 or more drug classes in 28 (4%), 71 (11%) and 9 (0.4%) patients (Figure 1).

Conclusions: Binary diagnosis flags used in cardiovascular registries do not capture the full variation of clinical phenotypes underlying cardiometabolic disease processes. Harnessing comprehensive EHR data may provide a fuller picture of cardiometabolic disease, enhance quality and utility of national registries and improve outcomes research.

Establishing a Global Standard for Wearable Devices in Sport and Exercise Medicine: Perspectives from Academic and Industry Stakeholders

Millions of consumer sport and fitness wearables (CSFWs) are used worldwide, and millions of datapoints are generated by each device. Moreover, these numbers are rapidly growing, and they contain a heterogeneity of devices, data types, and contexts for data collection. Companies and consumers would benefit from guiding standards on device quality and data formats. To address this growing need, we convened a virtual panel of industry and academic stakeholders, and this manuscript summarizes the outcomes of the discussion. Our objectives were to identify (1) key facilitators of and barriers to participation by CSFW manufacturers in guiding standards and (2) stakeholder priorities. The venues were the Yale Center for Biomedical Data Science Digital Health Monthly Seminar Series (62 participants) and the New England Chapter of the American College of Sports Medicine Annual Meeting (59 participants). In the discussion, stakeholders outlined both facilitators of (e.g., commercial return on investment in device quality, lucrative research partnerships, and transparent and multilevel evaluation of device quality) and barriers (e.g., competitive advantage conflict, lack of flexibility in previously developed devices) to participation in guiding standards. There was general agreement to adopt Keadle et al.'s standard pathway for testing devices (i.e., benchtop, laboratory, field-based, implementation) without consensus on the prioritization of these steps. Overall, there was enthusiasm not to add prescriptive or regulatory steps, but instead create a networking hub that connects companies to consumers and researchers for flexible guidance navigating the heterogeneity, multi-tiered development, dynamicity, and nebulousness of the CSFW field.

Performance Metrics for the Comparative Analysis of Clinical Risk Prediction Models Employing Machine Learning

BACKGROUND

New methods such as machine learning techniques have been increasingly used to enhance the performance of risk predictions for clinical decision-making. However, commonly reported performance metrics may not be sufficient to capture the advantages of these newly proposed models for their adoption by health care professionals to improve care. Machine learning models often improve risk estimation for certain subpopulations that may be missed by these metrics.

METHODS AND RESULTS

This article addresses the limitations of commonly reported metrics for performance comparison and proposes additional metrics. Our discussions cover metrics related to overall performance, discrimination, calibration, resolution, reclassification, and model implementation. Models for predicting acute kidney injury after percutaneous coronary intervention are used to illustrate the use of these metrics.

CONCLUSIONS

We demonstrate that commonly reported metrics may not have sufficient sensitivity to identify improvement of machine learning models and propose the use of a comprehensive list of performance metrics for reporting and comparing clinical risk prediction models.

Postprandial concentration of circulating branched chain amino acids are able to predict the carbohydrate content of the ingested mixed meal

BACKGROUND

The amount of the macronutrients protein and carbohydrate (CHO) in a mixed meal is known to affect each other's digestion, absorption, and subsequent metabolism. While the effect of the amount of dietary protein and fat on the glycemic response is well studied, the ability of postprandial plasma amino acid patterns to predict the meal composition is unknown.

OBJECTIVE

To study the postprandial plasma amino acid patterns in relation to the protein, CHO, and fat content of different mixed meals and to investigate if these patterns can predict the macronutrient meal composition.

DESIGN

Ten older adults were given 9 meals with 3 different levels (low, medium, and high) of protein, CHO, and fat in different combinations, taking the medium content as that of a standardized western meal. We monitored the postprandial plasma response for amino acids, glucose, insulin, and triglycerides for 8 h and the areas under the curve (AUC) were subsequently calculated. Multiple regression analysis was performed to determine if amino acid patterns could predict the meal composition.

RESULTS

Increasing meal CHO content reduced the postprandial plasma response of several amino acids including all branched chain amino acids (BCAA) (leucine; q < 0.0001, isoleucine; q = 0.0035, valine; q = 0.0022). The plasma BCAA patterns after the meal significantly predicted the meal's CHO content (leucine; p < 0.0001, isoleucine; p = 0.0003, valine; p = 0.0008) along with aspartate (p < 0.0001), tyrosine (p < 0.0001), methionine (p = 0.0159) and phenylalanine (p = 0.0332). Plasma citrulline predicted best the fat content of the meal (p = 0.0024).

CONCLUSIONS

The postprandial plasma BCAA patterns are lower with increasing meal CHO content and are strong predictors of a mixed meal protein and CHO composition, as are plasma citrulline for the fat content. We hypothesize that postprandial plasma amino acid concentrations can be used to predict the meal's macronutrient composition.

Toward Dynamic Risk Prediction of Outcomes After Coronary Artery Bypass Graft: Improving Risk Prediction With Intraoperative Events Using Gradient Boosting

BACKGROUND

Intraoperative data may improve models predicting postoperative events. We evaluated the effect of incorporating intraoperative variables to the existing preoperative model on the predictive performance of the model for coronary artery bypass graft.

METHODS

We analyzed 378 572 isolated coronary artery bypass graft cases performed across 1083 centers, using the national Society of Thoracic Surgeons Adult Cardiac Surgery Database between 2014 and 2016. Outcomes were operative mortality, 5 postoperative complications, and composite representation of all events. We fitted models by logistic regression or extreme gradient boosting (XGBoost). For each modeling approach, we used preoperative only, intraoperative only, or pre+intraoperative variables. We developed 84 models with unique combinations of the 3 variable sets, 2 variable selection methods, 2 modeling approaches, and 7 outcomes. Each model was tested in 20 iterations of 70:30 stratified random splitting into development/testing samples. Model performances were evaluated on the testing dataset using the C statistic, area under the precision-recall curve, and calibration metrics, including the Brier score.

RESULTS

The mean patient age was 65.3 years, and 24.7% were women. Operative mortality, excluding intraoperative death, occurred in 1.9%. In all outcomes, models that considered pre+intraoperative variables demonstrated significantly improved Brier score and area under the precision-recall curve compared with models considering pre or intraoperative variables alone. XGBoost without external variable selection had the best C statistics, Brier score, and area under the precision-recall curve values in 4 of the 7 outcomes (mortality, renal failure, prolonged ventilation, and composite) compared with logistic regression models with or without variable selection. Based on the calibration plots, risk restratification for mortality showed that the logistic regression model underestimated the risk in 11 114 patients (9.8%) and overestimated in 12 005 patients (10.6%). In contrast, the XGBoost model underestimated the risk in 7218 patients (6.4%) and overestimated in 0 patients (0%).

CONCLUSIONS

In isolated coronary artery bypass graft, adding intraoperative variables to preoperative variables resulted in improved predictions of all 7 outcomes. Risk models based on XGBoost may provide a better prediction of adverse events to guide clinical care.

Use of Machine Learning Models to Predict Death After Acute Myocardial Infarction

Importance

Accurate prediction of adverse outcomes after acute myocardial infarction (AMI) can guide the triage of care services and shared decision-making, and novel methods hold promise for using existing data to generate additional insights.

Objective

To evaluate whether contemporary machine learning methods can facilitate risk prediction by including a larger number of variables and identifying complex relationships between predictors and outcomes.

Design, Setting, and Participants

This cohort study used the American College of Cardiology Chest Pain-MI Registry to identify all AMI hospitalizations between January 1, 2011, and December 31, 2016. Data analysis was performed from February 1, 2018, to October 22, 2020.

Main Outcomes and Measures

Three machine learning models were developed and validated to predict in-hospital mortality based on patient comorbidities, medical history, presentation characteristics, and initial laboratory values. Models were developed based on extreme gradient descent boosting (XGBoost, an interpretable model), a neural network, and a meta-classifier model. Their accuracy was compared against the current standard developed using a logistic regression model in a validation sample.

Results

A total of 755 402 patients (mean [SD] age, 65 [13] years; 495 202 [65.5%] male) were identified during the study period. In independent validation, 2 machine learning models, gradient descent boosting and meta-classifier (combination including inputs from gradient descent boosting and a neural network), marginally improved discrimination compared with logistic regression (C statistic, 0.90 for best performing machine learning model vs 0.89 for logistic regression). Nearly perfect calibration in independent validation data was found in the XGBoost (slope of predicted to observed events, 1.01; 95% CI, 0.99-1.04) and the meta-classifier model (slope of predicted-to-observed events, 1.01; 95% CI, 0.99-1.02), with more precise classification across the risk spectrum. The XGBoost model reclassified 32 393 of 121 839 individuals (27%) and the meta-classifier model reclassified 30 836 of 121 839 individuals (25%) deemed at moderate to high risk for death in logistic regression as low risk, which were more consistent with the observed event rates.

Conclusions and Relevance

In this cohort study using a large national registry, none of the tested machine learning models were associated with substantive improvement in the discrimination of in-hospital mortality after AMI, limiting their clinical utility. However, compared with logistic regression, XGBoost and meta-classifier models, but not the neural network, offered improved resolution of risk for high-risk individuals.

Accuracy of Computable Phenotyping Approaches for SARS-CoV-2 Infection and COVID-19 Hospitalizations from the Electronic Health Record

Objective:

Real-world data have been critical for rapid-knowledge generation throughout the COVID-19 pandemic. To ensure high-quality results are delivered to guide clinical decision making and the public health response, as well as characterize the response to interventions, it is essential to establish the accuracy of COVID-19 case definitions derived from administrative data to identify infections and hospitalizations.

Methods:

Electronic Health Record (EHR) data were obtained from the clinical data warehouse of the Yale New Haven Health System (Yale, primary site) and 3 hospital systems of the Mayo Clinic (validation site). Detailed characteristics on demographics, diagnoses, and laboratory results were obtained for all patients with either a positive SARS-CoV-2 PCR or antigen test or ICD-10 diagnosis of COVID-19 (U07.1) between April 1, 2020 and March 1, 2021. Various computable phenotype definitions were evaluated for their accuracy to identify SARS-CoV-2 infection and COVID-19 hospitalizations.

Results:

Of the 69,423 individuals with either a diagnosis code or a laboratory diagnosis of a SARS-CoV-2 infection at Yale, 61,023 had a principal or a secondary diagnosis code for COVID-19 and 50,355 had a positive SARS-CoV-2 test. Among those with a positive laboratory test, 38,506 (76.5%) and 3449 (6.8%) had a principal and secondary diagnosis code of COVID-19, respectively, while 8400 (16.7%) had no COVID-19 diagnosis. Moreover, of the 61,023 patients with a COVID-19 diagnosis code, 19,068 (31.2%) did not have a positive laboratory test for SARS-CoV-2 in the EHR. Of the 20 cases randomly sampled from this latter group for manual review, all had a COVID-19 diagnosis code related to asymptomatic testing with negative subsequent test results. The positive predictive value (precision) and sensitivity (recall) of a COVID-19 diagnosis in the medical record for a documented positive SARS-CoV-2 test were 68.8% and 83.3%, respectively. Among 5,109 patients who were hospitalized with a principal diagnosis of COVID-19, 4843 (94.8%) had a positive SARS-CoV-2 test within the 2 weeks preceding hospital admission or during hospitalization. In addition, 789 hospitalizations had a secondary diagnosis of COVID-19, of which 446 (56.5%) had a principal diagnosis consistent with severe clinical manifestation of COVID-19 (e.g., sepsis or respiratory failure). Compared with the cohort that had a principal diagnosis of COVID-19, those with a secondary diagnosis had a more than 2-fold higher in-hospital mortality rate (13.2% vs 28.0%, P<0.001). In the validation sample at Mayo Clinic, diagnosis codes more consistently identified SARS-CoV-2 infection (precision of 95%) but had lower recall (63.5%) with substantial variation across the 3 Mayo Clinic sites. Similar to Yale, diagnosis codes consistently identified COVID-19 hospitalizations at Mayo, with hospitalizations defined by secondary diagnosis code with 2-fold higher in-hospital mortality compared to those with a primary diagnosis of COVID-19.

Conclusions:

COVID-19 diagnosis codes misclassified the SARS-CoV-2 infection status of many people, with implications for clinical research and epidemiological surveillance. Moreover, the codes had different performance across two academic health systems and identified groups with different risks of mortality. Real-world data from the EHR can be used to in conjunction with diagnosis codes to improve the identification of people infected with SARS-CoV-2.

Clinical characteristics and outcomes for 7,995 patients with SARS-CoV-2 infection

OBJECTIVE

Severe acute respiratory syndrome virus (SARS-CoV-2) has infected millions of people worldwide. Our goal was to identify risk factors associated with admission and disease severity in patients with SARS-CoV-2.

DESIGN

This was an observational, retrospective study based on real-world data for 7,995 patients with SARS-CoV-2 from a clinical data repository.

SETTING

Yale New Haven Health (YNHH) is a five-hospital academic health system serving a diverse patient population with community and teaching facilities in both urban and suburban areas.

POPULATIONS

The study included adult patients who had SARS-CoV-2 testing at YNHH between March 1 and April 30, 2020.

MAIN OUTCOME AND PERFORMANCE MEASURES

Primary outcomes were admission and in-hospital mortality for patients with SARS-CoV-2 infection as determined by RT-PCR testing. We also assessed features associated with the need for respiratory support.

RESULTS

Of the 28605 patients tested for SARS-CoV-2, 7995 patients (27.9%) had an infection (median age 52.3 years) and 2154 (26.9%) of these had an associated admission (median age 66.2 years). Of admitted patients, 2152 (99.9%) had a discharge disposition at the end of the study period. Of these, 329 (15.3%) required invasive mechanical ventilation and 305 (14.2%) expired. Increased age and male sex were positively associated with admission and in-hospital mortality (median age 80.7 years), while comorbidities had a much weaker association with the risk of admission or mortality. Black race (OR 1.43, 95%CI 1.14-1.78) and Hispanic ethnicity (OR 1.81, 95%CI 1.50-2.18) were identified as risk factors for admission, but, among discharged patients, age-adjusted in-hospital mortality was not significantly different among racial and ethnic groups.

CONCLUSIONS

This observational study identified, among people testing positive for SARS-CoV-2 infection, older age and male sex as the most strongly associated risks for admission and in-hospital mortality in patients with SARS-CoV-2 infection. While minority racial and ethnic groups had increased burden of disease and risk of admission, age-adjusted in-hospital mortality for discharged patients was not significantly different among racial and ethnic groups. Ongoing studies will be needed to continue to evaluate these risks, particularly in the setting of evolving treatment guidelines.

Temporal relationship of computed and structured diagnoses in electronic health record data

Background

The electronic health record (EHR) holds the prospect of providing more complete and timely access to clinical information for biomedical research, quality assessments, and quality improvement compared to other data sources, such as administrative claims. In this study, we sought to assess the completeness and timeliness of structured diagnoses in the EHR compared to computed diagnoses for hypertension (HTN), hyperlipidemia (HLD), and diabetes mellitus (DM).

Methods

We determined the amount of time for a structured diagnosis to be recorded in the EHR from when an equivalent diagnosis could be computed from other structured data elements, such as vital signs and laboratory results. We used EHR data for encounters from January 1, 2012 through February 10, 2019 from an academic health system. Diagnoses for HTN, HLD, and DM were computed for patients with at least two observations above threshold separated by at least 30 days, where the thresholds were outpatient blood pressure of ≥ 140/90 mmHg, any low-density lipoprotein ≥ 130 mg/dl, or any hemoglobin A1c ≥ 6.5%, respectively. The primary measure was the length of time between the computed diagnosis and the time at which a structured diagnosis could be identified within the EHR history or problem list.

Results

We found that 39.8% of those with HTN, 21.6% with HLD, and 5.2% with DM did not receive a corresponding structured diagnosis recorded in the EHR. For those who received a structured diagnosis, a mean of 389, 198, and 166 days elapsed before the patient had the corresponding diagnosis of HTN, HLD, or DM, respectively, recorded in the EHR.

Conclusions

We found a marked temporal delay between when a diagnosis can be computed or inferred and when an equivalent structured diagnosis is recorded within the EHR. These findings demonstrate the continued need for additional study of the EHR to avoid bias when using observational data and reinforce the need for computational approaches to identify clinical phenotypes.

Use of Mechanical Circulatory Support Devices Among Patients With Acute Myocardial Infarction Complicated by Cardiogenic Shock

IMPORTANCE

Mechanical circulatory support (MCS) devices, including intravascular microaxial left ventricular assist devices (LVADs) and intra-aortic balloon pumps (IABPs), are used in patients who undergo percutaneous coronary intervention (PCI) for acute myocardial infarction (AMI) complicated by cardiogenic shock despite limited evidence of their clinical benefit.

OBJECTIVE

To examine trends in the use of MCS devices among patients who underwent PCI for AMI with cardiogenic shock, hospital-level use variation, and factors associated with use.

DESIGN, SETTING, AND PARTICIPANTS

This cross-sectional study used the CathPCI and Chest Pain-MI Registries of the American College of Cardiology National Cardiovascular Data Registry. Patients who underwent PCI for AMI complicated by cardiogenic shock between October 1, 2015, and December 31, 2017, were identified from both registries. Data were analyzed from October 2018 to August 2020.

EXPOSURES

Therapies to provide hemodynamic support were categorized as intravascular microaxial LVAD, IABP, TandemHeart, extracorporeal membrane oxygenation, LVAD, other devices, combined IABP and intravascular microaxial LVAD, combined IABP and other device (defined as TandemHeart, extracorporeal membrane oxygenation, LVAD, or another MCS device), or medical therapy only.

MAIN OUTCOMES AND MEASURES

Use of MCS devices overall and specific MCS devices, including intravascular microaxial LVAD, at both patient and hospital levels and variables associated with use.

RESULTS

Among the 28 304 patients included in the study, the mean (SD) age was 65.4 (12.6) years and 18 968 were men (67.0%). The overall MCS device use was constant from the fourth quarter of 2015 to the fourth quarter of 2017, although use of intravascular microaxial LVADs significantly increased (from 4.1% to 9.8%; P < .001), whereas use of IABPs significantly decreased (from 34.8% to 30.0%; P < .001). A significant hospital-level variation in MCS device use was found. The median (interquartile range [IQR]) proportion of patients who received MCS devices was 42% (30%-54%), and the median proportion of patients who received intravascular microaxial LVADs was 1% (0%-10%). In multivariable analyses, cardiac arrest at first medical contact or during hospitalization (odds ratio [OR], 1.82; 95% CI, 1.58-2.09) and severe left main and/or proximal left anterior descending coronary artery stenosis (OR, 1.36; 95% CI, 1.20-1.54) were patient characteristics that were associated with higher odds of receiving intravascular microaxial LVADs only compared with IABPs only.

CONCLUSIONS AND RELEVANCE

This study found that, among patients who underwent PCI for AMI complicated by cardiogenic shock, overall use of MCS devices was constant, and a 2.5-fold increase in intravascular microaxial LVAD use was found along with a corresponding decrease in IABP use and a significant hospital-level variation in MCS device use. These trends were observed despite limited clinical trial evidence of improved outcomes associated with device use.

The Use of Telehealth Technology to Support Health Coaching for Older Adults: Literature Review

BACKGROUND

Health coaching is an intervention process for driving behavior change through goal-setting, education, encouragement, and feedback on health-related behaviors. Telehealth systems that include health coaching and remote monitoring are making inroads in managing chronic conditions and may be especially suited for older populations.

OBJECTIVE

This literature review aimed to investigate the current status of health coaching interventions incorporating telehealth technology and the associated effectiveness of this intervention to deliver health care with an emphasis on older adults (aged 65 and older).

METHODS

A literature review was conducted to identify the research conducted on health coaching combined with remote monitoring for delivering health care to older adults. The Ovid MEDLINE and CINAHL databases were queried using a combination of relevant search terms (including middle aged, aged, older adult, elderly, health coaching, and wellness coaching). The search retrieved 196 papers published from January 2010 to September 2019 in English. Following a systematic review process, the titles and abstracts of the papers retrieved were screened for applicability to health coaching for older adults to define a subset for further review. Papers were excluded if the studied population did not include older adults. The full text of the 42 papers in this subset was then reviewed, and 13 papers related to health coaching combined with remote monitoring for older adults were included in this review.

RESULTS

Of the 13 studies reviewed, 10 found coaching supported by telehealth technology to provide effective outcomes. Effectiveness outcomes assessed in the studies included hospital admissions/re-admissions, mortality, hemoglobin A_1c (HbA_1c) level, body weight, blood pressure, physical activity level, fatigue, quality of life, and user acceptance of the coaching program and technology.

CONCLUSIONS

Telehealth systems that include health coaching have been implemented in older populations as a viable intervention method for managing chronic conditions with mixed results. Health coaching combined with telehealth may be an effective solution for providing health care to older adults. However, health coaching is predominantly performed by human coaches with limited use of technology to augment or replace the human coach. The opportunity exists to expand health coaching to include automated coaching.

A Survey of Challenges and Opportunities in Sensing and Analytics for Risk Factors of Cardiovascular Disorders

Cardiovascular disorders cause nearly one in three deaths in the United States. Short- and long-term care for these disorders is often determined in short-term settings. However, these decisions are made with minimal longitudinal and long-term data. To overcome this bias towards data from acute care settings, improved longitudinal monitoring for cardiovascular patients is needed. Longitudinal monitoring provides a more comprehensive picture of patient health, allowing for informed decision making. This work surveys sensing and machine learning in the field of remote health monitoring for cardiovascular disorders. We highlight three needs in the design of new smart health technologies: (1) need for sensing technologies that track longitudinal trends of the cardiovascular disorder despite infrequent, noisy, or missing data measurements; (2) need for new analytic techniques designed in a longitudinal, continual fashion to aid in the development of new risk prediction techniques and in tracking disease progression; and (3) need for personalized and interpretable machine learning techniques, allowing for advancements in clinical decision making. We highlight these needs based upon the current state of the art in smart health technologies and analytics. We then discuss opportunities in addressing these needs for development of smart health technologies for the field of cardiovascular disorders and care.

Clinical Characteristics and Outcomes for 7,995 Patients with SARS-CoV-2 Infection

OBJECTIVE

Severe acute respiratory syndrome virus (SARS-CoV-2) has infected millions of people worldwide. Our goal was to identify risk factors associated with admission and disease severity in patients with SARS-CoV-2.

DESIGN

This was an observational, retrospective study based on real-world data for 7,995 patients with SARS-CoV-2 from a clinical data repository.

SETTING

Yale New Haven Health (YNHH) is a five-hospital academic health system serving a diverse patient population with community and teaching facilities in both urban and suburban areas.

POPULATIONS

The study included adult patients who had SARS-CoV-2 testing at YNHH between March 1 and April 30, 2020.

MAIN OUTCOME AND PERFORMANCE MEASURES

Primary outcomes were admission and in-hospital mortality for patients with SARS-CoV-2 infection as determined by RT-PCR testing. We also assessed features associated with the need for respiratory support.

RESULTS

Of the 28605 patients tested for SARS-CoV-2, 7995 patients (27.9%) had an infection (median age 52.3 years) and 2154 (26.9%) of these had an associated admission (median age 66.2 years). Of admitted patients, 2152 (99.9%) had a discharge disposition at the end of the study period. Of these, 329 (15.3%) required invasive mechanical ventilation and 305 (14.2%) expired. Increased age and male sex were positively associated with admission and in-hospital mortality (median age 80.7 years), while comorbidities had a much weaker association with the risk of admission or mortality. Black race (OR 1.43, 95%CI 1.14-1.78) and Hispanic ethnicity (OR 1.81, 95%CI 1.50-2.18) were identified as risk factors for admission, but, among discharged patients, age-adjusted in-hospital mortality was not significantly different among racial and ethnic groups.

CONCLUSIONS

This observational study identified, among people testing positive for SARSCoV-2 infection, older age and male sex as the most strongly associated risks for admission and in-hospital mortality in patients with SARS-CoV-2 infection. While minority racial and ethnic groups had increased burden of disease and risk of admission, age-adjusted in-hospital mortality for discharged patients was not significantly different among racial and ethnic groups. Ongoing studies will be needed to continue to evaluate these risks, particularly in the setting of evolving treatment guidelines.

Reinforcement Learning using EEG signals for Therapeutic Use of Music in Emotion Management

Prolonged influence of negative emotions can result in clinical depression or anxiety, and while many prescribed techniques exist, music therapy approaches, coupled with psychotherapy, have shown to help lower depressive symptoms, supplementing traditional treatment approaches. Identifying the appropriate choice of music, therefore, is of utmost importance. Selecting appropriate playlists, however, are challenged by user feedback that may inadvertently select songs that amplify the negative effects. Therefore, this work uses electroencephalogram (EEG) that automatically identifies the emotional impact of music and trains a reinforcement-learning approach to identify an adaptive personalized playlist of music to lead to improved emotional states. This work uses data from 32 users, collected in the publicly available DEAP dataset, to select songs for users that guide them towards joyful emotional states. Using a domain-specific reward-shaping function, a Q-learning agent is able to correctly guide a majority of users to the target emotional states, represented in a common emotion wheel. The average angular error of all users is 57°, with a standard deviation of 2.8 and the target emotional state is achieved.Clinical relevance- Music therapy for improving clinical depression and anxiety can be supplemented by additional emotion-guided music decisions in remote and personal settings by using automated techniques to capture emotional state and identify music that best guides users to target joyful states.

Recommendations for Reporting Machine Learning Analyses in Clinical Research

Use of machine learning (ML) in clinical research is growing steadily given the increasing availability of complex clinical data sets. ML presents important advantages in terms of predictive performance and identifying undiscovered subpopulations of patients with specific physiology and prognoses. Despite this popularity, many clinicians and researchers are not yet familiar with evaluating and interpreting ML analyses. Consequently, readers and peer-reviewers alike may either overestimate or underestimate the validity and credibility of an ML-based model. Conversely, ML experts without clinical experience may present details of the analysis that are too granular for a clinical readership to assess. Overwhelming evidence has shown poor reproducibility and reporting of ML models in clinical research suggesting the need for ML analyses to be presented in a clear, concise, and comprehensible manner to facilitate understanding and critical evaluation. We present a recommendation for transparent and structured reporting of ML analysis results specifically directed at clinical researchers. Furthermore, we provide a list of key reporting elements with examples that can be used as a template when preparing and submitting ML-based manuscripts for the same audience.

Protocol for project recovery after cardiac surgery: a single-center cohort study leveraging digital platform to characterise longitudinal patient-reported postoperative recovery patterns

INTRODUCTION

Improving postoperative patient recovery after cardiac surgery is a priority, but our current understanding of individual variations in recovery and factors associated with poor recovery is limited. We are using a health-information exchange platform to collect patient-reported outcome measures (PROMs) and wearable device data to phenotype recovery patterns in the 30-day period after cardiac surgery hospital discharge, to identify factors associated with these phenotypes and to investigate phenotype associations with clinical outcomes.

METHODS AND ANALYSIS

We designed a prospective cohort study to enrol 200 patients undergoing valve, coronary artery bypass graft or aortic surgery at a tertiary centre in the USA. We are enrolling patients postoperatively after the intensive care unit discharge and delivering electronic surveys directly to patients every 3 days for 30 days after hospital discharge. We will conduct medical record reviews to collect patient demographics, comorbidity, operative details and hospital course using the Society of Thoracic Surgeons data definitions. We will use phone interview and medical record review data for adjudication of survival, readmission and complications. We will apply group-based trajectory modelling to the time-series PROM and device data to classify patients into distinct categories of recovery trajectories. We will evaluate whether certain recovery pattern predicts death or hospital readmissions, as well as whether clinical factors predict a patient having poor recovery trajectories. We will evaluate whether early recovery patterns predict the overall trajectory at the patient-level.

ETHICS AND DISSEMINATION

The Yale Institutional Review Board approved this study. Following the description of the study procedure, we obtain written informed consent from all study participants. The consent form states that all personal information, survey response and any medical records are confidential, will not be shared and are stored in an encrypted database. We plan to publish our study findings in peer-reviewed journals.

Developing Personalized Models of Blood Pressure Estimation from Wearable Sensors Data Using Minimally-trained Domain Adversarial Neural Networks

Blood pressure monitoring is an essential component of hypertension management and in the prediction of associated comorbidities. Blood pressure is a dynamic vital sign with frequent changes throughout a given day. Capturing blood pressure remotely and frequently (also known as ambulatory blood pressure monitoring) has traditionally been achieved by measuring blood pressure at discrete intervals using an inflatable cuff. However, there is growing interest in developing a cuffless ambulatory blood pressure monitoring system to measure blood pressure continuously. One such approach is by utilizing bioimpedance sensors to build regression models. A practical problem with this approach is that the amount of data required to confidently train such a regression model can be prohibitive. In this paper, we propose the application of the domain-adversarial training neural network (DANN) method on our multitask learning (MTL) blood pressure estimation model, allowing for knowledge transfer between subjects. Our proposed model obtains average root mean square error (RMSE) of 4.80 ± 0.74 mmHg for diastolic blood pressure and 7.34 ± 1.88 mmHg for systolic blood pressure when using three minutes of training data, 4.64 ± 0.60 mmHg and 7.10 ± 1.79 respectively when using four minutes of training data, and 4.48±0.57 mmHg and 6.79±1.70 respectively when using five minutes of training data. DANN improves training with minimal data in comparison to both directly training and to training with a pretrained model from another subject, decreasing RMSE by 0.19 to 0.26 mmHg (diastolic) and by 0.46 to 0.67 mmHg (systolic) in comparison to the best baseline models. We observe that four minutes of training data is the minimum requirement for our framework to exceed ISO standards within this cohort of patients.

BoXHED: Boosted eXact Hazard Estimator with Dynamic covariates

The proliferation of medical monitoring devices makes it possible to track health vitals at high frequency, enabling the development of dynamic health risk scores that change with the underlying readings. Survival analysis, in particular hazard estimation, is well-suited to analyzing this stream of data to predict disease onset as a function of the time-varying vitals. This paper introduces the software package BoXHED (pronounced 'box-head') for nonparametrically estimating hazard functions via gradient boosting. BoXHED 1.0 is a novel tree-based implementation of the generic estimator proposed in Lee et al. (2017), which was designed for handling time-dependent covariates in a fully nonparametric manner. BoXHED is also the first publicly available software implementation for Lee et al. (2017). Applying it to a cardiovascular disease dataset from the Framingham Heart Study reveals novel interaction effects among known risk factors, potentially resolving an open question in clinical literature.

Association of Use of an Intravascular Microaxial Left Ventricular Assist Device vs Intra-aortic Balloon Pump With In-Hospital Mortality and Major Bleeding Among Patients With Acute Myocardial Infarction Complicated by Cardiogenic Shock

IMPORTANCE

Acute myocardial infarction (AMI) complicated by cardiogenic shock is associated with substantial morbidity and mortality. Although intravascular microaxial left ventricular assist devices (LVADs) provide greater hemodynamic support as compared with intra-aortic balloon pumps (IABPs), little is known about clinical outcomes associated with intravascular microaxial LVAD use in clinical practice.

OBJECTIVE

To examine outcomes among patients undergoing percutaneous coronary intervention (PCI) for AMI complicated by cardiogenic shock treated with mechanical circulatory support (MCS) devices.

DESIGN, SETTING, AND PARTICIPANTS

A propensity-matched registry-based retrospective cohort study of patients with AMI complicated by cardiogenic shock undergoing PCI between October 1, 2015, and December 31, 2017, who were included in data from hospitals participating in the CathPCI and the Chest Pain-MI registries, both part of the American College of Cardiology's National Cardiovascular Data Registry. Patients receiving an intravascular microaxial LVAD were matched with those receiving IABP on demographics, clinical history, presentation, infarct location, coronary anatomy, and clinical laboratory data, with final follow-up through December 31, 2017.

EXPOSURES

Hemodynamic support, categorized as intravascular microaxial LVAD use only, IABP only, other (such as use of a percutaneous extracorporeal ventricular assist system, extracorporeal membrane oxygenation, or a combination of MCS device use), or medical therapy only.

MAIN OUTCOMES AND MEASURES

The primary outcomes were in-hospital mortality and in-hospital major bleeding.

RESULTS

Among 28 304 patients undergoing PCI for AMI complicated by cardiogenic shock, the mean (SD) age was 65.0 (12.6) years, 67.0% were men, 81.3% had an ST-elevation myocardial infarction, and 43.3% had cardiac arrest. Over the study period among patients with AMI, an intravascular microaxial LVAD was used in 6.2% of patients, and IABP was used in 29.9%. Among 1680 propensity-matched pairs, there was a significantly higher risk of in-hospital death associated with use of an intravascular microaxial LVAD (45.0%) vs with an IABP (34.1% [absolute risk difference, 10.9 percentage points {95% CI, 7.6-14.2}; P < .001) and also higher risk of in-hospital major bleeding (intravascular microaxial LVAD [31.3%] vs IABP [16.0%]; absolute risk difference, 15.4 percentage points [95% CI, 12.5-18.2]; P < .001). These associations were consistent regardless of whether patients received a device before or after initiation of PCI.

CONCLUSIONS AND RELEVANCE

Among patients undergoing PCI for AMI complicated by cardiogenic shock from 2015 to 2017, use of an intravascular microaxial LVAD compared with IABP was associated with higher adjusted risk of in-hospital death and major bleeding complications, although study interpretation is limited by the observational design. Further research may be needed to understand optimal device choice for these patients.

The National Institutes of Health funding for clinical research applying machine learning techniques in 2017

Machine learning (ML) techniques have become ubiquitous and indispensable for solving intricate problems in most disciplines. To determine the extent of funding for clinical research projects applying ML techniques by the National Institutes of Health (NIH) in 2017, we searched the NIH Research Portfolio Online Reporting Tools Expenditures and Results (RePORTER) system using relevant keywords. We identified 535 projects, which together received a total of $264 million, accounting for 2% of the NIH extramural budget for clinical research.

Using Intelligent Personal Annotations to Improve Human Activity Recognition for Movements in Natural Environments

Personal tracking algorithms for health monitoring are critical for understanding an individual's life-style and personal choices in natural environments (NE). In order to train such tracking algorithms in NE, however, annotated data is needed, particularly when tracking a variety of activities of daily living. These algorithms are often trained in laboratory settings, with expectations that they will perform equally well in NE, which is often not the case; they must be trained on annotated data collected in NE and wearable computers provide opportunities to collect such data, though the process is burdensome. Therefore, we propose an intelligent scoring algorithm that limits the number of user annotation requests through the confidence of predictions generated by the tracking algorithm and automatically annotating data with high confidence. We enhance our scoring algorithm by providing improvements in our tracking algorithm by obtaining context data from nearable sensors. Each specific context of a user bounds the set of activities that can likely occur, which in turn improves the tracking algorithm and confidence. Finally, we propose a hierarchical annotation approach, where repeated use allows us to ask for detailed annotations that differentiate fine-grained differences in ways individuals perform activities. We validate our approach in a diet monitoring case study. We vary the number of annotations requested per day to evaluate model accuracy; we improve accuracy in NE by 8% when restricting requests to 20 per day and improve F1-score of activities by 11% with hierarchical annotations, while discussing implementation, accuracy, and power consumption in real-time use.

Development and Validation of a Model for Predicting the Risk of Acute Kidney Injury Associated With Contrast Volume Levels During Percutaneous Coronary Intervention

Importance

Determining the association of contrast volume during percutaneous coronary intervention (PCI) with the risk of acute kidney injury (AKI) is important for optimizing PCI safety.

Objective

To quantify how the risk of AKI is associated with contrast volume, accounting for the possibility of nonlinearity and heterogeneity among different baseline risks.

Design, Setting, and Participants

This prognostic study used data from the American College of Cardiology National Cardiovascular Data Registry CathPCI Registry for 1694 US hospitals. Derivation analysis included 2 076 694 individuals who underwent PCI from July 1, 2011, to June 30, 2015. Validation analysis included 961 863 individuals who underwent PCI from July 1, 2015, to March 31, 2017. Data analysis took place from July 2018 to May 2019.

Exposure

Contrast volume during PCI.

Main Outcomes and Measures

Acute kidney injury was defined using 3 thresholds for preprocedure to postprocedure creatinine level increase (ie, ≥0.3 mg/dL, ≥0.5 mg/dL, and ≥1.0 mg/dL). A model quantifying the association of contrast volume with AKI was developed, and the existence of nonlinearity and heterogeneity were examined by likelihood ratio tests. The model was derived in the training set (a random 50% of the derivation cohort), and performance was evaluated in the test set (the remaining 50% of the derivation cohort) and an independent validation set by area under the receiver operating characteristic curve (AUC) and calibration slope of observed vs predicted risks.

Results

The 2 076 694 patients in the derivation set had a mean (SD) age of 65.1 (12.1) years, and 662 525 (31.9%) were women; 133 306 (6.4%) had creatinine level increases of at least 0.3 mg/dL, 66 626 (3.2%) had creatinine level increases of at least 0.5 mg/dL, and 28 378 (1.4%) had creatinine level increases of at least 1.0 mg/dL. In the validation set of 961 843 patients (mean [SD] age, 65.7 [12.1] years; 305 577 [31.8%] women), these rates were 62 913 (6.5%), 34 229 (3.6%), and 15 555 (1.6%), respectively. The association of contrast volume and AKI risk was nonlinear (χ226 = 1436.2; P < .001) and varied by preprocedural risk (χ220 = 105.6; P < .001). In the test set, the model yielded an AUC of 0.777 (95% CI, 0.775-0.779) for predicting risk of a creatinine level increase of at least 0.3 mg/dL, 0.839 (95% CI, 0.837-0.841) for predicting risk of a creatinine level increase of at least 0.5 mg/dL, and 0.870 (95% CI, 0.867-0.873) for predicting risk of a creatinine level increase of at least 1.0 mg/dL; it achieved a calibration slope of 0.998 (95% CI, 0.989-1.007), 0.999 (95% CI, 0.989-1.008), and 0.986 (95% CI, 0.973-0.998), respectively, for the AKI severity levels. The model had similar performance in the validation set (creatinine level increase of ≥0.3 mg/dL: AUC, 0.794; 95% CI, 0.792-0.795; calibration slope, 1.039; 95% CI, 1.030-1.047; creatinine level increase of ≥0.5 mg/dL: AUC, 0.845; 95% CI, 0.843-0.848; calibration slope, 1.063; 95% CI, 1.054-1.074; creatinine level increase of ≥1.0 mg/dL: AUC, 0.872; 95% CI, 0.869-0.875; calibration slope, 1.103; 95% CI, 1.089-1.117).

Conclusions and Relevance

The association of contrast volume with AKI risk is complex, varies by baseline risk, and can be predicted by a model. Future research to evaluate the effect of the model on AKI is needed.

Monitoring Lung Mechanics during Mechanical Ventilation using Machine Learning Algorithms

Evaluation of lung mechanics is the primary component for designing lung protective optimal ventilation strategies. This paper presents a machine learning approach for bedside assessment of respiratory resistance (R) and compliance (C). We develop machine learning algorithms to track flow rate and airway pressure and estimate R and C continuously and in real-time. An experimental study is conducted, by connecting a pressure control ventilator to a test lung that simulates various R and C values, to gather sensor data for validation of the devised algorithms. We develop supervised learning algorithms based on decision tree, decision table, and Support Vector Machine (SVM) techniques to predict R and C values. Our experimental results demonstrate that the proposed algorithms achieve 90.3%, 93.1%, and 63.9% accuracy in assessing respiratory R and C using decision table, decision tree, and SVM, respectively. These results along with our ability to estimate R and C with 99.4% accuracy using a linear regression model demonstrate the potential of the proposed approach for constructing a new generation of ventilation technologies that leverage novel computational models to control their underlying parameters for personalized healthcare and context-aware interventions.

Machine Learning Prediction of Mortality and Hospitalization in Heart Failure With Preserved Ejection Fraction

OBJECTIVES

This study sought to develop models for predicting mortality and heart failure (HF) hospitalization for outpatients with HF with preserved ejection fraction (HFpEF) in the TOPCAT (Treatment of Preserved Cardiac Function Heart Failure with an Aldosterone Antagonist) trial.

BACKGROUND

Although risk assessment models are available for patients with HF with reduced ejection fraction, few have assessed the risks of death and hospitalization in patients with HFpEF.

METHODS

The following 5 methods: logistic regression with a forward selection of variables; logistic regression with a lasso regularization for variable selection; random forest (RF); gradient descent boosting; and support vector machine, were used to train models for assessing risks of mortality and HF hospitalization through 3 years of follow-up and were validated using 5-fold cross-validation. Model discrimination and calibration were estimated using receiver-operating characteristic curves and Brier scores, respectively. The top prediction variables were assessed by using the best performing models, using the incremental improvement of each variable in 5-fold cross-validation.

RESULTS

The RF was the best performing model with a mean C-statistic of 0.72 (95% confidence interval [CI]: 0.69 to 0.75) for predicting mortality (Brier score: 0.17), and 0.76 (95% CI: 0.71 to 0.81) for HF hospitalization (Brier score: 0.19). Blood urea nitrogen levels, body mass index, and Kansas City Cardiomyopathy Questionnaire (KCCQ) subscale scores were strongly associated with mortality, whereas hemoglobin level, blood urea nitrogen, time since previous HF hospitalization, and KCCQ scores were the most significant predictors of HF hospitalization.

CONCLUSIONS

These models predict the risks of mortality and HF hospitalization in patients with HFpEF and emphasize the importance of health status data in determining prognosis. (Treatment of Preserved Cardiac Function Heart Failure with an Aldosterone Antagonist [TOPCAT]; NCT00094302).

Comparison of Machine Learning Methods With National Cardiovascular Data Registry Models for Prediction of Risk of Bleeding After Percutaneous Coronary Intervention

IMPORTANCE

Better prediction of major bleeding after percutaneous coronary intervention (PCI) may improve clinical decisions aimed to reduce bleeding risk. Machine learning techniques, bolstered by better selection of variables, hold promise for enhancing prediction.

OBJECTIVE

To determine whether machine learning techniques better predict post-PCI major bleeding compared with the existing National Cardiovascular Data Registry (NCDR) models.

DESIGN, SETTING, AND PARTICIPANTS

This comparative effectiveness study used the NCDR CathPCI Registry data version 4.4 (July 1, 2009, to April 1, 2015), machine learning techniques were used (logistic regression with lasso regularization and gradient descent boosting [XGBoost, version 0.71.2]), and output was then compared with the existing simplified risk score and full NCDR models. The existing models were recreated, and then performance was evaluated through additional techniques and variables in a 5-fold cross-validation in analysis conducted from October 1, 2015, to October 27, 2017. The setting was retrospective modeling of a nationwide clinical registry of PCI. Participants were all patients undergoing PCI. Percutaneous coronary intervention procedures were excluded if they were not the index PCI of admission, if the hospital site had missing outcomes measures, or if the patient underwent subsequent coronary artery bypass grafting.

EXPOSURES

Clinical variables available at admission and diagnostic coronary angiography data were used to determine the severity and complexity of presentation.

MAIN OUTCOMES AND MEASURES

The main outcome was in-hospital major bleeding within 72 hours after PCI. Results were evaluated by comparing C statistics, calibration, and decision threshold-based metrics, including the F score (harmonic mean of positive predictive value and sensitivity) and the false discovery rate.

RESULTS

The post-PCI major bleeding rate among 3 316 465 procedures (patients' median age, 65 years; interquartile range, 56-73 years; 68.1% male) was 4.5%. The existing full model achieved a mean C statistic of 0.78 (95% CI, 0.78-0.78). The use of XGBoost and full range of selected variables achieved a C statistic of 0.82 (95% CI, 0.82-0.82), with an F score of 0.31 (95% CI, 0.30-0.31). XGBoost correctly identified an additional 3.7% of cases identified as high risk who experienced a bleeding event and an overall improvement of 1.0% of cases identified as low risk who did not experience a bleeding event. The data-driven decision threshold helped improve the false discovery rate of the existing techniques. The existing simplified risk score model improved the false discovery rate from more than 90% to 78.7%. Modifying the model and the data decision threshold improved this rate from 78.7% to 73.4%.

CONCLUSIONS AND RELEVANCE

Machine learning techniques improved the prediction of major bleeding after PCI. These techniques may help to better identify patients who would benefit most from strategies to reduce bleeding risk.

Sparse Embedding for Interpretable Hospital Admission Prediction

This paper introduces a sparse embedding for electronic health record (EHR) data in order to predict hospital admission. We use a k-sparse autoencoder to embed the original registry data into a much lower dimension, with sparsity as a goal. Then, t-SNE is used to show the embedding of each patient's data in a 2D plot. We then demonstrate the predictive accuracy in different existing machine learning algorithms. Our sparse embedding performs competitively against the original data and traditional embedding vectors with an AUROC of 0.878. In addition, we demonstrate the expressive power of our sparse embedding, i.e. interpretability. Sparse embedding can discover more phenotypes in t-SNE visualization than original data or traditional embedding. The discovered phenotypes can be regarded as different risk groups, through which we can study the driving risk factors for each patient phenotype.

Enhancing the prediction of acute kidney injury risk after percutaneous coronary intervention using machine learning techniques: A retrospective cohort study

BACKGROUND

The current acute kidney injury (AKI) risk prediction model for patients undergoing percutaneous coronary intervention (PCI) from the American College of Cardiology (ACC) National Cardiovascular Data Registry (NCDR) employed regression techniques. This study aimed to evaluate whether models using machine learning techniques could significantly improve AKI risk prediction after PCI.

METHODS AND FINDINGS

We used the same cohort and candidate variables used to develop the current NCDR CathPCI Registry AKI model, including 947,091 patients who underwent PCI procedures between June 1, 2009, and June 30, 2011. The mean age of these patients was 64.8 years, and 32.8% were women, with a total of 69,826 (7.4%) AKI events. We replicated the current AKI model as the baseline model and compared it with a series of new models. Temporal validation was performed using data from 970,869 patients undergoing PCIs between July 1, 2016, and March 31, 2017, with a mean age of 65.7 years; 31.9% were women, and 72,954 (7.5%) had AKI events. Each model was derived by implementing one of two strategies for preprocessing candidate variables (preselecting and transforming candidate variables or using all candidate variables in their original forms), one of three variable-selection methods (stepwise backward selection, lasso regularization, or permutation-based selection), and one of two methods to model the relationship between variables and outcome (logistic regression or gradient descent boosting). The cohort was divided into different training (70%) and test (30%) sets using 100 different random splits, and the performance of the models was evaluated internally in the test sets. The best model, according to the internal evaluation, was derived by using all available candidate variables in their original form, permutation-based variable selection, and gradient descent boosting. Compared with the baseline model that uses 11 variables, the best model used 13 variables and achieved a significantly better area under the receiver operating characteristic curve (AUC) of 0.752 (95% confidence interval [CI] 0.749-0.754) versus 0.711 (95% CI 0.708-0.714), a significantly better Brier score of 0.0617 (95% CI 0.0615-0.0618) versus 0.0636 (95% CI 0.0634-0.0638), and a better calibration slope of observed versus predicted rate of 1.008 (95% CI 0.988-1.028) versus 1.036 (95% CI 1.015-1.056). The best model also had a significantly wider predictive range (25.3% versus 21.6%, p < 0.001) and was more accurate in stratifying AKI risk for patients. Evaluated on a more contemporary CathPCI cohort (July 1, 2015-March 31, 2017), the best model consistently achieved significantly better performance than the baseline model in AUC (0.785 versus 0.753), Brier score (0.0610 versus 0.0627), calibration slope (1.003 versus 1.062), and predictive range (29.4% versus 26.2%). The current study does not address implementation for risk calculation at the point of care, and potential challenges include the availability and accessibility of the predictors.

CONCLUSIONS

Machine learning techniques and data-driven approaches resulted in improved prediction of AKI risk after PCI. The results support the potential of these techniques for improving risk prediction models and identification of patients who may benefit from risk-mitigation strategies.

Knowledge-driven dictionaries for sparse representation of continuous glucose monitoring signals

Continuous glucose monitoring (CGM) of patients with diabetes allows the effective management of the disease and reduces the risk of hypoglycemic or hyperglycemic episodes. Towards this goal, the development of reliable CGM models is essential for representing the corresponding signals and interpreting them with respect to factors and outcomes of interest. We propose a sparse decomposition model to approximate CGM time-series as a linear combination of a small set of exemplar atoms, appropriately designed through parametric functions to capture the main fluctuations of the CGM signal. Sparse decomposition is performed through the orthogonal matching pursuit (OMP). Results indicate that the proposed model provides 0.1 relative reconstruction error with 0.8 compression rate on a publicly available dataset containing 25 patients diagnosed with Type 1 diabetes. The atoms selected from the OMP procedure can be further interpreted in relation to the clinically meaningful components of the CGM signal (e.g. glucose spikes, hypoglycemic episodes, etc.