Clinical Time Series Data Research Articles

Predictive modeling of biomedical temporal data in healthcare applications: review and future directions.

Predictive modeling of clinical time series data is challenging due to various factors. One such difficulty is the existence of missing values, which leads to irregular data. Another challenge is capturing correlations across multiple dimensions in order to achieve accurate predictions. Additionally, it is essential to take into account the temporal structure, which includes both short-term and long-term recurrent patterns, to gain a comprehensive understanding of disease progression and to make accurate predictions for personalized healthcare. In critical situations, models that can make multi-step ahead predictions are essential for early detection. This review emphasizes the need for forecasting models that can effectively address the aforementioned challenges. The selection of models must also take into account the data-related constraints during the modeling process. Time series models can be divided into statistical, machine learning, and deep learning models. This review concentrates on the main models within these categories, discussing their capability to tackle the mentioned challenges. Furthermore, this paper provides a brief overview of a technique aimed at mitigating the limitations of a specific model to enhance its suitability for clinical prediction. It also explores ensemble forecasting methods designed to merge the strengths of various models while reducing their respective weaknesses, and finally discusses hierarchical models. Apart from the technical details provided in this document, there are certain aspects in predictive modeling research that have arisen as possible obstacles in implementing models using biomedical data. These obstacles are discussed leading to the future prospects of model building with artificial intelligence in healthcare domain.

Early neurological deterioration detection with a transformer convolutional auto-encoder model

This paper proposes an improved transformer convolutional auto-encoder model combined with the exponentially weighted moving average (EWMA) control chart to detect early neurological deterioration (END) of ischemic stroke patients after endovascular therapy in advance. In the proposed method, the transformer convolutional auto-encoder is used to extract crucial features of multivariate clinical monitoring time series data and obtain the reconfiguration loss while EWMA control chart is utilized to monitor the derived reconfiguration loss and identify anomalies. To verify the feasibility and effectiveness of the proposed END detection approach, multivariate clinical monitoring time series data of ischemic stroke patients in the neurocritical care unit from Beijing Tiantan hospital are collected. Meanwhile, the proposed approach is benchmarked with seven state-of-the-art models. The computation results show that the proposed approach achieves the best performance with the lowest false alarm rate and the highest detection rate. Therefore, the proposed END detection model is practical to guide doctors in conducting clinical interventions in advance to prevent deterioration in patients with ischemic stroke.

Predicting Major Bleeding in Patients with Venous Thromboembolism on Extended Anticoagulation Therapy Using Follow-up Data and Long Short-Term Memory Based Recurrent Neural Network

Background: It is often challenging for physicians to decide on the duration of anticoagulation treatment for patients with venous thromboembolism (VTE), as they need to weigh the risk of recurrent thrombosis and bleeding at the same time. Several clinical models have been developed to help physicians identify patients with high risk of bleeding. However, these tools use only the baseline clinical information and are not able to incorporate the clinical conditions and events that occur over time that may influence the risk of bleeding. Therefore, capturing the patterns and relationships in the continuously changing time series of clinical data could be a more powerful approach to developing predictor models than just relying on the baseline clinical information. Nonetheless, creating predictive models from follow-up clinical information is challenging given that time series clinical data are non-uniform, high-dimensional, multivariate observations. Here, we present the first attempt at creating a model that uses the time series follow-up information to predict major bleeding over time using a Recurrent Neural Network (RNN) with Long Short-Term Memory (LSTM) cells. Methods: 2542 patients diagnosed with VTE were enrolled in a prospective cohort study over 8 years. In addition to recording their clinical information at baseline, 6-month follow-up interviews were conducted using a standard script to monitor bleeding status and record clinical information. Major bleeding was defined by the International Society on Thrombosis and Haemostasis, with suspected bleeding events classified by an independent adjudication committee. Overall, 118 patients had major bleeding - a 4.6% incidence rate. The median and mode of the clinical variables were used to impute missing numerical and categorical values, respectively, and for patients who had no follow-up information, for whom bleeding occurred before the first follow-up, an artificial follow-up data point was generated from their corresponding baseline data. Thereafter, the data was divided into two stratified sets: 70% for training, and 30% for testing. Five supervised neural network-based machine learning models with different architectures were trained on the baseline dataset, or the follow-up dataset, or both to predict major bleeding. After training, these machine learning models were tested on the testing set and compared to the conventional clinical models, modified to make them compatible with the available predictor variables in our dataset, including the CHAP, the HAS-BLED, the VTE-BLEED, the RIETE, the ACCP, and the OBRI, which only use the baseline information. Results: Overall, the models that used the follow-up information had a higher area under the Receiver Operating Curve (AUROC) or c-statistic compared to the other models that only relied on the baseline dataset. In particular, the LSTM RNN model was able to achieve AUROC of 81.3% that is more than 10% higher compared to the best performing clinical model. We discovered that the LSTM RNN model mostly relied on features such as number of concomitant medications, years since baseline visit, use of specific antibiotics or antiplatelet agents, and presence of new hypertension to predict bleeding from the follow-up dataset. Furthermore, half of the bleeding events occurred within the first year after patients' baseline visits - a trend reflected in the predictions made by LSTM RNN model. Finally, the models that used both the baseline and the follow-up datasets showed different results depending on their architectures; that is, the simpler ensemble model achieved AUROC of 82.5% while the more complex model had AUROC of 70.8% due to overfitting. Conclusion: We have shown that using time series follow-up data can improve bleeding risk prediction in patients with VTE who are on extended anticoagulant therapy compared to just using the baseline data, and clinicians might benefit from using such an approach. Furthermore, our results indicate that LSTM RNN is a suitable architecture to model routine clinical follow-up data. Finally, we believe using time series data could improve the performance of the other clinical models that are currently based on one-time baseline measurements.

TADA: Temporal-aware Adversarial Domain Adaptation for patient outcomes forecasting

Patient outcomes forecasting (POF) has been shown to be an effective diagnostic assistant that can be used to predict disease progression and patient status in advance. In practice, the variation in patient progress and physical symptoms poses significant challenges for POF model training and testing. These differences result in POF models trained on one set of patients (source domain) not being efficiently self-adapted to the disease prediction task of another set of patients (target domain). Domain adaptive approaches can balance the overall distribution differences, named long-term divergence between patient training data and test data. However, the complex time series dependence and multivariate distribution properties of patient clinical data lead to the inability of existing domain adaptation methods to guarantee consistent distribution of patient clinical data at a finer granularity, named short-term divergence, e.g., clinical data distribution at different time intervals. In this paper, we propose a novel POF approach (TADA) based on clinical time series data using Temporal-aware Adversarial Domain Adaptation. In TADA, the traditional binary domain labels are generalized to act as the supervision signals to adversarially bridge the domain gap considering both long-term and short-term clinical information. TADA not only preserves the long-term representation of the clinical data but also captures the short-term data distribution at each period. The experimental results, based on both the publicly available and private clinical time series datasets, demonstrate that TADA can provide more reliable forecasting than state-of-the-art approaches.

An intelligent adverse delivery outcomes prediction model based on the fusion of multiple obstetric clinical data

Adverse delivery outcomes is a major re-productive health problem that affects the physical and mental health of pregnant women. Obviously, obstetric clinical data has periodically time series characteristics. This paper proposed a three stage adverse delivery outcomes prediction model via the fusion of multiple time series clinical data. The first stage is data aggregation, in which the data set is collected from the obstetric clinical data and divided based on time series features. In the second stage, a multi-channel gated cycle unit is used to solve the calculation error caused by irregular sampling of time series data. The hidden layer feature vector is connected with the fully connected layer, reshaped into a new one-dimensional feature, and fused with the non-time series data into a new data set. The third stage is the prediction stage of adverse delivery outcomes. By connecting the multi-channel gated cycle unit with the extreme gradient lift, the data transmitted in the corresponding channel is used in the feature extraction stage, in which the weighted entropy-based feature extraction is adopted. With the help of the extracted features, a hybrid artificial neural network architecture (MGRU-XGB) was developed to predict adverse delivery outcomes. The experimental results showed that the hybrid model had the best prediction performance for adverse delivery outcomes compared with other single models in terms of sensitivity, specificity, AUC and other evaluation indexes.

WindowSHAP: An efficient framework for explaining time-series classifiers based on Shapley values

Unpacking and comprehending how black-box machine learning algorithms (such as deep learning models) make decisions has been a persistent challenge for researchers and end-users. Explaining time-series predictive models is useful for clinical applications with high stakes to understand the behavior of prediction models, e.g., to determine how different variables and time points influence the clinical outcome. However, existing approaches to explain such models are frequently unique to architectures and data where the features do not have a time-varying component. In this paper, we introduce WindowSHAP, a model-agnostic framework for explaining time-series classifiers using Shapley values. We intend for WindowSHAP to mitigate the computational complexity of calculating Shapley values for long time-series data as well as improve the quality of explanations. WindowSHAP is based on partitioning a sequence into time windows. Under this framework, we present three distinct algorithms of Stationary, Sliding and Dynamic WindowSHAP, each evaluated against baseline approaches, KernelSHAP and TimeSHAP, using perturbation and sequence analyses metrics. We applied our framework to clinical time-series data from both a specialized clinical domain (Traumatic Brain Injury - TBI) as well as a broad clinical domain (critical care medicine). The experimental results demonstrate that, based on the two quantitative metrics, our framework is superior at explaining clinical time-series classifiers, while also reducing the complexity of computations. We show that for time-series data with 120 time steps (hours), merging 10 adjacent time points can reduce the CPU time of WindowSHAP by 80 % compared to KernelSHAP. We also show that our Dynamic WindowSHAP algorithm focuses more on the most important time steps and provides more understandable explanations. As a result, WindowSHAP not only accelerates the calculation of Shapley values for time-series data, but also delivers more understandable explanations with higher quality.

Deep multi-modal intermediate fusion of clinical record and time series data in mortality prediction.

In intensive care units (ICUs), mortality prediction is performed by combining information from these two sources of ICU patients by monitoring patient health. Respectively, time series data generated from each patient admission to the ICU and clinical records consisting of physician diagnostic summaries. However, existing mortality prediction studies mainly cascade the multimodal features of time series data and clinical records for prediction, ignoring thecross-modal correlation between the underlying features in different modal data. To address theseissues, we propose a multimodal fusion model for mortality prediction that jointly models patients' time-series data as well as clinical records. We apply a fine-tuned Bert model (Bio-Bert) to the patient's clinical record to generate a holistic embedding of the text part, which is then combined with the output of an LSTM model encoding the patient's time-series data to extract valid features. The global contextual information of each modal data is extracted using the improved fusion module to capture the correlation between different modal data. Furthermore, the improved fusion module can be easily added to the fusion features of any unimodal network and utilize existing pre-trained unimodal model weights. We use a real dataset containing 18904 ICU patients to train and evaluate our model, and the research results show that the representations obtained by themodel can achieve better prediction accuracy compared to the baseline.

A time-aware attention model for prediction of acute kidney injury after pediatric cardiac surgery.

Acute kidney injury (AKI) is a common complication after pediatric cardiac surgery, and the early detection of AKI may allow for timely preventive or therapeutic measures. However, current AKI prediction researches pay less attention to time information among time-series clinical data and model building strategies that meet complex clinical application scenario. This study aims to develop and validate a model for predicting postoperative AKI that operates sequentially over individual time-series clinical data. A retrospective cohort of 3386 pediatric patients extracted from PIC database was used for training, calibrating, and testing purposes. A time-aware deep learning model was developed and evaluated from 3 clinical perspectives that use different data collection windows and prediction windows to answer different AKI prediction questions encountered in clinical practice. We compared our model with existing state-of-the-art models from 3 clinical perspectives using the area under the receiver operating characteristic curve (ROC AUC) and the area under the precision-recall curve (PR AUC). Our proposed model significantly outperformed the existing state-of-the-art models with an improved average performance for any AKI prediction from the 3 evaluation perspectives. This model predicted 91% of all AKI episodes using data collected at 24h after surgery, resulting in a ROC AUC of 0.908 and a PR AUC of 0.898. On average, our model predicted 83% of all AKI episodes that occurred within the different time windows in the 3 evaluation perspectives. The calibration performance of the proposed model was substantially higher than the existing state-of-the-art models. This study showed that a deep learning model can accurately predict postoperative AKI using perioperative time-series data. It has the potential to be integrated into real-time clinical decision support systems to support postoperative care planning.

Personalized Spiking Neural Network Models of Clinical and Environmental Factors to Predict Stroke

The high incidence of stroke occurrence necessitates the understanding of its causes and possible ways for early prediction and prevention. In this respect, statistical methods offer the “big picture,” but they have a weak predictive ability at an individual level. This research proposes a new personalized modeling method based on computational spiking neural networks (SNN) for the identification of causal associations between clinical and environmental time series data that can be used to predict individual stroke events. The method is tested on 804 stroke patients. Given a clinical data set of patients who experienced a stroke in the past and the corresponding environmental time-series data for a selected time-window before the stroke event, the method identifies the clusters of individuals with a high risk for stroke under similar conditions. The methodology involves a pipeline of processes when creating a personalized model for an individual x: (1) selecting a group of individuals Gx with similar personal records to x; (2) training a personalized SNN x model of several days of environmental data related to the Gx group to predict the risk of stroke for x at least one day earlier; (3) model interpretability through 3D visualization; (4) discovery of personalized predictive markers. The results are twofold, first proposing a new computational methodology and second presenting new findings. It is found that certain environmental factors, such as SO2, PM10, CO, and PM2.5, increase the risk of stroke if an individual x belongs to a certain cluster of people, characterized by a combination of family history of stroke and diabetes, overweight, vascular/heart disease, age, and other. For the used population data, the proposed method can predict accurately individual risk of stroke before the day of the stroke. The paper presents a new methodology for personalized machine learning methods to define subgroups of the population with a high risk of stroke and to predict early individual risk of the stroke event. This makes the proposed cognitive computation method useful to reduce morbidity and mortality in society. The method is broadly applicable for predicting individual risk of other diseases and mental health conditions.

Multi-View Integrative Attention-Based Deep Representation Learning for Irregular Clinical Time-Series Data.

Electronic health record (EHR) data are sparse and irregular as they are recorded at irregular time intervals, and different clinical variables are measured at each observation point. In this work, to handle irregular multivariate time-series data, we consider the human knowledge of the aspects to be measured and time to measure them in different situations, known as multi-view features, which are indirectly represented in the data. We propose a scheme to realize multi-view features integration learning via a self-attention mechanism. Specifically, we devise a novel multi-integration attention module (MIAM) to extract complex information that is inherent in irregular time-series data. We explicitly learn the relationships among the observed values, missing indicators, and time interval between the consecutive observations in a simultaneous manner. In addition, we build an attention-based decoder as a missing value imputer that helps empower the representation learning of the interrelations among multi-view observations for the prediction task this decoder operates only in the training phase so that the final model is implemented in an imputation-free manner. We validated the effectiveness of our method over the public MIMIC-III and PhysioNet challenge 2012 datasets by comparing with and outperforming the state-of-the-art methods in three downstream tasks i.e., prediction of the in-hospital mortality, prediction of the length of stay, and phenotyping. Moreover, we conduct a layer-wise relevance propagation (LRP) analysis based on case studies to highlight the explainability of the trained model.

Learning to predict in-hospital mortality risk in the intensive care unit with attention-based temporal convolution network

BackgroundDynamic prediction of patient mortality risk in the ICU with time series data is limited due to high dimensionality, uncertainty in sampling intervals, and other issues. A new deep learning method, temporal convolution network (TCN), makes it possible to deal with complex clinical time series data in ICU. We aimed to develop and validate it to predict mortality risk using time series data from MIMIC III dataset.MethodsA total of 21,139 records of ICU stays were analysed and 17 physiological variables from the MIMIC III dataset were used to predict mortality risk. Then we compared the model performance of the attention-based TCN with that of traditional artificial intelligence (AI) methods.ResultsThe area under receiver operating characteristic (AUCROC) and area under precision-recall curve (AUC-PR) of attention-based TCN for predicting the mortality risk 48 h after ICU admission were 0.837 (0.824 -0.850) and 0.454, respectively. The sensitivity and specificity of attention-based TCN were 67.1% and 82.6%, respectively, compared to the traditional AI method, which had a low sensitivity (< 50%).ConclusionsThe attention-based TCN model achieved better performance in the prediction of mortality risk with time series data than traditional AI methods and conventional score-based models. The attention-based TCN mortality risk model has the potential for helping decision-making for critical patients.Trial registrationData used for the prediction of mortality risk were extracted from the freely accessible MIMIC III dataset. The project was approved by the Institutional Review Boards of Beth Israel Deaconess Medical Center (Boston, MA) and the Massachusetts Institute of Technology (Cambridge, MA). Requirement for individual patient consent was waived because the project did not impact clinical care and all protected health information was deidentified. The data were accessed via a data use agreement between PhysioNet, a National Institutes of Health–supported data repository (https://www.physionet.org/), and one of us (Yu-wen Chen, Certification Number: 28341490). All methods were carried out in accordance with the institutional guidelines and regulations.

Data Pre-Processing Using Neural Processes for Modeling Personalized Vital-Sign Time-Series Data

Clinical time-series data retrieved from electronic medical records are widely used to build predictive models of adverse events to support resource management. Such data is often sparse and irregularly-sampled, which makes it challenging to use many common machine learning methods. Missing values may be interpolated by carrying the last value forward, or through linear regression. Gaussian process (GP) regression is also used for performing imputation, and often re-sampling of time-series at regular intervals. The use of GPs can require extensive, and likely adhoc, investigation to determine model structure, such as an appropriate covariance function. This can be challenging for multivariate real-world clinical data, in which time-series variables exhibit different dynamics to one another. In this work, we construct generative models to estimate missing values in clinical time-series data using a neural latent variable model, known as a Neural Process (NP). The NP model employs a conditional prior distribution in the latent space to learn global uncertainty in the data by modelling variations at a local level. In contrast to conventional generative modelling, this prior is not fixed and is itself learned during the training process. Thus, NP model provides the flexibility to adapt to the dynamics of the available clinical data. We propose a variant of the NP framework for efficient modelling of the mutual information between the latent and input spaces, ensuring meaningful learned priors. Experiments using the MIMIC III dataset demonstrate the effectiveness of the proposed approach as compared to conventional methods.

A Multi-Headed Transformer Approach for Predicting the Patient’s Clinical Time-Series Variables From Charted Vital Signs

Deep learning has progressively been the spotlight of innovations that aim to leverage the clinical time-series data that are longitudinally recorded in the Electronic Health Records(EHR) to forecast the patient’s survival and vital signs deterioration. However, their recording velocity, as well as their noisiness, hinder the proper adoption of the recently proposed benchmarks. The Recurrent Neural Networks (RNN) especially the Long-short Term Memory (LSTMs) have achieved better results in recent studies but they are hard to train and interpret and fail to properly capture the long-term dependencies. Moreover, the RNNs suffer greatly with clinical time series due to their sequential processing which cripples the prospect of parallel processing. Recently the Transformer approach was proposed for Natural Language Processing (NLP) tasks and achieved state-of-the-art results. Hence to tackle the drawbacks that are suffered by the RNNs we propose a clinical time series Multi-head Transformer (MHT), which is a transformer-based model that forecasts the patient’s future time series variables using the vitals signs. To prove the generalization of the model we use the same model for other critical tasks that describe the Intensive Care Unit (ICU) patient’s progression and the associated risks like the remaining Length Of Stay(LoS), the In-hospital Mortality as well as the 24 hours mortality. Our model achieves an Area Under The Curve-Receiver Operating Characteristics( AUC-ROC) of 0.98 and an Area Under the Curve, Precision-Recall (AUC-PR) of 0.424 for vital time series prediction, and an AUC-ROC of 0.875 in the mortality prediction. The model performs well for the frequently recorded variables like the Heart Rate (HR) and performs barely like the LSTM counterparts for the intermittently captured records such as the White Blood Count (WBC).

ICU admission and mortality classifiers for COVID-19 patients based on subgroups of dynamically associated profiles across multiple timepoints

The coronavirus disease 2019 (COVID-19) which is caused by severe acute respiratory syndrome coronavirus type 2 (SARS-CoV-2) is consistently causing profound wounds in the global healthcare system due to its increased transmissibility. Currently, there is an urgent unmet need to identify the underlying dynamic associations among COVID-19 patients and distinguish patient subgroups with common clinical profiles towards the development of robust classifiers for ICU admission and mortality. To address this need, we propose a four step pipeline which: (i) enhances the quality of multiple timeseries clinical data through an automated data curation workflow, (ii) deploys Dynamic Bayesian Networks (DBNs) for the detection of features with increased connectivity based on dynamic association analysis across multiple points, (iii) utilizes Self Organizing Maps (SOMs) and trajectory analysis for the early identification of COVID-19 patients with common clinical profiles, and (iv) trains robust multiple additive regression trees (MART) for ICU admission and mortality classification based on the extracted homogeneous clusters, to identify risk factors and biomarkers for disease progression. The contribution of the extracted clusters and the dynamically associated clinical data improved the classification performance for ICU admission to sensitivity 0.83 and specificity 0.83, and for mortality to sensitivity 0.74 and specificity 0.76. Additional information was included to enhance the performance of the classifiers yielding an increase by 4% in sensitivity and specificity for mortality. According to the risk factor analysis, the number of lymphocytes, SatO2, PO2/FiO2, and O2 supply type were highlighted as risk factors for ICU admission and the percentage of neutrophils and lymphocytes, PO2/FiO2, LDH, and ALP for mortality, among others. To our knowledge, this is the first study that combines dynamic modeling with clustering analysis to identify homogeneous groups of COVID-19 patients towards the development of robust classifiers for ICU admission and mortality.

An Interpretable Early Dynamic Sequential Predictor for Sepsis-Induced Coagulopathy Progression in the Real-World Using Machine Learning.

Sepsis-associated coagulation dysfunction greatly increases the mortality of sepsis. Irregular clinical time-series data remains a major challenge for AI medical applications. To early detect and manage sepsis-induced coagulopathy (SIC) and sepsis-associated disseminated intravascular coagulation (DIC), we developed an interpretable real-time sequential warning model toward real-world irregular data. Eight machine learning models including novel algorithms were devised to detect SIC and sepsis-associated DIC 8n (1 ≤ n ≤ 6) hours prior to its onset. Models were developed on Xi'an Jiaotong University Medical College (XJTUMC) and verified on Beth Israel Deaconess Medical Center (BIDMC). A total of 12,154 SIC and 7,878 International Society on Thrombosis and Haemostasis (ISTH) overt-DIC labels were annotated according to the SIC and ISTH overt-DIC scoring systems in train set. The area under the receiver operating characteristic curve (AUROC) were used as model evaluation metrics. The eXtreme Gradient Boosting (XGBoost) model can predict SIC and sepsis-associated DIC events up to 48 h earlier with an AUROC of 0.929 and 0.910, respectively, and even reached 0.973 and 0.955 at 8 h earlier, achieving the highest performance to date. The novel ODE-RNN model achieved continuous prediction at arbitrary time points, and with an AUROC of 0.962 and 0.936 for SIC and DIC predicted 8 h earlier, respectively. In conclusion, our model can predict the sepsis-associated SIC and DIC onset up to 48 h in advance, which helps maximize the time window for early management by physicians.

Digital signatures for early traumatic brain injury outcome prediction in the intensive care unit

Traumatic brain injury (TBI) is a leading neurological cause of death and disability across the world. Early characterization of TBI severity could provide a window for therapeutic intervention and contribute to improved outcome. We hypothesized that granular electronic health record data available in the first 24 h following admission to the intensive care unit (ICU) can be used to differentiate outcomes at discharge. Working from two ICU datasets we focused on patients with a primary admission diagnosis of TBI whose length of stay in ICU was ≥ 24 h (N = 1689 and 127). Features derived from clinical, laboratory, medication, and physiological time series data in the first 24 h after ICU admission were used to train elastic-net regularized Generalized Linear Models for the prediction of mortality and neurological function at ICU discharge. Model discrimination, determined by area under the receiver operating characteristic curve (AUC) analysis, was 0.903 and 0.874 for mortality and neurological function, respectively. Model performance was successfully validated in an external dataset (AUC 0.958 and 0.878 for mortality and neurological function, respectively). These results demonstrate that computational analysis of data routinely collected in the first 24 h after admission accurately and reliably predict discharge outcomes in ICU stratum TBI patients.

Semisupervised Deep Learning Techniques for Predicting Acute Respiratory Distress Syndrome From Time-Series Clinical Data: Model Development and Validation Study.

BackgroundA high number of patients who are hospitalized with COVID-19 develop acute respiratory distress syndrome (ARDS).ObjectiveIn response to the need for clinical decision support tools to help manage the next pandemic during the early stages (ie, when limited labeled data are present), we developed machine learning algorithms that use semisupervised learning (SSL) techniques to predict ARDS development in general and COVID-19 populations based on limited labeled data.MethodsSSL techniques were applied to 29,127 encounters with patients who were admitted to 7 US hospitals from May 1, 2019, to May 1, 2021. A recurrent neural network that used a time series of electronic health record data was applied to data that were collected when a patient’s peripheral oxygen saturation level fell below the normal range (<97%) to predict the subsequent development of ARDS during the remaining duration of patients’ hospital stay. Model performance was assessed with the area under the receiver operating characteristic curve and area under the precision recall curve of an external hold-out test set.ResultsFor the whole data set, the median time between the first peripheral oxygen saturation measurement of <97% and subsequent respiratory failure was 21 hours. The area under the receiver operating characteristic curve for predicting subsequent ARDS development was 0.73 when the model was trained on a labeled data set of 6930 patients, 0.78 when the model was trained on the labeled data set that had been augmented with the unlabeled data set of 16,173 patients by using SSL techniques, and 0.84 when the model was trained on the entire training set of 23,103 labeled patients.ConclusionsIn the context of using time-series inpatient data and a careful model training design, unlabeled data can be used to improve the performance of machine learning models when labeled data for predicting ARDS development are scarce or expensive.

Deep dynamic imputation of clinical time series for mortality prediction

Missing values in clinical time-series data are pervasive and inevitable; they not only increase the complexity and difficulty of analyzing the data but also lead to biased results. To tackle these two problems, researchers have been exploring recurrent neural network (RNN)-based methods for detecting how well missing values are addressed with the aim of achieving state-of-the-art performance. However, these methods have two practical drawbacks. 1) Handling time-series data with multiple, irregular, abnormal values is difficult. 2) The patterns that may be present in the missing clinical data are not thoroughly considered. Moreover, to the best of our knowledge, none of these methods have been explicitly designed to dynamically optimize the imputation quality for better performance in the realm of clinical time-series analytics. By considering the quality of imputed values, we propose a 2-step integrated imputation-prediction model based on gated recurrent units (GRUs) for medical prediction tasks. In the first step, the missing values are imputed using a sophisticated model based on a replenished GRU with a hidden state decay mechanism (RGRU-D), which is followed by evaluation through two additional layers. In the second step, the optimized imputed values are used to predict the risk of mortality in critical patients. Our model effectively supplies missing values for the masking, time interval, bursty, and cumulative missing rate variables within an integrated deep architecture. Extensive experiments on a real-world ICU dataset demonstrate that our model performs better than the compared methods in terms of the imputation quality and prediction accuracy.

Utilizing the Heterogeneity of Clinical Data for Model Refinement and Rule Discovery Through the Application of Genetic Algorithms to Calibrate a High-Dimensional Agent-Based Model of Systemic Inflammation.

Introduction: Accounting for biological heterogeneity represents one of the greatest challenges in biomedical research. Dynamic computational and mathematical models can be used to enhance the study and understanding of biological systems, but traditional methods for calibration and validation commonly do not account for the heterogeneity of biological data, which may result in overfitting and brittleness of these models. Herein we propose a machine learning approach that utilizes genetic algorithms (GAs) to calibrate and refine an agent-based model (ABM) of acute systemic inflammation, with a focus on accounting for the heterogeneity seen in a clinical data set, thereby avoiding overfitting and increasing the robustness and potential generalizability of the underlying simulation model.Methods: Agent-based modeling is a frequently used modeling method for multi-scale mechanistic modeling. However, the same properties that make ABMs well suited to representing biological systems also present significant challenges with respect to their construction and calibration, particularly with respect to the selection of potential mechanistic rules and the large number of associated free parameters. We have proposed that machine learning approaches (such as GAs) can be used to more effectively and efficiently deal with rule selection and parameter space characterization; the current work applies GAs to the challenge of calibrating a complex ABM to a specific data set, while preserving biological heterogeneity reflected in the range and variance of the data. This project uses a GA to augment the rule-set for a previously validated ABM of acute systemic inflammation, the Innate Immune Response ABM (IIRABM) to clinical time series data of systemic cytokine levels from a population of burn patients. The genome for the GA is a vector generated from the IIRABM’s Model Rule Matrix (MRM), which is a matrix representation of not only the constants/parameters associated with the IIRABM’s cytokine interaction rules, but also the existence of rules themselves. Capturing heterogeneity is accomplished by a fitness function that incorporates the sample value range (“error bars”) of the clinical data.Results: The GA-enabled parameter space exploration resulted in a set of putative MRM rules and associated parameterizations which closely match the cytokine time course data used to design the fitness function. The number of non-zero elements in the MRM increases significantly as the model parameterizations evolve toward a fitness function minimum, transitioning from a sparse to a dense matrix. This results in a model structure that more closely resembles (at a superficial level) the structure of data generated by a standard differential gene expression experimental study.Conclusion: We present an HPC-enabled machine learning/evolutionary computing approach to calibrate a complex ABM to complex clinical data while preserving biological heterogeneity. The integration of machine learning, HPC, and multi-scale mechanistic modeling provides a pathway forward to more effectively representing the heterogeneity of clinical populations and their data.

Poster Abstract: Protecting User Data Privacy with Adversarial Perturbations.

The increased availability of on-body sensors gives researchers access to rich time-series data, many of which are related to human health conditions. Sharing such data can allow cross-institutional collaborations that create advanced data-driven models to make inferences on human well-being. However, such data are usually considered privacy-sensitive, and publicly sharing this data may incur significant privacy concerns. In this work, we seek to protect clinical time-series data against membership inference attacks, while maximally retaining the data utility. We achieve this by adding an imperceptible noise to the raw data. Known as adversarial perturbations, the noise is specially trained to force a deep learning model to make inference mistakes (in our case, mispredicting user identities). Our preliminary results show that our solution can better protect the data from membership inference attacks than the baselines, while succeeding in all the designed data quality checks.