Multi-step ahead predictive model for blood glucose concentrations of type-1 diabetic patients.

Continuous monitoring of blood glucose (BG) levels is a key aspect of diabetes management. Patients with Type-1 diabetes (T1D) require an effective tool to monitor these levels in order to make appropriate decisions regarding insulin administration and food intake to keep BG levels in target range. Effectively and accurately predicting future BG levels at multi-time steps ahead benefits a patient with diabetes by helping them decrease the risks of extremes in BG including hypo- and hyperglycemia. In this study, we present a novel multi-component deep learning model BG-Predict that predicts the BG levels in a multi-step look ahead fashion. The model is evaluated both quantitatively and qualitatively on actual blood glucose data for 97 patients. For the prediction horizon (PH) of 30 mins, the average values for root mean squared error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and normalized mean squared error (NRMSE) are [Formula: see text] mg/dL, 16.77 ± 4.87 mg/dL, [Formula: see text] and [Formula: see text] respectively. When Clarke and Parkes error grid analyses were performed comparing predicted BG with actual BG, the results showed average percentage of points in Zone A of [Formula: see text] and [Formula: see text] respectively. We offer this tool as a mechanism to enhance the predictive capabilities of algorithms for patients with T1D.

DST-Predict: Predicting Individual Mobility Patterns From Mobile Phone GPS Data.

Predicting spatial behaviors of an individual (e.g., frequent visits to specific locations) is important to improve our understanding of the complexity of human mobility patterns, and to capture anomalous behaviors in an individual's spatial movements, which can be particularly useful in situations such as those induced by the COVID-19 pandemic. We propose a system called Deep Spatio-Temporal Predictor (DST-Predict), that can predict the future visit frequency of an individual based on one's past mobility behaviour patterns using GPS trace data collected from mobile phones. Predicting such spatial behavior is challenging, primarily because individuals' patterns of location visits for each individual consists of both systematic and random components, which vary across the spatial and temporal scales of analysis. To address these issues, we propose a novel multi-view sequence-to-sequence model that uses Convolutional Long-short term memory (ConvLSTM) where the past history of frequent visit patterns features is used to predict individuals' future visit patterns in a multi-step manner. Using the GPS survey data obtained from 1,464 participants in western New York, US, we demonstrated that the proposed system is capable of predicting individuals' frequency of visits to common places in an urban setting, with high accuracy.

Identifying Patients Experiencing Opioid-Induced Respiratory Depression During Recovery From Anesthesia: The Application of Electronic Monitoring Devices.

BACKGROUND: Postsurgical patients experiencing opioid-related adverse drug events have 55% longer hospital stays, 47% higher costs associated with their care, 36% increased risk of 30-day readmission, and 3.4 times higher risk of inpatient mortality compared to those with no opioid-related adverse drug events. Most of the adverse events are preventable. GENERAL AIM: This study explored three types of electronic monitoring devices (pulse oximetry, capnography, and minute ventilation [MV]) to determine which were more effective at identifying the patient experiencing respiratory compromise and, further, to determine whether algorithms could be developed from the electronic monitoring data to aid in earlier detection of respiratory depression. MATERIALS AND METHODS: A study was performed in the postanesthesia care unit (PACU) in an inner city. Sixty patients were recruited in the preoperative admissions department on the day of their surgery. Forty-eight of the 60 patients wore three types of electronic monitoring devices while they were recovering from back, neck, hip, or knee surgery. Machine learning models were used for the analysis. RESULTS: Twenty-four of the 48 patients exhibited sustained signs of opioid-induced respiratory depression (OIRD). Although the SpO2 values did not change, end-tidal CO2 levels increased, and MV decreased, representing hypoventilation. A machine learning model was able to predict an OIRD event 10 min before the actual event occurred with 80% accuracy. LINKING EVIDENCE TO ACTION: Electronic monitoring devices are currently used as a tool to assess respiratory status using thresholds to distinguish when respiratory depression has occurred. This study introduces a potential paradigm shift from a reactive approach to a proactive approach that would identify a patient at high risk for OIRD. Capnography and MV were found to be effective tools in detecting respiratory compromise in the PACU.

longSil: an Evaluation Metric to Assess Quality of Clustering Longitudinal Clinical Data.

Longitudinal disease subtyping is an important problem within the broader scope of computational phenotyping. In this article, we discuss several data-driven unsupervised disease subtyping methods to obtain disease subtypes from longitudinal clinical data. The methods are analyzed in the context of chronic kidney disease, one of the leading health problems, both in the USA and worldwide. To provide a quantitative comparison of the different methods, we propose a novel evaluation metric that measures the cluster tightness and degree of separation between the various clusters produced by each method. Comparative results for two significantly large clinical datasets are provided, along with key insights that are possible due to the proposed evaluation metric.

Mortality Risk in Homebound Older Adults Predicted From Routinely Collected Nursing Data.

BACKGROUND: Newer analytic approaches for developing predictive models provide a method of creating decision support to translate findings into practice. OBJECTIVES: The aim of this study was to develop and validate a clinically interpretable predictive model for 12-month mortality risk among community-dwelling older adults. This is done by using routinely collected nursing assessment data to aide homecare nurses in identifying older adults who are at risk for decline, providing an opportunity to develop care plans that support patient and family goals for care. METHODS: A retrospective secondary analysis of Medicare and Medicaid data of 635,590 Outcome and Assessment Information Set (OASIS-C) start-of-care assessments from January 1, 2012, to December 31, 2012, was linked to the Master Beneficiary Summary File (2012-2013) for date of death. The decision tree was benchmarked against gold standards for predictive modeling, logistic regression, and artificial neural network (ANN). The models underwent k-fold cross-validation and were compared using area under the curve (AUC) and other data science metrics, including Matthews correlation coefficient (MCC). RESULTS: Decision tree variables associated with 12-month mortality risk included OASIS items: age, (M1034) overall status, (M1800-M1890) activities of daily living total score, cancer, frailty, (M1410) oxygen, and (M2020) oral medication management. The final models had good discrimination: decision tree, AUC = .71, 95% confidence interval (CI) [.705, .712], sensitivity = .73, specificity = .58, MCC = .31; ANN, AUC = .74, 95% CI [.74, .74], sensitivity = .68, specificity = .68, MCC = .35; and logistic regression, AUC = .74, 95% CI [.735, .742], sensitivity = .64, specificity = .70, MCC = .35. DISCUSSION: The AUC and 95% CI for the decision tree are slightly less accurate than logistic regression and ANN; however, the decision tree was more accurate in detecting mortality. The OASIS data set was useful to predict 12-month mortality risk. The decision tree is an interpretable predictive model developed from routinely collected nursing data that may be incorporated into a decision support tool to identify older adults at risk for death.

Extracting Deep Phenotypes for Chronic Kidney Disease Using Electronic Health Records.

INTRODUCTION: As chronic kidney disease (CKD) is among the most prevalent chronic diseases in the world with various rate of progression among patients, identifying its phenotypic subtypes is important for improving risk stratification and providing more targeted therapy and specific treatments for patients having different trajectories of the disease progression. PROBLEM DEFINITION AND DATA: The rapid growth and adoption of electronic health records (EHR) technology has created a unique opportunity to leverage the abundant clinical data, available as EHRs, to find meaningful phenotypic subtypes for CKD. In this study, we focus on extracting disease severity profiles for CKD while accounting for other confounding factors. PROBABILISTIC SUBTYPING MODEL: We employ a probabilistic model to identify precise phenotypes from EHR data of patients who have chronic kidney disease. Using this model, patient's eGFR trajectory is decomposed as a combination of four different components including disease subtype effect, covariate effect, individual long-term effect and individual short-term effect. EXPERIMENTAL RESULTS: The discovered disease subtypes obtained by Probabilistic Subtyping Model for CKD are presented and their clinical relevance is analyzed. DISCUSSION: Several clinical health markers that were found associated with disease subtypes are presented with suggestion for further investigation on their use as risk predictors. Several assumptions in the study are also clarified and discussed. CONCLUSION: The large dataset of EHRs can be used to identify deep phenotypes retrospectively. Directions for further expansion of the model are also discussed.