A Laplacian regularized graph neural network for predictive modeling of multiple chronic conditions.

BACKGROUND AND GOALS: One of the biggest difficulties facing healthcare systems today is the prevalence of multiple chronic diseases (MCC). Mortality and the development of new chronic illnesses are more likely in those with MCC. Pre-existing diseases and risk factors specific to the patient have an impact on the complex stochastic process that guides the evolution of MCC. This study's goal is to use a brand-new Graph Neural Network (GNN) model to examine the connections between specific chronic illnesses, patient-level risk factors, and pre-existing conditions. METHODS: We propose a graph neural network model to analyze the relationship between five chronic conditions (diabetes, obesity, cognitive impairment, hyperlipidemia, and hypertension). The proposed model adds a graph Laplacian regularization term to the loss function, which aims to improve the parameter learning process and accuracy of the GNN based on the graph structure. For validation, we used historical data from the Cameron County Hispanic Cohort (CCHC). RESULTS: Evaluating the Laplacian regularized GNN on data from 600 patients, we expanded our analysis from two chronic conditions to five chronic conditions. The proposed model consistently surpassed a baseline GNN model, achieving an average accuracy of ≥89% across all combinations. In contrast, the performance of the standard model declined more markedly with the addition of more chronic conditions. The Laplacian regularization provided consistent predictions for adjacent nodes, beneficial in cases with shared attributes among nodes. CONCLUSIONS: The incorporation of Laplacian regularization in our GNN model is essential, resulting in enhanced node categorization and better predictive performance by harnessing the graph structure. This study underscores the significance of considering graph structure when designing neural networks for graph data. Future research might further explore and refine this regularization method for various tasks using graph-structured data.

A technology assisted precision ketogenic diet intervention for cardio-renal-metabolic health in overweight or obese adults: Protocol for a randomized controlled trial.

BACKGROUND: The obesity epidemic is a public health concern, as it is associated with a variety of chronic conditions. The ketogenic diet has drawn much scientific and public attention. However, implementation is challenging and its effect on cardio-renal-metabolic health is inconclusive. This study will assess the feasibility, acceptability, and preliminary efficacy of a technology-assisted ketogenic diet on cardio-renal-metabolic health. METHODS: This is a single center, 6-month, stratified, randomized controlled trial. A total of 60 overweight/obese adults (18+ years old) will be enrolled, including 20 without type 2 diabetes (T2D) and without chronic kidney disease (CKD); 20 with T2D, but without CKD; and 20 with early-stage CKD. Participants will be stratified based on health conditions and randomized into a ketogenic diet (n = 30) or a low-fat diet group (n = 30). Health education involving diet and physical activity will be delivered both digitally and in-person. Mobile and connected health technologies will be used to track lifestyle behaviors and health indicators, as well as provide weekly feedback. The primary outcome (weight) and the secondary outcomes (e.g., blood pressure, glycemic control, renal health) will be assessed with traditional measurements and metabolomics. DISCUSSION: Mobile and connected health technologies provide new opportunities to improve chronic condition management, health education attendance, planned lifestyle changes and engagement, and health outcomes. The advancement of bioinformatics technology offers the possibility to profile and analyze omics data which may advance our understanding of the underlying mechanisms of intervention effects on health outcomes at the molecular level for personalized and precision lifestyle interventions.

Dynamic Functional Continuous Time Bayesian Networks for Prediction and Monitoring of the Impact of Patients' Modifiable Lifestyle Behaviors on the Emergence of Multiple Chronic Conditions.

More than a quarter of all Americans are estimated to have multiple chronic conditions (MCC). It is known that shared modifiable lifestyle behaviors account for many common MCC. What is not precisely known is the dynamic effect of changes in lifestyle behaviors on the trajectories of MCC emergence. This paper proposes dynamic functional continuous time Bayesian networks to effectively formulate the dynamic effect of patients' modifiable lifestyle behaviors and their interaction with non-modifiable demographics and preexisting conditions on the emergence of MCC. The proposed method considers the parameters of the conditional dependencies of MCC as a nonlinear state-space model and develops an extended Kalman filter to capture the dynamics of the modifiable risk factors on the MCC evolution. It also develops a tensor-based control chart based on the integration of multilinear principal component analysis and multivariate exponentially weighted moving average chart to monitor the effect of changes in the modifiable risk factors on the risk of new MCC. We validate the proposed method based on a combination of simulation and a real dataset of 385 patients from the Cameron County Hispanic Cohort. The dataset examines the emergence of 5 chronic conditions (Diabetes, Obesity, Cognitive Impairment, Hyperlipidemia, Hypertension) based on 4 modifiable lifestyle behaviors representing (Diet, Exercise, Smoking Habits, Drinking Habits) and 3 non-modifiable demographic risk factors (Age, Gender, Education). For the simulated study, the proposed algorithm shows a run-length of 4 samples (4 months) to identify behavioral changes with significant impacts on the risk of new MCC. For the real data study, the proposed algorithm shows a run-length of one sample (one year) to identify behavioral changes with significant impacts on the risk of new MCC. The results demonstrate the sensitivity of the proposed methodology for dynamic prediction and monitoring of the risk of MCC emergence in individual patients.

A Functional Model for Structure Learning and Parameter Estimation in Continuous Time Bayesian Network: An Application in Identifying Patterns of Multiple Chronic Conditions.

Bayesian networks are powerful statistical models to study the probabilistic relationships among sets of random variables with significant applications in disease modeling and prediction. Here, we propose a continuous time Bayesian network with conditional dependencies represented as regularized Poisson regressions to model the impact of exogenous variables on the conditional intensities of the network. We also propose an adaptive group regularization method with an intuitive early stopping feature based on Gaussian mixture model clustering for efficient learning of the structure and parameters of the proposed network. Using a dataset of patients with multiple chronic conditions extracted from electronic health records of the Department of Veterans Affairs, we compare the performance of the proposed network with some of the existing methods in the literature for both short-term (one-year ahead) and long-term (multi-year ahead) predictions. The proposed model provides a sparse intuitive representation of the complex functional relationships between multiple chronic conditions. It also provides the capability of analyzing multiple disease trajectories over time, given any combination of preexisting conditions.

Summarizing Complex Graphical Models of Multiple Chronic Conditions Using the Second Eigenvalue of Graph Laplacian: Algorithm Development and Validation.

BACKGROUND: It is important but challenging to understand the interactions of multiple chronic conditions (MCC) and how they develop over time in patients and populations. Clinical data on MCC can now be represented using graphical models to study their interaction and identify the path toward the development of MCC. However, the current graphical models representing MCC are often complex and difficult to analyze. Therefore, it is necessary to develop improved methods for generating these models. OBJECTIVE: This study aimed to summarize the complex graphical models of MCC interactions to improve comprehension and aid analysis. METHODS: We examined the emergence of 5 chronic medical conditions (ie, traumatic brain injury [TBI], posttraumatic stress disorder [PTSD], depression [Depr], substance abuse [SuAb], and back pain [BaPa]) over 5 years among 257,633 veteran patients. We developed 3 algorithms that utilize the second eigenvalue of the graph Laplacian to summarize the complex graphical models of MCC by removing less significant edges. The first algorithm learns a sparse probabilistic graphical model of MCC interactions directly from the data. The second algorithm summarizes an existing probabilistic graphical model of MCC interactions when a supporting data set is available. The third algorithm, which is a variation of the second algorithm, summarizes the existing graphical model of MCC interactions with no supporting data. Finally, we examined the coappearance of the 100 most common terms in the literature of MCC to validate the performance of the proposed model. RESULTS: The proposed summarization algorithms demonstrate considerable performance in extracting major connections among MCC without reducing the predictive accuracy of the resulting graphical models. For the model learned directly from the data, the area under the curve (AUC) performance for predicting TBI, PTSD, BaPa, SuAb, and Depr, respectively, during the next 4 years is as follows-year 2: 79.91%, 84.04%, 78.83%, 82.50%, and 81.47%; year 3: 76.23%, 80.61%, 73.51%, 79.84%, and 77.13%; year 4: 72.38%, 78.22%, 72.96%, 77.92%, and 72.65%; and year 5: 69.51%, 76.15%, 73.04%, 76.72%, and 69.99%, respectively. This demonstrates an overall 12.07% increase in the cumulative sum of AUC in comparison with the classic multilevel temporal Bayesian network. CONCLUSIONS: Using graph summarization can improve the interpretability and the predictive power of the complex graphical models of MCC.

IDENTIFYING THE NUMBER OF COMPONENTS IN GAUSSIAN MIXTURE MODELS USING NUMERICAL ALGEBRAIC GEOMETRY.

Using Gaussian mixture models for clustering is a statistically mature method for clustering in data science with numerous successful applications in science and engineering. The parameters for a Gaussian mixture model are typically estimated from training data using the iterative expectation-maximization algorithm, which requires the number of Gaussian components a priori. In this study, we propose two algorithms rooted in numerical algebraic geometry, namely an area-based algorithm and a local maxima algorithm, to identify the optimal number of components. The area-based algorithm transforms several Gaussian mixture models with varying number of components into sets of equivalent polynomial regression splines. Next, it uses homotopy continuation methods for evaluating the resulting splines to identify the number of components that results in the best fit. The local maxima algorithm forms a set of polynomials by fitting a smoothing spline to a kernel density estimate of the data. Next, it uses numerical algebraic geometry to solve the system of the first derivatives for finding the local maxima of the resulting smoothing spline, which estimates the number of mixture components. The local maxima algorithm also identifies the location of the centers of Gaussian components. Using a real-world case study in automotive manufacturing and multiple simulations, we compare the performance of the proposed algorithms with that of Akaike information criterion (AIC) and Bayesian information criterion (BIC), which are popular methods in the literature. We show the proposed algorithms are more robust than AIC and BIC when the Gaussian assumption is violated.

An Integrated Framework for Reducing Hospital Readmissions using Risk Trajectories Characterization and Discharge Timing Optimization.

When patients leave the hospital for lower levels of care, they experience a risk of adverse events on a daily basis. The advent of value-based purchasing among other major initiatives has led to an increasing emphasis on reducing the occurrences of these post-discharge adverse events. This has spurred the development of new prediction technologies to identify which patients are at risk for an adverse event as well as actions to mitigate those risks. Those actions include pre-discharge and post-discharge interventions to reduce risk. However, traditional prediction models have been developed to support only post-discharge actions; predicting risk of adverse events at the time of discharge only. In this paper we develop an integrated framework of risk prediction and discharge optimization that supports both types of interventions: discharge timing and post-discharge monitoring. Our method combines a kernel approach for capturing the non-linear relationship between length of stay and risk of an adverse event, with a Principle Component Analysis method that makes the resulting estimation tractable. We then demonstrate how this prediction model could be used to support both types of interventions by developing a simple and easily implementable discharge timing optimization.

Development of a Deep Learning Model for Dynamic Forecasting of Blood Glucose Level for Type 2 Diabetes Mellitus: Secondary Analysis of a Randomized Controlled Trial.

BACKGROUND: Type 2 diabetes mellitus (T2DM) is a major public health burden. Self-management of diabetes including maintaining a healthy lifestyle is essential for glycemic control and to prevent diabetes complications. Mobile-based health data can play an important role in the forecasting of blood glucose levels for lifestyle management and control of T2DM. OBJECTIVE: The objective of this work was to dynamically forecast daily glucose levels in patients with T2DM based on their daily mobile health lifestyle data including diet, physical activity, weight, and glucose level from the day before. METHODS: We used data from 10 T2DM patients who were overweight or obese in a behavioral lifestyle intervention using mobile tools for daily monitoring of diet, physical activity, weight, and blood glucose over 6 months. We developed a deep learning model based on long short-term memory-based recurrent neural networks to forecast the next-day glucose levels in individual patients. The neural network used several layers of computational nodes to model how mobile health data (food intake including consumed calories, fat, and carbohydrates; exercise; and weight) were progressing from one day to another from noisy data. RESULTS: The model was validated based on a data set of 10 patients who had been monitored daily for over 6 months. The proposed deep learning model demonstrated considerable accuracy in predicting the next day glucose level based on Clark Error Grid and ±10% range of the actual values. CONCLUSIONS: Using machine learning methodologies may leverage mobile health lifestyle data to develop effective individualized prediction plans for T2DM management. However, predicting future glucose levels is challenging as glucose level is determined by multiple factors. Future study with more rigorous study design is warranted to better predict future glucose levels for T2DM management.

Analysis of Feature Extraction Methods for Prediction of 30-Day Hospital Readmissions.

OBJECTIVES: This article aims to determine possible improvements made by feature extraction methods to the machine learning prediction methods for predicting 30-day hospital readmissions. METHODS: The study evaluates five feature extraction methods including principal component analysis (PCA), kernel principal component analysis (KPCA), isomap, Laplacian eigenmaps, and locality preserving projections (LPPs) for improving the accuracy of nine machine learning prediction methods in predicting 30-day hospital readmissions. The specific prediction methods considered include logistic regression, Cox regression, linear discriminant analysis, k-nearest neighbor (KNN), support vector machines (SVMs), bagged trees, boosted trees, random forest, and artificial neural networks. All models are developed in MATLAB and validated using area under the curve based on two population-based data sets from partner hospitals. RESULTS: Laplacian eigenmaps and isomap feature extraction provide the most improvement to the readmission predictive accuracy of KNN, SVM, bagged trees, boosted trees, and linear discriminant analysis methods. The results for artificial neural networks, random forest, Cox regression, and logistic regression show improvement for only one of the data sets. Also, PCA and LPP provided the best computation efficiency followed by KPCA, Laplacian eigenmaps, and isomap. CONCLUSION: Feature extraction methods can improve the predictive performance of machine learning methods for predicting readmissions. However, the improvement depended on the specific choice of the prediction method, feature extraction method, and the complexity of the data set features.

Mining patterns of comorbidity evolution in patients with multiple chronic conditions using unsupervised multi-level temporal Bayesian network.

Over the past few decades, the rise of multiple chronic conditions has become a major concern for clinicians. However, it is still not known precisely how multiple chronic conditions emerge among patients. We propose an unsupervised multi-level temporal Bayesian network to provide a compact representation of the relationship among emergence of multiple chronic conditions and patient level risk factors over time. To improve the efficiency of the learning process, we use an extension of maximum weight spanning tree algorithm and greedy search algorithm to study the structure of the proposed network in three stages, starting with learning the inter-relationship of comorbidities within each year, followed by learning the intra-relationship of comorbidity emergence between consecutive years, and finally learning the hierarchical relationship of comorbidities and patient level risk factors. We also use a longest path algorithm to identify the most likely sequence of comorbidities emerging from and/or leading to specific chronic conditions. Using a de-identified dataset of more than 250,000 patients receiving care from the U.S. Department of Veterans Affairs for a period of five years, we compare the performance of the proposed unsupervised Bayesian network in comparison with those of Bayesian networks developed based on supervised and semi-supervised learning approaches, as well as multivariate probit regression, multinomial logistic regression, and latent regression Markov mixture clustering focusing on traumatic brain injury (TBI), post-traumatic stress disorder (PTSD), depression (Depr), substance abuse (SuAb), and back pain (BaPa). Our findings show that the unsupervised approach has noticeably accurate predictive performance that is comparable to the best performing semi-supervised and the second-best performing supervised approaches. These findings also revealed that the unsupervised approach has improved performance over multivariate probit regression, multinomial logistic regression, and latent regression Markov mixture clustering.

A Multi-way Multi-task Learning Approach for Multinomial Logistic Regression*. An Application in Joint Prediction of Appointment Miss-opportunities across Multiple Clinics.

OBJECTIVES: Whether they have been engineered for it or not, most healthcare systems experience a variety of unexpected events such as appointment miss-opportunities that can have significant impact on their revenue, cost and resource utilization. In this paper, a multi-way multi-task learning model based on multinomial logistic regression is proposed to jointly predict the occurrence of different types of miss-opportunities at multiple clinics. METHODS: An extension of L1 / L2 regularization is proposed to enable transfer of information among various types of miss-opportunities as well as different clinics. A proximal algorithm is developed to transform the convex but non-smooth likelihood function of the multi-way multi-task learning model into a convex and smooth optimization problem solvable using gradient descent algorithm. RESULTS: A dataset of real attendance records of patients at four different clinics of a VA medical center is used to verify the performance of the proposed multi-task learning approach. Additionally, a simulation study, investigating more general data situations is provided to highlight the specific aspects of the proposed approach. Various individual and integrated multinomial logistic regression models with/without LASSO penalty along with a number of other common classification algorithms are fitted and compared against the proposed multi-way multi-task learning approach. Fivefold cross validation is used to estimate comparing models parameters and their predictive accuracy. The multi-way multi-task learning framework enables the proposed approach to achieve a considerable rate of parameter shrinkage and superior prediction accuracy across various types of miss-opportunities and clinics. CONCLUSIONS: The proposed approach provides an integrated structure to effectively transfer knowledge among different miss-opportunities and clinics to reduce model size, increase estimation efficacy, and more importantly improve predictions results. The proposed framework can be effectively applied to medical centers with multiple clinics, especially those suffering from information scarcity on some type of disruptions and/or clinics.

Mining Major Transitions of Chronic Conditions in Patients with Multiple Chronic Conditions.

OBJECTIVES: Evolution of multiple chronic conditions (MCC) follows a complex stochastic process, influenced by several factors including the inter-relationship of existing conditions, and patient-level risk factors. Nearly 20% of citizens aged 18 years and older are burdened with two or more (multiple) chronic conditions (MCC). Treatment for people living with MCC currently accounts for an estimated 66% of the Nation's healthcare costs. However, it is still not known precisely how MCC emerge and accumulate among individuals or in the general population. This study investigates major patterns of MCC transitions in a diverse population of patients and identifies the risk factors affecting the transition process. METHODS: A Latent regression Markov clustering (LRMCL) algorithm is proposed to identify major transitions of four MCC that include hypertension (HTN), depression, Post- Traumatic Stress Disorder (PTSD), and back pain. A cohort of 601,805 individuals randomly selected from the population of Iraq and Afghanistan war Veterans (IAVs) who received VA care during three or more years between 2002-2015, is used for training the proposed LRMCL algorithm. RESULTS: Two major clusters of MCC transition patterns with 78% and 22% probability of membership respectively were identified. The primary cluster demonstrated the possibility of improvement when the number of MCC is small and an increase in probability of MCC accumulation as the number of co- morbidities increased. The second cluster showed stability (no change) of MCC overtime as the major pattern. Age was the most significant risk factor associated with the most probable cluster for each IAV. CONCLUSIONS: These findings suggest that our proposed LRMCL algorithm can be used to describe and understand MCC transitions, which may ultimately allow healthcare systems to support optimal clinical decision- making. This method will be used to describe a broader range of MCC transitions in this and non-VA populations, and will add treatment information to see if models including treatments and MCC emergence can be used to support clinical decision-making in patient care.

A probabilistic model for predicting the probability of no-show in hospital appointments.

The number of no-shows has a significant impact on the revenue, cost and resource utilization for almost all healthcare systems. In this study we develop a hybrid probabilistic model based on logistic regression and empirical Bayesian inference to predict the probability of no-shows in real time using both general patient social and demographic information and individual clinical appointments attendance records. The model also considers the effect of appointment date and clinic type. The effectiveness of the proposed approach is validated based on a patient dataset from a VA medical center. Such an accurate prediction model can be used to enable a precise selective overbooking strategy to reduce the negative effect of no-shows and to fill appointment slots while maintaining short wait times.