A Bayesian model for estimating multi-state disease progression

Forecasting disease trajectories in critical illness: comparison of probabilistic dynamic systems to static models to predict patient status in the intensive care unit

OBJECTIVE

Conventional prediction models fail to integrate the constantly evolving nature of critical illness. Alternative modelling approaches to study dynamic changes in critical illness progression are needed. We compare static risk prediction models to dynamic probabilistic models in early critical illness.

DESIGN

We developed models to simulate disease trajectories of critically ill COVID-19 patients across different disease states. Eighty per cent of cases were randomly assigned to a training and 20% of the cases were used as a validation cohort. Conventional risk prediction models were developed to analyse different disease states for critically ill patients for the first 7 days of intensive care unit (ICU) stay. Daily disease state transitions were modelled using a series of multivariable, multinomial logistic regression models. A probabilistic dynamic systems modelling approach was used to predict disease trajectory over the first 7 days of an ICU admission. Forecast accuracy was assessed and simulated patient clinical trajectories were developed through our algorithm.

SETTING AND PARTICIPANTS

We retrospectively studied patients admitted to a Cleveland Clinic Healthcare System in Ohio, for the treatment of COVID-19 from March 2020 to December 2022.

RESULTS

5241 patients were included in the analysis. For ICU days 2-7, the static (conventional) modelling approach, the accuracy of the models steadily decreased as a function of time, with area under the curve (AUC) for each health state below 0.8. But the dynamic forecasting approach improved its ability to predict as a function of time. AUC for the dynamic forecasting approach were all above 0.90 for ICU days 4-7 for all states.

CONCLUSION

We demonstrated that modelling critical care outcomes as a dynamic system improved the forecasting accuracy of the disease state. Our model accurately identified different disease conditions and trajectories, with a <10% misclassification rate over the first week of critical illness.

Estimating Transition Probabilities Across the Alzheimer's Disease Continuum Using a Nationally Representative Real-World Database in the United States

INTRODUCTION

Clinical Alzheimer's disease (AD) begins with mild cognitive impairment (MCI) and progresses to mild, moderate, or severe dementia, constituting a disease continuum that eventually leads to death. This study aimed to estimate the probabilities of transitions across those disease states.

METHODS

We developed a mixed-effects multi-state Markov model to estimate the transition probabilities, adjusted for 5 baseline covariates, using the Health and Retirement Study (HRS) database. HRS surveys older adults in the United States bi-annually. Alzheimer states were defined using the modified Telephone Interview of Cognitive Status (TICS-m).

RESULTS

A total of 11,292 AD patients were analyzed. Patients were 70.8 ± 9.0 years old, 54.9% female, and with 12.0 ± 3.3 years of education. Within 1 year from the initial state, the model estimated a higher probability of transition to the next AD state in earlier disease: 12.8% from MCI to mild AD and 5.0% from mild to moderate AD, but < 1% from moderate to severe AD. After 10 years, the probability of transition to the next state was markedly higher for all states, but still higher in earlier disease: 29.8% from MCI to mild AD, 23.5% from mild to moderate AD, and 5.7% from moderate to severe AD. Across all AD states, the probability of transition to death was < 5% after 1 year and > 15% after 10 years. Older age, fewer years of education, unemployment, and nursing home stay were associated with a higher risk of disease progression (p < 0.01).

CONCLUSIONS

This analysis shows that the risk of progression is greater in earlier AD states, increases over time, and is higher in patients who are older, with fewer years of education, unemployed, or in a nursing home at baseline. The estimated transition probabilities can provide guidance for future disease management and clinical trial design optimization, and can be used to refine existing cost-effectiveness frameworks.

Moth-Flame Optimization for Early Prediction of Heart Diseases

Heart disease is among the leading causes of mortality globally. Predicting cardiovascular disease is a major difficulty in clinical data analysis. AI has been demonstrated to be powerful in deciding and anticipating an enormous measure of information created by the health domain. We provide a unique method for finding essential traits employing machine learning approaches in this paper, which enhances the effectiveness of identifying heart diseases. Decision tree (DT), support vector machine (SVM), artificial neural network (ANN), and K-nearest neighbor (KNN) are the classification techniques used to create the proposed system. Ensemble stacking integrates the four classification models to create a single best-fit predictive model using logistic regression. Many explorations have been directed at the identification of cardiac infection; however, the exactness of the outcomes is poor. Accordingly, to further enhance the efficiency, Moth-Flame Optimization (MFO) algorithm is proposed. The feature selection strategies are used to improve the classification accuracy while shortening the execution time of the classification system. Medical data are used to assess the probability of heart disease based on BP, age, gender, chest ache, cholesterol, blood sugar, and other variables. Results revealed that the proposed system excelled other existing models, obtaining 99% accuracy in the Cleveland dataset.

A Machine Learning Approach to Classifying Self-Reported Health Status in a Cohort of Patients With Heart Disease Using Activity Tracker Data

Constructing statistical models using personal sensor data could allow for tracking health status over time, thereby enabling the possibility of early intervention. The goal of this study was to use machine learning algorithms to classify patient-reported outcomes (PROs) using activity tracker data in a cohort of patients with stable ischemic heart disease (SIHD). A population of 182 patients with SIHD were monitored over a period of 12 weeks. Each subject received a Fitbit Charge 2 device to record daily activity data, and each subject completed eight Patient-Reported Outcomes Measurement Information Systems short form at the end of each week as a self-assessment of their health status. Two models were built to classify PRO scores using activity tracker data. The first model treated each week independently, whereas the second used a hidden Markov model (HMM) to take advantage of correlations between successive weeks. Retrospective analysis compared the classification accuracy of the two models and the importance of each feature. In the independent model, a random forest classifier achieved a mean area under curve (AUC) of 0.76 for classifying the physical function PRO. The HMM model achieved significantly better AUCs for all PROs (p < 0.05) other than Fatigue and Sleep Disturbance, with a highest mean AUC of 0.79 for the physical function-short form 10a. Our study demonstrates the ability of activity tracker data to classify health status over time. These results suggest that patient outcomes can be monitored in real time using activity trackers.

An Interpretable Deep Hierarchical Semantic Convolutional Neural Network for Lung Nodule Malignancy Classification

While deep learning methods have demonstrated performance comparable to human readers in tasks such as computer-aided diagnosis, these models are difficult to interpret, do not incorporate prior domain knowledge, and are often considered as a "black-box." The lack of model interpretability hinders them from being fully understood by end users such as radiologists. In this paper, we present a novel interpretable deep hierarchical semantic convolutional neural network (HSCNN) to predict whether a given pulmonary nodule observed on a computed tomography (CT) scan is malignant. Our network provides two levels of output: 1) low-level semantic features; and 2) a high-level prediction of nodule malignancy. The low-level outputs reflect diagnostic features often reported by radiologists and serve to explain how the model interprets the images in an expert-interpretable manner. The information from these low-level outputs, along with the representations learned by the convolutional layers, are then combined and used to infer the high-level output. This unified architecture is trained by optimizing a global loss function including both low- and high-level tasks, thereby learning all the parameters within a joint framework. Our experimental results using the Lung Image Database Consortium (LIDC) show that the proposed method not only produces interpretable lung cancer predictions but also achieves significantly better results compared to using a 3D CNN alone.

A Bayesian semiparametric Markov regression model for juvenile dermatomyositis

Juvenile dermatomyositis (JDM) is a rare autoimmune disease that may lead to serious complications, even to death. We develop a 2-state Markov regression model in a Bayesian framework to characterise disease progression in JDM over time and gain a better understanding of the factors influencing disease risk. The transition probabilities between disease and remission state (and vice versa) are a function of time-homogeneous and time-varying covariates. These latter types of covariates are introduced in the model through a latent health state function, which describes patient-specific health over time and accounts for variability among patients. We assume a nonparametric prior based on the Dirichlet process to model the health state function and the baseline transition intensities between disease and remission state and vice versa. The Dirichlet process induces a clustering of the patients in homogeneous risk groups. To highlight clinical variables that most affect the transition probabilities, we perform variable selection using spike and slab prior distributions. Posterior inference is performed through Markov chain Monte Carlo methods. Data were made available from the UK JDM Cohort and Biomarker Study and Repository, hosted at the UCL Institute of Child Health.