Sparse multi-output Gaussian processes for online medical time series prediction

Revealing chronic disease progression patterns using Gaussian process for stage inference

OBJECTIVE

The early stages of chronic disease typically progress slowly, so symptoms are usually only noticed until the disease is advanced. Slow progression and heterogeneous manifestations make it challenging to model the transition from normal to disease status. As patient conditions are only observed at discrete timestamps with varying intervals, an incomplete understanding of disease progression and heterogeneity affects clinical practice and drug development.

MATERIALS AND METHODS

We developed the Gaussian Process for Stage Inference (GPSI) approach to uncover chronic disease progression patterns and assess the dynamic contribution of clinical features. We tested the ability of the GPSI to reliably stratify synthetic and real-world data for osteoarthritis (OA) in the Osteoarthritis Initiative (OAI), bipolar disorder (BP) in the Adolescent Brain Cognitive Development Study (ABCD), and hepatocellular carcinoma (HCC) in the UTHealth and The Cancer Genome Atlas (TCGA).

RESULTS

First, GPSI identified two subgroups of OA based on image features, where these subgroups corresponded to different genotypes, indicating the bone-remodeling and overweight-related pathways. Second, GPSI differentiated BP into two distinct developmental patterns and defined the contribution of specific brain region atrophy from early to advanced disease stages, demonstrating the ability of the GPSI to identify diagnostic subgroups. Third, HCC progression patterns were well reproduced in the two independent UTHealth and TCGA datasets.

CONCLUSION

Our study demonstrated that an unsupervised approach can disentangle temporal and phenotypic heterogeneity and identify population subgroups with common patterns of disease progression. Based on the differences in these features across stages, physicians can better tailor treatment plans and medications to individual patients.

Prediction and diagnosis of depression using machine learning with electronic health records data: a systematic review

BACKGROUND

Depression is one of the most significant health conditions in personal, social, and economic impact. The aim of this review is to summarize existing literature in which machine learning methods have been used in combination with Electronic Health Records for prediction of depression.

METHODS

Systematic literature searches were conducted within arXiv, PubMed, PsycINFO, Science Direct, SCOPUS and Web of Science electronic databases. Searches were restricted to information published after 2010 (from 1st January 2011 onwards) and were updated prior to the final synthesis of data (27th January 2022).

RESULTS

Following the PRISMA process, the initial 744 studies were reduced to 19 eligible for detailed evaluation. Data extraction identified machine learning methods used, types of predictors used, the definition of depression, classification performance achieved, sample size, and benchmarks used. Area Under the Curve (AUC) values more than 0.9 were claimed, though the average was around 0.8. Regression methods proved as effective as more developed machine learning techniques.

LIMITATIONS

The categorization, definition, and identification of the numbers of predictors used within models was sometimes difficult to establish, Studies were largely Western Educated Industrialised, Rich, Democratic (WEIRD) in demography.

CONCLUSION

This review supports the potential use of machine learning techniques with Electronic Health Records for the prediction of depression. All the selected studies used clinically based, though sometimes broad, definitions of depression as their classification criteria. The reported performance of the studies was comparable to or even better than that found in primary care. There are concerns with generalizability and interpretability.

Continuous time recurrent neural networks: Overview and benchmarking at forecasting blood glucose in the intensive care unit

OBJECTIVE

Blood glucose measurements in the intensive care unit (ICU) are typically made at irregular intervals. This presents a challenge in choice of forecasting model. This article gives an overview of continuous time autoregressive recurrent neural networks (CTRNNs) and evaluates how they compare to autoregressive gradient boosted trees (GBT) in forecasting blood glucose in the ICU.

METHODS

Continuous time autoregressive recurrent neural networks (CTRNNs) are a deep learning model that account for irregular observations through incorporating continuous evolution of the hidden states between observations. This is achieved using a neural ordinary differential equation (ODE) or neural flow layer. In this manuscript, we give an overview of these models, including the varying architectures that have been proposed to account for issues such as ongoing medical interventions. Further, we demonstrate the application of these models to probabilistic forecasting of blood glucose in a critical care setting using electronic medical record and simulated data and compare with GBT and linear models.

RESULTS

The experiments confirm that addition of a neural ODE or neural flow layer generally improves the performance of autoregressive recurrent neural networks in the irregular measurement setting. However, several CTRNN architecture are outperformed by a GBT model (Catboost), with only a long short-term memory (LSTM) and neural ODE based architecture (ODE-LSTM) achieving comparable performance on probabilistic forecasting metrics such as the continuous ranked probability score (ODE-LSTM: 0.118 ± 0.001; Catboost: 0.118 ± 0.001), ignorance score (0.152 ± 0.008; 0.149 ± 0.002) and interval score (175 ± 1; 176 ± 1).

CONCLUSION

The application of deep learning methods for forecasting in situations with irregularly measured time series such as blood glucose shows promise. However, appropriate benchmarking by methods such as GBT approaches (plus feature transformation) are key in highlighting whether novel methodologies are truly state of the art in tabular data settings.

Graph-guided Gaussian Process-based Diagnosis of CVD Severity with Uncertainty Measures

The severity of coronary artery disease can be assessed invasively using the Fractional Flow Reserve (FFR) index which is a useful diagnostic tool for the clinicians to select the treatment approach. The present work capitalizes a Gaussian process (GP) framework over graphs for the prediction of FFR index using only non-invasive imaging and clinical features. More specifically, taking the per-node one-hop connectivity vector as input, we employed a regression-based task by applying an ensemble of graph-adapted Gaussian process experts, with a data-adaptive fashion via online training. The main novelty of the work lies in the fact that for the first time in a medical field the inference model considers only the similarity condition of the patients, instead of their features. Our results demonstrate the impressive merits of the proposed medical EGP (MedEGP) method, in comparison to the single GP, and Linear Regression (LR) models to predict the FFR index, with well-calibrated uncertainty.Clinical Relevance- This paper establishes an accurate non-invasive approach to predict the FFR for the diagnosis of coronary artery disease.

Oxygenation thresholds for invasive ventilation in hypoxemic respiratory failure: a target trial emulation in two cohorts

BACKGROUND

The optimal thresholds for the initiation of invasive ventilation in patients with hypoxemic respiratory failure are unknown. Using the saturation-to-inspired oxygen ratio (SF), we compared lower versus higher hypoxemia severity thresholds for initiating invasive ventilation.

METHODS

This target trial emulation included patients from the Medical Information Mart for Intensive Care (MIMIC-IV, 2008-2019) and the Amsterdam University Medical Centers (AmsterdamUMCdb, 2003-2016) databases admitted to intensive care and receiving inspired oxygen fraction ≥ 0.4 via non-rebreather mask, noninvasive ventilation, or high-flow nasal cannula. We compared the effect of using invasive ventilation initiation thresholds of SF < 110, < 98, and < 88 on 28-day mortality. MIMIC-IV was used for the primary analysis and AmsterdamUMCdb for the secondary analysis. We obtained posterior means and 95% credible intervals (CrI) with nonparametric Bayesian G-computation.

RESULTS

We studied 3,357 patients in the primary analysis. For invasive ventilation initiation thresholds SF < 110, SF < 98, and SF < 88, the predicted 28-day probabilities of invasive ventilation were 72%, 47%, and 19%. Predicted 28-day mortality was lowest with threshold SF < 110 (22.2%, CrI 19.2 to 25.0), compared to SF < 98 (absolute risk increase 1.6%, CrI 0.6 to 2.6) or SF < 88 (absolute risk increase 3.5%, CrI 1.4 to 5.4). In the secondary analysis (1,279 patients), the predicted 28-day probability of invasive ventilation was 50% for initiation threshold SF < 110, 28% for SF < 98, and 19% for SF < 88. In contrast with the primary analysis, predicted mortality was highest with threshold SF < 110 (14.6%, CrI 7.7 to 22.3), compared to SF < 98 (absolute risk decrease 0.5%, CrI 0.0 to 0.9) or SF < 88 (absolute risk decrease 1.9%, CrI 0.9 to 2.8).

CONCLUSION

Initiating invasive ventilation at lower hypoxemia severity will increase the rate of invasive ventilation, but this can either increase or decrease the expected mortality, with the direction of effect likely depending on baseline mortality risk and clinical context.

Do Thresholds for Invasive Ventilation in Hypoxemic Respiratory Failure Exist? A Cohort Study

Rationale: Invasive ventilation is a significant event for patients with respiratory failure. Physiologic thresholds standardize the use of invasive ventilation in clinical trials, but it is unknown whether thresholds prompt invasive ventilation in clinical practice. Objectives: To measure, in patients with hypoxemic respiratory failure, the probability of invasive ventilation within 3 hours after meeting physiologic thresholds. Methods: We studied patients admitted to intensive care receiving Fi_O2 of 0.4 or more via nonrebreather mask, noninvasive positive pressure ventilation, or high-flow nasal cannula, using data from the Medical Information Mart for Intensive Care (MIMIC)-IV database (2008-2019) and the Amsterdam University Medical Centers Database (AmsterdamUMCdb) (2003-2016). We evaluated 17 thresholds, including the ratio of arterial to inspired oxygen, the ratio of saturation to inspired oxygen ratio, composite scores, and criteria from randomized trials. We report the probability of invasive ventilation within 3 hours of meeting each threshold and its association with covariates using odds ratios (ORs) and 95% credible intervals (CrIs). Measurements and Main Results: We studied 4,726 patients (3,365 from MIMIC, 1,361 from AmsterdamUMCdb). Invasive ventilation occurred in 28% (1,320). In MIMIC, the highest probability of invasive ventilation within 3 hours of meeting a threshold was 20%, after meeting prespecified neurologic or respiratory criteria while on vasopressors, and 19%, after a ratio of arterial to inspired oxygen of <80 mm Hg. In AmsterdamUMCdb, the highest probability was 34%, after vasopressor initiation, and 25%, after a ratio of saturation to inspired oxygen of <90. The probability after meeting the threshold from randomized trials was 9% (MIMIC) and 13% (AmsterdamUMCdb). In MIMIC, a race/ethnicity of Black (OR, 0.75; 95% CrI, 0.57-0.96) or Asian (OR, 0.6; 95% CrI, 0.35-0.95) compared with White was associated with decreased probability of invasive ventilation after meeting a threshold. Conclusions: The probability of invasive ventilation within 3 hours of meeting physiologic thresholds was low and associated with patient race/ethnicity.

Study on the Prediction Method of Long-term Benign and Malignant Pulmonary Lesions Based on LSTM

In order to more accurately and comprehensively characterize the changes and development rules of lesion characteristics in pulmonary medical images in different periods, the study was conducted to predict the evolution of pulmonary nodules in the longitudinal dimension of time, and a benign and malignant prediction model of pulmonary lesions in different periods was constructed under multiscale three-dimensional (3D) feature fusion. According to the sequence of computed tomography (CT) images of patients at different stages, 3D interpolation was conducted to generate 3D lung CT images. The 3D features of different size lesions in the lungs were extracted using 3D convolutional neural networks for fusion features. A time-modulated long short-term memory was constructed to predict the benign and malignant lesions by using the improved time-length memory method to learn the feature vectors of lung lesions with temporal and spatial characteristics in different periods. The experiment shows that the area under the curve of the proposed method is 92.71%, which is higher than that of the traditional method.