Using Machine Learning Algorithms to Develop a Clinical Decision-Making Tool for COVID-19 Inpatients

Establishment of machine learning-based tool for early detection of pulmonary embolism

BACKGROUND AND OBJECTIVES

Pulmonary embolism (PE) is a complex disease with high mortality and morbidity rate, leading to increasing society burden. However, current diagnosis is solely based on symptoms and laboratory data despite its complex pathology, which easily leads to misdiagnosis and missed diagnosis by inexperienced doctors. Especially, CT pulmonary angiography, the gold standard method, is not widely available. In this study, we aim to establish a rapid and accurate screening model for pulmonary embolism using machine learning technology. Importantly, data required for disease prediction are easily accessed, including routine laboratory data and medical record information of patients.

METHODS

We extracted features from patients' routine laboratory results and medical records, including blood routine, biochemical group, blood coagulation routine and other test results, as well as symptoms and medical history information. Samples with a feature loss rate greater than 0.8 were deleted from the original database. Data from 4723 cases were retained, 231 of which were positive for pulmonary embolism. 50 features were retained through the positive and negative statistical hypothesis testing which was used to build the predictive model. In order to avoid identification as majority-class samples caused by the imbalance of sample proportion, we used the method of Synthetic Minority Oversampling Technique (SMOTE) to increase the amount of information on minority samples. Five typical machine learning algorithms were used to model the screening of pulmonary embolism, including Support Vector Machines, Logistic Regression, Random Forest, XGBoost, and Back Propagation Neural Networks. To evaluate model performance, sensitivity, specificity and AUC curve were analyzed as the main evaluation indicators. Furthermore, a baseline model was established using the characteristics of the pulmonary embolism guidelines as a comparison model.

RESULTS

We found that XGBoost showed better performance compared to other models, with the highest sensitivity and specificity (0.99 and 0.99, respectively). Moreover, it showed significant improvement in performance compared to the baseline model (sensitivity and specificity were 0.76 and 0.76 respectively). More important, our model showed low missed diagnosis rate (0.46) and high AUC value (0.992). Finally, the calculation time of our model is only about 0.05 s to obtain the possibility of pulmonary embolism.

CONCLUSIONS

In this study, five machine learning classification models were established to assess the likelihood of patients suffering from pulmonary embolism, and the XGBoost model most significantly improved the precision, sensitivity, and AUC for pulmonary embolism screening. Collectively, we have established an AI-based model to accurately predict pulmonary embolism at early stage.

Prediction of Clinical Remission with Adalimumab Therapy in Patients with Ulcerative Colitis by Fourier Transform-Infrared Spectroscopy Coupled with Machine Learning Algorithms

We aimed to develop prediction models for clinical remission associated with adalimumab treatment in patients with ulcerative colitis (UC) using Fourier transform-infrared (FT-IR) spectroscopy coupled with machine learning (ML) algorithms. This prospective, observational, multicenter study enrolled 62 UC patients and 30 healthy controls. The patients were treated with adalimumab for 56 weeks, and clinical remission was evaluated using the Mayo score. Baseline fecal samples were collected and analyzed using FT-IR spectroscopy. Various data preprocessing methods were applied, and prediction models were established by 10-fold cross-validation using various ML methods. Orthogonal partial least squares-discriminant analysis (OPLS-DA) showed a clear separation of healthy controls and UC patients, applying area normalization and Pareto scaling. OPLS-DA models predicting short- and long-term remission (8 and 56 weeks) yielded area-under-the-curve values of 0.76 and 0.75, respectively. Logistic regression and a nonlinear support vector machine were selected as the best prediction models for short- and long-term remission, respectively (accuracy of 0.99). In external validation, prediction models for short-term (logistic regression) and long-term (decision tree) remission performed well, with accuracy values of 0.73 and 0.82, respectively. This was the first study to develop prediction models for clinical remission associated with adalimumab treatment in UC patients by fecal analysis using FT-IR spectroscopy coupled with ML algorithms. Logistic regression, nonlinear support vector machines, and decision tree were suggested as the optimal prediction models for remission, and these were noninvasive, simple, inexpensive, and fast analyses that could be applied to personalized treatments.

Clinical Features and Risk Stratification of Multiple Myeloma Patients with COVID-19

SARS-CoV-2 infection often results in a more severe COVID-19 disease course in multiple myeloma (MM) patients compared to immunocompetent individuals. The aim of this report is to summarize the clinical features of the MM patients with COVID-19 and the impact of MM treatment on outcomes to guide risk stratification and ensure the appropriate management of the patients. Serological responses in MM patients post-infection or -vaccination are also reviewed to better understand the strategy of prevention. Along with reports from the literature, we presented findings from a retrospective analysis of the clinical characteristics and outcomes of COVID-19 infection in MM patients in our institution. Study population includes 34 MM patients with a median age of 61 (range: 35-82 years) who tested positive for SARS-CoV-2 between 1 March 2020-15 August 2021. We examined the effect of chemotherapy, the benefit of neutralizing monoclonal antibody (Bamlanivimab) and the impact of anti-CD38 antibody (daratumumab) on the hospitalization and mortality of the patients, as well as the efficacy of native antibody production. Our results showed that MM patients have increased hospitalization and mortality rates from COVID-19 compared with that of general population, especially those on active chemotherapy. Advanced age, high-risk myeloma, renal disease, and suboptimal disease control are independent predictors of adverse outcomes. The use of daratumumab does not increase the disease severity/hospitalization or the post-infection/vaccination seropositivity of SARS-CoV-2. The neutralizing antibody decreases overall mortality. Evidence from the current study and previous publications suggest that testing of neutralizing antibody post-SARS-CoV-2 vaccination in MM patients may be needed in reducing COVID-19 risk.

Explainable Machine Learning Techniques To Predict Amiodarone-Induced Thyroid Dysfunction Risk: Multicenter, Retrospective Study With External Validation

BACKGROUND

Machine learning offers new solutions for predicting life-threatening, unpredictable amiodarone-induced thyroid dysfunction. Traditional regression approaches for adverse-effect prediction without time-series consideration of features have yielded suboptimal predictions. Machine learning algorithms with multiple data sets at different time points may generate better performance in predicting adverse effects.

OBJECTIVE

We aimed to develop and validate machine learning models for forecasting individualized amiodarone-induced thyroid dysfunction risk and to optimize a machine learning-based risk stratification scheme with a resampling method and readjustment of the clinically derived decision thresholds.

METHODS

This study developed machine learning models using multicenter, delinked electronic health records. It included patients receiving amiodarone from January 2013 to December 2017. The training set was composed of data from Taipei Medical University Hospital and Wan Fang Hospital, while data from Taipei Medical University Shuang Ho Hospital were used as the external test set. The study collected stationary features at baseline and dynamic features at the first, second, third, sixth, ninth, 12th, 15th, 18th, and 21st months after amiodarone initiation. We used 16 machine learning models, including extreme gradient boosting, adaptive boosting, k-nearest neighbor, and logistic regression models, along with an original resampling method and 3 other resampling methods, including oversampling with the borderline-synthesized minority oversampling technique, undersampling-edited nearest neighbor, and over- and undersampling hybrid methods. The model performance was compared based on accuracy; Precision, recall, F₁-score, geometric mean, area under the curve of the receiver operating characteristic curve (AUROC), and the area under the precision-recall curve (AUPRC). Feature importance was determined by the best model. The decision threshold was readjusted to identify the best cutoff value and a Kaplan-Meier survival analysis was performed.

RESULTS

The training set contained 4075 patients from Taipei Medical University Hospital and Wan Fang Hospital, of whom 583 (14.3%) developed amiodarone-induced thyroid dysfunction, while the external test set included 2422 patients from Taipei Medical University Shuang Ho Hospital, of whom 275 (11.4%) developed amiodarone-induced thyroid dysfunction. The extreme gradient boosting oversampling machine learning model demonstrated the best predictive outcomes among all 16 models. The accuracy; Precision, recall, F₁-score, G-mean, AUPRC, and AUROC were 0.923, 0.632, 0.756, 0.688, 0.845, 0.751, and 0.934, respectively. After readjusting the cutoff, the best value was 0.627, and the F₁-score reached 0.699. The best threshold was able to classify 286 of 2422 patients (11.8%) as high-risk subjects, among which 275 were true-positive patients in the testing set. A shorter treatment duration; higher levels of thyroid-stimulating hormone and high-density lipoprotein cholesterol; and lower levels of free thyroxin, alkaline phosphatase, and low-density lipoprotein were the most important features.

CONCLUSIONS

Machine learning models combined with resampling methods can predict amiodarone-induced thyroid dysfunction and serve as a support tool for individualized risk prediction and clinical decision support.

Molecular sensitization pattern to house dust mites is formed from the first years of life and includes group 1, 2, Der p 23, Der p 5, Der p 7 and Der p 21 allergens

BACKGROUND

As the process and nature of developing sensitivity to house dust mites (HDMs) remain not fully studied, our goal was to establish the pattern, nature and timeframe of house dust mite (HDM) sensitization development in patients in Ukraine as well as the period when treatment of such patients would be most effective.

METHODS

The data of the multiplex allergy test Alex² was collected from 20,033 patients. To determine age specifics of sensitization, descriptive statistics were used. Bayesian Network analysis was used to build probabilistic patterns of individual sensitization.

RESULTS

Patients from Ukraine were most often sensitized to HDM allergens of group 1 (Der p 1, Der f 1) and group 2 (Der p 2, Der f 2) as well as to Der p 23 (55%). A considerable sensitivity to Der p 5, Der p 7 and Der p 21 allergens was also observed. The overall nature of sensitization to HDM allergens among the population of Ukraine is formed within the first year of life. By this time, there is a pronounced sensitization to HDM allergens of groups 1 and 2 as well as to Der p 23. Significance of sensitization to Der p 5, Der p 7 and Der p 21 allergens grows starting from the age of 3-6. Bayesian Network data analysis indicated the leading role of sensitization to Der p 1 and Der f 2. While developing the sensitivity to group 5 allergens, the leading role may belong to Der p 21 allergen.

CONCLUSION

The results obtained indicate the importance of determining the sensitization profile using the multi-component approach. A more detailed study of the optimal age for AIT prescription is required as the pattern of sensitization to HDMs is formed during the first year of life.

Hybrid Bayesian Network-Based Modeling: COVID-19-Pneumonia Case

The primary goal of this paper is to develop an approach for predicting important clinical indicators, which can be used to improve treatment. Using mathematical predictive modeling algorithms, we examined the course of COVID-19-based pneumonia (CP) with inpatient treatment. Algorithms used include dynamic and ordinary Bayesian networks (OBN and DBN), popular ML algorithms, the state-of-the-art auto ML approach and our new hybrid method based on DBN and auto ML approaches. Predictive targets include treatment outcomes, length of stay, dynamics of disease severity indicators, and facts of prescribed drugs for different time intervals of observation. Models are validated using expert knowledge, current clinical recommendations, preceding research and classic predictive metrics. The characteristics of the best models are as follows: MAE of 3.6 days of predicting LOS (DBN plus FEDOT auto ML framework), 0.87 accuracy of predicting treatment outcome (OBN); 0.98 F1 score for predicting facts of prescribed drug (DBN). Moreover, the advantage of the proposed approach is Bayesian network-based interpretability, which is very important in the medical field. After the validation of other CP datasets for other hospitals, the proposed models can be used as part of the decision support systems for improving COVID-19-based pneumonia treatment. Another important finding is the significant differences between COVID-19 and non-COVID-19 pneumonia.

COVID-Transformer: Interpretable COVID-19 Detection Using Vision Transformer for Healthcare

In the recent pandemic, accurate and rapid testing of patients remained a critical task in the diagnosis and control of COVID-19 disease spread in the healthcare industry. Because of the sudden increase in cases, most countries have faced scarcity and a low rate of testing. Chest X-rays have been shown in the literature to be a potential source of testing for COVID-19 patients, but manually checking X-ray reports is time-consuming and error-prone. Considering these limitations and the advancements in data science, we proposed a Vision Transformer-based deep learning pipeline for COVID-19 detection from chest X-ray-based imaging. Due to the lack of large data sets, we collected data from three open-source data sets of chest X-ray images and aggregated them to form a 30 K image data set, which is the largest publicly available collection of chest X-ray images in this domain to our knowledge. Our proposed transformer model effectively differentiates COVID-19 from normal chest X-rays with an accuracy of 98% along with an AUC score of 99% in the binary classification task. It distinguishes COVID-19, normal, and pneumonia patient's X-rays with an accuracy of 92% and AUC score of 98% in the Multi-class classification task. For evaluation on our data set, we fine-tuned some of the widely used models in literature, namely, EfficientNetB0, InceptionV3, Resnet50, MobileNetV3, Xception, and DenseNet-121, as baselines. Our proposed transformer model outperformed them in terms of all metrics. In addition, a Grad-CAM based visualization is created which makes our approach interpretable by radiologists and can be used to monitor the progression of the disease in the affected lungs, assisting healthcare.

Comorbidity Patterns of Mood Disorders in Adult Inpatients: Applying Association Rule Mining

This study explored physical and psychiatric comorbidities of mood disorders using association rule mining. There were 7709 subjects who were patients (≥19 years old) diagnosed with mood disorders and included in the data collected by the Korean National Hospital Discharge In-depth Injury Survey (KNHDS) between 2006 and 2018. Physical comorbidities (46.17%) were higher than that of psychiatric comorbidities (27.28%). The frequent comorbidities of mood disorders (F30-F39) were hypertensive diseases (I10-I15), neurotic, stress-related and somatoform disorders (F40-F48), diabetes mellitus (E10-E14), and diseases of esophagus, stomach, and duodenum (K20-K31). The bidirectional association path of mood disorders (F30-F39) with hypertensive diseases (I10-I15) and diabetes mellitus (E10-E14) were the strongest. Depressive episodes (F32) and recurrent depressive disorders (F33) revealed strong bidirectional association paths with other degenerative diseases of the nervous system (G30-G32) and organic, including symptomatic and mental disorders (F00-F09). Bipolar affective disorders (F31) revealed strong bidirectional association paths with diabetes mellitus (E10-E14) and hypertensive diseases (I10-I15). It was found that different physical and psychiatric disorders are comorbid according to the sub-classification of mood disorders. Understanding the comorbidity patterns of major comorbidities for each mood disorder can assist mental health providers in treating and managing patients with mood disorders.

Economic Evaluation of Mental Health Effects of Flooding Using Bayesian Networks

The appraisal of appropriate levels of investment for devising flooding mitigation and to support recovery interventions is a complex and challenging task. Evaluation must account for social, political, environmental and other conditions, such as flood state expectations and local priorities. The evaluation method should be able to quickly identify evolving investment needs as the incidence and magnitude of flood events continue to grow. Quantification is essential and must consider multiple direct and indirect effects on flood related outcomes. The method proposed is this study is a Bayesian network, which may be used ex-post for evaluation, but also ex-ante for future assessment, and near real-time for the reallocation of investment into interventions. The particular case we study is the effect of flood interventions upon mental health, which is a gap in current investment analyses. Natural events such as floods expose people to negative mental health disorders including anxiety, distress and post-traumatic stress disorder. Such outcomes can be mitigated or exacerbated not only by state funded interventions, but by individual and community skills and experience. Success is also dampened when vulnerable and previously exposed victims are affected. Current measures evaluate solely the effectiveness of interventions to reduce physical damage to people and assets. This paper contributes a design for a Bayesian network that exposes causal pathways and conditional probabilities between interventions and mental health outcomes as well as providing a tool that can readily indicate the level of investment needed in alternative interventions based on desired mental health outcomes.

Development of an Early Alert System for an Additional Wave of COVID-19 Cases Using a Recurrent Neural Network with Long Short-Term Memory

The impact of the still ongoing “Coronavirus Disease 2019” (COVID-19) pandemic has been and is still vast, affecting not only global human health and stretching healthcare facilities, but also profoundly disrupting societal and economic systems worldwide. The nature of the way the virus spreads causes cases to come in further recurring waves. This is due a complex array of biological, societal and environmental factors, including the novel nature of the emerging pathogen. Other parameters explaining the epidemic trend consisting of recurring waves are logistic–organizational challenges in the implementation of the vaccine roll-out, scarcity of doses and human resources, seasonality, meteorological drivers, and community heterogeneity, as well as cycles of strengthening and easing/lifting of the mitigation interventions. Therefore, it is crucial to be able to have an early alert system to identify when another wave of cases is about to occur. The availability of a variety of newly developed indicators allows for the exploration of multi-feature prediction models for case data. Ten indicators were selected as features for our prediction model. The model chosen is a Recurrent Neural Network with Long Short-Term Memory. This paper documents the development of an early alert/detection system that functions by predicting future daily confirmed cases based on a series of features that include mobility and stringency indices, and epidemiological parameters. The model is trained on the intermittent period in between the first and the second wave, in all of the South African provinces.