Comparison of Different Machine Learning Classifiers for Glaucoma Diagnosis Based on Spectralis OCT

Development of machine learning-based predictors for early diagnosis of hepatocellular carcinoma

Hepatocellular carcinoma (HCC) remains a formidable malignancy that significantly impacts human health, and the early diagnosis of HCC holds paramount importance. Therefore, it is imperative to develop an efficacious signature for the early diagnosis of HCC. In this study, we aimed to develop early HCC predictors (eHCC-pred) using machine learning-based methods and compare their performance with existing methods. The enhancements and advancements of eHCC-pred encompassed the following: (i) utilization of a substantial number of samples, including an increased representation of cirrhosis tissues without HCC (CwoHCC) samples for model training and augmented numbers of HCC and CwoHCC samples for model validation; (ii) incorporation of two feature selection methods, namely minimum redundancy maximum relevance and maximum relevance maximum distance, along with the inclusion of eight machine learning-based methods; (iii) improvement in the accuracy of early HCC identification, elevating it from 78.15 to 97% using identical independent datasets; and (iv) establishment of a user-friendly web server. The eHCC-pred is freely accessible at http://www.dulab.com.cn/eHCC-pred/ . Our approach, eHCC-pred, is anticipated to be robustly employed at the individual level for facilitating early HCC diagnosis in clinical practice, surpassing currently available state-of-the-art techniques.

Artificial intelligence in the diagnosis of glaucoma and neurodegenerative diseases

Artificial Intelligence is a rapidly expanding field within computer science that encompasses the emulation of human intelligence by machines. Machine learning and deep learning - two primary data-driven pattern analysis approaches under the umbrella of artificial intelligence - has created considerable interest in the last few decades. The evolution of technology has resulted in a substantial amount of artificial intelligence research on ophthalmic and neurodegenerative disease diagnosis using retinal images. Various artificial intelligence-based techniques have been used for diagnostic purposes, including traditional machine learning, deep learning, and their combinations. Presented here is a review of the literature covering the last 10 years on this topic, discussing the use of artificial intelligence in analysing data from different modalities and their combinations for the diagnosis of glaucoma and neurodegenerative diseases. The performance of published artificial intelligence methods varies due to several factors, yet the results suggest that such methods can potentially facilitate clinical diagnosis. Generally, the accuracy of artificial intelligence-assisted diagnosis ranges from 67-98%, and the area under the sensitivity-specificity curve (AUC) ranges from 0.71-0.98, which outperforms typical human performance of 71.5% accuracy and 0.86 area under the curve. This indicates that artificial intelligence-based tools can provide clinicians with useful information that would assist in providing improved diagnosis. The review suggests that there is room for improvement of existing artificial intelligence-based models using retinal imaging modalities before they are incorporated into clinical practice.

Artificial intelligence in glaucoma: opportunities, challenges, and future directions

Artificial intelligence (AI) has shown excellent diagnostic performance in detecting various complex problems related to many areas of healthcare including ophthalmology. AI diagnostic systems developed from fundus images have become state-of-the-art tools in diagnosing retinal conditions and glaucoma as well as other ocular diseases. However, designing and implementing AI models using large imaging data is challenging. In this study, we review different machine learning (ML) and deep learning (DL) techniques applied to multiple modalities of retinal data, such as fundus images and visual fields for glaucoma detection, progression assessment, staging and so on. We summarize findings and provide several taxonomies to help the reader understand the evolution of conventional and emerging AI models in glaucoma. We discuss opportunities and challenges facing AI application in glaucoma and highlight some key themes from the existing literature that may help to explore future studies. Our goal in this systematic review is to help readers and researchers to understand critical aspects of AI related to glaucoma as well as determine the necessary steps and requirements for the successful development of AI models in glaucoma.

Using Machine Learning to Identify the Relationships between Demographic, Biochemical, and Lifestyle Parameters and Plasma Vitamin D Concentration in Healthy Premenopausal Chinese Women

INTRODUCTION

Vitamin D plays a vital role in maintaining homeostasis and enhancing the absorption of calcium, an essential component for strengthening bones and preventing osteoporosis. There are many factors known to relate to plasma vitamin D concentration (PVDC). However, most of these studies were performed with traditional statistical methods. Nowadays, machine learning methods (Mach-L) have become new tools in medical research. In the present study, we used four Mach-L methods to explore the relationships between PVDC and demographic, biochemical, and lifestyle factors in a group of healthy premenopausal Chinese women. Our goals were as follows: (1) to evaluate and compare the predictive accuracy of Mach-L and MLR, and (2) to establish a hierarchy of the significance of the aforementioned factors related to PVDC.

METHODS

Five hundred ninety-three healthy Chinese women were enrolled. In total, there were 35 variables recorded, including demographic, biochemical, and lifestyle information. The dependent variable was 25-OH vitamin D (PVDC), and all other variables were the independent variables. Multiple linear regression (MLR) was regarded as the benchmark for comparison. Four Mach-L methods were applied (random forest (RF), stochastic gradient boosting (SGB), extreme gradient boosting (XGBoost), and elastic net). Each method would produce several estimation errors. The smaller these errors were, the better the model was.

RESULTS

Pearson's correlation, age, glycated hemoglobin, HDL-cholesterol, LDL-cholesterol, and hemoglobin were positively correlated to PVDC, whereas eGFR was negatively correlated to PVDC. The Mach-L methods yielded smaller estimation errors for all five parameters, which indicated that they were better methods than the MLR model. After averaging the importance percentage from the four Mach-L methods, a rank of importance could be obtained. Age was the most important factor, followed by plasma insulin level, TSH, spouse status, LDH, and ALP.

CONCLUSIONS

In a healthy Chinese premenopausal cohort using four different Mach-L methods, age was found to be the most important factor related to PVDC, followed by plasma insulin level, TSH, spouse status, LDH, and ALP.

Comparison between multiple logistic regression and machine learning methods in prediction of abnormal thallium scans in type 2 diabetes

BACKGROUND

The prevalence of type 2 diabetes (T2D) has been increasing dramatically in recent decades, and 47.5% of T2D patients will die of cardiovascular disease. Thallium-201 myocardial perfusion scan (MPS) is a precise and non-invasive method to detect coronary artery disease (CAD). Most previous studies used traditional logistic regression (LGR) to evaluate the risks for abnormal CAD. Rapidly developing machine learning (Mach-L) techniques could potentially outperform LGR in capturing non-linear relationships.

AIM

To aims were: (1) Compare the accuracy of Mach-L methods and LGR; and (2) Found the most important factors for abnormal TMPS.

METHODS

556 T2D were enrolled in the study (287 men and 269 women). Demographic and biochemistry data were used as independent variables and the sum of stressed score derived from MPS scan was the dependent variable. Subjects with a MPS score ≥ 9 were defined as abnormal. In addition to traditional LGR, classification and regression tree (CART), random forest, Naïve Bayes, and eXtreme gradient boosting were also applied. Sensitivity, specificity, accuracy and area under the receiver operation curve were used to evaluate the respective accuracy of LGR and Mach-L methods.

RESULTS

Except for CART, the other Mach-L methods outperformed LGR, with gender, body mass index, age, low-density lipoprotein cholesterol, glycated hemoglobin and smoking emerging as the most important factors to predict abnormal MPS.

CONCLUSION

Four Mach-L methods are found to outperform LGR in predicting abnormal TMPS in Chinese T2D, with the most important risk factors being gender, body mass index, age, low-density lipoprotein cholesterol, glycated hemoglobin and smoking.

Comparison between linear regression and four different machine learning methods in selecting risk factors for osteoporosis in a Chinese female aged cohort

BACKGROUND

Population aging is emerging as an increasingly acute challenge for countries around the world. One particular manifestation of this phenomenon is the impact of osteoporosis on individuals and national health systems. Previous studies of risk factors for osteoporosis were conducted using traditional statistical methods, but more recent efforts have turned to machine learning approaches. Most such efforts, however, treat the target variable (bone mineral density [BMD] or fracture rate) as a categorical one, which provides no quantitative information. The present study uses five different machine learning methods to analyze the risk factors for T-score of BMD, seeking to (1) compare the prediction accuracy between different machine learning methods and traditional multiple linear regression (MLR) and (2) rank the importance of 25 different risk factors.

METHODS

The study sample includes 24 412 women older than 55 years with 25 related variables, applying traditional MLR and five different machine learning methods: classification and regression tree, Naïve Bayes, random forest, stochastic gradient boosting, and eXtreme gradient boosting. The metrics used for model performance comparisons are the symmetric mean absolute percentage error, relative absolute error, root relative squared error, and root mean squared error.

RESULTS

Machine learning approaches outperformed MLR for all four prediction errors. The average importance ranking of each factor generated by the machine learning methods indicates that age is the most important factor determining T-score, followed by estimated glomerular filtration rate (eGFR), body mass index (BMI), uric acid (UA), and education level.

CONCLUSION

In a group of women older than 55 years, we demonstrated that machine learning methods provide superior performance in estimating T-Score, with age being the most important impact factor, followed by eGFR, BMI, UA, and education level.

Roles of biochemistry data, lifestyle, and inflammation in identifying abnormal renal function in old Chinese

BACKGROUND

The incidence of chronic kidney disease (CKD) has dramatically increased in recent years, with significant impacts on patient mortality rates. Previous studies have identified multiple risk factors for CKD, but they mostly relied on the use of traditional statistical methods such as logistic regression and only focused on a few risk factors.

AIM

To determine factors that can be used to identify subjects with a low estimated glomerular filtration rate (L-eGFR < 60 mL/min per 1.73 m2) in a cohort of 1236 Chinese people aged over 65.

METHODS

Twenty risk factors were divided into three models. Model 1 consisted of demographic and biochemistry data. Model 2 added lifestyle data to Model 1, and Model 3 added inflammatory markers to Model 2. Five machine learning methods were used: Multivariate adaptive regression splines, eXtreme Gradient Boosting, stochastic gradient boosting, Light Gradient Boosting Machine, and Categorical Features + Gradient Boosting. Evaluation criteria included accuracy, sensitivity, specificity, area under the receiver operating characteristic curve (AUC), F-1 score, and balanced accuracy.

RESULTS

A trend of increasing AUC of each was observed from Model 1 to Model 3 and reached statistical significance. Model 3 selected uric acid as the most important risk factor, followed by age, hemoglobin (Hb), body mass index (BMI), sport hours, and systolic blood pressure (SBP).

CONCLUSION

Among all the risk factors including demographic, biochemistry, and lifestyle risk factors, along with inflammation markers, UA is the most important risk factor to identify L-eGFR, followed by age, Hb, BMI, sport hours, and SBP in a cohort of elderly Chinese people.

Relationships of Macular Functional Impairment With Structural and Vascular Changes According to Glaucoma Severity

Purpose

To determine the pointwise relationships of central visual field (VF) defects with macular ganglion cell loss and macular vessel density (VD) loss during various stages of glaucoma.

Methods

Eyes with primary open-angle glaucoma (POAG) were subjected to optical coherence tomography (OCT) and OCT angiography (OCTA) to evaluate macular ganglion cell layer (GCL) thickness and macular VD in the superficial and deep vascular complexes (SVC and DVC). OCT, OCTA, and VF locations were matched after correcting for retinal ganglion cell (RGC) displacement. Pointwise correlations of GCL thickness and VDs of the SVC and DVC with central VF sensitivity (VFS) were evaluated by Pearson's correlation analysis and compared in eyes with early and advanced POAG by Meng's test.

Results

Of the 100 eyes, 52 and 48 were classified as early and advanced POAG. Macular VD showed overall better correlation with central VFS than GCL thickness in both the early and advanced groups. SVC density showed the strongest correlation with central VFS in all groups (R = 0.327 in early group, R = 0.325 in advanced group, all P < 0.001). Although DVC density showed better correlation with VFS (R = 0.311) than GCL thickness (R = 0.212) in the early group (P < 0.001), the correlation was comparable in the advanced group (R = 0.199 and 0.176, respectively, P = 0.254).

Conclusions

After adjustment for RGC displacement, macular SVC density was better correlated with central VFS than macular GCL thickness in both early and advanced POAG. Macular DVC density showed better correlation with VFS than GCL thickness in early but not in advanced POAG, indicating that DVC loss may be involved in early central VF loss.

Early Detection of Primary Open Angle, Angle Closure, and Normal Tension Glaucoma in an Asian Population Using Optical Coherence Tomography

PRCIS

Spectral-domain optical coherence tomography (SD-OCT) facilitates early glaucoma detection in the Chinese population in Taiwan. The best parameters for primary open angle glaucoma (POAG), primary angle closure glaucoma (PACG), normal tension glaucoma (NTG), and suspected glaucoma (GS) detection are temporal inferior Bruch's membrane opening-minimum rim width (BMO-MRW), inner temporal macular ganglion cell layer (mGCL), temporal superior Circumpapillary retinal nerve fiber layer (cpRNFL), and mean global BMO-MRW, respectively.

PURPOSE

We investigated the diagnostic capability of SD-OCT for different types of early glaucoma among the Chinese population in Taiwan.

PARTICIPANTS AND METHODS

One eye each was assessed from 113 individuals with healthy eyes, 125 individuals with suspected glaucoma (GS), and 156 patients with early glaucoma (POAG, 87; PACG, 50; and NTG, 19). Circumpapillary (cp) RNFL thickness (global and sectoral), BMO-MRW, and macular parameters, including the macular RNFL (mRNFL), mGCL, and macular inner plexiform layer (mIPL), were assessed using SD-OCT. The areas under receiver operator characteristic curves (AUCs) were calculated to evaluate the diagnostic capacity of the parameters to differentiate between healthy and early glaucomatous eyes.

RESULTS

The parameters most suitable for detecting early POAG, PACG, NTG, and GS were temporal inferior BMO-MRW (AUC, 0.847), inner temporal mGCL (AUC, 0.770), temporal superior cpRNFL (AUC, 0.861), and mean global BMO-MRW (AUC, 0.768), respectively. Among the macular parameters, the mGCL exhibited the highest diagnostic capacity. The diagnostic capacity of the mGCL was lower than that of cpRNFL and BMO-MRW for POAG and NTG but not PACG. After adjusting for confounding variables in multivariable analysis, the AUC was determined to be 0.935 for POAG and 0.787 for GS.

CONCLUSION

SD-OCT facilitates the detection of early POAG, PACG, and NTG. Using a combination of cpRNFL, BMO-MRW, and macular parameters may enhance their diagnostic capacities. Further studies are necessary to validate these findings.

Machine Learning Predictive Models for Evaluating Risk Factors Affecting Sperm Count: Predictions Based on Health Screening Indicators

In many countries, especially developed nations, the fertility rate and birth rate have continually declined. Taiwan's fertility rate has paralleled this trend and reached its nadir in 2022. Therefore, the government uses many strategies to encourage more married couples to have children. However, couples marrying at an older age may have declining physical status, as well as hypertension and other metabolic syndrome symptoms, in addition to possibly being overweight, which have been the focus of the studies for their influences on male and female gamete quality. Many previous studies based on infertile people are not truly representative of the general population. This study proposed a framework using five machine learning (ML) predictive algorithms-random forest, stochastic gradient boosting, least absolute shrinkage and selection operator regression, ridge regression, and extreme gradient boosting-to identify the major risk factors affecting male sperm count based on a major health screening database in Taiwan. Unlike traditional multiple linear regression, ML algorithms do not need statistical assumptions and can capture non-linear relationships or complex interactions between dependent and independent variables to generate promising performance. We analyzed annual health screening data of 1375 males from 2010 to 2017, including data on health screening indicators, sourced from the MJ Group, a major health screening center in Taiwan. The symmetric mean absolute percentage error, relative absolute error, root relative squared error, and root mean squared error were used as performance evaluation metrics. Our results show that sleep time (ST), alpha-fetoprotein (AFP), body fat (BF), systolic blood pressure (SBP), and blood urea nitrogen (BUN) are the top five risk factors associated with sperm count. ST is a known risk factor influencing reproductive hormone balance, which can affect spermatogenesis and final sperm count. BF and SBP are risk factors associated with metabolic syndrome, another known risk factor of altered male reproductive hormone systems. However, AFP has not been the focus of previous studies on male fertility or semen quality. BUN, the index for kidney function, is also identified as a risk factor by our established ML model. Our results support previous findings that metabolic syndrome has negative impacts on sperm count and semen quality. Sleep duration also has an impact on sperm generation in the testes. AFP and BUN are two novel risk factors linked to sperm counts. These findings could help healthcare personnel and law makers create strategies for creating environments to increase the country's fertility rate. This study should also be of value to follow-up research.

Integrated Machine Learning Decision Tree Model for Risk Evaluation in Patients with Non-Valvular Atrial Fibrillation When Taking Different Doses of Dabigatran

The new generation of nonvitamin K antagonists are broadly applied for stroke prevention due to their notable efficacy and safety. Our study aimed to develop a suggestive utilization of dabigatran through an integrated machine learning (ML) decision-tree model. Participants taking different doses of dabigatran in the Randomized Evaluation of Long-Term Anticoagulant Therapy trial were included in our analysis and defined as the 110 mg and 150 mg groups. The proposed scheme integrated ML methods, namely naive Bayes, random forest (RF), classification and regression tree (CART), and extreme gradient boosting (XGBoost), which were used to identify the essential variables for predicting vascular events in the 110 mg group and bleeding in the 150 mg group. RF (0.764 for 110 mg; 0.747 for 150 mg) and XGBoost (0.708 for 110 mg; 0.761 for 150 mg) had better area under the receiver operating characteristic curve (AUC) values than logistic regression (benchmark model; 0.683 for 110 mg; 0.739 for 150 mg). We then selected the top ten important variables as internal nodes of the CART decision tree. The two best CART models with ten important variables output tree-shaped rules for predicting vascular events in the 110 mg group and bleeding in the 150 mg group. Our model can be used to provide more visualized and interpretable suggestive rules to clinicians managing NVAF patients who are taking dabigatran.

Classification Algorithms Used in Predicting Glaucoma Progression

In this paper, various machine learning algorithms were used in order to predict the evolution of open-angle glaucoma (POAG). The datasets were built containing clinical observations and objective measurements made at the Countess of Chester Hospital in the UK and at the "St. Spiridon" Hospital of Iași, Romania. Using these datasets, different classification problems were proposed. The evaluation of glaucoma progression was conducted based on parameters such as VFI (Visual field index), MD (Mean Deviation), PSD (Pattern standard deviation), and RNFL (Retinal Nerve Fiber Layer). As classification tools, the following algorithms were used: Multilayer Perceptron, Random Forest, Random Tree, C4.5, k-Nearest Neighbors, Support Vector Machine, and Non-Nested Generalized Exemplars. The best results, with an accuracy of over 90%, were obtained with Multilayer Perceptron and Random Forest algorithms. The NNGE algorithm also proved very useful in creating a hierarchy of the input values according to their influence (weight) on the considered outputs. On the other hand, the decision tree algorithms gave us insight into the logic used in their classification, which is of practical importance in obtaining additional information regarding the rationale behind a certain rule or decision.

Comparison between Machine Learning and Multiple Linear Regression to Identify Abnormal Thallium Myocardial Perfusion Scan in Chinese Type 2 Diabetes

Type 2 diabetes mellitus (T2DM) patients have a high risk of coronary artery disease (CAD). Thallium-201 myocardial perfusion scan (Th-201 scan) is a non-invasive and extensively used tool in recognizing CAD in clinical settings. In this study, we attempted to compare the predictive accuracy of evaluating abnormal Th-201 scans using traditional multiple linear regression (MLR) with four machine learning (ML) methods. From the study, we can determine whether ML surpasses traditional MLR and rank the clinical variables and compare them with previous reports.In total, 796 T2DM, including 368 men and 528 women, were enrolled. In addition to traditional MLR, classification and regression tree (CART), random forest (RF), stochastic gradient boosting (SGB) and eXtreme gradient boosting (XGBoost) were also used to analyze abnormal Th-201 scans. Stress sum score was used as the endpoint (dependent variable). Our findings show that all four root mean square errors of ML are smaller than with MLR, which implies that ML is more precise than MLR in determining abnormal Th-201 scans by using clinical parameters. The first seven factors, from the most important to the least are:body mass index, hemoglobin, age, glycated hemoglobin, Creatinine, systolic and diastolic blood pressure. In conclusion, ML is not inferior to traditional MLR in predicting abnormal Th-201 scans, and the most important factors are body mass index, hemoglobin, age, glycated hemoglobin, creatinine, systolic and diastolic blood pressure. ML methods are superior in these kinds of studies.

The Development and Clinical Application of Innovative Optical Ophthalmic Imaging Techniques

The field of ophthalmic imaging has grown substantially over the last years. Massive improvements in image processing and computer hardware have allowed the emergence of multiple imaging techniques of the eye that can transform patient care. The purpose of this review is to describe the most recent advances in eye imaging and explain how new technologies and imaging methods can be utilized in a clinical setting. The introduction of optical coherence tomography (OCT) was a revolution in eye imaging and has since become the standard of care for a plethora of conditions. Its most recent iterations, OCT angiography, and visible light OCT, as well as imaging modalities, such as fluorescent lifetime imaging ophthalmoscopy, would allow a more thorough evaluation of patients and provide additional information on disease processes. Toward that goal, the application of adaptive optics (AO) and full-field scanning to a variety of eye imaging techniques has further allowed the histologic study of single cells in the retina and anterior segment. Toward the goal of remote eye care and more accessible eye imaging, methods such as handheld OCT devices and imaging through smartphones, have emerged. Finally, incorporating artificial intelligence (AI) in eye images has the potential to become a new milestone for eye imaging while also contributing in social aspects of eye care.

Decision Trees for Glaucoma Screening Based on the Asymmetry of the Retinal Nerve Fiber Layer in Optical Coherence Tomography

Purpose: The aim of this study was to analyze the relevance of asymmetry features between both eyes of the same patient for glaucoma screening using optical coherence tomography. Methods: Spectral-domain optical coherence tomography was used to estimate the thickness of the peripapillary retinal nerve fiber layer in both eyes of the patients in the study. These measurements were collected in a dataset from healthy and glaucoma patients. Several metrics for asymmetry in the retinal nerve fiber layer thickness between the two eyes were then proposed. These metrics were evaluated using the dataset by performing a statistical analysis to assess their significance as relevant features in the diagnosis of glaucoma. Finally, the usefulness of these asymmetry features was demonstrated by designing supervised machine learning models that can be used for the early diagnosis of glaucoma. Results: Machine learning models were designed and optimized, specifically decision trees, based on the values of proposed asymmetry metrics. The use of these models on the dataset provided good classification of the patients (accuracy 88%, sensitivity 70%, specificity 93% and precision 75%). Conclusions: The obtained machine learning models based on retinal nerve fiber layer asymmetry are simple but effective methods which offer a good trade-off in classification of patients and simplicity. The fast binary classification relies on a few asymmetry values of the retinal nerve fiber layer thickness, allowing their use in the daily clinical practice for glaucoma screening.

Comparing Multiple Linear Regression and Machine Learning in Predicting Diabetic Urine Albumin–Creatinine Ratio in a 4-Year Follow-Up Study

The urine albumin–creatinine ratio (uACR) is a warning for the deterioration of renal function in type 2 diabetes (T2D). The early detection of ACR has become an important issue. Multiple linear regression (MLR) has traditionally been used to explore the relationships between risk factors and endpoints. Recently, machine learning (ML) methods have been widely applied in medicine. In the present study, four ML methods were used to predict the uACR in a T2D cohort. We hypothesized that (1) ML outperforms traditional MLR and (2) different ranks of the importance of the risk factors will be obtained. A total of 1147 patients with T2D were followed up for four years. MLR, classification and regression tree, random forest, stochastic gradient boosting, and eXtreme gradient boosting methods were used. Our findings show that the prediction errors of the ML methods are smaller than those of MLR, which indicates that ML is more accurate. The first six most important factors were baseline creatinine level, systolic and diastolic blood pressure, glycated hemoglobin, and fasting plasma glucose. In conclusion, ML might be more accurate in predicting uACR in a T2D cohort than the traditional MLR, and the baseline creatinine level is the most important predictor, which is followed by systolic and diastolic blood pressure, glycated hemoglobin, and fasting plasma glucose in Chinese patients with T2D.

Important Risk Factors in Patients with Nonvalvular Atrial Fibrillation Taking Dabigatran Using Integrated Machine Learning Scheme—A Post Hoc Analysis

Our study aims to develop an effective integrated machine learning (ML) scheme to predict vascular events and bleeding in patients with nonvalvular atrial fibrillation taking dabigatran and identify important risk factors. This study is a post-hoc analysis from the Randomized Evaluation of Long-Term Anticoagulant Therapy trial database. One traditional prediction method, logistic regression (LGR), and four ML techniques—naive Bayes, random forest (RF), classification and regression tree, and extreme gradient boosting (XGBoost)—were combined to construct our scheme. Area under the receiver operating characteristic curve (AUC) of RF (0.780) and XGBoost (0.717) was higher than that of LGR (0.674) in predicting vascular events. In predicting bleeding, AUC of RF (0.684) and XGBoost (0.618) showed higher values than those generated by LGR (0.605). Our integrated ML feature selection scheme based on the two convincing prediction techniques identified age, history of congestive heart failure and myocardial infarction, smoking, kidney function, and body mass index as major variables of vascular events; age, kidney function, smoking, bleeding history, concomitant use of specific drugs, and dabigatran dosage as major variables of bleeding. ML is an effective data analysis algorithm for solving complex medical data. Our results may provide preliminary direction for precision medicine.

Exploratory Study of Palliative Care Utilization and Medical Expense for Inpatients at the End-of-Life

Background: Previous research mostly analyzed the utilization of palliative care for patients with cancer, and data regarding non-cancer inpatients are limited. Objectives: This research aimed to investigate the current situation regarding palliative care and the important factors that influence its utilization by inpatients (including inpatients with and without cancer) at the end of their lives. We also explored the feasibility of establishing a prediction model of palliative care utilization for inpatients at the end of their lives. These findings will allow medical staff to monitor and focus on those who may require palliative care, resulting in more end-of-life patients receiving palliative care and thereby reducing medical expense and improving their quality of life. Methods: This was a retrospective study based on real-world health information system (HIS) data from 5 different branches of Taipei City Hospital between 1 January 2018 and 31 December 2018 that enrolled a total of 1668 deceased inpatients. To explore palliative care utilization at the end of life, we used 5-fold cross-validation in four different statistical models to obtain the performance of predictive accuracy: logistic regression (LGR), classification and regression tree (CART), multivariate adaptive regression spline (MARS), and gradient boosting (GB). The important variables that may affect palliative care utilization by inpatients were also identified. Results: The results were as follows: (1) 497 (29.8%) inpatients received palliative care; (2) the average daily hospitalization cost of patients with cancer who received palliative care (NTD 5789 vs. NTD 12,115; p ≤ 0.001) and all patients who received palliative care (NTD 91,527 vs. NTD 186,981; p = 0.0037) were statistically significantly lower than patients who did not receive palliative care; (3) diagnosis, hospital, and length of stay (LOS) may affect palliative care utilization of inpatient; diagnosis, hospitalization unit, and length of hospitalization were statistically significant by LGR; (4) 51.5% of patients utilized palliative consultation services, and 48.5% utilized palliative care units; and (5) MARS had the most consistent results; its accuracy was 0.751, and the main predictors of palliative care utilization are hospital, medical expense, LOS, diagnosis, and Palliative Care Screening Tool-Taiwan version (TW-PCST) scores. Conclusions: The results reveal that palliative care utilization by inpatients remains low, and it is necessary to educate patients without cancer of the benefits and advantages of palliative care. Although data were limited, the predictability of the MARS model was 0.751; a better prediction model with more data is necessary for further research. Precisely predicting the need for palliative care may encourage patients and their family members to consider palliative care, which may balance both physical and mental care. Therefore, unnecessary medical care can be avoided and limited medical resources can be allocated to more patients in need.

An Adaptive Deep Ensemble Learning Method for Dynamic Evolving Diagnostic Task Scenarios

Increasingly, machine learning methods have been applied to aid in diagnosis with good results. However, some complex models can confuse physicians because they are difficult to understand, while data differences across diagnostic tasks and institutions can cause model performance fluctuations. To address this challenge, we combined the Deep Ensemble Model (DEM) and tree-structured Parzen Estimator (TPE) and proposed an adaptive deep ensemble learning method (TPE-DEM) for dynamic evolving diagnostic task scenarios. Different from previous research that focuses on achieving better performance with a fixed structure model, our proposed model uses TPE to efficiently aggregate simple models more easily understood by physicians and require less training data. In addition, our proposed model can choose the optimal number of layers for the model and the type and number of basic learners to achieve the best performance in different diagnostic task scenarios based on the data distribution and characteristics of the current diagnostic task. We tested our model on one dataset constructed with a partner hospital and five UCI public datasets with different characteristics and volumes based on various diagnostic tasks. Our performance evaluation results show that our proposed model outperforms other baseline models on different datasets. Our study provides a novel approach for simple and understandable machine learning models in tasks with variable datasets and feature sets, and the findings have important implications for the application of machine learning models in computer-aided diagnosis.