Feature Importance Analysis Research Articles

Improving the Detection of Potential Cases of Familial Hypercholesterolemia: Could Machine Learning Be Part of the Solution?

Familial hypercholesterolemia (FH), while highly prevalent, is a significantly underdiagnosed monogenic disorder. Improved detection could reduce the large number of cardiovascular events attributable to poor case finding. We aimed to assess whether machine learning algorithms outperform clinical diagnostic criteria (signs, history, and biomarkers) and the recommended screening criteria in the United Kingdom in identifying individuals with FH-causing variants, presenting a scalable screening criteria for general populations. Analysis included UK Biobank participants with whole exome sequencing, classifying them as having FH when (likely) pathogenic variants were detected in their LDLR, APOB, or PCSK9 genes. Data were stratified into 3 data sets for (1) feature importance analysis; (2) deriving state-of-the-art statistical and machine learning models; (3) evaluating models' predictive performance against clinical diagnostic and screening criteria: Dutch Lipid Clinic Network, Simon Broome, Make Early Diagnosis to Prevent Early Death, and Familial Case Ascertainment Tool. One thousand and three of 454 710 participants were classified as having FH. A Stacking Ensemble model yielded the best predictive performance (sensitivity, 74.93%; precision, 0.61%; accuracy, 72.80%, area under the receiver operating characteristic curve, 79.12%) and outperformed clinical diagnostic criteria and the recommended screening criteria in identifying FH variant carriers within the validation data set (figures for Familial Case Ascertainment Tool, the best baseline model, were 69.55%, 0.44%, 65.43%, and 71.12%, respectively). Our model decreased the number needed to screen compared with the Familial Case Ascertainment Tool (164 versus 227). Our machine learning-derived model provides a higher pretest probability of identifying individuals with a molecular diagnosis of FH compared with current approaches. This provides a promising, cost-effective scalable tool for implementation into electronic health records to prioritize potential FH cases for genetic confirmation.

Digital-twin-driven intelligent tracking error compensation of ultra-precision machining

In ultra-precision machining (UPM), linear axis tracking affects contour accuracy and final machining quality. Traditional error modeling is complicated by the identification of numerous unknown parameters linked to nonlinear characteristics in the linear feed axes. To fill this gap, this study proposed a digital-twin-driven framework integrating the developed G-code interpreter and the deep learning model to achieve real-time tracking error compensation for UPM. To enhance the prediction accuracy of the tracking error of each axis of UPM machines, Bayesian hyperparameter optimization and feature importance analysis were conducted in the proposed TCN-BiLSTM model using high-quality training datasets from well-designed experiments. Ultimately, validation of the proposed system on a three-axis ultra-precision milling machine demonstrated its excellent performance. The experimental results showed that the optimized TCN-BiLSTM model exhibited an excellent capacity to predict the tracking error of the X-axis and Y-axis with minimal mean absolute error values of 0.000009 and 0.000023, respectively. Implementing the customized application reduced X-axis and Y-axis tracking errors by approximately 45–75% and 40–70%, respectively. This study first validates the feasibility of deep learning to improve accuracy in the UPM field, which will provide significant insight into speeding up the digitalization and intellectualization of the UPM scenario.

Two-stage feature selection for machine learning-aided DFT-based surface reactivity study on single-atom alloys

This paper presents a feature-centric strategy for predicting adsorption energies of key CO2 reduction reaction (CO2RR) adsorbates, CO and H species, utilizing density functional theory-based calculations for eight adsorption sites and considering alloying effects of nine transition metals at single-atom concentrations. Here, we explore a class of materials consisting of a majority host metal where individual atoms of a different element are dispersed called single-atom alloys (SAA). A total of eight feature selection methods are assessed within Gradient Boosting Regression and Linear Regression models. This study proposes a practical and effective two-stage approach that narrows down the initial 86 features to subsets of 10 and 7 for CO and H adsorption energy predictions, respectively, with the arithmetic mean of valence electrons (VE-am) feature consistently emerging as highly influential, validated through permutation and Shapley additive explanations-based feature importance analyses. The models exhibit robust performance on unseen data, indicating their generalization capability. The findings emphasize VE-am as a potential key machine learning feature for CO2RR on SAA surfaces and underline the effectiveness of the feature-centric approach in understanding feature impacts in machine learning models for CO2RR on SAA systems. Additionally, while other features based on structural, electronic and elemental properties may not individually impact the model significantly, their collective contribution plays a vital role in achieving more accurate adsorption energy predictions.

A fast neural emulator for interstellar chemistry

ABSTRACT Astrochemical models are important tools to interpret observations of molecular and atomic species in different environments. However, these models are time-consuming, precluding a thorough exploration of the parameter space, leading to uncertainties and biased results. Using neural networks to simulate the behaviour of astrochemical models is a way to circumvent this problem, providing fast calculations that are based on real astrochemical models. In this paper, we present a fast neural emulator of the astrochemical code Nautilus based on conditional neural fields. The resulting model produces the abundance of 192 species for arbitrary times between 1 and 107 yr. Uncertainties well below 0.2 dex are found for all species, while the computing time is of the order of 104 smaller than Nautilus. This will open up the possibility of performing much more complex forward models to better understand the physical properties of the interstellar medium. As an example of the power of these models, we ran a feature importance analysis on the electron abundance predicted by Nautilus. We found that the electron density is coupled to the initial sulphur abundance in a low-density gas. Increasing the initial sulphur abundance from a depleted scenario to the cosmic abundance leads to an enhancement of an order of magnitude of the electron density. This enhancement can potentially influence the dynamics of the gas in star formation sites.

In-Hospital Adverse Events of Pheochromocytoma-Induced Takotsubo Syndrome: A Literature Review and Cluster Analysis of 172 Cases.

Pheochromocytoma-induced takotsubo syndrome (Pheo-TTS) significantly increases the risk of adverse events for inpatient. The early identification of risk factors at admission is crucial for effective risk stratification and minimizing complications in Pheo-TTS patients. We conducted a systematic review combined with hierarchical cluster and feature importance analysis of demographic, clinical and laboratory data upon admission, alongside in-hospital complication data for Pheo-TTS patients. We analyzed cases published in PubMed and Embase from 2 May 2006 to 27 April 2023. Among 172 Pheo-TTS patients, cluster analysis identified two distinct groups: a chest pain dominant (CPD) group (n = 86) and a non-chest pain dominant (non-CPD) group (n = 86). The non-CPD group was characterized by a younger age (44.0 15.2 vs. 52.4 14.4, p 0.001), a higher prevalence of neurological/psychiatric disorders (53.5% vs. 32.6%), and increased presentation of dyspnea (87.2% vs. 17.4%), pulmonary rales (59.3% vs. 8.1%), and tachycardia (77.9% vs. 30.2%). Additionally, they exhibited more atypical takotsubo syndrome (TTS) imaging phenotypes (55.8% vs. 36.5%, all p 0.05). The non-CPD group experienced more than a 2-fold increase for in-hospital adverse events compared to the CPD group (70.9% vs. 30.2%, p 0.001). After adjusting for confounding factors, the absence of chest pain (odds ratio [OR] = 0.407, 95% confidence interval [CI] 0.169-0.979, p = 0.045), the presence of abdominal symptoms (OR = 3.939, 95% CI 1.770-8.766, p = 0.001), pulmonary rales (OR = 4.348, 95% CI 1.857-10.179, p = 0.001), and atypical TTS imaging phenotype (OR = 3.397, 95% CI 1.534-7.525, p = 0.003) remained as independent predictors of in-hospital complications. Clinical manifestations and imaging features at admission help to predict in-hospital complications for Pheo-TTS patients.

Predicting Cd(II) adsorption capacity of biochar materials using typical machine learning models for effective remediation of aquatic environments

The screening and design of “green” biochar materials with high adsorption capacity play a pivotal role in promoting the sustainable treatment of Cd(II)-containing wastewater. In this study, six typical machine learning (ML) models, namely Linear Regression, Random Forest, Gradient Boosting Decision Tree, CatBoost, K-Nearest Neighbors, and Backpropagation Neural Network, were employed to accurately predict the adsorption capacity of Cd(II) onto biochars. A large dataset with 1051 data points was generated using 21 input variables obtained from batch adsorption experiments, including preparation conditions for biochar (2 features), physical properties of biochar (4 features), chemical composition of biochar (9 features), and adsorption experiment conditions (6 features). The rigorous evaluation and comparison of the ML models revealed that the CatBoost model exhibited the highest test R2 value (0.971) and the lowest RMSE (20.54 mg/g), significantly outperforming all other models. The feature importance analysis using Shapley Additive Explanations (SHAP) indicated that biochar chemical compositions had the greatest impact on model predictions of adsorption capacity (42.2 %), followed by adsorption conditions (37.57 %), biochar physical characteristics (12.38 %), and preparation conditions (7.85 %). The optimal experimental conditions optimized by partial dependence plots (PDP) are as follows: as high Cd(II) concentration as possible, C(%) of 33 %, N(%) of 0.3 %, adsorption time of 600 min, pyrolysis time of 50 min, biochar dosage of less than 2 g/L, O(%) of 42 %, biochar pH value of 11.2, and DBE of 1.15. This study unveils novel insights into the adsorption of Cd(II) and provides a comprehensive reference for the sustainable engineering of biochars in Cd(II) wastewater treatment.

Developing and comparing deep learning and machine learning algorithms for osteoporosis risk prediction.

Osteoporosis, characterized by low bone mineral density (BMD), is an increasingly serious public health issue. So far, several traditional regression models and machine learning (ML) algorithms have been proposed for predicting osteoporosis risk. However, these models have shown relatively low accuracy in clinical implementation. Recently proposed deep learning (DL) approaches, such as deep neural network (DNN), which can discover knowledge from complex hidden interactions, offer a new opportunity to improve predictive performance. In this study, we aimed to assess whether DNN can achieve a better performance in osteoporosis risk prediction. By utilizing hip BMD and extensive demographic and routine clinical data of 8,134 subjects with age more than 40 from the Louisiana Osteoporosis Study (LOS), we developed and constructed a novel DNN framework for predicting osteoporosis risk and compared its performance in osteoporosis risk prediction with four conventional ML models, namely random forest (RF), artificial neural network (ANN), k-nearest neighbor (KNN), and support vector machine (SVM), as well as a traditional regression model termed osteoporosis self-assessment tool (OST). Model performance was assessed by area under 'receiver operating curve' (AUC) and accuracy. By using 16 discriminative variables, we observed that the DNN approach achieved the best predictive performance (AUC = 0.848) in classifying osteoporosis (hip BMD T-score ≤ -1.0) and non-osteoporosis risk (hip BMD T-score > -1.0) subjects, compared to the other approaches. Feature importance analysis showed that the top 10 most important variables identified by the DNN model were weight, age, gender, grip strength, height, beer drinking, diastolic pressure, alcohol drinking, smoke years, and economic level. Furthermore, we performed subsampling analysis to assess the effects of varying number of sample size and variables on the predictive performance of these tested models. Notably, we observed that the DNN model performed equally well (AUC = 0.846) even by utilizing only the top 10 most important variables for osteoporosis risk prediction. Meanwhile, the DNN model can still achieve a high predictive performance (AUC = 0.826) when sample size was reduced to 50% of the original dataset. In conclusion, we developed a novel DNN model which was considered to be an effective algorithm for early diagnosis and intervention of osteoporosis in the aging population.

Prediction and explanation of debris flow velocity based on multi-strategy fusion Stacking ensemble learning model

The debris flow velocity fundamentally determines its intensity, thereby rendering its prediction a crucial aspect of disaster prevention and control strategies. However, accurate velocity prediction has consistently posed significant challenges due to the intricate interplay of various influential factors. To address the limitations of existing models, an explainable multi-strategy fusion of Stacking ensemble learning is proposed. Initially, an improved snake optimization (ISO) algorithm is employed to adjust parameters within the model’s learners. Benchmark function comparison tests are then conducted to validate the reliability of the ISO algorithm. Subsequently, a learner selection method based on predictive performance and degree of difference is established to facilitate the selection of basic learners and meta-learners. This leads to the construction of the Stacking ensemble learning model, achieved through the integration of parameter optimization strategies from the improved swarm intelligence algorithm strategy, the error weighting strategy, and the decomposition strategy. To assess the model, a case study of the Jiangjiagou Gulley debris flow is undertaken, focusing on the prediction of the debris flow velocity. The results demonstrate high predictive accuracy, with RMSE, MAE, and MAPE values of 0.19, 0.17, and 2.46% respectively. Furthermore, under the SHAP framework, global and local explanations of the predictions are provided. Through feature importance analysis, the bed slope gradient is identified as the most crucial feature in the velocity prediction of the Jiangjiagou Gulley debris flow. Coupling effects and contributions of input features to the debris flow velocity prediction are further analyzed and explained through feature interaction analysis and single sample analysis. This study not only provides a new method for debris flow velocity prediction but also provides guiding suggestions for debris flow monitoring and control.

Hybridized machine learning models for phosphate pollution modeling in water systems for multiple uses

Phosphate pollution in water bodies is a significant environmental concern, especially in regions with extensive agricultural practices. Hence, a tool for accurately assessing the phosphate concentration is essential. This research paper explores the effectiveness of machine learning (ML) models combined with nature-inspired optimization algorithms for predicting phosphate levels in water systems. The novelty consists of integrating the power of machine learning models, which have been presenting excellent performance at capturing complex relationships in environmental pollution data, with the Harris Hawks Optimizer (HHO) optimization capabilities inspired by hawks' hunting behavior. Four hybrid implementations combining the HHO were evaluated, and four feature subsets were assessed to identify the most influential variables in the modeling process. Using water quality data from Brazilian upstream watersheds, the hybrid models were trained and validated, enabling accurate and robust predictions of phosphate concentrations. The elastic net (EN) model optimized by Harris Hawk Optimizer (HHO-EN) produced the best-averaged performance among all experiments (R = 0.825, R2 = 0.670, Root Mean Square Error (RMSE) = 0.049 mg/L, Mean Absolute Error (MAE) = 0.037 mg/L). A parametric and feature importance analysis identified the most influential parameters in the contamination modeling process. Hybrid machine learning models represent a novel and efficient strategy for water quality monitoring and environmental management, supporting the preservation of aquatic ecosystems.

Unraveling sex differences in Parkinson's disease through explainable machine learning

Sex differences affect Parkinson's disease (PD) development and manifestation. Yet, current PD identification and treatments underuse these distinctions. Sex-focused PD literature often prioritizes prevalence rates over feature importance analysis. However, underlying aspects could make a feature significant for predicting PD, despite its score. Interactions between features require consideration, as do distinctions between scoring disparities and actual feature importance. For instance, a higher score in males for a certain feature doesn't necessarily mean it's less important for characterizing PD in females. This article proposes an explainable Machine Learning (ML) model to elucidate these underlying factors, emphasizing the importance of features. This insight could be critical for personalized medicine, suggesting the need to tailor data collection and analysis for males and females. The model identifies sex-specific differences in PD, aiding in predicting outcomes as “Healthy” or “Pathological”. It adopts a system-level approach, integrating heterogeneous data - clinical, imaging, genetics, and demographics - to study new biomarkers for diagnosis. The explainable ML approach aids non-ML experts in understanding model decisions, fostering trust and facilitating interpretation of complex ML outcomes, thus enhancing usability and translational research. The ML model identifies muscle rigidity, autonomic and cognitive assessments, and family history as key contributors to PD diagnosis, with sex differences noted. The genetic variant SNCA-rs356181 may be more significant in characterizing PD in males. Interaction analysis reveals a greater occurrence of feature interplay among males compared to females. These disparities offer insights into PD pathophysiology and could guide the development of sex-specific diagnostic and therapeutic approaches.

Investigation of machine learning models in predicting compressive strength for ultra-high-performance geopolymer concrete: A comparative study

Ultra-high-performance geopolymer concrete (UHPGC) is a new category of traditional UHPC developed to meet the desire for ultra-high-strength and green building materials. In the current study, random forest (RF), support vector regression (SVR), and extreme gradient boosting (XGB) are used to forecast the compressive strength (CS) of UHPGC. Firstly, the findings of the 113 CS tests available in the previous studies were extracted. Twelve feature variables, including GGBS, silica fume, fly ash, and rice husk ash contents as precursors, the Na2SiO3, NaOH, KOH, and extra water content, polypropylene fiber, steel fiber, liquid-to-binder (L/B) ratio, and curing temperature, were investigated. After analyzing the extracted data, it was found that there were more mixtures of steel fiber-based UHPGCs and synthetic fibers compared to mixtures without fibers. This may reduce the accuracy and comprehensiveness of the predictive models used. To address this issue, several experiments were designed, performed, and tested. Overall, the dataset of 128 CS results was used to develop the machine learning (ML) models. The findings validate the effectiveness of the RF, SVR, and XGB models in accurately predicting the strength of the UHPGC, as constructed by their excellent predictive accuracy (R2 > 0.84). The XGB model performance is superior to the RF and SVR models. The feature importance analysis determined that the steel fiber content and L/B ratio were the top two elements that might profoundly impact the CS. Additionally, NaOH and silica fume also have a positive correlation with CS. Conversely, the extra water and percentage of GGBS exhibit a low correlation with the CS. Through the application of ML models, this study not only ascertains the significance of algorithms including RF, SVR, and XGB in precisely forecasting the CS of UHPGC, but also reveals essential understandings regarding the importance of steel fiber content, L/B ratio, and various other pivotal variables, consequently facilitating the development of refined formulations and improved functionalities in eco-friendly construction materials.

Enhancing machine learning models for total organic carbon prediction by integrating geospatial parameters in river watersheds

This study utilizes machine learning (ML) algorithms to develop a robust total organic carbon (TOC) prediction model for river waters in the Geumho River sub-basins, South Korea, considering both non-rain and rain events. The model incorporates geospatial parameters such as land use, slope, flow rate, and basic water quality metrics including biochemical oxygen demand (BOD), chemical oxygen demand (COD), total nitrogen (TN), total phosphorus (TP), and suspended solids (SS). A key aspect of this research is examining how land use information enhances the model's predictive accuracy. We compared two ML algorithms—extreme gradient boosting (XGBoost) and deep neural networks (DNN)—with a traditional multiple linear regression (MLR) approach. XGBoost outperformed the others, achieving an R2 value between 0.61 and 0.68 in the test dataset and demonstrating significant improvement during rain events with an R2 of 0.77 when including land use data. In contrast, this enhancement was not observed with the MLR model. Feature importance analysis using Shapley values highlighted COD as the primary predictor for non-rain events, while during rain events, COD, TP, TN, SS and agricultural land collectively influenced TOC levels. This study significantly advances understanding of TOC variability across different land use scenarios in river systems and underscores the importance of integrating geospatial and water quality parameters to enhance TOC prediction, particularly during rain events. This methodology provides a valuable framework for developing river management strategies and monitoring long-term TOC trends, especially in scenarios with gaps in essential monitoring data.

QupKake: Integrating Machine Learning and Quantum Chemistry for Micro-pKa Predictions.

Accurate prediction of micro-pKa values is crucial for understanding and modulating the acidity and basicity of organic molecules, with applications in drug discovery, materials science, and environmental chemistry. This work introduces QupKake, a novel method that combines graph neural network models with semiempirical quantum mechanical (QM) features to achieve exceptional accuracy and generalization in micro-pKa prediction. QupKake outperforms state-of-the-art models on a variety of benchmark data sets, with root-mean-square errors between 0.5 and 0.8 pKa units on five external test sets. Feature importance analysis reveals the crucial role of QM features in both the reaction site enumeration and micro-pKa prediction models. QupKake represents a significant advancement in micro-pKa prediction, offering a powerful tool for various applications in chemistry and beyond.

Short-duration prediction of urban storm-water levels using the residual-error ensemble correction technique

ABSTRACT Predicting water levels in urban storm-water sewer systems is a critical study that could provide vital information to help reduce the risk of flooding. This study proposed a new ensemble model based on the integration of a meta-learner model, residual-error corrections, and a multiple-output framework. To achieve the meta-learner model, three multiple-output data-driven-based (MOD) sewer flooding models employing support vector regression (SVR), k-nearest neighbor regression (KNR), and categorical gradient boosting regression (CGBR) techniques were constructed and applied to predict the short-duration evolution of water levels at seven storm-water gauging sites in Taipei city, Taiwan, considering 10-min datasets spanning nearly 6 years (2016–2021). The Bayesian optimization algorithm was utilized in the training phases for all the models to avoid overfitting or underfitting. Enhancing the analysis of feature importance was also conducted to explore model interpretability based on the SHapley Additive exPlanation (SHAP) algorithm. The outputs of storm-water management model (SWMM) were used as benchmark solutions. For the model validation phase, the proposed integrated model improved the lead-time-averaged Nash–Sutcliffe efficiency of single KNR, SVR, and CGBR models by 174.5, 42.4, and 69.4%, respectively, showing that the proposed accurate model could be useful for urban flood warning systems.

Exploring CrossFit performance prediction and analysis via extensive data and machine learning.

The analysis of athletic performance has always aroused great interest from sport scientist. This study utilized machine learning methods to build predictive models using a comprehensive CrossFit (CF) dataset, aiming to reveal valuable insights into the factors influencing performance and emerging trends. Random forest (RF) and multiple linear regression (MLR) were employed to predict performance in four key weightlifting exercises within CF: clean and jerk, snatch, back squat, and deadlift. Performance was evaluated using R-squared (R2) values and mean squared error (MSE). Feature importance analysis was conducted using RF, XGBoost, and AdaBoost models. The RF model excelled in deadlift performance prediction (R2=0.80), while the MLR model demonstrated remarkable accuracy in clean and jerk (R2=0.93). Across exercises, clean and jerk consistently emerged as a crucial predictor. The feature importance analysis revealed intricate relationships among exercises, with gender significantly impacting deadlift performance. This research advances our understanding of performance prediction in CF through machine learning techniques. It provides actionable insights for practitioners, optimize performance, and demonstrates the potential for future advancements in data-driven sports analytics.

Workplace access, burnout, and prescription drug misuse among Korean hospital nurses: a cross-sectional study

BackgroundPrescription drug misuse has been identified as a global issue of concern. Nurses’ prescription drug misuse is linked to personal health problems and impaired nursing care. This study explored the level of South Korean hospital nurses’ prescription drug misuse and examined associations with workplace access and burnout.MethodsThis cross-sectional study used data from 1142 nurses working in South Korean hospital settings. Nurses completed the online survey anonymously. Descriptive analysis, logistic regression, and Shapley value assessment were conducted.ResultsPain relievers (44.2%), IV drips (26.8%), and antibiotics (13.5%) were the most commonly misused drugs among hospital nurses. Accessibility in the workplace was high, with nurses reporting frequent administration of IV drips, pain relievers, and antibiotics and perceiving these drugs as easily accessible. Logistic regression demonstrated that perceived availability was significantly related to misuse across all drug types. Burnout was associated with IV drips, sleeping pills, and steroids, increasing the likelihood of misuse. Shapley feature importance analysis highlighted perceived availability as the most influential factor for IV drips, pain relievers, and steroids, while burnout emerged as crucial for antibiotics and sleeping pills. Notably, age played a significant role in appetite suppressant misuse, distinguishing it from other drugs.ConclusionOur results revealed that workplace access and burnout are associated with nurses’ prescription drug misuse. Effective educational strategies are essential for enhancing nurses’ willingness to seek help for personal health issues. Hospital organizations play a crucial role in facilitating access to healthcare and fostering a supportive environment for nurses to seek treatment when necessary. Additionally, governmental policies should prioritize the implementation of active surveillance systems to monitor medication usage in healthcare settings, thereby mitigating drug misuse among healthcare professionals. By addressing these issues, we can ensure the well-being of nurses and promote a safer healthcare environment.

Applications of deep learning models in precision prediction of survival rates for heart failure patients.

Heart failure poses a significant challenge in the global health domain, and accurate prediction of mortality is crucial for devising effective treatment plans. In this study, we employed a Seq2Seq model from deep learning, integrating 12 patient features. By finely modeling continuous medical records, we successfully enhanced the accuracy of mortality prediction. The objective of this research was to leverage the Seq2Seq model in conjunction with patient features for precise mortality prediction in heart failure cases, surpassing the performance of traditional machine learning methods. The study utilized a Seq2Seq model in deep learning, incorporating 12 patient features, to intricately model continuous medical records. The experimental design aimed to compare the performance of Seq2Seq with traditional machine learning methods in predicting mortality rates. The experimental results demonstrated that the Seq2Seq model outperformed conventional machine learning methods in terms of predictive accuracy. Feature importance analysis provided critical patient risk factors, offering robust support for formulating personalized treatment plans. This research sheds light on the significant applications of deep learning, specifically the Seq2Seq model, in enhancing the precision of mortality prediction in heart failure cases. The findings present a valuable direction for the application of deep learning in the medical field and provide crucial insights for future research and clinical practices.

A secure and light‐weight patient survival prediction in Internet of Medical Things framework

SummaryThoracic surgeries in major lung resections for primary lung cancer are fraught with potential risks, emphasising the need to understand factors contributing to postoperative mortality. This study investigates the interplay of objective and subjective data in predicting postoperative outcomes to reduce data transmission costs in the Internet of Medical Things (IoMT). Objective metrics, such as forced vital capacity (FVC), offer consistent, quantifiable insights essential for predictive modelling. Conversely, subjective data derived from patient self‐reports suggest that the patient's personal experiences are crucial for assessing the quality of life postsurgery. Utilising a dataset from the University of California, Irvine's Machine Learning Repository (UCI), 17 distinct attributes were examined. Using ensemble learning classifiers, the extra trees classifier is superior when utilising all features, achieving an accuracy of 0.92. Combining select subjective features, specifically PRE6, PRE8 and AGE (demographic), with objective data, yielded a comparable accuracy of 0.91. Feature importance analysis further highlights the significance of features like PRE5, PRE4 and AGE. This suggests potential redundancies in the full feature set, emphasising the importance of feature selection. Importantly, when compared with existing literature, this study's findings offer insights into the future of predictive modelling in thoracic surgeries, with implications for the rapidly evolving field of the IoMT.

Machine Learning-powered 28-day Mortality Prediction Model for Hospitalized Patients with Acute Decompensation of Liver Cirrhosis.

Chronic liver disease and cirrhosis are persistent global health threats, ranking among the top causes of death. Despite medical advancements, their mortality rates have remained stagnant for decades. Existing scoring systems such as Child-Turcotte-Pugh and Mayo End-Stage Liver Disease have limitations, prompting the exploration of more accurate predictive methods using artificial intelligence and machine learning (ML). We retrospectively reviewed the data of all adult patients with acute decompensated liver cirrhosis admitted to a tertiary hospital during 2015-2021. The dataset underwent preprocessing to handle missing values and standardize continuous features. Traditional ML and deep learning algorithms were applied to build a 28-day mortality prediction model. The subjects were 173 cirrhosis patients, whose medical records were examined. We developed and evaluated multiple models for 28-day mortality prediction. Among traditional ML algorithms, logistic regression outperformed was achieving an accuracy of 82.9%, precision of 55.6%, recall of 71.4%, and an F1-score of 0.625. Naive Bayes and Random Forest models also performed well, both achieving the same accuracy (82.9%) and precision (54.5%). The deep learning models (multilayer artificial neural network, recurrent neural network, and Long Short-Term Memory) exhibited mixed results, with the multilayer artificial neural network achieving an accuracy of 74.3% but lower precision and recall. The feature importance analysis identified key predictability contributors, including admission in the intensive care unit (importance: 0.112), use of mechanical ventilation (importance: 0.095), and mean arterial pressure (importance: 0.073). Our study demonstrates the potential of ML in predicting 28-day mortality following hospitalization with acute decompensation of liver cirrhosis. Logistic Regression, Naive Bayes, and Random Forest models proved effective, while deep learning models exhibited variable performance. These models can serve as useful tools for risk stratification and timely intervention. Implementing these models in clinical practice has the potential to improve patient outcomes and resource allocation.

Machine Learning from Veno-Venous Extracorporeal Membrane Oxygenation Identifies Factors Associated with Neurological Outcomes.

Neurological complications are common in patients receiving veno-venous extracorporeal membrane oxygenation (VV-ECMO) support. We used machine learning (ML) algorithms to identify predictors for neurological outcomes for these patients. All demographic, clinical, and circuit-related variables were extracted for adults with VV-ECMO support at a tertiary care center from 2016 to 2022. The primary outcome was good neurological outcome (GNO) at discharge defined as a modified Rankin Scale of 0-3. Of 99 total VV-ECMO patients (median age = 48years; 65% male), 37% had a GNO. The best performing ML model achieved an area under the receiver operating characteristic curve of 0.87. Feature importance analysis identified down-trending gas/sweep/blender flow, FiO2, and pump speed as the most salient features for predicting GNO. Utilizing pre- as well as post-initiation variables, ML identified on-ECMO physiologic and pulmonary conditions that best predicted neurological outcomes.