Performance Of Prediction Models Research Articles

Rapid quantitative analysis of coal composition using laser-induced breakdown spectroscopy coupled with random forest algorithm.

Coal is the primary energy source in China, widely used in energy production, industrial processes, and chemical engineering. Due to the complexity and diversity of coal quality, there is an urgent need for new technologies to achieve rapid and accurate detection and analysis of coal, aiming to improve coal resource utilization and reduce pollutant emissions. This study proposes a rapid quantitative analysis of coal using laser-induced breakdown spectroscopy combined with the random forest algorithm. Firstly, a Q-switched Nd: YAG laser at 1064nm was employed to ablate coal samples, generating plasma, and spectral data were collected using a spectrometer. Secondly, the study explores the impact of different parameters in the preprocessing method (wavelet transform) on the predictive performance of the random forest model. It identifies elements related to coal ash content and calorific value along with their spectral information. Subsequently, to further validate the predictive performance of the model, a comparison is made with models established using support vector machine, artificial neural network, and partial least squares. Finally, under optimal parameters for spectral information preprocessing (wavelet transform with Db4 as the base function and 3 decomposition levels), a model combining wavelet transform with Random Forest is established to predict and analyze the ash content and calorific value of coal. The results demonstrate that the Wavelet Transform-Random Forest model exhibits excellent predictive performance (coal ash content: R2 = 0.9470, RMSECV = 4.8594, RMSEP = 4.8450; coal calorific value: R2 = 0.9485, RMSECV = 1.5996, RMSEP = 1.5949). Therefore, laser-induced breakdown spectroscopy combined with the random forest algorithm is an effective method for rapid and accurate detection and analysis of coal. The predicted coal composition values show high accuracy, providing insights and methods for coal composition monitoring and analysis.

Predicting treatment response in multicenter non-small cell lung cancer patients based on federated learning

BackgroundMulticenter non-small cell lung cancer (NSCLC) patient data is information-rich. However, its direct integration becomes exceptionally challenging due to constraints involving different healthcare organizations and regulations. Traditional centralized machine learning methods require centralizing these sensitive medical data for training, posing risks of patient privacy leakage and data security issues. In this context, federated learning (FL) has attracted much attention as a distributed machine learning framework. It effectively addresses this contradiction by preserving data locally, conducting local model training, and aggregating model parameters. This approach enables the utilization of multicenter data with maximum benefit while ensuring privacy safeguards. Based on pre-radiotherapy planning target volume images of NSCLC patients, a multicenter treatment response prediction model is designed by FL for predicting the probability of remission of NSCLC patients. This approach ensures medical data privacy, high prediction accuracy and computing efficiency, offering valuable insights for clinical decision-making.MethodsWe retrospectively collected CT images from 245 NSCLC patients undergoing chemotherapy and radiotherapy (CRT) in four Chinese hospitals. In a simulation environment, we compared the performance of the centralized deep learning (DL) model with that of the FL model using data from two sites. Additionally, due to the unavailability of data from one hospital, we established a real-world FL model using data from three sites. Assessments were conducted using measures such as accuracy, receiver operating characteristic curve, and confusion matrices.ResultsThe model’s prediction performance obtained using FL methods outperforms that of traditional centralized learning methods. In the comparative experiment, the DL model achieves an AUC of 0.718/0.695, while the FL model demonstrates an AUC of 0.725/0.689, with real-world FL model achieving an AUC of 0.698/0.672.ConclusionsWe demonstrate that the performance of a FL predictive model, developed by combining convolutional neural networks (CNNs) with data from multiple medical centers, is comparable to that of a traditional DL model obtained through centralized training. It can efficiently predict CRT treatment response in NSCLC patients while preserving privacy.

Optimizing clinico-genomic disease prediction across ancestries: a machine learning strategy with Pareto improvement

BackgroundAccurate prediction of an individual’s predisposition to diseases is vital for preventive medicine and early intervention. Various statistical and machine learning models have been developed for disease prediction using clinico-genomic data. However, the accuracy of clinico-genomic prediction of diseases may vary significantly across ancestry groups due to their unequal representation in clinical genomic datasets.MethodsWe introduced a deep transfer learning approach to improve the performance of clinico-genomic prediction models for data-disadvantaged ancestry groups. We conducted machine learning experiments on multi-ancestral genomic datasets of lung cancer, prostate cancer, and Alzheimer’s disease, as well as on synthetic datasets with built-in data inequality and distribution shifts across ancestry groups.ResultsDeep transfer learning significantly improved disease prediction accuracy for data-disadvantaged populations in our multi-ancestral machine learning experiments. In contrast, transfer learning based on linear frameworks did not achieve comparable improvements for these data-disadvantaged populations.ConclusionsThis study shows that deep transfer learning can enhance fairness in multi-ancestral machine learning by improving prediction accuracy for data-disadvantaged populations without compromising prediction accuracy for other populations, thus providing a Pareto improvement towards equitable clinico-genomic prediction of diseases.

Multifractal Characteristics and Displacement Prediction of Deformation on Tunnel Portal Slope of Shallow Buried Tunnel Adjacent to Important Structures

The tunnel portal section is often in extremely weak and fragmented strata, and the deformation of the portal side and slope will affect the stability of the surrounding rock and the tunnel-supporting structure. However, the deformation characteristics and displacement development patterns of slopes in the tunnel portal section are not clear. In this paper, the multifractal characteristics and displacement prediction of the deformation sequence of the tunnel portal slope at of a weak and water-rich shallow buried tunnel adjacent to an important structure are studied in depth. Combined with the deformation characteristics of the tunnel portal slope, a suitable slope monitoring and measurement scheme is designed to analyze the deformation pattern of the tunnel portal slope. Based on the multifractal detrended fluctuation analysis (MF-DFA) method, the multifractal characteristics of the deformation monitoring sequences at each monitoring point of the tunnel portal slope are analyzed. The multifractal characteristics of displacement sequences at different monitoring points of the tunnel portal slope are consistent with the actual monitoring results. Furthermore, the Long Short-Term Memory (LSTM) model is optimized using the Particle Swarm Optimization (PSO) algorithm to predict the deformation of the tunnel portal slope. The results show that the maximum mean square error (MSE) of the horizontal displacement test set prediction results is 0.142, and the coefficient of determination (R2) is higher than 91%. The maximum value of MSE for vertical displacement test set prediction is 0.069, and the R2 are higher than 91%. The study shows that the performance of the PSO-LSTM prediction model can meet the requirements for predicting the displacement of the tunnel portal slope. Based on the MF-DFA method and PSO-LSTM prediction model, the fluctuation characteristics of the displacement value of the tunnel portal section can be accurately identified and the displacement development pattern can be effectively predicted. The conclusions of the study are of great practical significance for the safe construction of the tunnel portal section.

Enabling active visitor management: local, short-term occupancy prediction at a touristic point of interest

AbstractAfter the temporary shock of the Covid-19 pandemic, the rapid recovery and resumed growth of the tourism sectors accelerates unsustainable tourism, resulting in local (over-)crowding, environmental damage, increased emissions, and diminished tourism acceptance. Addressing these challenges requires an active visitor management system at points of interest (POI), which requires local and timely POI-specific occupancy predictions to predict and mitigate crowding. Therefore, we present a new approach to measure visitor movement at an open-spaced, and freely accessible POI and evaluate the prediction performance of multiple occupancy and visitor count machine learning prediction models. We analyze multiple case combinations regarding spatial granularity, time granularity, and prediction time horizons. With an analysis of the SHAP values we determine the influence of the most important features on the prediction and extract transferable knowledge for similar regions lacking visitor movement data. The results underline that POI-specific prediction is achievable with a moderate relation for occupancy prediction and a strong relation for visitor count prediction. Across all cases, XGBoost and Random Forest outperform other models, with prediction accuracy increasing as the prediction time horizon shortens. For effective active visitor management, combining multiple models with different spatial aggregations and prediction time horizons provides the best information basis to identify appropriate steering measures. This innovative application of digital technologies facilitates information exchange between destination management organizations and tourists, promoting sustainable destination development and enhancing tourism experience.

AI-based approach to enable proactive identification of early lung cancer: A retrospective population health study and economic model.

10536 Background: Approximately 48,500 new cases of lung cancer are diagnosed each year in the UK with a five-year survival rate which lags behind other European countries. Most patients are diagnosed at stage III or IV disease, which has a significantly worse prognosis. The integration of real-world data and AI techniques presents a promising avenue for improvement, using novel solutions not achievable through traditional care models. This study explores proactive patient identification at scale, aligning with the "left shift" paradigm in population health. The approach uses innovative technologies and novel data sources to identify patients with early-stage disease, enhancing both clinical and economic outcomes. Importantly, these interventions should be feasible on a large scale without overburdening already stretched clinical services. Methods: A retrospective study of real-world data for an urban population with high socioeconomic disadvantage and health needs was undertaken. Two endpoints were determined: detection potential and health economic impact. For detection potential (DP), patients aged 40 and above who underwent chest X-rays (CXR) at a large academic medical centre between 2016-2022 were included. Natural Language Processing (NLP) analysed unstructured free-text data from Electronic Health Records, combined with ICD-10 codes, for predictive model (PM) development using Machine Learning (ML) classification techniques. The PM's performance was assessed using the area under the receiver operating characteristic curve (AUCROC). Health economic impact was evaluated through a theoretical extrapolation of DP. Assuming a 9% left shift from late (stage III or IV) to early (stage I or II) disease, a health economic model (HEM) compared the impact of the current care model to a potential future state facilitated by the digital medicine care model (DMCM). Results: 75,342 patients were included, with 755 lung cancer diagnoses occurring within 6 months of CXR. The PM achieved a peak AUROC of 0.75, exceeding the performance of currently used risk prediction models (Q Cancer Risk, Risk Assessment Tools). A 9% left shift equated to 4.5 years mean increase in survival. The DMCM was associated with a net healthcare cost benefit of £6k per patient diagnosed with lung cancer, improved quality-adjusted life years of £21M for the year 1 cohort over 10 years and increased economic productivity (from additional years in workforce) of £2.4M. Conclusions: The study demonstrates the practical utility of AI-supported early lung cancer detection at population scale. Real-world evidence is needed to prospectively validate the model as cost-effective, scalable, and efficient. Future work will seek to improve the model's performance against a broader at-risk patient population and explore its applicability to other late-presenting tumours.

Clinical T1/2 renal cell carcinoma: multiparametric dynamic contrast-enhanced MRI features-based model for the prediction of individual adverse pathology

BackgroundThe detection of renal cell carcinoma (RCC) has been rising due to the enhanced utilization of cross-sectional imaging and incidentally discovered lesions with adverse pathology demonstrate potential for metastasis. The purpose of our study was to determine the clinical and multiparametric dynamic contrast-enhanced magnetic resonance imaging (CEMRI) associated independent predictors of adverse pathology for cT1/2 RCC and develop the predictive model.MethodsWe recruited 105 cT1/2 RCC patients between 2018 and 2022, all of whom underwent preoperative CEMRI and had complete clinicopathological data. Adverse pathology was defined as RCC patients with nuclear grade III-IV; pT3a upstage; type II papillary RCC, collecting duct or renal medullary carcinoma, unclassified RCC; sarcomatoid/rhabdoid features. The qualitative and quantitative CEMRI parameters were independently reviewed by two radiologists. Univariate and multivariate binary logistic regression analyses were utilized to determine the independent predictors of adverse pathology for cT1/2 RCC and construct the predictive model. The receiver operating characteristic (ROC) curve, confusion matrix, calibration plot, and decision curve analysis (DCA) were conducted to compare the diagnostic performance of different predictive models. The individual risk scores and linear predicted probabilities were calculated for risk stratification, and the Kaplan–Meier curve and log-rank tests were used for survival analysis.ResultsOverall, 45 patients were pathologically confirmed as RCC with adverse pathology. Clinical characteristics, including gender, and CEMRI parameters, including RENAL score, tumor margin irregularity, necrosis, and tumor apparent diffusion coefficient (ADC) value were identified as independent predictors of adverse pathology for cT1/2 RCC. The clinical-CEMRI predictive model yielded an area under the curve (AUC) of the ROC curve of 0.907, which outperformed the clinical model or CEMRI signature model alone. Good calibration, better clinical usefulness, excellent risk stratification ability of adverse pathology and prognosis were also achieved for the clinical-CEMRI predictive model.ConclusionsThe proposed clinical-CEMRI predictive model offers the potential for preoperative prediction of adverse pathology for cT1/2 RCC. With the ability to forecast adverse pathology, the predictive model could significantly benefit patients and clinicians alike by providing enhanced guidance for treatment planning and decision-making.

Independent validation of 30-day emergency department visit after initial chemotherapy A-PROACCT model.

e23243 Background: This study assessed model performance and, more importantly, the clinical usefulness of the PROACCT model created by Grant et al. (JAMA, 2019), refined as A-PROACCT by Stein et al. (JOP, 2023) with data from Orlando Health. The predictive performance of models anticipating acute care events within thirty days (ACE30) of initial chemotherapy administration (ICA) were evaluated. Methods: Stein et al. models classified high-risk patients as those whose model output risk score was at or above the 90th percentile. They studied 5153 ICAs between July and November of 2021, of which 559 (10.8%) had an ACE30. Their first model used drug category, age group, ED visits and hospitalizations in the prior year, and cancer type, and had sensitivity (Sen), specificity, (Spec), C-Statistic (C-Stat), and positive predictive value (PPV) of 0.28, 0.93, 0.79 and 0.18 respectively. Their second model added insurance category, number of anti-cancer agents, race, and BMI category and reported Sen, Spec, C-Stat, and PPV of 0.30, 0.94, 0.80 and 0.23 respectively. Of the high-risk patients, 33.6% and 34.1% had an ACE30 respectively. Of the low-risk patients, 8.3% and 8.1% had an ACE30 respectively. Our data were propagated through the two models. Results: Between February of 2012 and April 2021, Orlando Health had 15,483 distinct patient visits with ICAs. Overall, 2046 (13.2%) had an ACE30. Model one had Sen, Spec, C-Stat, and PPV of 0.19, 0.92, 0.65 and 0.27 respectively. Model two had Sen, Spec, C-Stat, and PPV of 0.19, 0.91, 0.65 and 0.25 respectively. Both models had an accuracy of 0.82. This demonstrated our data fit the Stein model. Of the 1460 ICAs identified as high-risk by model one, 387 (26.5%) had an ACE30, and of the 14,023 ICAs identified as low-risk, 1659 (11.8%) had an ACE30. Of the 1549 ICAs identified as high-risk by model two, 393 (25.4%) had an ACE30, and of the 13,934 ICAs identified as low-risk, 1653 (11.9%) had an ACE30. The difference of ACE30 rates between high risk and low risk groups were statistically significant (p < 0.0001) in both models. This demonstrated both models were clinically useful in identifying chemotherapy patients at high risk of acute care events. Conclusions: These models allowed for clinically relevant stratification of patients into meaningful high-risk and low-risk groups (26.5% vs 11.8% with our data) for risk of acute care events within 30 days of initial chemotherapy administration. Nevertheless, only 26.5% of these high-risk patients actually had an acute care event within thirty days of initial chemotherapy administration. Further research is needed to improve the precision of identifying high-risk patients, by potentially adding more clinical variables, using alternative sampling methods, and employing nonlinear and more complex machine learning methodologies.

A genetic algorithm-based feature selection approach for diabetes prediction

<p>Genetic algorithms have emerged as a powerful optimization technique for feature selection due to their ability to search through a vast feature space efficiently. This study discusses the importance of feature selection for prediction in healthcare and prominently focuses on diabetes mellitus. Feature selection is essential for improving the performance of prediction models, by finding significant features and removing unnecessary among them. The study aims to identify the most informative subset of features. Diabetes is a chronic metabolic disorder that poses significant health challenges worldwide. For the experiment, two datasets related to diabetes were downloaded from Kaggle and the results of both (datasets) with and without feature selection using the genetic algorithm were compared. Machine learning classifiers and genetic algorithms were combined to increase the precision of diabetes risk prediction. In the preprocessing phase, feature selection, machine learning classifiers, and performance metrics methods were applied to make this study feasible. The results of the experiment showed that genetic algorithm + logistic regression i.e., 80% (accuracy) works better for PIMA diabetes, and for Germany diabetes dataset genetic algorithm + random forest and genetic algorithm + K-Nearest Neighbor i.e., 98.5% performed better than other chosen classifiers. The researchers can better comprehend the importance of feature selection in healthcare through this study.</p>

MPSA: Multi-Position Supervised Soft Attention-based convolutional neural network for histopathological image classification

In recent years, significant achievements have been made in the field of histopathological image analysis using convolutional neural networks (CNNs). However, existing CNNs fail to fully capture the important local structures and regional information in histopathological images due to the complex tissue structures and variable pathological features present in these images. They often treat all regions equally, which further exacerbates the challenge of accurately analyzing such images. Current network model can’t extract deep layer features efficiently without guiding. To alleviate this problem, we propose a novel network model called Multi-Position Supervised Soft Attention (MPSA). MPSA adds regions of interest (RoI) labels at multiple feature layers for deep supervision, and then uses the supervised layers as soft attention to guide the learning of the classification network, enabling the network to accurately extract features of the lesion target. Additionally, we design a Multi-level Attention Feature Enhancement Module (MAFEM), which combines multiple levels of attention mechanisms to enhance the performance of the convolutional neural network in histopathological image classification. MAFEM includes spatial attention, soft attention of the main branch, and our proposed soft attention for multi-branch feature fusion. The proposed soft attention for multi-branch feature fusion aims to enhance the predictive performance of the classification model by activating relevant neurons in the diagnostic area in a highly activated state, while effectively avoiding noise activation. This innovative approach ensures that the model can focus on the most pertinent information, leading to improved classification accuracy. We conducted classification experiments on the liver cancer histopathological images dataset and the results showed that our method achieved a classification accuracy of 95.79%, indicating that it is very effective in the analysis of liver histopathological images. Our proposed network architecture has also demonstrated good generalization ability in other medical datasets, achieving a classification accuracy of 84.41% on the ultrasound carotid plaque dataset.

Tracking COPD exacerbation patterns and forecasting readmission risks utilizing electronic medical records

IntroductionAccurate evaluation of exacerbation frequency is an essential part of COPD assessment. But relying on just the prior-year exacerbation history may not capture the full picture of risk given the inherent year-to-year fluctuations in exacerbation rates. This study aimed to evaluate the predictive performance of models incorporating the 3-year exacerbation history based on electronic medical record. Materials and MethodsThis retrospective cohort study included 86,501 COPD hospitalized patients in Beijing from 2008 to 2014. The annual frequency of COPD exacerbation hospitalizations over a 3-year period after the index hospitalization was calculated, with patients segmented into seven distinct exacerbation trajectory groups. Logistic regression was used to evaluate the predictive capability of the 3-year exacerbation history for exacerbation readmission in the fourth year. Predictors included age, sex, comorbidities, and exacerbation hospitalization in previous 1–3 years. Model performance was evaluated using area under the receiver operating characteristic curve (AUC). ResultsOf the studied patients, 56.5% were men, and the mean age (SD) was 73.8 (10.3) years. The overall readmission rate for COPD exacerbation was 0.31 per person-year, with only 3.8% of patients persistently readmitted over three consecutive years. The 3-year trajectory of exacerbation frequency was associated with exacerbation risk in the fourth year. Compared to just the prior year, the inclusion of a 3-year exacerbation hospitalization history notably improved prediction accuracy, with AUC elevating from 0.731 (0.724–0.739) to 0.786 (0.779–0.792). ConclusionThese results unveil the fluctuating nature of COPD exacerbation hospitalization frequency across years and demonstrate that integrating a more comprehensive 3-year exacerbation history significantly refines the prediction model for future risk, thus providing a more nuanced and actionable insight for clinical care.

A deep learning approach for acute liver failure prediction with combined fully connected and convolutional neural networks.

Acute Liver Failure (ALF) is a critical medical condition with rapid development, often caused by viral infections, hepatotoxic drug abuse, or other severe liver diseases. Timely and accurate prediction of ALF occurrence is clinically crucial. However, predicting ALF poses challenges due to the diverse physiological differences among patients and the dynamic nature of the disease. This study introduces a deep learning approach that combines fully connected and convolutional neural networks for effective ALF prediction. The goal is to overcome limitations of traditional machine learning methods and enhance predictive model performance and generalization. The proposed model integrates a fully connected neural network for handling basic patient features and a convolutional neural network dedicated to capturing temporal patterns in patient data. The combination allows automatic learning of complex patterns and abstract features present in highly dynamic medical data associated with ALF. The model's effectiveness is demonstrated through comprehensive experiments and performance evaluations. It outperforms traditional machine learning methods, achieving 94.8% accuracy and superior generalization capabilities. The study highlights the potential of deep learning in ALF prediction, emphasizing the importance of considering individualized medical factors. Future research should focus on improving model robustness, addressing imbalanced data, and further exploring personalized features for enhanced predictive accuracy in real-world clinical scenarios.

Improving electricity demand forecasting accuracy: a novel grey-genetic programming approach using GMC(1,N) and residual sign estimation

PurposeThis paper addresses the challenges associated with forecasting electricity consumption using limited data without making prior assumptions on normality. The study aims to enhance the predictive performance of grey models by proposing a novel grey multivariate convolution model incorporating residual modification and residual genetic programming sign estimation.Design/methodology/approachThe research begins by constructing a novel grey multivariate convolution model and demonstrates the utilization of genetic programming to enhance prediction accuracy by exploiting the signs of forecast residuals. Various statistical criteria are employed to assess the predictive performance of the proposed model. The validation process involves applying the model to real datasets spanning from 2001 to 2019 for forecasting annual electricity consumption in Cameroon.FindingsThe novel hybrid model outperforms both grey and non-grey models in forecasting annual electricity consumption. The model's performance is evaluated using MAE, MSD, RMSE, and R2, yielding values of 0.014, 101.01, 10.05, and 99% respectively. Results from validation cases and real-world scenarios demonstrate the feasibility and effectiveness of the proposed model. The combination of genetic programming and grey convolution model offers a significant improvement over competing models. Notably, the dynamic adaptability of genetic programming enhances the model's accuracy by mimicking expert systems' knowledge and decision-making, allowing for the identification of subtle changes in electricity demand patterns.Originality/valueThis paper introduces a novel grey multivariate convolution model that incorporates residual modification and genetic programming sign estimation. The application of genetic programming to enhance prediction accuracy by leveraging forecast residuals represents a unique approach. The study showcases the superiority of the proposed model over existing grey and non-grey models, emphasizing its adaptability and expert-like ability to learn and refine forecasting rules dynamically. The potential extension of the model to other forecasting fields is also highlighted, indicating its versatility and applicability beyond electricity consumption prediction in Cameroon.

Artificial neural network modelling of natural gas dehydration process

A multi-input–multi-output artificial neuron network (MIMO-ANN) model has been developed for process monitoring and improvement on a natural gas glycol dehydration process. The MIMO-ANN model was based on a steady-state process simulation model constructed in commercial software Aspen HYSYS. A set of training data was generated with the converged simulation model for the training of the MIMO-ANN model in Python. The process input of the model includes lean glycol recirculation rate and purity along with wet gas inlet pressure. On the other hand, the process output considered includes dehydrated gas water and aromatics content, dehydrated gas hydrate formation temperature and water dew point, stripping gas flow rate as well as reboiler duty. The overall mean squared error (MSE) of the MIMO-ANN model was calculated as 1.79. The best-fit line that highly overlaps with a 45° diagonal line is constructed with a correlation coefficient (R2) score of more than 0.999 for all studied process output testing datasets. The mean absolute percentage error (MAPE) of the predicted outputs is generally less than 1%, except for dehydrated gas water dew point (16.63%) and stripping gas flow rate (11.55%), due to error predictions attempted on pseudo-zero values. This successfully displays an exceptional predictive performance of the MIMO-ANN model developed in this work which can be further deployed as an online dashboard for real-time monitoring tool.

Population pharmacokinetics and dosage optimization of linezolid in Chinese older patients.

To assess the pharmacokinetics and pharmacodynamics of linezolid in a retrospective cohort of hospitalized Chinese older patients. Patients > 60years of age, who received intravenous linezolid (600mg), were included. A population pharmacokinetics (PPK) model was established using nonlinear mixed-effects modeling. The predictive performance of the final model was assessed using goodness-of-fit plots, bootstrap analyses, and visual predictive checks. Monte Carlo simulations were used to evaluate the achievement of a pharmacodynamics target for the area under the serum concentration-time curve/minimum inhibitory concentration (AUC0-24/MIC). A total of 210 samples were collected from 120 patients. A one-compartment PPK model with linear elimination best predicted the linezolid plasma concentrations. Linezolid clearance (CL) was 4.22 L h-1 and volume of distribution (Vd) was 45.80 L; serum uric acid (SUA) was a significant covariate of CL. The results of this study indicated that the standard dose was associated with a risk of overexposure in older patients, particularly those with high SUA values; these patients would benefit from a lower dose (300mg every 12h).

Machine learning for development of ECG-based prediction models for atrial fibrillation, thromboembolism and mortality in patients with heart failure

Abstract Background Heart failure (HF) is a worldwide epidemic affecting more than 64 million people. Given its massive burden both socially and economically, early identification of individuals at risk for complications is important. Purpose To test the predictive performance of non-invasive models from 12-lead digitalized electrocardiography (ECG) and cycle ergometer for adverse outcome in individuals with heart failure. Methods 598 Patients scheduled for right heart catheterization due to unexplained dyspnea were screened. All ECG recorded at admission were further analysed using the software "ECG-Precision-Analysis V-1-2023" to measure the duration of amplified p-wave (APWD at 175mm/sec, 80mm/mV) and to confirm the diagnosis of inter-atrial block. Ergometer was performed in a supine position with increasing intensity adapted to individual capacity (25 or 50W). The maximum physical capacity (Watt-max) was determined. Follow-up (FU) was scheduled at 12-,36-, 60- and 120-month after discharge with the primary endpoint as MACCE (a composite of new onset of AF, ischemic stroke and peripheral embolism) in at baseline event-free patients and the secondary endpoint as all-cause mortality in the total cohort. Three machine learning (ML) algorithms (COX regression, BSR and LASSO regression) were used to select candidate variables and establish predictive models for MACCE and all-cause mortality. Model evaluation was performed in terms of discriminatory power using C-statistics for each endpoint. Results Prediction models developed using machine learning (ML) algorithms (COX regression, BSR and LASSO regression) identified Watt-max and/or APWD as core components for MACCE and mortality prediction . APWD and Watt-max were key components of ML models for prediction of thromboembolic risk and AF (MACCE) and reached accurate diagnostic performances (C-statistics up to 0.74), moreover, another model consisted of APWD and aIAB (advanced inter-atrial block) was established to serve as a reference to ML models with moderate performance for MACCE (mean C-statistics <0.70). All ML models for prediction of all-cause mortality included Watt-max and achieved C-statistics > 0.75. Conclusions In patients with heart failure, the combination of APWD, aIAB and maximum excercice capacity allow to identify patients at risk for AF and stroke, whereas mortality is predominantly predicted by maximum exercise capacity but not APWD.Kaplan-Meier curves for MACCE/MortalitytimeROC curves for MACCE/Mortality

Using honey bee flight activity data and a deep learning model as a toxicovigilance tool

Automatic monitoring devices placed at the entrances of honey bee hives have facilitated the detection of various sublethal effects related to pesticide exposure, such as homing failure and reduced flight activity. These devices have further demonstrated that different neurotoxic pesticide molecules produce similar sublethal impacts on flight activity. The detection of these effects was conducted a posteriori, following the recording of flight activity data. This study introduces a method using an artificial intelligence model, specifically a recurrent neural network, to detect the sublethal effects of pesticides in real-time based on honey bee flight activity. This model was trained on a flight activity dataset comprising 42,092 flight records from 1107 control and 1689 pesticide-exposed bees. The model was able to classify honey bees as healthy or pesticide-exposed based on the number of flights and minutes spent foraging per day. The model was the least accurate (68.46%) when only five days of records per bee were used for training. However, the highest classification accuracy of 99%, a Cohen Kappa of 0.9766, a precision of 0.99, a recall of 0.99, and an F1-score of 0.99 was achieved with the model trained on 25 days of activity data, signifying near-perfect classification ability. These results underscore the highly predictive performance of AI models for toxicovigilance and highlight the potential of our approach for real-time and cost-effective monitoring of risks due to exposure to neurotoxic pesticide in honey bee populations.

A deep learning-based radiomics model for predicting lymph node status from lung adenocarcinoma

ObjectivesAt present, there are many limitations in the evaluation of lymph node metastasis of lung adenocarcinoma. Currently, there is a demand for a safe and accurate method to predict lymph node metastasis of lung cancer. In this study, radiomics was used to accurately predict the lymph node status of lung adenocarcinoma patients based on contrast-enhanced CT.MethodsA total of 503 cases that fulfilled the analysis requirements were gathered from two distinct hospitals. Among these, 287 patients exhibited lymph node metastasis (LNM +) while 216 patients were confirmed to be without lymph node metastasis (LNM-). Using both traditional and deep learning methods, 22,318 features were extracted from the segmented images of each patient's enhanced CT. Then, the spearman test and the least absolute shrinkage and selection operator were used to effectively reduce the dimension of the feature data, enabling us to focus on the most pertinent features and enhance the overall analysis. Finally, the classification model of lung adenocarcinoma lymph node metastasis was constructed by machine learning algorithm. The Accuracy, AUC, Specificity, Precision, Recall and F1 were used to evaluate the efficiency of the model.ResultsBy incorporating a comprehensively selected set of features, the extreme gradient boosting method (XGBoost) effectively distinguished the status of lymph nodes in patients with lung adenocarcinoma. The Accuracy, AUC, Specificity, Precision, Recall and F1 of the prediction model performance on the external test set were 0.765, 0.845, 0.705, 0.784, 0.811 and 0.797, respectively. Moreover, the decision curve analysis, calibration curve and confusion matrix of the model on the external test set all indicated the stability and accuracy of the model.ConclusionsLeveraging enhanced CT images, our study introduces a noninvasive classification prediction model based on the extreme gradient boosting method. This approach exhibits remarkable precision in identifying the lymph node status of lung adenocarcinoma patients, offering a safe and accurate alternative to invasive procedures. By providing clinicians with a reliable tool for diagnosing and assessing disease progression, our method holds the potential to significantly improve patient outcomes and enhance the overall quality of clinical practice.

#1469 Predicting mortality risk in incident dialysis patients using clinical predictors and patient-reported predictors

Abstract Background and Aims Prognostic prediction models can be used to identify dialysis patients that are at high risk of dying. By doing so, vulnerable patients can be monitored more closely and treatment can be intensified where needed. In addition, individualized predictions can help inform patients about their prognosis. Traditionally, prognostic models primarily rely on known clinical predictors such as age, comorbidities and laboratory markers. Although it is still unchartered territory in the field of nephrology, studies in other fields have shown that patient-reported predictors, like pain and fatigue, can be of use as predictors of survival. Predictive performance of current prognostic models for mortality risk amongst dialysis patients is subpar and we hypothesized that the addition of patient-reported predictors may improve predictions. Therefore, we aimed to develop a novel prognostic model for the prediction of two-year mortality in patients starting dialysis using both clinical predictors and patient-reported predictors. Method Data from the Netherlands Cooperative Study on the Adequacy of Dialysis (NECOSAD) were used to develop two prognostic models for the prediction of 2-year all-cause mortality in incident dialysis patients: a base model consisting of clinical predictors and an updated model containing additional patient-reported predictors. Missing data were handled using a tenfold multiple imputation with fully conditional specification. The imputation model included all candidate predictors as well as the outcome variable of 2-year mortality. Predictors for the base model were selected based on existing mortality models and clinical expertise. Patient-reported predictors to add to the base model were chosen based on availability in NECOSAD, clinical expertise and literature describing patient-reported predictors. The models were developed using multivariate logistic regression and their discriminative abilities were assessed by calculating the area under the curve (AUC). Results A total of 1956 patients from NECOSAD were used for the analyses. Of these, 447 (22.9%) died within two years after baseline (3 months after dialysis initiation). The base model contained the following predictors: age, sex, serum albumin, C-reactive protein, diabetes mellitus, hemoglobin, residual kidney function (GFR at 3 months), body mass index, cardiovascular disease, dialysis modality, malignancy and primary kidney disease. The discriminative abilities of the base model were moderate with a AUC of 0.79 (95% CI: 0.77-0.82). The base model was updated using the following patient-reported predictors: mental component score, physical component score, general health perception, number of symptoms, symptom burden (measured with the KDQOL instrument) depressive symptoms, fatigue and pain. The addition of the patient-reported predictors marginally improved the predictive performance of the model achieving an AUC of 0.81 (95% CI: 0.79-0.84). Conclusion The base model including clinical predictors was enhanced slightly by the addition of patient-reported predictors.

Construction and validation of a novel tumor morphology immune inflammatory nutritional score (TIIN score) for intrahepatic cholangiocarcinoma: a multicenter study

BackgroundTumor morphology, immune function, inflammatory levels, and nutritional status play critical roles in the progression of intrahepatic cholangiocarcinoma (ICC). This multicenter study aimed to investigate the association between markers related to tumor morphology, immune function, inflammatory levels, and nutritional status with the prognosis of ICC patients. Additionally, a novel tumor morphology immune inflammatory nutritional score (TIIN score), integrating these factors was constructed.MethodsA retrospective analysis was performed on 418 patients who underwent radical surgical resection and had postoperative pathological confirmation of ICC between January 2016 and January 2020 at three medical centers. The cohort was divided into a training set (n = 272) and a validation set (n = 146). The prognostic significance of 16 relevant markers was assessed, and the TIIN score was derived using LASSO regression. Subsequently, the TIIN-nomogram models for OS and RFS were developed based on the TIIN score and the results of multivariate analysis. The predictive performance of the TIIN-nomogram models was evaluated using ROC survival curves, calibration curves, and clinical decision curve analysis (DCA).ResultsThe TIIN score, derived from albumin-to-alkaline phosphatase ratio (AAPR), albumin–globulin ratio (AGR), monocyte-to-lymphocyte ratio (MLR), and tumor burden score (TBS), effectively categorized patients into high-risk and low-risk groups using the optimal cutoff value. Compared to individual metrics, the TIIN score demonstrated superior predictive value for both OS and RFS. Furthermore, the TIIN score exhibited strong associations with clinical indicators including obstructive jaundice, CEA, CA19-9, Child–pugh grade, perineural invasion, and 8th edition AJCC N stage. Univariate and multivariate analysis confirmed the TIIN score as an independent risk factor for postoperative OS and RFS in ICC patients (p < 0.05). Notably, the TIIN-nomogram models for OS and RFS, constructed based on the multivariate analysis and incorporating the TIIN score, demonstrated excellent predictive ability for postoperative survival in ICC patients.ConclusionThe development and validation of the TIIN score, a comprehensive composite index incorporating tumor morphology, immune function, inflammatory level, and nutritional status, significantly contribute to the prognostic assessment of ICC patients. Furthermore, the successful application of the TIIN-nomogram prediction model underscores its potential as a valuable tool in guiding individualized treatment strategies for ICC patients. These findings emphasize the importance of personalized approaches in improving the clinical management and outcomes of ICC.