Conventional Logistic Regression Models Research Articles

Ultrasound-based radiomics machine learning models for diagnosing cervical lymph node metastasis in patients with non-small cell lung cancer: a multicentre study

BackgroundCervical lymph node metastasis (LNM) is an important prognostic factor for patients with non-small cell lung cancer (NSCLC). We aimed to develop and validate machine learning models that use ultrasound radiomic and descriptive semantic features to diagnose cervical LNM in patients with NSCLC.MethodsThis study included NSCLC patients who underwent neck ultrasound examination followed by cervical lymph node (LN) biopsy between January 2019 and January 2022 from three institutes. Radiomic features were extracted from the ultrasound images at the maximum cross-sectional areas of cervical LNs. Logistic regression (LR) and random forest (RF) models were developed. Model performance was assessed by the area under the curve (AUC) and accuracy, validated internally and externally by fivefold cross-validation and hold-out method, respectively.ResultsIn total, 313 patients with a median age of 64 years were included, and 276 (88.18%) had cervical LNM. Three descriptive semantic features, including long diameter, shape, and corticomedullary boundary, were selected by multivariate analysis. Out of the 474 identified radiomic features, 9 were determined to fit the LR model, while 15 fit the RF model. The average AUCs of the semantic and radiomics models were 0.876 (range: 0.781–0.961) and 0.883 (range: 0.798–0.966), respectively. However, the average AUC was higher for the semantic-radiomics combined LR model (0.901; range: 0.862–0.927). When the RF algorithm was applied, the average AUCs of the radiomics and semantic-radiomics combined models were improved to 0.908 (range: 0.837–0.966) and 0.922 (range: 0.872–0.982), respectively. The models tested by the hold-out method had similar results, with the semantic-radiomics combined RF model achieving the highest AUC value of 0.901 (95% CI, 0.886–0.968).ConclusionsThe ultrasound radiomic models showed potential for accurately diagnosing cervical LNM in patients with NSCLC when integrated with descriptive semantic features. The RF model outperformed the conventional LR model in diagnosing cervical LNM in NSCLC patients.

Bayesian compositional models for ordinal response.

Ordinal response is commonly found in medicine, biology, and other fields. In many situations, the predictors for this ordinal response are compositional, which means that the sum of predictors for each sample is fixed. Examples of compositional data include the relative abundance of species in microbiome data and the relative frequency of nutrition concentrations. Moreover, the predictors that are strongly correlated tend to have similar influence on the response outcome. Conventional cumulative logistic regression models for ordinal responses ignore the fixed-sum constraint on predictors and their associated interrelationships, and thus are not appropriate for analyzing compositional predictors.To solve this problem, we proposed Bayesian Compositional Models for Ordinal Response to analyze the relationship between compositional data and an ordinal response with a structured regularized horseshoe prior for the compositional coefficients and a soft sum-to-zero restriction on coefficients through the prior distribution. The method was implemented with R package rstan using efficient Hamiltonian Monte Carlo algorithm. We performed simulations to compare the proposed approach and existing methods for ordinal responses. Results revealed that our proposed method outperformed the existing methods in terms of parameter estimation and prediction. We also applied the proposed method to a microbiome study HMP2Data, to find microorganisms linked to ordinal inflammatory bowel disease levels. To make this work reproducible, the code and data used in this paper are available at https://github.com/Li-Zhang28/BCO.

Compare deep learning model and conventional logistic regression model for the identification of unstable saccular intracranial aneurysms in computed tomography angiography.

It is crucial to distinguish unstable from stable intracranial aneurysms (IAs) as early as possible to derive optimal clinical decision-making for further treatment or follow-up. The aim of this study was to investigate the value of a deep learning model (DLM) in identifying unstable IAs from computed tomography angiography (CTA) images and to compare its discriminatory ability with that of a conventional logistic regression model (LRM). From August 2011 to May 2021, a total of 1,049 patients with 681 unstable IAs and 556 stable IAs were retrospectively analyzed. IAs were randomly divided into training (64%), internal validation (16%), and test sets (20%). Convolutional neural network (CNN) analysis and conventional logistic regression (LR) were used to predict which IAs were unstable. The area under the curve (AUC), sensitivity, specificity and accuracy were calculated to evaluate the discriminating ability of the models. One hundred and ninety-seven patients with 229 IAs from Banan Hospital were used for external validation sets. The conventional LRM showed 11 unstable risk factors, including clinical and IA characteristics. The LRM had an AUC of 0.963 [95% confidence interval (CI): 0.941-0.986], a sensitivity, specificity and accuracy on the external validation set of 0.922, 0.906, and 0.913, respectively, in predicting unstable IAs. In predicting unstable IAs, the DLM had an AUC of 0.771 (95% CI: 0.582-0.960), a sensitivity, specificity and accuracy on the external validation set of 0.694, 0.929, and 0.782, respectively. The CNN-based DLM applied to CTA images did not outperform the conventional LRM in predicting unstable IAs. The patient clinical and IA morphological parameters remain critical factors for ensuring IA stability. Further studies are needed to enhance the diagnostic accuracy.

LightGBM outperforms other machine learning techniques in predicting graft failure after liver transplantation: Creation of a predictive model through large-scale analysis.

The incidence of graft failure following liver transplantation (LTx) is consistent. While traditional risk scores for LTx have limited accuracy, the potential of machine learning (ML) in this area remains uncertain, despite its promise in other transplant domains. This study aims to determine ML's predictive limitations in LTx by replicating methods used in previous heart transplant research. This study utilized the UNOS STAR database, selecting 64,384 adult patients who underwent LTx between 2010 and 2020. Gradient boosting models (XGBoost and LightGBM) were used to predict 14, 30, and 90-day graft failure compared to conventional logistic regression model. Models were evaluated using both shuffled and rolling cross-validation (CV) methodologies. Model performance was assessed using the AUC across validation iterations. In a study comparing predictive models for 14-day, 30-day and 90-day graft survival, LightGBM consistently outperformed other models, achieving the highest AUC of.740,.722, and.700 in shuffled CV methods. However, in rolling CV the accuracy of the model declined across every ML algorithm. The analysis revealed influential factors for graft survival prediction across all models, including total bilirubin, medical condition, recipient age, and donor AST, among others. Several features like donor age and recipient diabetes history were important in two out of three models. LightGBM enhances short-term graft survival predictions post-LTx. However, due to changing medical practices and selection criteria, continuous model evaluation is essential. Future studies should focus on temporal variations, clinical implications, and ensure model transparency for broader medical utility.

Machine learning models for predicting in-hospital outcomes after non-surgical treatment among patients with moderate-to-severe traumatic brain injury

AimThis study aims to develop prediction models for in-hospital outcomes after non-surgical treatment among patients with moderate-to-severe traumatic brain injury (TBI). MethodWe conducted a retrospective review of patients hospitalized for moderate-to-severe TBI in our department from 2011 to 2020. Five machine learning (ML) algorithms and the conventional logistic regression (LR) model were employed to predict in-hospital mortality and the Glasgow Outcome Scale (GOS) functional outcomes. These models utilized clinical and routine blood data collected upon admission. ResultsThis study included a total of 196 patients who received only non-surgical treatment after moderate-to-severe TBI. When predicting mortality, ML models achieved area under the curve (AUC) values of 0.921 to 0.994 using clinical and routine blood data, and 0.877 to 0.982 using only clinical data. In comparison, LR models yielded AUCs of 0.762 and 0.730 respectively. When predicting the GOS outcome, ML models achieved AUCs of 0.870 to 0.915 using clinical and routine blood data, and 0.858 to 0.927 using only clinical data. In comparison, the LR model yielded AUCs of 0.798 and 0.787 respectively. Repeated internal validation showed that the contributions of routine blood data for prediction models may depend on different prediction algorithms and different outcome measurements. ConclusionThe study reported ML-based prediction models that provided rapid and accurate predictions on short-term outcomes after non-surgical treatment among patients with moderate-to-severe TBI. The study also highlighted the superiority of ML models over conventional LR models and proposed the complex contributions of routine blood data in such predictions.

RADT-41. PREDICTION OF TREATMENT OUTCOME IN GAMMA KNIFE RADIOSURGERY OF BENIGN MENINGIOMAS

Abstract Objective This study aimed to assess the predictability of machine learning (ML) ensemble models for the success of Gamma Knife radiosurgery (GKRS) on non-skull base benign meningiomas. METHODS A total of 331 non-skull base benign meningiomas treated primarily with GKRS between 1997 and 2015 were retrospectively analyzed. When a tumor remained not larger than 125% of the initial volume at the last imaging follow-up and no significant change in the high signal intensity area around the tumor, it was classified as 'treatment success (TS).' Univariate correlation and multivariate logistic regression analyses were performed to identify statistically significant prognostic factors. Four ensemble algorithms were trained to predict the outcome of GKRS. The importance of features related to TS was analyzed. RESULTS The median MR imaging follow-up period was 8.6 years. The tumor control and signal change rates were 91.5% and 16.3%, respectively, and the TS rate was 77.6%. The conventional multivariate logistic regression model showed 83.7% accuracy and 37.8% specificity. The ML models had accuracies between 78.0% and 82.2% and specificities between 30.3% and 80.7%, and they were similar to those of conventional logistic regression analyses. The smaller brain volume absorbed more than 14 Gy, and the absence of signal change before GKRS was a favorable prognostic factor for TS in all models. The bagging algorithm with selected features showed the best specificity of 80.7% and the most significant AUC of 0.723. CONCLUSION A bagging ensemble algorithm with selected features had better specificity with similar accuracy than conventional multivariate logistic regression in predicting GKRS outcomes for non-skull base meningiomas.

Identifying the severity of diabetic retinopathy by visual function measures using both traditional statistical methods and interpretable machine learning: a cross-sectional study

Aims/hypothesisTo determine the extent to which diabetic retinopathy severity stage may be classified using machine learning (ML) and commonly used clinical measures of visual function together with age and sex.MethodsWe measured the visual function of 1901 eyes from 1032 participants in the Northern Ireland Sensory Ageing Study, deriving 12 variables from nine visual function tests. Missing values were imputed using chained equations. Participants were divided into four groups using clinical measures and grading of ophthalmic images: no diabetes mellitus (no DM), diabetes but no diabetic retinopathy (DM no DR), diabetic retinopathy without diabetic macular oedema (DR no DMO) and diabetic retinopathy with DMO (DR with DMO). Ensemble ML models were fitted to classify group membership for three tasks, distinguishing (A) the DM no DR group from the no DM group; (B) the DR no DMO group from the DM no DR group; and (C) the DR with DMO group from the DR no DMO group. More conventional multiple logistic regression models were also fitted for comparison. An interpretable ML technique was used to rank the contribution of visual function variables to predictions and to disentangle associations between diabetic eye disease and visual function from artefacts of the data collection process.ResultsThe performance of the ensemble ML models was good across all three classification tasks, with accuracies of 0.92, 1.00 and 0.84, respectively, for tasks A–C, substantially exceeding the accuracies for logistic regression (0.84, 0.61 and 0.80, respectively). Reading index was highly ranked for tasks A and B, whereas near visual acuity and Moorfields chart acuity were important for task C. Microperimetry variables ranked highly for all three tasks, but this was partly due to a data artefact (a large proportion of missing values).Conclusions/interpretationEnsemble ML models predicted status of diabetic eye disease with high accuracy using just age, sex and measures of visual function. Interpretable ML methods enabled us to identify profiles of visual function associated with different stages of diabetic eye disease, and to disentangle associations from artefacts of the data collection process. Together, these two techniques have great potential for developing prediction models using untidy real-world clinical data.Graphical

Effect of alcohol consumption on breast cancer: probabilistic bias analysis for adjustment of exposure misclassification bias and confounders

PurposeThis study was conducted to evaluate the effect of alcohol consumption on breast cancer, adjusting for alcohol consumption misclassification bias and confounders.MethodsThis was a case-control study of 932 women with breast cancer and 1000 healthy control. Using probabilistic bias analysis method, the association between alcohol consumption and breast cancer was adjusted for the misclassification bias of alcohol consumption as well as a minimally sufficient set of adjustment of confounders derived from a causal directed acyclic graph. Population attributable fraction was estimated using the Miettinen’s Formula.ResultsBased on the conventional logistic regression model, the odds ratio estimate between alcohol consumption and breast cancer was 1.05 (95% CI: 0.57, 1.91). However, the adjusted estimates of odds ratio based on the probabilistic bias analysis ranged from 1.82 to 2.29 for non-differential and from 1.93 to 5.67 for differential misclassification. Population attributable fraction ranged from 1.51 to 2.57% using non-differential bias analysis and 1.54–3.56% based on differential bias analysis.ConclusionA marked measurement error was in self-reported alcohol consumption so after correcting misclassification bias, no evidence against independence between alcohol consumption and breast cancer changed to a substantial positive association.

Mathematical Approach in Image Classification using Regression

This study presents a comprehensive evaluation of logistic regression in contrast to a hybrid model combining VGG16 with logistic regression for image classification tasks. The research findings illuminate a striking performance disparity between these two approaches, shedding light on the profound impact of integrating deep learning techniques into image classification. The transition from logistic regression to the VGG16-based hybrid model marks a notable turning point in our investigation. The VGG16 architecture, renowned for its prowess as a feature extractor, showcases an impressive 53.33% surge in accuracy compared to the conventional logistic regression model. This substantial leap underscores the model's capacity to decipher complex image characteristics that elude traditional statistical methods. Furthermore, precision, a crucial metric in classification tasks, experiences a substantial 53% augmentation when adopting the VGG16 hybrid approach. This enhancement signifies the hybrid model's ability to minimize false positives, making it particularly valuable in scenarios where precision holds paramount importance. Equally noteworthy is the substantial 54% improvement observed in both recall and F1-score, emphasizing the VGG16 hybrid model's remarkable capacity to identify and retrieve a higher proportion of true positives while maintaining a balance between precision and recall. This not only amplifies the model's ability to correctly classify images but also mitigates the risk of overlooking relevant instances. These compelling findings underscore the critical role of deep learning, specifically convolutional neural networks (CNNs), in the realm of image classification. The utilization of CNNs, exemplified by the VGG16 architecture, emerges as a game-changer, enabling the capture of intricate image features and patterns that traditional logistic regression struggles to discern. Generally, this study advocates for the integration of advanced deep learning techniques, like VGG16, in image classification endeavors. The substantial performance gains witnessed in accuracy, precision, recall, and F1-score reinforce the pivotal role of convolutional neural networks in enhancing the effectiveness of image classification tasks. By harnessing the power of deep learning, we unlock new horizons in image analysis, paving the way for more accurate and efficient classification systems

Modeling pedestrian behavior at urban signalised intersections using statistical-ANN hybrid approach – Case study of New Delhi

Traffic crashes at intersections are major hazard all over the world. The victims of road crashes are mostly pedestrians. A large share of accidents can be related to unsafe crossing behavior by pedestrians. The accidents are mostly fatal for pedestrians especially if the unsafe act is signal violation. This study primarily focused on cognizing factors which are significantly affecting the signal violation behavior of pedestrians’ at urban signalised intersections. Data is collected at 11 signalised intersections in New Delhi, India using video recording technique. Significant variables are identified using conventional statistics. The variables are then used as input neuron for the Artificial Neural Network (ANN) for modeling signal violation behavior. Sensitivity analysis is performed to identify the relative importance of factors related to the pedestrian signal violation. Host of different factors such as pedestrian demographic, behavioral attributes, crossing state and mobile usage are notably influencing signal violations. Pedestrian crossing speed, crossing path and waiting time are the top three predictors of violation behavior. The accuracy of ANN to predict signal violation behaviour is found to be about 85% and is considerably higher than conventional binary logistic regression (BLR) model. The area under the receiver operating characteristic (ROC) curve (AUC) is found to be 0.753, suggesting good model performance. The results suggest that ANN is a robust alternative to conventional regression, for studying and modeling pedestrian behavior. The findings from this study would assist engineers and policy makers to take proactive measures for designing pedestrian friendly facilities to reduce signal violations. This would ultimately improve pedestrian safety at urban signalised intersections.

Predicting hyperkalemia in patients with advanced chronic kidney disease using the XGBoost model

BackgroundHyperkalemia is a common complication of chronic kidney disease (CKD). Hyperkalemia is associated with mortality, CKD progression, hospitalization, and high healthcare costs in patients with CKD. We developed a machine learning model to predict hyperkalemia in patients with advanced CKD at an outpatient clinic.MethodsThis retrospective study included 1,965 advanced CKD patients between January 1, 2010, and December 31, 2020 in Taiwan. We randomly divided all patients into the training (75%) and testing (25%) datasets. The primary outcome was to predict hyperkalemia (K+ > 5.5 mEq/L) in the next clinic vist. Two nephrologists were enrolled in a human-machine competition. The area under the receiver operating characteristic curves (AUCs), sensitivity, specificity, and accuracy were used to evaluate the performance of XGBoost and conventional logistic regression models with that of these physicians.ResultsIn a human-machine competition of hyperkalemia prediction, the AUC, PPV, and accuracy of the XGBoost model were 0.867 (95% confidence interval: 0.840–0.894), 0.700, and 0.933, which was significantly better than that of our clinicians. There were four variables that were chosen as high-ranking variables in XGBoost and logistic regression models, including hemoglobin, the serum potassium level in the previous visit, angiotensin receptor blocker use, and calcium polystyrene sulfonate use.ConclusionsThe XGBoost model provided better predictive performance for hyperkalemia than physicians at the outpatient clinic.

Using Machine Learning to Predict Abnormal Carotid Intima-Media Thickness in Type 2 Diabetes.

Carotid intima-media thickness (c-IMT) is a reliable risk factor for cardiovascular disease risk in type 2 diabetes (T2D) patients. The present study aimed to compare the effectiveness of different machine learning methods and traditional multiple logistic regression in predicting c-IMT using baseline features and to establish the most significant risk factors in a T2D cohort. We followed up with 924 patients with T2D for four years, with 75% of the participants used for model development. Machine learning methods, including classification and regression tree, random forest, eXtreme gradient boosting, and Naïve Bayes classifier, were used to predict c-IMT. The results showed that all machine learning methods, except for classification and regression tree, were not inferior to multiple logistic regression in predicting c-IMT in terms of higher area under receiver operation curve. The most significant risk factors for c-IMT were age, sex, creatinine, body mass index, diastolic blood pressure, and duration of diabetes, sequentially. Conclusively, machine learning methods could improve the prediction of c-IMT in T2D patients compared to conventional logistic regression models. This could have crucial implications for the early identification and management of cardiovascular disease in T2D patients.

Identifying the Local Impacts of Speed-Related Factors on Tunnel Entrance Crash Severity

According to crash statistics, freeway tunnel entrance zones have higher injury severity rates than other freeway sections. The concentration of traffic injuries is a combination of high speed and low driver alertness. The government of South Korea has deployed the variable speed limit (VSL) as a safety countermeasure to mitigate the consequences of tunnel entrance crashes. This measure has not been quantitatively reviewed, therefore, the effectiveness of implementing VSL at tunnel entrance zones is unknown. The aim of this was to predict the injury severity of crashes occurring at the tunnel entrance zone given the nature of the crash, the size of the tunnel, the roadway geometric characteristics, and environmental factors. It was anticipated that a statistical model calibrated with historical crash data and tunnel features would be able to justify the current locations of VSL implementation and prioritize new VSL implementations. A geographically weighted logistic regression was employed and benchmarked by the conventional logistic regression model, considering the varying geographic locations of freeway tunnel entrances. The geographically weighted logistic regression model presented consistent findings and a slightly better statistical goodness-of-fit than the conventional logistic model. Moreover, the geographically weighted regression captured the local impact of low-level traffic speed on crash injury severities. The findings in this study could help researchers and practitioners evaluate site-specific traffic safety improvement strategies.

Abstract 951: Clinical model for predicting pathological complete response in patients with esophageal and gastroesophageal junction adenocarcinoma after trimodality therapy

Abstract Background: Approximately 25% of patients with esophageal adenocarcinoma achieve pathological complete response (pathCR) after chemoradiotherapy with following surgery (trimodality therapy). The prediction for pathCR would provide an important guidance for decision making regarding surgical strategy. Previous studies have used the logistic regression model to predict pathCR, however, few studies utilized the machine learning models. In this study, we aimed to establish a predictive model for pathCR using the machine learning models using clinical data in patients with GEJ cancer treated by trimodality therapy. Methods: 512 patients with localized esophageal and GEJ adenocarcinoma who received trimodality therapy in MD Anderson Cancer Center between 2002 to 2020 were included. We first performed survival analysis to confirm the survival benefit of pathCR. Then, several prediction models commonly used in machine learning, including logistic regression, LASSO, Random Forest, BART, and xgboost were used to predict pathCR. Each method was trained and validated using 10-fold cross-validation. Results: A total of 125 patients (24 %) achieved a path CR. Patients who achieved path CR had significantly longer overall survival (OS) and relapse free survival (RFS) than <pathCR 387 patients (76%) (median OS, 132 months vs 58 months, respectively, p < 0.002; median RFS, 63 months vs 27 months, respectively, p < 0.001). The logistic regression model indicated non-signet ring cell subtype, T1/T2 stage, negative biopsy after chemoradiation were independent predictive factors for pathCR. Finally, we created machine learning models and LASSO showing the highest predictive ability with AUC value of 0.644 among those models including conventional logistic regression model. The percent partial AUC at lower false positive rate of 0.25 was higher than that of the conventional logistic regression model (% pAUC, 23.7 vs 21.8, respectively) Conclusion: We established the clinical predictive model for pathCR in patients with GEJ cancer treated by trimodality therapy using machine learning method with acceptable predictive ability. Addition of molecular features might further refine the models. Citation Format: Kohei Yamashita, Jacob M. Maronge, Melissa Pool Pizzi, Wayne L. Hofstetter, Aileen Chen, Matheus Sewastjanow-Silva, Ernesto Rosa Vicentini, Meita S. Hirschmann, Anh Ta, Namita Shanbhag, Ying Yuan, Jaffer A. Ajani. Clinical model for predicting pathological complete response in patients with esophageal and gastroesophageal junction adenocarcinoma after trimodality therapy [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 951.

Modeling of occupant behavior considering spatial variation: Geostatistical analysis and application based on American time use survey data

Numerous occupant behavior (OB) models that simulate occupancy, activity and action at home have been developed to improve the accuracy and quality of energy demand estimations. Previous studies have revealed that the consideration of inter-occupant diversity improves the performance of OB models. However, existing models ignore spatial variation in OBs or partially consider it using a simple method without evaluating whether it is sufficient. Moreover, the modeling method to reproduce the spatial variation is missing. This study aims to develop a modeling method that can effectively reproduce spatial variation in OBs using American time use data. The global Moran’s index test confirmed that spatial variations exist in OBs; however, they differ by time of day and activities for studied population. Subsequently, two spatial variation representations were generated using the ordinary kriging and spatial autoregressive methods. Finally, three spatial logistic regression models that consider spatial variations were developed and evaluated. The developed models generated smaller errors and higher inter-occupant diversity than the conventional logistic regression models at the state level. The established method is applicable to any country and region. Using higher spatial resolution and richer time use datasets may further improve OB models to model region-specific characteristics of building energy demand.

Variable Selection of Spatial Logistic Autoregressive Model with Linear Constraints.

In recent years, spatial data widely exist in various fields such as finance, geology, environment, and natural science. These data collected by many scholars often have geographical characteristics. The spatial autoregressive model is a general method to describe the spatial correlations among observation units in spatial econometrics. The spatial logistic autoregressive model augments the conventional logistic regression model with an extra network structure when the spatial response variables are discrete, which enhances classification precision. In many application fields, prior knowledge can be formulated as constraints on the parameters to improve the effectiveness of variable selection and estimation. This paper proposes a variable selection method with linear constraints for the high-dimensional spatial logistic autoregressive model in order to integrate the prior information into the model selection. Monte Carlo experiments are provided to analyze the performance of our proposed method under finite samples. The results show that the method can effectively screen out insignificant variables and give the corresponding coefficient estimates of significant variables simultaneously. As an empirical illustration, we apply our method to land area data.

Tortuosity of parent artery predicts in-stent stenosis after pipeline flow-diverter stenting for internal carotid artery aneurysms.

Background and purposeThe relationship between the tortuosity of the parent artery and treatment outcomes is not well established. We investigate the association between parent artery tortuosity and flow diverter (FD) treatment outcomes in patients with internal carotid artery aneurysms in this study.MethodsA retrospective review study was conducted to identify all patients with internal carotid artery aneurysms who were implanted with Pipeline embolization device (PED) between 2016 and 2020. The relationship between parent artery tortuosity and aneurysm complete occlusion (CO) and in-stent stenosis (ISS) was analyzed. The mathematical parameters “Curvature”, “torsion”, and “DM” extracted from the parent artery were utilized to quantify the parent artery tortuosity. A vascular narrowing of greater than 25% was categorized as ISS. Logistic regression analysis was used to identify significant independent predictors. Furthermore, we compared the performance of four machine learning algorithms and Logistic Regression model in predicting ISS.ResultsThis research included 62 patients who with internal carotid artery aneurysms. In 49 (79%) cases, follow-up angiography (mean follow-up duration 11.7 ±7.3 months) revealed CO of the aneurysm. ISS was detected in 22 (35.5%) cases. According to univariate analysis, parent artery tortuosity and other variables were not associated with CO (p > 0.1). Maximum curvature (OR = 1.084; 95% CI, 1.008–1.165; p = 0.03) and DM (OR = 0.01; 95% CI, 0–0.488; p = 0.02) exhibited strong independent associations with ISS in multivariate analysis. The SVM model is superior to the conventional Logistic Regression model and the other models in predicting ISS.ConclusionsThe tortuosity of the parent artery may affect the treatment outcome of FD stenting. We found that parent artery tortuosity was associated with ISS, but not with aneurysm complete occlusion following PED stenting for internal carotid artery aneurysms in this study. Parent arteries with higher maximum curvature and lower DM were more likely to develop ISS.

Machine learning for early prediction of sepsis-associated acute brain injury.

BackgroundSepsis-associated encephalopathy (SAE) is defined as diffuse brain dysfunction associated with sepsis and leads to a high mortality rate. We aimed to develop and validate an optimal machine-learning model based on clinical features for early predicting sepsis-associated acute brain injury.MethodsWe analyzed adult patients with sepsis from the Medical Information Mart for Intensive Care (MIMIC III) clinical database. Candidate models were trained using random forest, support vector machine (SVM), decision tree classifier, gradients boosting machine (GBM), multiple layer perception (MLP), extreme gradient boosting (XGBoost), light gradients boosting machine (LGBM) and a conventional logistic regression model. These methods were applied to develop and validate the optimal model based on its accuracy and area under curve (AUC).ResultsIn total, 12,460 patients with sepsis met inclusion criteria, and 6,284 (50.4%) patients suffered from sepsis-associated acute brain injury. Compared other models, the LGBM model achieved the best performance. The AUC for both train set and test set indicated excellent validity (Trainset AUC 0.91, Testset AUC 0.87). Feature importance analysis showed that glucose, age, mean arterial pressure, heart rate, hemoglobin, and length of ICU stay were the top 6 important clinical factors to predict occurrence of sepsis-associated acute brain injury.ConclusionAlmost half of patients admitted to ICU with sepsis had sepsis-associated acute brain injury. The LGBM model better identify patients with sepsis-associated acute brain injury than did other machine-learning models. Glucose, age, and mean arterial pressure were the three most important clinical factors to predict occurrence of sepsis-associated acute brain injury.

Application of machine learning algorithms in predicting HIV infection among men who have sex with men: Model development and validation.

BackgroundContinuously growing of HIV incidence among men who have sex with men (MSM), as well as the low rate of HIV testing of MSM in China, demonstrates a need for innovative strategies to improve the implementation of HIV prevention. The use of machine learning algorithms is an increasing tendency in disease diagnosis prediction. We aimed to develop and validate machine learning models in predicting HIV infection among MSM that can identify individuals at increased risk of HIV acquisition for transmission-reduction interventions.MethodsWe extracted data from MSM sentinel surveillance in Zhejiang province from 2018 to 2020. Univariate logistic regression was used to select significant variables in 2018–2019 data (P < 0.05). After data processing and feature selection, we divided the model development data into two groups by stratified random sampling: training data (70%) and testing data (30%). The Synthetic Minority Oversampling Technique (SMOTE) was applied to solve the problem of unbalanced data. The evaluation metrics of model performance were comprised of accuracy, precision, recall, F-measure, and the area under the receiver operating characteristic curve (AUC). Then, we explored three commonly-used machine learning algorithms to compare with logistic regression (LR), including decision tree (DT), support vector machines (SVM), and random forest (RF). Finally, the four models were validated prospectively with 2020 data from Zhejiang province.ResultsA total of 6,346 MSM were included in model development data, 372 of whom were diagnosed with HIV. In feature selection, 12 variables were selected as model predicting indicators. Compared with LR, the algorithms of DT, SVM, and RF improved the classification prediction performance in SMOTE-processed data, with the AUC of 0.778, 0.856, 0.887, and 0.942, respectively. RF was the best-performing algorithm (accuracy = 0.871, precision = 0.960, recall = 0.775, F-measure = 0.858, and AUC = 0.942). And the RF model still performed well on prospective validation (AUC = 0.846).ConclusionMachine learning models are substantially better than conventional LR model and RF should be considered in prediction tools of HIV infection in Chinese MSM. Further studies are needed to optimize and promote these algorithms and evaluate their impact on HIV prevention of MSM.

Prediction of mortality risk of health checkup participants using machine learning-based models: the J-SHC study

Early detection and treatment of diseases through health checkups are effective in improving life expectancy. In this study, we compared the predictive ability for 5-year mortality between two machine learning-based models (gradient boosting decision tree [XGBoost] and neural network) and a conventional logistic regression model in 116,749 health checkup participants. We built prediction models using a training dataset consisting of 85,361 participants in 2008 and evaluated the models using a test dataset consisting of 31,388 participants from 2009 to 2014. The predictive ability was evaluated by the values of the area under the receiver operating characteristic curve (AUC) in the test dataset. The AUC values were 0.811 for XGBoost, 0.774 for neural network, and 0.772 for logistic regression models, indicating that the predictive ability of XGBoost was the highest. The importance rating of each explanatory variable was evaluated using the SHapley Additive exPlanations (SHAP) values, which were similar among these models. This study showed that the machine learning-based model has a higher predictive ability than the conventional logistic regression model and may be useful for risk assessment and health guidance for health checkup participants.