Network or regression-based methods for disease discrimination: a comparison study

Flow diverter stents for endovascular treatment of aneurysms: a comparative study of efficacy and safety between FREDX and FRED

BACKGROUND

The FRED X flow diverter (FREDX), as the second generation in the FRED series, aims to improve the treatment of cerebral aneurysms. This study compares the efficacy and safety of FREDX with its predecessor, FRED.

METHODS

This prospective registry included patients treated with FRED and FREDX devices. Efficacy was assessed using digital subtraction angiography with 3D volumetric reconstruction at immediate and 1 year follow-ups. Safety was evaluated by recording complications, analyzed through univariate contrasts, generalized mixed models, and Bayesian network analyses.

RESULTS

We treated 287 patients with 385 aneurysms, with 77.9% receiving FRED and 22.1% FREDX. The median age was 55 years (IQR 47-65) and 78.4% were women. The FREDX group showed a higher prevalence of saccular-like aneurysms (70.6% vs 52.7%, P=0.012) and a higher rate of complete occlusion compared with FRED interventions (79.4% vs 59.3%, P=0.022). After adjusting for confounders, these differences represented a 3.04-fold increased likelihood (95% CI 1.44 to 6.41, P=0.003) of achieving complete occlusion at 1 year with FREDX interventions. Regarding safety, two (3.5%) complications (both non-symptomatic) were observed in the FREDX group and 23 (10.4%) in the FRED group (P=0.166). Bayesian network analysis suggested a trend towards fewer complications for FREDX, with a median reduction of 5.5% in the posterior distribution of the prevalence of complications compared with FRED interventions.

CONCLUSIONS

The FREDX device shows improved complete occlusion rates at 1 year compared with the FRED device while maintaining a favourable safety profile, indicating its potential advantage in the treatment of cerebral aneurysms.

Prediction of the occurrence of leprosy reactions based on Bayesian networks

Introduction

Leprosy reactions (LR) are severe episodes of intense activation of the host inflammatory response of uncertain etiology, today the leading cause of permanent nerve damage in leprosy patients. Several genetic and non-genetic risk factors for LR have been described; however, there are limited attempts to combine this information to estimate the risk of a leprosy patient developing LR. Here we present an artificial intelligence (AI)-based system that can assess LR risk using clinical, demographic, and genetic data.

Methods

The study includes four datasets from different regions of Brazil, totalizing 1,450 leprosy patients followed prospectively for at least 2 years to assess the occurrence of LR. Data mining using WEKA software was performed following a two-step protocol to select the variables included in the AI system, based on Bayesian Networks, and developed using the NETICA software.

Results

Analysis of the complete database resulted in a system able to estimate LR risk with 82.7% accuracy, 79.3% sensitivity, and 86.2% specificity. When using only databases for which host genetic information associated with LR was included, the performance increased to 87.7% accuracy, 85.7% sensitivity, and 89.4% specificity.

Conclusion

We produced an easy-to-use, online, free-access system that identifies leprosy patients at risk of developing LR. Risk assessment of LR for individual patients may detect candidates for close monitoring, with a potentially positive impact on the prevention of permanent disabilities, the quality of life of the patients, and upon leprosy control programs.

Risk factors assessment and a Bayesian network model for predicting ischemic stroke in patients with cardiac myxoma

Objective

This study aims to identify relevant risk factors, assess the interactions between variables, and establish a predictive model for ischemic stroke (IS) in patients with cardiac myxoma (CM) using the Bayesian network (BN) approach.

Methods

Data of patients with CM were collected from three tertiary comprehensive hospitals in Beijing from January 2002 to January 2022. Age, sex, medical history, and information related to CM were extracted from the electronic medical record system. The BN model was constructed using the tabu search algorithm, and the conditional probability of each node was calculated using the maximum likelihood estimation method. The probability of each node of the network and the interrelationship between IS and its related factors were qualitatively and quantitatively analyzed. A receiver operating characteristic (ROC) curve was also plotted. Sensitivity, specificity, and area under the curve (AUC) values were calculated and compared between the BN and logistic regression models to evaluate the efficiency of the predictive model.

Results

A total of 416 patients with CM were enrolled in this study, including 61 with and 355 without IS. The BN model found that cardiac symptoms, systemic embolic symptoms, platelet counts, and tumor with high mobility were directly associated with the occurrence of IS in patients with CM. The BN model for predicting CM-IS achieved higher scores on AUC {0.706 [95% confidence interval (CI), 0.639-0.773]} vs. [0.697 (95% CI, 0.629-0.766)] and sensitivity (99.44% vs. 98.87%), but lower scores on accuracies (85.82% vs. 86.06%) and specificity (6.56% vs. 11.48%) than the logistic regression model.

Conclusion

Cardiac symptoms, systemic embolic symptoms, platelet counts, and tumor with high mobility are candidate predictors of IS in patients with CM. The BN model was superior or at least non-inferior to the traditional logistic regression model, and hence is potentially useful for early IS detection, diagnosis, and prevention in clinical practice.

Use of machine learning to predict medication adherence in individuals at risk for atherosclerotic cardiovascular disease

Background

Medication nonadherence is a critical problem with severe implications in individuals at risk for atherosclerotic cardiovascular disease. Many studies have attempted to predict medication adherence in this population, but few, if any, have been effective in prediction, sug-gesting that essential risk factors remain unidentified.

Objective

This study's objective was to (1) establish an accurate prediction model of medi-cation adherence in individuals at risk for atherosclerotic cardiovascular disease and (2) identify significant contributing factors to the predictive accuracy of medication adherence. In particular, we aimed to use only the baseline questionnaire data to assess medication adherence prediction feasibility.

Methods

A sample of 40 individuals at risk for atherosclerotic cardiovascular disease was recruited for an eight-week feasibility study. After collecting baseline data, we recorded data from a pillbox that sent events to a cloud-based server. Health measures and medication use events were analyzed using machine learning algorithms to identify variables that best predict medication adherence.

Results

Our adherence prediction model, based on only the ten most relevant variables, achieved an average error rate of 12.9%. Medication adherence was closely correlated with being encouraged to play an active role in their treatment, having confidence about what to do in an emergency, knowledge about their medications, and having a special person in their life.

Conclusions

Our results showed the significance of clinical and psychosocial factors for predicting medication adherence in people at risk for atherosclerotic cardiovascular diseases. Clini-cians and researchers can use these factors to stratify individuals to make evidence-based decisions to reduce the risks.

Risk factors and a Bayesian network model to predict ischemic stroke in patients with dilated cardiomyopathy

OBJECTIVE

This study aimed to identify risk factors and create a predictive model for ischemic stroke (IS) in patients with dilated cardiomyopathy (DCM) using the Bayesian network (BN) approach.

MATERIALS AND METHODS

We collected clinical data of 634 patients with DCM treated at three referral management centers in Beijing between 2016 and 2021, including 127 with and 507 without IS. The patients were randomly divided into training (441 cases) and test (193 cases) sets at a ratio of 7:3. A BN model was established using the Tabu search algorithm with the training set data and verified with the test set data. The BN and logistic regression models were compared using the area under the receiver operating characteristic curve (AUC).

RESULTS

Multivariate logistic regression analysis showed that hypertension, hyperlipidemia, atrial fibrillation/flutter, estimated glomerular filtration rate (eGFR), and intracardiac thrombosis were associated with IS. The BN model found that hyperlipidemia, atrial fibrillation (AF) or atrial flutter, eGFR, and intracardiac thrombosis were closely associated with IS. Compared to the logistic regression model, the BN model for IS performed better or equally well in the training and test sets, with respective accuracies of 83.7 and 85.5%, AUC of 0.763 [95% confidence interval (CI), 0.708-0.818] and 0.822 (95% CI, 0.748-0.896), sensitivities of 20.2 and 44.2%, and specificities of 98.3 and 97.3%.

CONCLUSION

Hypertension, hyperlipidemia, AF or atrial flutter, low eGFR, and intracardiac thrombosis were good predictors of IS in patients with DCM. The BN model was superior to the traditional logistic regression model in predicting IS in patients with DCM and is, therefore, more suitable for early IS detection and diagnosis, and could help prevent the occurrence and recurrence of IS in this patient cohort.

Application of Network Analysis to Uncover Variables Contributing to Functional Recovery after Stroke

To estimate network structures to discover the interrelationships among variables and distinguish the difference between networks. Three hundred and forty-eight stroke patients were enrolled in this retrospective study. A network analysis was used to investigate the association between those variables. A Network Comparison Test was performed to compare the correlation of variables between networks. Three hundred and twenty-five connections were identified, and 22 of these differed significantly between the high- and low-Functional Independence Measurement (FIM) groups. In the high-FIM network structure, brain-derived neurotrophic factor (BDNF) and length of stay (LOS) had associations with other nodes. However, there was no association with BDNF and LOS in the low-FIM network. In addition, the use of amantadine was associated with shorter LOS and lower FIM motor subscores in the high-FIM network, but there was no such connection in the low-FIM network. Centrality indices revealed that amantadine use had high centrality with others in the high-FIM network but not the low-FIM network. Coronary artery disease (CAD) had high centrality in the low-FIM network structure but not the high-FIM network. Network analysis revealed a new correlation of variables associated with stroke recovery. This approach might be a promising method to facilitate the discovery of novel factors important for stroke recovery.

Machine-learning models predicting osteoarthritis associated with the lead blood level

Lead is one of the most hazardous environmental pollutants in industrialized countries; lead exposure is a risk factor for osteoarthritis (OA) in older women. Here, the performance of several machine-learning (ML) algorithms in terms of predicting the prevalence of OA associated with lead exposure was compared. A total of 2224 women aged 50 years and older who participated in the Korea National Health and Nutrition Examination Surveys from 2005 to 2017 were divided into a training dataset (70%) for generation of ML models, and a test dataset (30%). We built and tested five ML algorithms, including logistic regression (LR), a k-nearest neighbor model, a decision tree, a random forest, and a support vector machine. All afforded acceptable predictive accuracy; the LR model was the most accurate and yielded the greatest area under the receiver operating characteristic curve. We found that various ML models can be used to predict the risk of OA associated with lead exposure effectively, using data from population-based survey.

The influence of single nucleotide polymorphisms of NOD2 or CD14 on the risk of Mycobacterium tuberculosis diseases: a systematic review

BACKGROUND

Tuberculosis (TB) is still one of the leading causes of death worldwide. Genetic studies have pointed to the relevance of the NOD2 and CD14 polymorphic alleles in association with the risk of diseases caused by Mycobacterium tuberculosis (Mtb) infection.

METHODS

A systematic review was performed on PubMed, EMBASE, Scientific Electronic Library Online (SciELO), and Literatura Latino-Americana e do Caribe em Ciências da Saúde (Lilacs) to examine the association between single nucleotide polymorphisms (SNP) and risk of Mtb diseases. Study quality was evaluated using the Newcastle-Ottawa Quality Scale (NOQS), and the linkage disequilibrium was calculated for all SNPs using a webtool (Package LDpop).

RESULTS

Thirteen studies matched the selection criteria. Of those, 9 investigated CD14 SNPs, and 6 reported a significant association between the T allele and TT genotypes of the rs2569190 SNP and increased risk of Mtb diseases. The genotype CC was found to be protective against TB disease. Furthermore, in two studies, the CD14 rs2569191 SNP with the G allele was significantly associated with increased risk of Mtb diseases. Four studies reported data uncovering the relationship between NOD2 SNPs and risk of Mtb diseases, with two reporting significant associations of rs1861759 and rs7194886 and higher risk of Mtb diseases in a Chinese Han population. Paradoxically, minor allele carriers (CG or GG) of rs2066842 and rs2066844 NOD2 SNPs were associated with lower risk of Mtb diseases in African Americans.

CONCLUSIONS

The CD14 rs2569190 and rs2569191 polymorphisms may influence risk of Mtb diseases depending on the allele. Furthermore, there is significant association between NOD2 SNPs rs1861759 and rs7194886 and augmented risk of Mtb diseases, especially in persons of Chinese ethnicity. The referred polymorphisms of CD14 and NOD2 genes likely play an important role in risk of Mtb diseases and pathology and may be affected by ethnicity.

SYSTEMATIC REVIEW REGISTRATION

CRD42020186523.

How demographic and clinical characteristics contribute to the recovery of post-stroke dysphagia?

According to the analysis to find out how demographic and clinical characteristics influent the dysphagia outcome after stroke, furthermore, giving some insights to clinical treatment.One hundred eighty post-stroke dysphagia (PSD) patients were enrolled in this retrospective study at the stroke rehabilitation department. The outcome measurements are beside water swallow test at discharge and length of stay at hospital. Twenty-five demographic and clinical variables were collected for this study. Logistic regression and multilinear regression were utilized to estimate models to identify the risk and protect predictors of PSD outcome.Mouth-opening degree, drooling severity scale (DSS) level, mini-mental state exam (MMSE) level, Barthel index and Berg balance scale were significant different between recovered and unrecovered group. Type of stroke, MMSE degree, DSS and hemoglobin level shown significant predictive value for PSD outcome in logistic regression. In addition, obstructive sleep apnea (OSA) and DSS degree were important risk factors for PSD outcome. Gender, body mass index, drinking, hypertension, recurrent stroke, water swallow test level on admission, Berg balance scale, DSS and days between onset to admission shown significant predictive value for length of stay of PSD patients.PSD outcome was influenced by type of stroke, MMSE degree, DSS and hemoglobin level significantly and obstructive sleep apnea act as an important risk role for PSD recovery.

Contributions and limitations of using machine learning to predict noise-induced hearing loss

Purpose

Noise-induced hearing loss (NIHL) is a global issue that impacts people’s life and health. The current review aims to clarify the contributions and limitations of applying machine learning (ML) to predict NIHL by analyzing the performance of different ML techniques and the procedure of model construction.

Methods

The authors searched PubMed, EMBASE and Scopus on November 26, 2020.

Results

Eight studies were recruited in the current review following defined inclusion and exclusion criteria. Sample size in the selected studies ranged between 150 and 10,567. The most popular models were artificial neural networks (n = 4), random forests (n = 3) and support vector machines (n = 3). Features mostly correlated with NIHL and used in the models were: age (n = 6), duration of noise exposure (n = 5) and noise exposure level (n = 4). Five included studies used either split-sample validation (n = 3) or ten-fold cross-validation (n = 2). Assessment of accuracy ranged in value from 75.3% to 99% with a low prediction error/root-mean-square error in 3 studies. Only 2 studies measured discrimination risk using the receiver operating characteristic (ROC) curve and/or the area under ROC curve.

Conclusion

In spite of high accuracy and low prediction error of machine learning models, some improvement can be expected from larger sample sizes, multiple algorithm use, completed reports of model construction and the sufficient evaluation of calibration and discrimination risk.

Comparison of Multivariable Logistic Regression and Machine Learning Models for Predicting Bronchopulmonary Dysplasia or Death in Very Preterm Infants

Bronchopulmonary dysplasia (BPD) is the most prevalent and clinically significant complication of prematurity. Accurate identification of at-risk infants would enable ongoing intervention to improve outcomes. Although postnatal exposures are known to affect an infant's likelihood of developing BPD, most existing BPD prediction models do not allow risk to be evaluated at different time points, and/or are not suitable for use in ethno-diverse populations. A comprehensive approach to developing clinical prediction models avoids assumptions as to which method will yield the optimal results by testing multiple algorithms/models. We compared the performance of machine learning and logistic regression models in predicting BPD/death. Our main cohort included infants <33 weeks' gestational age (GA) admitted to a Canadian Neonatal Network site from 2016 to 2018 (n = 9,006) with all analyses repeated for the <29 weeks' GA subcohort (n = 4,246). Models were developed to predict, on days 1, 7, and 14 of admission to neonatal intensive care, the composite outcome of BPD/death prior to discharge. Ten-fold cross-validation and a 20% hold-out sample were used to measure area under the curve (AUC). Calibration intercepts and slopes were estimated by regressing the outcome on the log-odds of the predicted probabilities. The model AUCs ranged from 0.811 to 0.886. Model discrimination was lower in the <29 weeks' GA subcohort (AUCs 0.699-0.790). Several machine learning models had a suboptimal calibration intercept and/or slope (k-nearest neighbor, random forest, artificial neural network, stacking neural network ensemble). The top-performing algorithms will be used to develop multinomial models and an online risk estimator for predicting BPD severity and death that does not require information on ethnicity.

Application of tabu search-based Bayesian networks in exploring related factors of liver cirrhosis complicated with hepatic encephalopathy and disease identification

This study aimed to explore the related factors and strengths of hepatic cirrhosis complicated with hepatic encephalopathy (HE) by multivariate logistic regression analysis and tabu search-based Bayesian networks (BNs), and to deduce the probability of HE in patients with cirrhosis under different conditions through BN reasoning. Multivariate logistic regression analysis indicated that electrolyte disorders, infections, poor spirits, hepatorenal syndrome, hepatic diabetes, prothrombin time, and total bilirubin are associated with HE. Inferences by BNs found that infection, electrolyte disorder and hepatorenal syndrome are closely related to HE. Those three variables are also related to each other, indicating that the occurrence of any of those three complications may induce the other two complications. When those three complications occur simultaneously, the probability of HE may reach 0.90 or more. The BN constructed by the tabu search algorithm can analyze not only how the correlative factors affect HE but also their interrelationships. Reasoning using BNs can describe how HE is induced on the basis of the order in which doctors acquire patient information, which is consistent with the sequential process of clinical diagnosis and treatment.

Identification of swallowing disorders in early and mid-stage Parkinson's disease using pattern recognition of pharyngeal high-resolution manometry data

BACKGROUND

Parkinson's disease (PD) can cause severe dysphagia, especially later in disease progression. Early identification of swallowing dysfunction may lead to earlier intervention. Pharyngeal high-resolution manometry (HRM) provides complementary information to videofluoroscopy, with advantages of being quantitative and objective. Artificial neural network (ANN) classification can examine non-linear relationships among multiple variables with relatively low bias. We evaluated if ANN techniques could differentiate between patients with PD and healthy controls.

METHODS

Simultaneous videofluoroscopy and pharyngeal HRM were performed on 31 patients with early to mid-stage PD and 31 age- and sex-matched controls during thin-liquid swallows of 2 cc, 10 cc and comfortable sip volume. We performed multilayer-perceptron analyses on only videofluoroscopic data, only HRM data or a combination of the two. We also evaluated variability-based parameters, representing variability in manometric parameters across multiple swallows. We hypothesized that patients with PD and controls would be classified with at least 80% accuracy, and that combined videofluoroscopic and HRM data would classify participants better than either alone.

KEY RESULTS

Classification rates were highest with all parameters considered. Maximum classification rate was 82.3 ± 5.2%, recorded for 2 cc swallows. Inclusion of variability-based parameters improved classification rates. Classification rates using only manometric parameters were similar to those using all parameters, and rates were substantially lower for the comfortable sip volumes.

CONCLUSIONS & INFERENCES

Results from these classifications highlight the differences between swallowing function in patients with early and mid-stage PD and healthy controls. Early identification of swallowing dysfunction is key to developing preventative swallowing treatments for those with PD.

Predicting the hearing outcome in sudden sensorineural hearing loss via machine learning models

OBJECTIVE

Sudden sensorineural hearing loss (SSHL) is a multifactorial disorder with high heterogeneity, thus the outcomes vary widely. This study aimed to develop predictive models based on four machine learning methods for SSHL, identifying the best performer for clinical application.

DESIGN

Single-centre retrospective study.

SETTING

Chinese People's liberation army (PLA) hospital, Beijing, China.

PARTICIPANTS

A total of 1220 in-patient SSHL patients were enrolled between June 2008 and December 2015.

MAIN OUTCOME MEASURES

An advanced deep learning technique, deep belief network (DBN), together with the conventional logistic regression (LR), support vector machine (SVM) and multilayer perceptron (MLP) were developed to predict the dichotomised hearing outcome of SSHL by inputting six feature collections derived from 149 potential predictors. Accuracy, precision, recall, F-score and the area under the receiver operator characteristic curves (ROC-AUC) were exploited to compare the prediction performance of different models.

RESULTS

Overall the best predictive ability was provided by the DBN model when tested in the raw data set with 149 variables, achieving an accuracy of 77.58% and AUC of 0.84. Nevertheless, DBN yielded inferior performance after feature pruning. In contrast, the LR, SVM and MLP models demonstrated opposite trend as the greatest individual prediction powers were obtained when included merely three variables, with the ROC-AUC ranging from 0.79 to 0.81, and then decreased with the increasing size of input features combinations.

CONCLUSIONS

With the input of enough features, DBN can be a robust prediction tool for SSHL. But LR is more practical for early prediction in routine clinical application using three readily available variables, that is time elapse between symptom onset and study entry, initial hearing level and audiogram.