Machine learning to predict rapid progression of carotid atherosclerosis in patients with impaired glucose tolerance

Construction of machine learning models for recognizing comorbid anxiety in epilepsy patients based on their clinical and quantitative EEG features

BACKGROUND

This study aimed to construct prediction models for the recognizing of anxiety disorders (AD) in patients with epilepsy (PWEs) by combining clinical features with quantitative electroencephalogram (qEEG) features and using machine learning (ML).

METHODS

Nineteen clinical features and 20-min resting-state EEG were collected from 71 PWEs comorbid with AD and another 60 PWEs without AD who met the inclusion-exclusion criteria of this study. The EEG were preprocessed and 684 Phase Locking Value (PLV) and 76 Lempel-Ziv Complexity (LZC) features on four bands were extracted. The Fisher score method was used to rank all the derived features. We constructed four models for recognizing AD in PWEs, whether PWEs based on different combinations of features using eXtreme gradient boosting (XGboost) and evaluated these models using the five-fold cross-validation method.

RESULTS

The prediction model constructed by combining the clinical, PLV, and LZC features showed the best performance, with an accuracy of 96.18%, precision of 94.29%, sensitivity of 98.33%, F1-score of 96.06%, and Area Under the Curve (AUC) of 0.96. The Fisher score ranking results displayed that the top ten features were depression, educational attainment, α_P3LZC, α_T6-PzPLV, α_F7LZC, β_Fp2-O1PLV, θ_T4-CzPLV, θ_F7-PzPLV, α_Fp2LZC, and θ_T4-PzPLV.

CONCLUSIONS

The model, constructed by combining the clinical and qEEG features PLV and LZC, efficiently identified the presence of AD comorbidity in PWEs and might have the potential to complement the clinical diagnosis. Our findings suggest that LZC features in the α band and PLV features in Fp2-O1 may be potential biomarkers for diagnosing AD in PWEs.

Identification of Prediabetes Discussions in Unstructured Clinical Documentation: Validation of a Natural Language Processing Algorithm

Background

Prediabetes affects 1 in 3 US adults. Most are not receiving evidence-based interventions, so understanding how providers discuss prediabetes with patients will inform how to improve their care.

Objective

This study aimed to develop a natural language processing (NLP) algorithm using machine learning techniques to identify discussions of prediabetes in narrative documentation.

Methods

We developed and applied a keyword search strategy to identify discussions of prediabetes in clinical documentation for patients with prediabetes. We manually reviewed matching notes to determine which represented actual prediabetes discussions. We applied 7 machine learning models against our manual annotation.

Results

Machine learning classifiers were able to achieve classification results that were close to human performance with up to 98% precision and recall to identify prediabetes discussions in clinical documentation.

Conclusions

We demonstrated that prediabetes discussions can be accurately identified using an NLP algorithm. This approach can be used to understand and identify prediabetes management practices in primary care, thereby informing interventions to improve guideline-concordant care.

Machine learning models for screening carotid atherosclerosis in asymptomatic adults

Carotid atherosclerosis (CAS) is a risk factor for cardiovascular and cerebrovascular events, but duplex ultrasonography isn’t recommended in routine screening for asymptomatic populations according to medical guidelines. We aim to develop machine learning models to screen CAS in asymptomatic adults. A total of 2732 asymptomatic subjects for routine physical examination in our hospital were included in the study. We developed machine learning models to classify subjects with or without CAS using decision tree, random forest (RF), extreme gradient boosting (XGBoost), support vector machine (SVM) and multilayer perceptron (MLP) with 17 candidate features. The performance of models was assessed on the testing dataset. The model using MLP achieved the highest accuracy (0.748), positive predictive value (0.743), F1 score (0.742), area under receiver operating characteristic curve (AUC) (0.766) and Kappa score (0.445) among all classifiers. It’s followed by models using XGBoost and SVM. In conclusion, the model using MLP is the best one to screen CAS in asymptomatic adults based on the results from routine physical examination, followed by using XGBoost and SVM. Those models may provide an effective and applicable method for physician and primary care doctors to screen asymptomatic CAS without risk factors in general population, and improve risk predictions and preventions of cardiovascular and cerebrovascular events in asymptomatic adults.

The prediction of asymptomatic carotid atherosclerosis with electronic health records: a comparative study of six machine learning models

BACKGROUND

Screening carotid B-mode ultrasonography is a frequently used method to detect subjects with carotid atherosclerosis (CAS). Due to the asymptomatic progression of most CAS patients, early identification is challenging for clinicians, and it may trigger ischemic stroke. Recently, machine learning has shown a strong ability to classify data and a potential for prediction in the medical field. The combined use of machine learning and the electronic health records of patients could provide clinicians with a more convenient and precise method to identify asymptomatic CAS.

METHODS

Retrospective cohort study using routine clinical data of medical check-up subjects from April 19, 2010 to November 15, 2019. Six machine learning models (logistic regression [LR], random forest [RF], decision tree [DT], eXtreme Gradient Boosting [XGB], Gaussian Naïve Bayes [GNB], and K-Nearest Neighbour [KNN]) were used to predict asymptomatic CAS and compared their predictability in terms of the area under the receiver operating characteristic curve (AUCROC), accuracy (ACC), and F1 score (F1).

RESULTS

Of the 18,441 subjects, 6553 were diagnosed with asymptomatic CAS. Compared to DT (AUCROC 0.628, ACC 65.4%, and F1 52.5%), the other five models improved prediction: KNN + 7.6% (0.704, 68.8%, and 50.9%, respectively), GNB + 12.5% (0.753, 67.0%, and 46.8%, respectively), XGB + 16.0% (0.788, 73.4%, and 55.7%, respectively), RF + 16.6% (0.794, 74.5%, and 56.8%, respectively) and LR + 18.1% (0.809, 74.7%, and 59.9%, respectively). The highest achieving model, LR predicted 1045/1966 cases (sensitivity 53.2%) and 3088/3566 non-cases (specificity 86.6%). A tenfold cross-validation scheme further verified the predictive ability of the LR.

CONCLUSIONS

Among machine learning models, LR showed optimal performance in predicting asymptomatic CAS. Our findings set the stage for an early automatic alarming system, allowing a more precise allocation of CAS prevention measures to individuals probably to benefit most.

Application of machine learning in understanding atherosclerosis: Emerging insights

Biological processes are incredibly complex—integrating molecular signaling networks involved in multicellular communication and function, thus maintaining homeostasis. Dysfunction of these processes can result in the disruption of homeostasis, leading to the development of several disease processes including atherosclerosis. We have significantly advanced our understanding of bioprocesses in atherosclerosis, and in doing so, we are beginning to appreciate the complexities, intricacies, and heterogeneity atherosclerosi. We are also now better equipped to acquire, store, and process the vast amount of biological data needed to shed light on the biological circuitry involved. Such data can be analyzed within machine learning frameworks to better tease out such complex relationships. Indeed, there has been an increasing number of studies applying machine learning methods for patient risk stratification based on comorbidities, multi-modality image processing, and biomarker discovery pertaining to atherosclerotic plaque formation. Here, we focus on current applications of machine learning to provide insight into atherosclerotic plaque formation and better understand atherosclerotic plaque progression in patients with cardiovascular disease.

Pioglitazone for prevention or delay of type 2 diabetes mellitus and its associated complications in people at risk for the development of type 2 diabetes mellitus

BACKGROUND

The term prediabetes is used to describe a population with an elevated risk of developing type 2 diabetes mellitus (T2DM). With projections of an increase in the incidence of T2DM, prevention or delay of the disease and its complications is paramount. It is currently unknown whether pioglitazone is beneficial in the treatment of people with increased risk of developing T2DM.

OBJECTIVES

To assess the effects of pioglitazone for prevention or delay of T2DM and its associated complications in people at risk of developing T2DM.

SEARCH METHODS

We searched CENTRAL, MEDLINE, Chinese databases, ICTRP Search Portal and ClinicalTrials.gov. We did not apply any language restrictions. Further, we investigated the reference lists of all included studies and reviews. We tried to contact all study authors. The date of the last search of databases was November 2019 (March 2020 for Chinese databases).

SELECTION CRITERIA

We included randomised controlled trials (RCTs) with a minimum duration of 24 weeks, and participants diagnosed with intermediate hyperglycaemia with no concomitant diseases, comparing pioglitazone as monotherapy or part of dual therapy with other glucose-lowering drugs, behaviour-changing interventions, placebo or no intervention.

DATA COLLECTION AND ANALYSIS

Two review authors independently screened abstracts, read full-text articles and records, assessed risk of bias and extracted data. We performed meta-analyses with a random-effects model and calculated risk ratios (RRs) for dichotomous outcomes and mean differences (MDs) for continuous outcomes, with 95% confidence intervals (CIs) for effect estimates. We evaluated the certainty of the evidence with the GRADE.

MAIN RESULTS

We included 27 studies with a total of 4186 randomised participants. The size of individual studies ranged between 43 and 605 participants and the duration varied between 6 and 36 months. We judged none of the included studies as having low risk of bias across all 'Risk of bias' domains. Most studies identified people at increased risk of T2DM by impaired fasting glucose or impaired glucose tolerance (IGT), or both. Our main outcome measures were all-cause mortality, incidence of T2DM, serious adverse events (SAEs), cardiovascular mortality, nonfatal myocardial infarction or stroke (NMI/S), health-related quality of life (QoL) and socioeconomic effects. The following comparisons mostly reported only a fraction of our main outcome set. Three studies compared pioglitazone with metformin. They did not report all-cause and cardiovascular mortality, NMI/S, QoL or socioeconomic effects. Incidence of T2DM was 9/168 participants in the pioglitazone groups versus 9/163 participants in the metformin groups (RR 0.98, 95% CI 0.40 to 2.38; P = 0.96; 3 studies, 331 participants; low-certainty evidence). No SAEs were reported in two studies (201 participants; low-certainty evidence). One study compared pioglitazone with acarbose. Incidence of T2DM was 1/50 participants in the pioglitazone group versus 2/46 participants in the acarbose group (very low-certainty evidence). No participant experienced a SAE (very low-certainty evidence).One study compared pioglitazone with repaglinide. Incidence of T2DM was 2/48 participants in the pioglitazone group versus 1/48 participants in the repaglinide group (low-certainty evidence). No participant experienced a SAE (low-certainty evidence). One study compared pioglitazone with a personalised diet and exercise consultation. All-cause and cardiovascular mortality, NMI/S, QoL or socioeconomic effects were not reported. Incidence of T2DM was 2/48 participants in the pioglitazone group versus 5/48 participants in the diet and exercise consultation group (low-certainty evidence). No participant experienced a SAE (low-certainty evidence). Six studies compared pioglitazone with placebo. No study reported on QoL or socioeconomic effects. All-cause mortality was 5/577 participants the in the pioglitazone groups versus 2/579 participants in the placebo groups (Peto odds ratio 2.38, 95% CI 0.54 to 10.50; P = 0.25; 4 studies, 1156 participants; very low-certainty evidence). Incidence of T2DM was 80/700 participants in the pioglitazone groups versus 131/695 participants in the placebo groups (RR 0.40, 95% CI 0.17 to 0.95; P = 0.04; 6 studies, 1395 participants; low-certainty evidence). There were 3/93 participants with SAEs in the pioglitazone groups versus 1/94 participants in the placebo groups (RR 3.00, 95% CI 0.32 to 28.22; P = 0.34; 2 studies, 187 participants; very low-certainty evidence). However, the largest study for this comparison did not distinguish between serious and non-serious adverse events. This study reported that 121/303 (39.9%) participants in the pioglitazone group versus 151/299 (50.5%) participants in the placebo group experienced an adverse event (P = 0.03). One study observed cardiovascular mortality in 2/181 participants in the pioglitazone group versus 0/186 participants in the placebo group (RR 5.14, 95% CI 0.25 to 106.28; P = 0.29; very low-certainty evidence). One study observed NMI in 2/303 participants in the pioglitazone group versus 1/299 participants in the placebo group (RR 1.97: 95% CI 0.18 to 21.65; P = 0.58; very low-certainty evidence). Twenty-one studies compared pioglitazone with no intervention. No study reported on cardiovascular mortality, NMI/S, QoL or socioeconomic effects. All-cause mortality was 11/441 participants in the pioglitazone groups versus 12/425 participants in the no-intervention groups (RR 0.85, 95% CI 0.38 to 1.91; P = 0.70; 3 studies, 866 participants; very low-certainty evidence). Incidence of T2DM was 60/1034 participants in the pioglitazone groups versus 197/1019 participants in the no-intervention groups (RR 0.31, 95% CI 0.23 to 0.40; P < 0.001; 16 studies, 2053 participants; moderate-certainty evidence). Studies reported SAEs in 16/610 participants in the pioglitazone groups versus 21/601 participants in the no-intervention groups (RR 0.71, 95% CI 0.38 to 1.32; P = 0.28; 7 studies, 1211 participants; low-certainty evidence). We identified two ongoing studies, comparing pioglitazone with placebo and with other glucose-lowering drugs. These studies, with 2694 participants. may contribute evidence to future updates of this review.

AUTHORS' CONCLUSIONS

Pioglitazone reduced or delayed the development of T2DM in people at increased risk of T2DM compared with placebo (low-certainty evidence) and compared with no intervention (moderate-certainty evidence). It is unclear whether the effect of pioglitazone is sustained once discontinued. Pioglitazone compared with metformin neither showed advantage nor disadvantage regarding the development of T2DM in people at increased risk (low-certainty evidence). The data and reporting of all-cause mortality, SAEs, micro- and macrovascular complications were generally sparse. None of the included studies reported on QoL or socioeconomic effects.

Artificial intelligence in drug combination therapy

Currently, the development of medicines for complex diseases requires the development of combination drug therapies. It is necessary because in many cases, one drug cannot target all necessary points of intervention. For example, in cancer therapy, a physician often meets a patient having a genomic profile including more than five molecular aberrations. Drug combination therapy has been an area of interest for a while, for example the classical work of Loewe devoted to the synergism of drugs was published in 1928-and it is still used in calculations for optimal drug combinations. More recently, over the past several years, there has been an explosion in the available information related to the properties of drugs and the biomedical parameters of patients. For the drugs, hundreds of 2D and 3D molecular descriptors for medicines are now available, while for patients, large data sets related to genetic/proteomic and metabolomics profiles of the patients are now available, as well as the more traditional data relating to the histology, history of treatments, pretreatment state of the organism, etc. Moreover, during disease progression, the genetic profile can change. Thus, the ability to optimize drug combinations for each patient is rapidly moving beyond the comprehension and capabilities of an individual physician. This is the reason, that biomedical informatics methods have been developed and one of the more promising directions in this field is the application of artificial intelligence (AI). In this review, we discuss several AI methods that have been successfully implemented in several instances of combination drug therapy from HIV, hypertension, infectious diseases to cancer. The data clearly show that the combination of rule-based expert systems with machine learning algorithms may be promising direction in this field.

A Special Report on Changing Trends in Preventive Stroke/Cardiovascular Risk Assessment Via B-Mode Ultrasonography

PURPOSE OF REVIEW

Cardiovascular disease (CVD) and stroke risk assessment have been largely based on the success of traditional statistically derived risk calculators such as Pooled Cohort Risk Score or Framingham Risk Score. However, over the last decade, automated computational paradigms such as machine learning (ML) and deep learning (DL) techniques have penetrated into a variety of medical domains including CVD/stroke risk assessment. This review is mainly focused on the changing trends in CVD/stroke risk assessment and its stratification from statistical-based models to ML-based paradigms using non-invasive carotid ultrasonography.

RECENT FINDINGS

In this review, ML-based strategies are categorized into two types: non-image (or conventional ML-based) and image-based (or integrated ML-based). The success of conventional (non-image-based) ML-based algorithms lies in the different data-driven patterns or features which are used to train the ML systems. Typically these features are the patients' demographics, serum biomarkers, and multiple clinical parameters. The integrated (image-based) ML-based algorithms integrate the features derived from the ultrasound scans of the arterial walls (such as morphological measurements) with conventional risk factors in ML frameworks. Even though the review covers ML-based system designs for carotid and coronary ultrasonography, the main focus of the review is on CVD/stroke risk scores based on carotid ultrasound. There are two key conclusions from this review: (i) fusion of image-based features with conventional cardiovascular risk factors can lead to more accurate CVD/stroke risk stratification; (ii) the ability to handle multiple sources of information in big data framework using artificial intelligence-based paradigms (such as ML and DL) is likely to be the future in preventive CVD/stroke risk assessment.

Applying Machine Learning Techniques to Identify Undiagnosed Patients with Exocrine Pancreatic Insufficiency

BACKGROUND

Exocrine pancreatic insufficiency (EPI) is a serious condition characterized by a lack of functional exocrine pancreatic enzymes and the resultant inability to properly digest nutrients. EPI can be caused by a variety of disorders, including chronic pancreatitis, pancreatic cancer, and celiac disease. EPI remains underdiagnosed because of the nonspecific nature of clinical symptoms, lack of an ideal diagnostic test, and the inability to easily identify affected patients using administrative claims data.

OBJECTIVES

To develop a machine learning model that identifies patients in a commercial medical claims database who likely have EPI but are undiagnosed.

METHODS

A machine learning algorithm was developed in Scikit-learn, a Python module. The study population, selected from the 2014 Truven MarketScan® Commercial Claims Database, consisted of patients with EPI-prone conditions. Patients were labeled with 290 condition category flags and split into actual positive EPI cases, actual negative EPI cases, and unlabeled cases. The study population was then randomly divided into a training subset and a testing subset. The training subset was used to determine the performance metrics of 27 models and to select the highest performing model, and the testing subset was used to evaluate performance of the best machine learning model.

RESULTS

The study population consisted of 2088 actual positive EPI cases, 1077 actual negative EPI cases, and 437 530 unlabeled cases. In the best performing model, the precision, recall, and accuracy were 0.91, 0.80, and 0.86, respectively. The best-performing model estimated that the number of patients likely to have EPI was about 12 times the number of patients directly identified as EPI-positive through a claims analysis in the study population. The most important features in assigning EPI probability were the presence or absence of diagnosis codes related to pancreatic and digestive conditions.

CONCLUSIONS

Machine learning techniques demonstrated high predictive power in identifying patients with EPI and could facilitate an enhanced understanding of its etiology and help to identify patients for possible diagnosis and treatment.