Unsupervised Learning Research Articles

Six clinical phenotypes with prognostic implications were identified by unsupervised machine learning in children and adolescents with SARS-CoV-2 infection: results from a German nationwide registry

ObjectivePhenotypes are important for patient classification, disease prognostication, and treatment customization. We aimed to identify distinct clinical phenotypes of children and adolescents hospitalized with SARS-CoV-2 infection, and to evaluate their prognostic differences.MethodsThe German Society of Pediatric Infectious Diseases (DGPI) registry is a nationwide, prospective registry for children and adolescents hospitalized with a SARS-CoV-2 infection in Germany. We applied hierarchical clustering for phenotype identification with variables including sex, SARS-CoV-2-related symptoms on admission, pre-existing comorbidities, clinically relevant coinfection, and SARS-CoV-2 risk factors. Outcomes of this study were: discharge status and ICU admission. Discharge status was categorized as: full recovery, residual symptoms, and unfavorable prognosis (including consequential damage that has already been identified as potentially irreversible at the time of discharge and SARS-CoV-2-related death). After acquiring the phenotypes, we evaluated their correlation with discharge status by multinomial logistic regression model, and correlation with ICU admission by binary logistic regression model. We conducted an analogous subgroup analysis for those aged < 1 year (infants) and those aged ⩾ 1 year (non-infants).ResultsThe DGPI registry enrolled 6983 patients, through which we identified six distinct phenotypes for children and adolescents with SARS-CoV-2 which can be characterized by their symptom pattern: phenotype A had a range of symptoms, while predominant symptoms of patients with other phenotypes were gastrointestinal (95.9%, B), asymptomatic (95.9%, C), lower respiratory tract (49.8%, D), lower respiratory tract and ear, nose and throat (86.2% and 41.7%, E), and neurological (99.2%, F). Regarding discharge status, patients with D and E phenotype had the highest odds of having residual symptoms (OR: 1.33 [1.11, 1.59] and 1.91 [1.65, 2.21], respectively) and patients with phenotype D were significantly more likely (OR: 4.00 [1.95, 8.19]) to have an unfavorable prognosis. Regarding ICU, patients with phenotype D had higher possibility of ICU admission than staying in normal ward (OR: 4.26 [3.06, 5.98]), compared to patients with phenotype A. The outcomes observed in the infants and non-infants closely resembled those of the entire registered population, except infants did not exhibit typical neurological/neuromuscular phenotypes.ConclusionsPhenotypes enable pediatric patient stratification by risk and thus assist in personalized patient care. Our findings in SARS-CoV-2-infected population might also be transferable to other infectious diseases.

Effective Machine Learning Solution for State Classification and Productivity Identification: Case of Pneumatic Pressing Machine

The fourth industrial revolution (Industry 4.0) brought significant changes in manufacturing, driven by technologies like artificial intelligence (AI), Internet of Things (IoT), 5G, robotics, and big data analytics. For industries to remain competitive, the primary goals must be the improvement of the efficiency and safety of machinery, the reduction of production costs, and the enhancement of product quality. Predictive maintenance (PdM) utilizes historical data and AI models to diagnose equipment’s health and predict the remaining useful life (RUL), providing critical insights for machinery effectiveness and product manufacturing. This prediction is a critical strategy to maximize the useful life of equipment, especially in large-scale and important infostructures. This study focuses on developing an unsupervised machine state-classification solution utilizing real-world industrial measurements collected from a pneumatic pressing machine. Unsupervised machine learning (ML) models were tested to diagnose and output the working state of the pressing machine at each given point (offline, idle, pressing, defective). Our research contributes to extracting valuable insights regarding real-world industrial settings for PdM and production efficiency using unsupervised ML, promoting operation safety, cost reduction, and productivity enhancement in modern industries.

Supervised and unsupervised machine learning for elemental changes evaluation of torrefied biochars

This study implements a comprehensive analysis of supervised and unsupervised learning to evaluate elemental changes and investigate the merits of the machine learning method for torrefied biochar property analysis. The focus is on data analysis using artificial neural networks (ANNs) and k-means algorithms for analyzing the carbonization index (CI) and deoxygenation index (DI) after biomass torrefaction. The predictive response surfaces for CI and DI are formulated and analyzed accordingly by preprocessing the raw data as either lignocellulosic or microalgal datatype. Based on ANNs, the relative importance weights of the factors of temperature, duration, and biomass type are 45.87 %, 23.19 %, and 30.94 %, respectively, for the CI response, while 35.27 %, 26.62 %, and 38.11 %, respectively for the DI response. For k-means analysis, the optimal number of clusters for lignocellulosic and microalgal datasets are two and three, respectively. The R2 value of ANNs is 0.9565. The distribution and percentage of the dataset within the clusters are influenced by the time for the lignocellulosic datatype, while they are influenced by temperature for the microalgal datatype. The obtained results are conducive to improving the cognition of the machine learning method on torrefaction performance analysis.

Selection of Optimum Enhancement Technique for the Discrimination of Alkali Granites and Syenites around Suryamalai Batholith of Central Tamil Nadu, India

Image processing techniques such as Band Ratio, Principal Component Analysis, Supervised, Unsupervised classification and Reclassification techniques are generally used as an enhancement technique for alteration zone demarcation and lithological discrimination in geosciences environment. In the present study, Landsat OLI satellite imagery has been utilized to produce various combinations of image enhancement to discriminate alkali syenite and alkali granite of Proterozoic age. These techniques were also used to differentiate three phases of granites reported in Suryamalai Batholith which is very difficult to map by conventional method due to its rugged topographic nature. 12 types of enhancement techniques were applied to differentiate rock alkali syenite from granite, however only few methods such as Kaufmanns ratio, ratio of 5/7, 3/1 and 3/5 and supervised classification were found to be suitable for the discrimination of those rocky types. Band ratio 5/7, 3/1 and 3/5, Crosta analysis and thermal reclassification techniques significantly brought to light the difference between the metamorphic rocks such as fissile hornblende biotite gneiss and hornblende biotite gneiss of Archaean age. Similarly, phase I and phase III granites were perfectly discriminated from all the image enhancement techniques. It is differentiated mainly based on its grain size and mineral variation between these two rock types. Study revealed that standard digital image processing technique can be effectively used for lithological mapping.

Discretionary dissemination on Twitter

AbstractThe study provides large‐scale descriptive evidence on the timing and nature of corporate financial tweeting. Using an unsupervised machine learning approach to analyze 24 million tweets posted by S&amp;P 1500 firms from 2012 to 2020, we find that firms are more likely to tweet financial information around significantly negative or positive news events, such as earnings announcements and the filing of financial statements. This convex U‐shaped relation between the likelihood of posting financial tweets and the materiality of accounting events becomes stronger over time. Whereas research based on early samples concludes that firms are less likely to disseminate financial information on Twitter when the news is bad and material, the symmetric dissemination behavior we find suggests that these conclusions should be revised. We also show that a machine learning algorithm (Twitter‐Latent Dirichlet Allocation) is superior to a dictionary approach in classifying short messages like tweets.

Early Season Crop Acreage Estimation for Pigeon Pea using SAR Data: A Case Study of Kalaburagi District, Karnataka

Crop acreage estimates using space-based observations during mid-season and pre-harvest periods are operationally provided for major Kharif and Rabi crops in India. These are critical inputs for planning and decision-making concerning food security, storage, pricing, and procurement. This work explores the potential of Satellite C-band Synthetic Aperture Radar (SAR) data in VV and VH polarization, acquired during monsoon season, for early crop acreage assessment of Pigeon Pea. The present study evaluates two classification approaches based on multi-temporal SAR data to generate a spatial distribution map of the pigeon pea crop in the Kalaburagi region of Karnataka. The temporal profiles of sigma nought (backscatter coefficient) and Entropy vs. Alpha Angle were studied with reference to crop phenology. Random forest algorithm has been used to classify multi-temporal SAR Ground Range Detected (GRD) data from June to November 2019. Temporal SAR data from June to early November with VV+VH polarization was optimal for tur area assessment with 85% accuracy. This study also examines phase and intensity information from 3-date SAR Single Look Complex (SLC) data with an unsupervised Wishart Classification approach for improved crop acreage estimation. The results emphasize the effectiveness of dual-polarimetric SAR data for timely crop inventory of Pigeon Pea during the Kharif season.

An Unsupervised Image Fusion Approach Using Convolution Neural Network for Feature Enhancement of Medical Brain Images

Image fusion is the process of amalgamation of two or more images into a single image that contains composite, enriched with diverse details from the original images. In the medical field, image fusion serves as an indispensable tool for elevating the precision of medical imagery and facilitating diagnostic processes. With the advent of deep learning, there has been a significant increase in the accuracy and effectiveness of image fusion techniques. This paper presents a deeplearningbased approach for medical image fusion that combines the advantages of deep-learning techniques with traditional image fusion methods. The proposed method is evaluated on medical data from different modalities, and the experimental results show that the proposed method outperforms existing state-of-the-art image fusion techniques.

Physics-guided deep learning-based real-time image reconstruction of Fourier-domain optical coherence tomography

In this paper, we introduce a physics-guided deep learning approach for high-quality, real-time Fourier-domain optical coherence tomography (FD-OCT) image reconstruction. Unlike traditional supervised deep learning methods, the proposed method employs unsupervised learning. It leverages the underlying OCT imaging physics to guide the neural networks, which could thus generate high-quality images and provide a physically sound solution to the original problem. Evaluations on synthetic and experimental datasets demonstrate the superior performance of our proposed physics-guided deep learning approach. The method achieves the highest image quality metrics compared to the inverse discrete Fourier transform (IDFT), the optimization-based methods, and several state-of-the-art methods based on deep learning. Our method enables real-time frame rates of 232 fps for synthetic images and 87 fps for experimental images, which represents significant improvements over existing techniques. Our physics-guided deep learning-based approach could offer a promising solution for FD-OCT image reconstruction, which potentially paves the way for leveraging the power of deep learning in real-world OCT imaging applications.

A GIS-Based Framework to Analyze the Behavior of Urban Greenery During Heatwaves Using Satellite Data

This work proposes a new unsupervised method to evaluate the behavior of urban green areas in the presence of heatwave scenarios by analyzing three indices extracted from satellite data: the Normalized Difference Vegetation Index (NDVI), the Normalized Difference Moisture Index (NDMI), and Land Surface Temperature (LST). The aim of this research is to analyze the behavior of urban vegetation types during heatwaves through the analysis of these three indices. To evaluate how these indices characterize urban green areas during heatwaves, an unsupervised classification method of the three indices is proposed that uses the Elbow method to determine the optimal number of classes and the Jenks classification algorithm. Each class is assigned a Gaussian fuzzy set and the green urban areas are classified using zonal statistics operators. The membership degree of the corresponding fuzzy set is calculated to assess the reliability of the classification. Finally, for each type of greenery, the frequencies of types of green areas belonging to NDVI, NDMI, and LST classes are analyzed to evaluate their behavior during heatwaves. The framework was tested in an urban area consisting of the city of Naples (Italy). The results show that some types of greenery, such as deciduous forests and olive groves, are more efficient, in terms of health status and cooling effect, than other types of urban green areas during heatwaves; they are classified with NDVI and NDMI values of mainly High and Medium High, and maximum LST values of Medium Low. Conversely, uncultivated areas show critical behaviors during heatwaves; they are classified with maximum NDVI and NDMI values of Medium Low and maximum LST values of Medium High. The research results represent a support to urban planners and local municipalities in designing effective strategies and nature-based solutions to deal with heat waves in urban settlements.

Segment-Based Unsupervised Deep Learning for Human Activity Recognition using Accelerometer Data and SBOA based Channel Attention Networks

The deep learning community has increasingly focused on the critical challenges of human activity segmentation and detection based on sensors, which have numerous real-world applications. In most prior efforts, activity segmentation and recognition have been treated as separate processes, relying on pre-segmented sensor streams. This research proposes an unsupervised deep learning approach for Human Activity Recognition (HAR) that is segment-based, with an emphasis on activity continuity. The approach integrates segment-based SimCLR with Segment Feature Decorrelation (SDFD) and a new framework that leverages pairs of segment data for contrastive learning of visual representations. Furthermore, the Secretary Bird Optimization Algorithm (SBOA) and Channel Attention with Spatial Attention Network (CASANet) are utilized to enhance the performance of sensor-based human activity detection. CASANet effectively extracts key features and spatial dependencies in sensor data, while SBOA optimizes the model for greater accuracy and generalization. Evaluations on two publicly available datasets—Mhealth and PAMAP2—demonstrated an average F1 score of 98%, highlighting the approach’s efficacy in improving activity recognition performance.

Credit Anomaly Detection Method based on Bayesian Networks

Due to the virtual nature of online transactions and the underdeveloped personal credit risk assessment system of Internet finance lending platforms in China, these platforms often struggle to verify the creditworthiness of applicants prone to fraud and default, leading to loan losses. As a result, risk control has become the core of Internet lending platforms. Modern risk control relies on statistical and data mining techniques to analyze and model data, uncover patterns that indicate loan defaults, and identify anomalies. This paper proposes a Bayesian network-based credit anomaly detection method, which detects anomalies by calculating and ranking the joint anomaly probabilities of sample instances. The technique operates under unsupervised learning and can effectively handle missing and imbalanced data. This method analyzes a loan credit dataset with 32,581 samples and 12 features of the customers. The result of the analysis demonstrates the feasibility and effectiveness of this method in detecting anomalies in online loans.

Unsupervised end-to-end training with a self-defined target

Abstract Designing algorithms for versatile AI hardware that can learn on the edge using both labeled and unlabeled data is challenging. Deep end-to-end training methods incorporating phases of self-supervised and supervised learning are accurate and adaptable to input data but self-supervised learning requires even more computational and memory resources than supervised learning, too high for current embedded hardware. Conversely, unsupervised layer-by-layer training, such as Hebbian learning, is more compatible with existing hardware but doesn't integrate well with supervised learning. &amp;#xD;To address this, we propose a method enabling networks or hardware designed for end-to-end supervised learning to also perform high-performance unsupervised learning by adding two simple elements to the output layer: Winner-Take-All (WTA) selectivity and homeostasis regularization. These mechanisms introduce a 'self-defined target' for unlabeled data, allowing purely unsupervised training for both fully-connected and convolutional layers using backpropagation or equilibrium propagation on datasets like MNIST (up to 99.2%), Fashion-MNIST (up to 90.3%), and SVHN (up to 81.5%).&amp;#xD;We extend this method to semi-supervised learning, adjusting targets based on data type, achieving 96.6% accuracy with only 600 labeled MNIST samples in a multi-layer perceptron. Our results show that this approach can effectively enable networks and hardware initially dedicated to supervised learning to also perform unsupervised learning, adapting to varying availability of labeled data.

Early Fault Detection and Operator-Based MIMO Fault-Tolerant Temperature Control of Microreactor

A microreactor is a chemical reaction device that mixes liquids in a very narrow channel and continuously generates reactions. They are attracting attention as next-generation chemical reaction devices because of their ability to achieve small-scale and highly efficient reactions compared to the conventional badge method. However, the challenge is to design a control system that is tolerant of faults in some of the enormous number of sensors in order to achieve parallel production by numbering up. In a previous study, a simultaneous control system for two different temperatures was proposed in an experimental system that imitated the microreactor cooled by Peltier devices. In addition, a fault-tolerant control system for one area has also been proposed. However, the fault-tolerant control system could not be applied to the control system of two temperatures in the previous study. In this paper, we extend it to a two-input, two-output fault-tolerant control system. We also use a fault detection system that combines ChangeFinder, a time-series data analysis method, and One-Class SVM, an unsupervised learning method. Finally, the effectiveness of the proposed method is confirmed by experiments.

Review of Convolutional Neural Network

Over the past decade, the domain of deep learning has emerged as a swiftly expanding area of interest among researchers globally. It offers substantial benefits over traditional shallow networks, particularly in the realms of feature extraction and model fitting. Deep learning excels at uncovering intricate, distributed features from raw data, which exhibit robust generalization capabilities. It has triumphantly addressed challenges that were once deemed intractable within the field of artificial intelligence. With the exponential growth in the volume of training data and a marked increase in computational power, deep learning has achieved remarkable strides and found applications across a spectrum of domains, including but not limited to object detection, computer vision, natural language processing, speech recognition, and semantic parsing, thereby accelerating the evolution of artificial intelligence. Deep learning encompasses a series of non-linear transformations organized hierarchically, with deep neural networks representing the predominant form of contemporary deep learning methodologies. Inspired by the neural connections found in the animal visual cortex, these networks are characterized by local connectivity, shared weights, and pooling operations. Such attributes not only simplify the network model and curtail the number of trainable parameters but also engender invariance to shifts, distortions, and scaling, thereby endowing the model with enhanced robustness and fault tolerance. This makes the training and optimization of the network structure more manageable. This paper commences with a retrospective on the evolution of convolutional neural networks (CNNs). Subsequently, it delineates the architectures of neuron models and multilayer perceptron. It proceeds to dissect the CNN architecture, which typically encompasses a sequence of convolutional and pooling layers, culminating in fully connected layers, each serving distinct functions. The paper also elaborates on various enhanced algorithms for CNNs, such as the “Network in Network” and “Spatial Transformer Networks,” while also introducing supervised and unsupervised learning methodologies pertinent to CNNs and highlighting several prevalent open-source tools. Further, the paper scrutinizes the application of CNNs in various fields, including image classification, facial recognition, audio retrieval, electrocardiogram analysis, and object detection. It posits the amalgamation of CNNs with recurrent neural networks as a potential alternative for training datasets. The research culminates in the design of CNN structures with varying parameters and depths, supported by experimental analysis that elucidates the interplay among these parameters and the ramifications of their configuration. Ultimately, the paper synthesizes the merits and lingering challenges associated with CNNs and their practical applications.

A novel classification of dyslipidaemia through the analysis of five million lipid profiles: an unsupervised machine learning approach

Abstract Background Dyslipidemia encompasses a wide range of lipoprotein disorders categorised through two classifications (Fredrickson-Levy [FL] or Sniderman).(1,2) However, both classifications are criticised for relying on incomplete knowledge of lipoprotein metabolism, especially with the emergence of novel treatment options and variations in individual treatment responses.(3) Clustering, an unsupervised machine learning (ML) algorithm that can process a wide range of variables, has the potential to unmask patient groups with distinct molecular profiles and unique therapeutic targets that can inform more effective prevention strategies for cardiovascular disease (CVD).(4) Aim We aimed to use unsupervised ML algorithms to discover intrinsic dyslipidaemia categories from lipoprotein measurements, recognise the necessary components of lipid panels for classification, and analyse the similarities between the newly formed clusters, FL and Sniderman classifications. Methods Lipid profiles of 5,080,248 patients were obtained from the ‘Very Large Database of Lipids’ database. This yielded up to 78 blood components per patient, including at least 31 lipoprotein variables. The analysis involved unsupervised K-means clustering with optimised values for K and the subset of variables, determined in an unsupervised manner using a suitable measure of complexity. We then interpreted our clusters using probabilistic decision trees to provide compact and interpretable representations. Finally, we compared the clusters with Sniderman and FL categories. Results In a completely unsupervised fashion, we identified 14 clusters that could be matched to Sniderman categories. The confusion matrix showed total agreement of 76% (see Figure 1, left panel), relative Cohen’s kappa of 0.78 (the relative version captures accuracy on categories containing smaller numbers of patient profiles) and an accuracy of 96% on the small Type III class. Similar results were observed when matching to FL types. We accurately represented our clusters using probabilistic decision trees of small depth (see Figure 2). We discovered that the data had low intrinsic dimension and a manifold-like structure in which the different clusters could be illustrated (see Figure 1, right panel). Specifically, only 3 variables were needed to obtain our classification: apolipoprotein b, total cholesterol and triglycerides. Conclusion We showed that completely unsupervised ML techniques can uncover dyslipidaemia categories in lipoprotein profiles from a large patient population. The categories largely align with existing classifications based on prior knowledge of lipoprotein metabolism. Furthermore, few lipoprotein variables were required for categorisation (low-dimension data), which could aid in determining which lipoproteins should be measured in a clinical setting. Further analysis of the differences between ML clusters and traditional classifications is needed, which may enhance CVD risk management.

Using Unsupervised Learning Methods to Explore the Trajectory of Long-Term Care Service in Taiwan

Abstract Background Understanding the temporal trends in long-term care (LTC) utilization enables more effective planning by government and service providers. This study employed machine learning methods to explore LTC utilization trajectories among care recipients in a southern county of Taiwan. Methods Administrative data from the government-funded LTC programme was utilized, with a total of 24,614 participants included after excluding those with less than six months of data. Daily service utilization records, mainly focusing on caregiving services, were aggregated, and the number of days each participant used LTC services within every 30-day period post-application was calculated. Utilization rates during 6 and 12 months post-application were separately analyzed. K-means clustering, an unsupervised machine learning method, was employed to select the optimal classification and identify LTC utilization trajectories using 10-fold cross validation. Temporal frequency variations of each caregiving service within each trajectory group were visualized using heatmaps. Results In the analysis of LTC utilization within the first 6 months post-application, trajectories were classified into 7 groups; while within 12 months, trajectories were divided into 4 groups. These trajectories can be categorized into 4 distinct patterns: High-Stable (6 mo: 21.3%; 12 mo: 35.7%), Low-Stable (6 mo: 33.1%; 12 mo: 37.2%), High-Decrease (6 mo: 12.8%; 12 mo: 10.7%), and Low-Increase (6 mo: 32.7%; 12 mo: 16.5%). We further explored differences in socio-demographic characteristics and needs factors among trajectory groups. Heatmaps were used to illustrate temporal variations in the utilization rates of specific service items, which played a significant role in determining the trajectories. Conclusions K-means clustering is useful in identifying distinct patterns among LTC users. Tailored service delivery strategies can be designed to meet the diverse needs of these different user groups. Key messages • The unsupervised machine learning method distinguished various temporal dynamic patterns in LTC utilization. • Tailored service delivery strategies can be designed to meet the diverse needs of these different user groups.

A Strategy for Predicting the Performance of Supervised and Unsupervised Tabular Data Classifiers

AbstractMachine Learning algorithms that perform classification are increasingly been adopted in Information and Communication Technology (ICT) systems and infrastructures due to their capability to profile their expected behavior and detect anomalies due to ongoing errors or intrusions. Deploying a classifier for a given system requires conducting comparison and sensitivity analyses that are time-consuming, require domain expertise, and may even not achieve satisfactory classification performance, resulting in a waste of money and time for practitioners and stakeholders. This paper predicts the expected performance of classifiers without needing to select, craft, exercise, or compare them, requiring minimal expertise and machinery. Should classification performance be predicted worse than expectations, the users could focus on improving data quality and monitoring systems instead of wasting time in exercising classifiers, saving key time and money. The prediction strategy uses scores of feature rankers, which are processed by regressors to predict metrics such as Matthews Correlation Coefficient (MCC) and Area Under ROC-Curve (AUC) for quantifying classification performance. We validate our prediction strategy through a massive experimental analysis using up to 12 feature rankers that process features from 23 public datasets, creating additional variants in the process and exercising supervised and unsupervised classifiers. Our findings show that it is possible to predict the value of performance metrics for supervised or unsupervised classifiers with a mean average error (MAE) of residuals lower than 0.1 for many classification tasks. The predictors are publicly available in a Python library whose usage is straightforward and does not require domain-specific skill or expertise.

Identifying clinical phenotype clusters for coronary artery disease: a novel approach using clustering by unsupervised machine learning

Abstract Background Guideline recommendations for the prevention of cardiovascular (CV) events in patients with coronary artery disease (CAD) are one-size-fits-all. Clinically identifiable phenotypes needing specific considerations might exist. Purpose To identify phenotypes in patients with CAD based on clinical characteristics and their relationship with the risk of new CV events and treatment benefits. Methods Unsupervised machine learning through latent class analysis (LCA) was performed on patients with CAD included in the UCC-SMART cohort (n = 5,888) and SWEDEHEART registry (n = 67,428), to identify patient clusters (i.e., phenotypes) based on clinical characteristics: age, sex, body-mass-index, diastolic blood pressure, history of diabetes, myocardial infarction, atrial fibrillation, polyvascular disease, current smoking, non-high density lipoprotein cholesterol, estimated glomerular filtration rate, and C-reactive protein. Number of clusters was determined using the Bayesian information criterion. Event free survival and hazard ratios (HR) for major adverse cardiovascular events (MACE) as primary endpoint, including non-fatal myocardial infarction, non-fatal stroke, and cardiovascular death, were estimated per cluster. The LCA were performed in each respective cohort and cross-validated in the other cohort to assess impact of variations in clustering variables on cluster reproducibility. In UCC-SMART, bootstrapping was performed the assess stability of the LCA model and impact on cluster assignment. Results In UCC-SMART four distinct phenotypes within the CAD population could be distinguished: Phenotype A (16%) of mostly males with isolated CAD, phenotype B (30%) of older patients with polyvascular disease, Phenotype C (28%) of only women with limited traditional risk factors and Phenotype D (27%) young, current smokers with a premature event (Figure 1). Cross-validation in SWEDEHEART identified roughly the same four phenotypes, i.e., phenotype A (39%) with 100% males and 3.8% polyvascular disease, phenotype B (8%) with average age of 71 years and 30% polyvascular disease, phenotype C (19%) with 100% women, and phenotype D (34%) 54 years and 25% current smokers. Notably, although age was comparable between phenotype A and C, MACE-free survival was worse for the female-only cluster. Similarly, although patients where on average 9 years younger in phenotype D when compared to phenotype A, MACE-free survival was worse (Figure 2). No significant changes in clustering models were observed in the bootstrap samples. Conclusion Phenotypes can be distinguished within patients with CAD. These phenotypes are related to differences in the risk of recurrent events. These findings suggest that specific guideline recommendations for these groups could be considered.Cluster interpretation in SMARTSurvival along cluster lines

Time-series clustering of raw recordings of 24-hour ambulatory blood presssure for risk stratification in general population

Abstract Objective 24-hour ambulatory blood pressure measurements (ABPM) is the best out-of-office BP measurement used to refine cardiovascular (CV) risk stratification. Currently, only summary ABPM indexes are used in the clinical practice while the temporal characteristics of the raw BP recordings have not been explored. Therefore, in this study we employed an unsupervised machine learning (ML) algorithm suitable for multivariate time-series analysis to identify distinct BP and heart rate (HR) profiles associated with the incidence of adverse CV events. Design and Method In 1391 community-dwelling individuals (mean age, 46.9 years; 50.97% women), we acquired 24-hour ABPM and clinical data and collected CV events (n=399) on average 22 years later. We employed multivariate dynamic time warping and k-medoids algorithm on the raw recordings of systolic and diastolic BP and HR. To assess the clinical significance of the derived clusters, we compared the clinical characteristics and the incidence of adverse events. We validated the trained model using 24-hour ABPM obtained from external population cohort (n=1137). Results The silhouette score and Dunn index showed that the optimal number of clusters was 4. Across all clusters we observed significant difference in age and sex distribution. Cluster 4 had the worst CV risk factors profiling, with higher office BP and anti-hypertensive medication intake compared to the other clusters. We observed differences in the incidence of adverse events between clusters (Figure), with cluster 1 representing the lowest risk group and cluster 4 the highest. After adjusting for traditional CV risk factors, including office systolic BP, cluster 1 had the lowest risk (HR:0.92; 95%CI:0.84-0.99; P=0.036) and cluster 4 the highest (HR:1.13; 95%CI:1.03-1.24; P=0.006) as compared to the average population risk. The analysis of the external validation cohort revealed similar ABPM patterns and clinical characteristics. Conclusion Employing an unsupervised ML model on the raw ABPM recordings we identified clinically meaningful clusters associated with an increasing cumulative incidence of CV events. This ML analysis paves the way towards utilization of temporal BP and HR information and subsequently optimization of CV risk stratification.

Association between cardiogenic shock phenotype, vasopressor requirements and mortality

Abstract Background Greater vasopressor requirements predict higher mortality in patients with cardiogenic shock (CS).(1) The extent to which CS phenotype modifies the relationship between vasopressor dose and mortality risk is unknown. Purpose We investigated the association between in-hospital mortality and vasopressor dosing (expressed as norepinephrine equivalents [NEE]) during the first 24 hours of hospitalization across distinct CS phenotypes. Methods We included patients with CS admitted to our cardiac intensive care unit (CICU) between 2007 and 2015. Peak combined vasopressor doses during the first 24 hours of hospitalization were recorded as NEE. High dose vasopressor (HDV) use was defined as peak 24 hour NEE &amp;gt;0.3 mcg/kg/min. Unsupervised machine learning was used to assign patients to one of three phenotypes (non-congested, cardiorenal, and haemometabolic), as reported previously by our group and others.(2,3) The primary outcome was in-hospital mortality. Multivariable logistic regression was used to adjust for relevant covariates. Results A total of 1078 patients with CS (median age 69 years, 37% female, 59% with acute myocardial infarction) were admitted to our cardiac intensive care unit (CICU) between 2017 and 2015. The number of patients assigned to each CS phenotype were: haemometabolic (n=325, 30%), cardiorenal (n=314, 29%), non-congested (439, 41%). There was a strong incremental association between in-hospital mortality and NEE (Fig. 1, adjusted OR 1.104 per 0.1 mcg/kg/min, 95% CI 1.134–1.231, P&amp;lt;0.001), and patients receiving HDV (n=234, 22%) had significantly higher in-hospital mortality (57% vs 28%, p&amp;lt;0.001). Among patients who did not receive HDV, mortality was higher in haemometabolic CS (49%) compared to 24% in cardiorenal CS and 21% in non-congested CS (adjusted OR 3.58 relative to non-congested CS, 95% CI 2.46-5.23, p&amp;lt;0.001, Fig. 1). Among patients receiving HDV, hemometabolic CS was also associated with higher mortality (68%) than those with cardiorenal (46.2%) or non-congested CS (39.3%), (adjusted OR 3.26 relative to non-congested CS, 95% CI 1.75-6.18, P&amp;lt;0.001, Fig. 1). Conclusion Integration of CS phenotype with vasopressor requirements may provide incremental prognostic information in CS during the first 24 hours of admission, with haemometabolic CS portending a greater mortality risk even at lower NEE. Patients with HDV (particularly those requiring &amp;gt;=1 mcg/kg/min NEE) had high mortality irrespective of CS phenotype.Figure 1