Search

Background: Waste and fraud are important problems for health insurers to deal with. With the advent of big data, these insurers are looking more and more towards data mining and machine learning methods to help in detecting waste and fraud. However, labeled data is costly and difficult to acquire as it requires expert investigators and known care providers with atypical behavior. Methods: In this work we show how recent advances in machine learning can be used to set up a workflow that can aid investigators in discovering practitioners or groups of practitioners with unusual resource use in order to more efficiently combat waste and fraud. We combine three different techniques, which have not been used in the context of healthcare insurance anomaly detection: categorical embeddings to deal with high-cardinality categorical variables, state-of-the-art unsupervised anomaly detection techniques to detect anomalies and Shapley additive explanations (SHAP) to explain the model output. Results: The method has been evaluated on providers with a known anomalous profile and with the help of experts of the largest health insurance fund in Belgium. The quantitative experiments show that categorical embeddings offer a significant improvement compared to standard methods and that the state-of-the-art unsupervised anomaly detection techniques generally show an improvement over traditional methods. In a practical setting, the proposed workflow with SHAP was able to detect a previously unknown, anomalous trend among general practitioners. Conclusions: The proposed workflow is able to detect known care providers with atypical behaviour and helps expert investigators in making informed decisions concerning possible fraud or overconsumption in the health insurance field. [ABSTRACT FROM AUTHOR]

Background: Systemic lupus erythematosus (SLE) has numerous symptoms across organs and an unpredictable flare-remittance pattern. This has made it challenging to understand drivers of long-term SLE outcomes. Our objective was to identify whether changes in DNA methylation over time, in an actively flaring SLE cohort, were associated with remission and whether these changes meaningfully subtype SLE patients. Methods: Fifty-nine multi-ethnic SLE patients had clinical visits and DNA methylation profiles at a flare and approximately 3 months later. Methylation was measured using the Illumina EPIC array. We identified sites where methylation change between visits was associated with remission at the follow-up visit using limma package and a time x remission interaction term. Models adjusted for batch, age at diagnosis, time between visits, age at flare, sex, medications, and cell-type proportions. Separately, a paired T-test identified Bonferroni significant methylation sites with ≥ 3% change between visits (n = 546). Methylation changes at these sites were used for unsupervised consensus hierarchical clustering. Associations between clusters and patient features were assessed. Results: Nineteen patients fully remitted at the follow-up visit. For 1,953 CpG sites, methylation changed differently for remitters vs. non-remitters (Bonferroni p < 0.05). Nearly half were within genes regulated by interferon. The largest effect was at cg22873177; on average, remitters had 23% decreased methylation between visits while non-remitters had no change. Three SLE patient clusters were identified using methylation differences agnostic of clinical outcomes. All Cluster 1 subjects (n = 12) experienced complete flare remission, despite similar baseline disease activity scores, medications, and demographics as other clusters. Methylation changes at six CpG sites, including within immune-related CD45 and IFI genes, were particularly distinct for each cluster, suggesting these may be good candidates for stratifying patients in the future. Conclusions: Changes in DNA methylation during active SLE were associated with remission status and identified subgroups of SLE patients with several distinct clinical and biological characteristics. DNA methylation patterns might help inform SLE subtypes, leading to targeted therapies based on relevant underlying biological pathways. [ABSTRACT FROM AUTHOR]

Background: Specific food preferences can determine an individual's dietary patterns and therefore, may be associated with certain health risks and benefits. Methods: Using food preference questionnaire (FPQ) data from a subset comprising over 180,000 UK Biobank participants, we employed Latent Profile Analysis (LPA) approach to identify the main patterns or profiles among participants. blood biochemistry across groups/profiles was compared using the non-parametric Kruskal–Wallis test. We applied the Limma algorithm for differential abundance analysis on 168 metabolites and 2923 proteins, and utilized the Database for Annotation, Visualization and Integrated Discovery (DAVID) to identify enriched biological processes and pathways. Relative risks (RR) were calculated for chronic diseases and mental conditions per group, adjusting for sociodemographic factors. Results: Based on their food preferences, three profiles were termed: the putative Health-conscious group (low preference for animal-based or sweet foods, and high preference for vegetables and fruits), the Omnivore group (high preference for all foods), and the putative Sweet-tooth group (high preference for sweet foods and sweetened beverages). The Health-conscious group exhibited lower risk of heart failure (RR = 0.86, 95%CI 0.79–0.93) and chronic kidney disease (RR = 0.69, 95%CI 0.65–0.74) compared to the two other groups. The Sweet-tooth group had greater risk of depression (RR = 1.27, 95%CI 1.21–1.34), diabetes (RR = 1.15, 95%CI 1.01–1.31), and stroke (RR = 1.22, 95%CI 1.15–1.31) compared to the other two groups. Cancer (overall) relative risk showed little difference across the Health-conscious, Omnivore, and Sweet-tooth groups with RR of 0.98 (95%CI 0.96–1.01), 1.00 (95%CI 0.98–1.03), and 1.01 (95%CI 0.98–1.04), respectively. The Health-conscious group was associated with lower levels of inflammatory biomarkers (e.g., C-reactive Protein) which are also known to be elevated in those with common metabolic diseases (e.g., cardiovascular disease). Other markers modulated in the Health-conscious group, ketone bodies, insulin-like growth factor-binding protein (IGFBP), and Growth Hormone 1 were more abundant, while leptin was less abundant. Further, the IGFBP pathway, which influences IGF1 activity, may be significantly enhanced by dietary choices. Conclusions: These observations align with previous findings from studies focusing on weight loss interventions, which include a reduction in leptin levels. Overall, the Health-conscious group, with preference to healthier food options, has better health outcomes, compared to Sweet-tooth and Omnivore groups. [ABSTRACT FROM AUTHOR]

Background: Septic patients who develop acute respiratory failure (ARF) requiring mechanical ventilation represent a heterogenous subgroup of critically ill patients with widely variable clinical characteristics. Identifying distinct phenotypes of these patients may reveal insights about the broader heterogeneity in the clinical course of sepsis, considering multi-organ dynamics. We aimed to derive novel phenotypes of sepsis-induced ARF using observational clinical data and investigate the generalizability of the derived phenotypes. Methods: We performed a multi-center retrospective study of ICU patients with sepsis who required mechanical ventilation for ≥ 24 h. Data from two different high-volume academic hospital centers were used, where all phenotypes were derived in MICU of Hospital-I (N = 3225). The derived phenotypes were validated in MICU of Hospital-II (N = 848), SICU of Hospital-I (N = 1112), and SICU of Hospital-II (N = 465). Clinical data from 24 h preceding intubation was used to derive distinct phenotypes using an explainable machine learning-based clustering model interpreted by clinical experts. Results: Four distinct ARF phenotypes were identified: A (severe multi-organ dysfunction (MOD) with a high likelihood of kidney injury and heart failure), B (severe hypoxemic respiratory failure [median P/F = 123]), C (mild hypoxia [median P/F = 240]), and D (severe MOD with a high likelihood of hepatic injury, coagulopathy, and lactic acidosis). Patients in each phenotype showed differences in clinical course and mortality rates despite similarities in demographics and admission co-morbidities. The phenotypes were reproduced in external validation utilizing the MICU of Hospital-II and SICUs from Hospital-I and -II. Kaplan–Meier analysis showed significant difference in 28-day mortality across the phenotypes (p < 0.01) and consistent across MICU and SICU of both Hospital-I and -II. The phenotypes demonstrated differences in treatment effects associated with high positive end-expiratory pressure (PEEP) strategy. Conclusion: The phenotypes demonstrated unique patterns of organ injury and differences in clinical outcomes, which may help inform future research and clinical trial design for tailored management strategies. [ABSTRACT FROM AUTHOR]

The aim of this study was to develop a machine learning-assisted rapid determination methodology for traditional Chinese Medicine Constitution. Based on the Constitution in Chinese Medicine Questionnaire (CCMQ), the most applied diagnostic instrument for assessing individuals' constitutions, we employed automated supervised machine learning algorithms (i.e., Tree-based Pipeline Optimization Tool; TPOT) on all the possible item combinations for each subscale and an unsupervised machine learning algorithm (i.e., variable clustering; varclus) on the whole scale to select items that can best predict body constitution (BC) classifications or BC scores. By utilizing subsets of items selected based on TPOT and corresponding machine learning algorithms, the accuracies of BC classifications prediction ranged from 0.819 to 0.936, with the root mean square errors of BC scores prediction stabilizing between 6.241 and 9.877. Overall, the results suggested that the automated machine learning algorithms performed better than the varclus algorithm for item selection. Additionally, based on an automated machine learning item selection procedure, we provided the top three ranked item combinations with each possible subscale length, along with their corresponding algorithms for predicting BC classification and severity. This approach could accommodate the needs of different practitioners in traditional Chinese medicine for rapid constitution determination. [ABSTRACT FROM AUTHOR]

Background: Physical activity is a crucial aspect of health benefits in the public society. Although studies on the temporal physical activity patterns might lead to the protocol for efficient intervention/program, a standardized procedure to determine and analyze the temporal physical activity patterns remains to be developed. Here, we attempted to develop a procedure to cluster 24-hour patterns of physical activity as step counts measured with an accelerometer-based wearable sensor. Methods: The 1 Hz step count data was collected by a hip-worn triaxial accelerometer from 42 healthy participants, comprising 35 males and 7 females, at the Sendai Oroshisho center in 2008. This is a cross sectional study using unsupervised machine learning, specifically the kernel k-means algorithm with the global alignment kernel was applied on a total of 815 days from 42 participants, and 6 activity patterns were identified. Further, the probability of each 24-hour step-counting pattern was calculated for every participant and used for spectral clustering of step-behavioral patterns. Results: We could identify six 24-hour step-counting patterns and five daily step-behavioral clusters. We could further identify five step-behavioral clusters, all-day dominant (21 participants), all-day + bi-phasic dominant (8 participants), bi-phasic dominant (6 participants), all-day + evening dominant (4 participants), and morning dominant (3 participants). When the amount of physical activity was categorized into tertile groups reflecting highly active, moderately active, and low active, each tertile group consisted of different proportions of six 24-hour step-counting patterns Conclusions: Our study introduces a novel approach using an unsupervised machine learning method to categorize daily hourly activity, revealing six distinct step counting patterns and five clusters representing daily step behaviors. Our procedure would be reliable for finding and clustering physical activity patterns/behaviors and reveal diversity in the categorization by a traditional tertile procedure using total step amount. [ABSTRACT FROM AUTHOR]

Background: The use of machine learning in medical diagnosis and treatment has grown significantly in recent years with the development of computer-aided diagnosis systems, often based on annotated medical radiology images. However, the lack of large annotated image datasets remains a major obstacle, as the annotation process is time-consuming and costly. This study aims to overcome this challenge by proposing an automated method for annotating a large database of medical radiology images based on their semantic similarity. Results: An automated, unsupervised approach is used to create a large annotated dataset of medical radiology images originating from the Clinical Hospital Centre Rijeka, Croatia. The pipeline is built by data-mining three different types of medical data: images, DICOM metadata and narrative diagnoses. The optimal feature extractors are then integrated into a multimodal representation, which is then clustered to create an automated pipeline for labelling a precursor dataset of 1,337,926 medical images into 50 clusters of visually similar images. The quality of the clusters is assessed by examining their homogeneity and mutual information, taking into account the anatomical region and modality representation. Conclusions: The results indicate that fusing the embeddings of all three data sources together provides the best results for the task of unsupervised clustering of large-scale medical data and leads to the most concise clusters. Hence, this work marks the initial step towards building a much larger and more fine-grained annotated dataset of medical radiology images. [ABSTRACT FROM AUTHOR]

Background: There exists consistent empirical evidence in the literature pointing out ample heterogeneity in terms of the clinical evolution of patients with COVID-19. The identification of specific phenotypes underlying in the population might contribute towards a better understanding and characterization of the different courses of the disease. The aim of this study was to identify distinct clinical phenotypes among hospitalized patients with SARS-CoV-2 pneumonia using machine learning clustering, and to study their association with subsequent clinical outcomes as severity and mortality. Methods: Multicentric observational, prospective, longitudinal, cohort study conducted in four hospitals in Spain. We included adult patients admitted for in-hospital stay due to SARS-CoV-2 pneumonia. We collected a broad spectrum of variables to describe exhaustively each case: patient demographics, comorbidities, symptoms, physiological status, baseline examinations (blood analytics, arterial gas test), etc. For the development and internal validation of the clustering/phenotype models, the dataset was split into training and test sets (50% each). We proposed a sequence of machine learning stages: feature scaling, missing data imputation, reduction of data dimensionality via Kernel Principal Component Analysis (KPCA), and clustering with the k-means algorithm. The optimal cluster model parameters –including k, the number of phenotypes– were chosen automatically, by maximizing the average Silhouette score across the training set. Results: We enrolled 1548 patients, each of them characterized by 92 clinical attributes (d=109 features after variable encoding). Our clustering algorithm identified k=3 distinct phenotypes and 18 strongly informative variables: Phenotype A (788 cases [50.9% prevalence] – age ∼ 57, Charlson comorbidity ∼ 1, pneumonia CURB-65 score ∼ 0 to 1, respiratory rate at admission ∼ 18 min-1, FiO2 ∼ 21%, C-reactive protein CRP ∼ 49.5 mg/dL [median within cluster]); phenotype B (620 cases [40.0%] – age ∼ 75, Charlson ∼ 5, CURB-65 ∼ 1 to 2, respiration ∼ 20 min-1, FiO2 ∼ 21%, CRP ∼ 101.5 mg/dL); and phenotype C (140 cases [9.0%] – age ∼ 71, Charlson ∼ 4, CURB-65 ∼ 0 to 2, respiration ∼ 30 min-1, FiO2 ∼ 38%, CRP ∼ 152.3 mg/dL). Hypothesis testing provided solid statistical evidence supporting an interaction between phenotype and each clinical outcome: severity and mortality. By computing their corresponding odds ratios, a clear trend was found for higher frequencies of unfavourable evolution in phenotype C with respect to B, as well as more unfavourable in phenotype B than in A. Conclusion: A compound unsupervised clustering technique (including a fully-automated optimization of its internal parameters) revealed the existence of three distinct groups of patients – phenotypes. In turn, these showed strong associations with the clinical severity in the progression of pneumonia, and with mortality. [ABSTRACT FROM AUTHOR]

Objectives: The traditional understanding of craniocervical alignment emphasizes specific anatomical landmarks. However, recent research has challenged the reliance on forward head posture as the primary diagnostic criterion for neck pain. An advanced relationship exists between neck pain and craniocervical alignment, which requires a deeper exploration of diverse postures and movement patterns using advanced techniques, such as clustering analysis. We aimed to explore the complex relationship between craniocervical alignment, and neck pain and to categorize alignment patterns in individuals with nonspecific neck pain using the K-means algorithm. Methods: This study included 229 office workers with nonspecific neck pain who applied unsupervised machine learning techniques. The craniocervical angles (CCA) during rest, protraction, and retraction were measured using two-dimensional video analysis, and neck pain severity was assessed using the Northwick Park Neck Pain Questionnaire (NPQ). CCA during sitting upright in a comfortable position was assessed to evaluate the resting CCA. The average of midpoints between repeated protraction and retraction measures was considered as the midpoint CCA. The K-means algorithm helped categorize participants into alignment clusters based on age, sex and CCA data. Results: We found no significant correlation between NPQ scores and CCA data, challenging the traditional understanding of neck pain and alignment. We observed a significant difference in age (F = 140.14, p < 0.001), NPQ total score (F = 115.83, p < 0.001), resting CCA (F = 79.22, p < 0.001), CCA during protraction (F = 33.98, p < 0.001), CCA during retraction (F = 40.40, p < 0.001), and midpoint CCA (F = 66.92, p < 0.001) among the three clusters and healthy controls. Cluster 1 was characterized by the lowest resting and midpoint CCA, and CCA during pro- and -retraction, indicating a significant forward head posture and a pattern of retraction restriction. Cluster 2, the oldest group, showed CCA measurements similar to healthy controls, yet reported the highest NPQ scores. Cluster 3 exhibited the highest CCA during protraction and retraction, suggesting a limitation in protraction movement. Discussion: Analyzing 229 office workers, three distinct alignment patterns were identified, each with unique postural characteristics; therefore, treatments addressing posture should be individualized and not generalized across the population. [ABSTRACT FROM AUTHOR]

Background: Depression is a global burden with profound personal and economic consequences. Previous studies have reported that the amount of physical activity is associated with depression. However, the relationship between the temporal patterns of physical activity and depressive symptoms is poorly understood. In this exploratory study, we hypothesize that a particular temporal pattern of daily physical activity could be associated with depressive symptoms and might be a better marker than the total amount of physical activity. Methods: To address the hypothesis, we investigated the association between depressive symptoms and daily dominant activity behaviors based on 24-h temporal patterns of physical activity. We conducted a cross-sectional study on NHANES 2011–2012 data collected from the noninstitutionalized civilian resident population of the United States. The number of participants that had the whole set of physical activity data collected by the accelerometer is 6613. Among 6613 participants, 4242 participants had complete demography and Patient Health Questionnaire-9 (PHQ-9) questionnaire, a tool to quantify depressive symptoms. The association between activity-count behaviors and depressive symptoms was analyzed using multivariable logistic regression to adjust for confounding factors in sequential models. Results: We identified four physical activity-count behaviors based on five physical activity-counting patterns classified by unsupervised machine learning. Regarding PHQ-9 scores, we found that evening dominant behavior was positively associated with depressive symptoms compared to morning dominant behavior as the control group. Conclusions: Our results might contribute to monitoring and identifying individuals with latent depressive symptoms, emphasizing the importance of nuanced activity patterns and their probability of assessing depressive symptoms effectively. [ABSTRACT FROM AUTHOR]

Background: Optical coherence tomography angiography (OCTA) is an innovative technology providing visual and quantitative data on retinal microvasculature in a non-invasive manner. Main body: Due to variations in the technical specifications of different OCTA devices, there are significant inter-device differences in OCTA data, which can limit their comparability and generalizability. These variations can also result in a domain shift problem that may interfere with applicability of machine learning models on data obtained from different OCTA machines. One possible approach to address this issue may be unsupervised deep image-to-image translation leveraging systems such as Cycle-Consistent Generative Adversarial Networks (Cycle-GANs) and Denoising Diffusion Probabilistic Models (DDPMs). Through training on unpaired images from different device domains, Cycle-GANs and DDPMs may enable cross-domain translation of images. They have been successfully applied in various medical imaging tasks, including segmentation, denoising, and cross-modality image-to-image translation. In this commentary, we briefly describe how Cycle-GANs and DDPMs operate, and review the recent experiments with these models on medical and ocular imaging data. We then discuss the benefits of applying such techniques for inter-device translation of OCTA data and the potential challenges ahead. Conclusion: Retinal imaging technologies and deep learning-based domain adaptation techniques are rapidly evolving. We suggest exploring the potential of image-to-image translation methods in improving the comparability of OCTA data from different centers or devices. This may facilitate more efficient analysis of heterogeneous data and broader applicability of machine learning models trained on limited datasets in this field. [ABSTRACT FROM AUTHOR]

Background: Functional gastrointestinal disorders (FGIDs), as a group of syndromes with no identified structural or pathophysiological biomarkers, are currently classified by Rome criteria based on gastrointestinal symptoms (GI). However, the high overlap among FGIDs in patients makes treatment and identifying underlying mechanisms challenging. Furthermore, disregarding psychological factors in the current classification, despite their approved relationship with GI symptoms, underlines the necessity of more investigation into grouping FGID patients. We aimed to provide more homogenous and well-separated clusters based on both GI and psychological characteristics for patients with FGIDs using an unsupervised machine learning algorithm. Methods: Based on a cross-sectional study, 3765 (79%) patients with at least one FGID were included in the current study. In the first step, the clustering utilizing a machine learning algorithm was merely executed based on GI symptoms. In the second step, considering the previous step's results and focusing on the clusters with a diverse combination of GI symptoms, the clustering was re-conducted based on both GI symptoms and psychological factors. Results: The first phase clustering of all participants based on GI symptoms resulted in the formation of pure and non-pure clusters. Pure clusters exactly illustrated the properties of most pure Rome syndromes. Re-clustering the members of the non-pure clusters based on GI and psychological factors (i.e., the second clustering step) resulted in eight new clusters, indicating the dominance of multiple factors but well-discriminated from other clusters. The results of the second step especially highlight the impact of psychological factors in grouping FGIDs. Conclusions: In the current study, the existence of Rome disorders, which were previously defined by expert opinion-based consensus, was approved, and, eight new clusters with multiple dominant symptoms based on GI and psychological factors were also introduced. The more homogeneous clusters of patients could lead to the design of more precise clinical experiments and further targeted patient care. [ABSTRACT FROM AUTHOR]

Rheumatoid arthritis (RA) is an autoimmune disease that exhibits a high degree of heterogeneity, marked by unpredictable disease flares and significant variations in the response to available treatments. The lack of optimal stratification for RA patients may be a contributing factor to the poor efficacy of current treatment options. The objective of this study is to elucidate the molecular characteristics of RA through the utilization of mitochondrial genes and subsequently construct and authenticate a diagnostic framework for RA. Mitochondrial proteins were obtained from the MitoCarta database, and the R package limma was employed to filter for differentially expressed mitochondrial genes (MDEGs). Metascape was utilized to perform enrichment analysis, followed by an unsupervised clustering algorithm using the ConsensuClusterPlus package to identify distinct subtypes based on MDEGs. The immune microenvironment, biological pathways, and drug response were further explored in these subtypes. Finally, a multi-biomarker-based diagnostic model was constructed using machine learning algorithms. Utilizing 88 MDEGs present in transcript profiles, it was possible to classify RA patients into three distinct subtypes, each characterized by unique molecular and cellular signatures. Subtype A exhibited a marked activation of inflammatory cells and pathways, while subtype C was characterized by the presence of specific innate lymphocytes. Inflammatory and immune cells in subtype B displayed a more modest level of activation (Wilcoxon test P < 0.05). Notably, subtype C demonstrated a stronger correlation with a superior response to biologics such as infliximab, anti-TNF, rituximab, and methotrexate/abatacept (P = 0.001) using the fisher test. Furthermore, the mitochondrial diagnosis SVM model demonstrated a high degree of discriminatory ability in distinguishing RA in both training (AUC = 100%) and validation sets (AUC = 80.1%). This study presents a pioneering analysis of mitochondrial modifications in RA, offering a novel framework for patient stratification and potentially enhancing therapeutic decision-making. [ABSTRACT FROM AUTHOR]

Background: The Progress Test Medizin (PTM) is a 200-question formative test that is administered to approximately 11,000 students at medical universities (Germany, Austria, Switzerland) each term. Students receive feedback on their knowledge (development) mostly in comparison to their own cohort. In this study, we use the data of the PTM to find groups with similar response patterns. Methods: We performed k-means clustering with a dataset of 5,444 students, selected cluster number k = 5, and answers as features. Subsequently, the data was passed to XGBoost with the cluster assignment as target enabling the identification of cluster-relevant questions for each cluster with SHAP. Clusters were examined by total scores, response patterns, and confidence level. Relevant questions were evaluated for difficulty index, discriminatory index, and competence levels. Results: Three of the five clusters can be seen as "performance" clusters: cluster 0 (n = 761) consisted predominantly of students close to graduation. Relevant questions tend to be difficult, but students answered confidently and correctly. Students in cluster 1 (n = 1,357) were advanced, cluster 3 (n = 1,453) consisted mainly of beginners. Relevant questions for these clusters were rather easy. The number of guessed answers increased. There were two "drop-out" clusters: students in cluster 2 (n = 384) dropped out of the test about halfway through after initially performing well; cluster 4 (n = 1,489) included students from the first semesters as well as "non-serious" students both with mostly incorrect guesses or no answers. Conclusion: Clusters placed performance in the context of participating universities. Relevant questions served as good cluster separators and further supported our "performance" cluster groupings. [ABSTRACT FROM AUTHOR]

Background: Electronic medical records (EMR)-trained machine learning models have the potential in CVD risk prediction by integrating a range of medical data from patients, facilitate timely diagnosis and classification of CVDs. We tested the hypothesis that unsupervised ML approach utilizing EMR could be used to develop a new model for detecting prevalent CVD in clinical settings., Methods: We included 155,894 patients (aged ≥ 18 years) discharged between January 2014 and July 2022, from Xuhui Hospital, Shanghai, China, including 64,916 CVD cases and 90,979 non-CVD cases. K-means clustering was used to generate the clustering models with k = 2, 4, and 8 as predetermined number of clusters k = 2, 4, and 8. Bayesian theorem was used to estimate the models' predictive accuracy., Results: The overall predictive accuracy of the 2-, 4-, and 8-classification clustering models in the training set was 0.856, 0.8634, and 0.8506, respectively. Similarly, the predictive accuracy of the 2-, 4-, and 8-classification clustering models in the testing set was 0.8598, 0.8659, and 0.8525, respectively. After reducing from 19 dimensions to 2 dimensions by principal component analysis, significant separation was observed for CVD cases and non-CVD cases in both training and testing sets., Conclusion: Our findings indicate that the utilization of EMR data can support the development of a robust model for CVD detection through an unsupervised ML approach. Further investigation using longitudinal design is needed to refine the model for its applications in clinical settings., Competing Interests: Declarations. Ethics approval and consent to participate: This study was performed in accordance with the guidelines of the Declaration of Helsinki. The study design was approved by the Ethics Committee of Shanghai Xuhui Central Hospital (approval no.: 2023033), and the institutional review board waived the requirement to obtain the informed consent. Consent for publication: All authors have approved for its publication. Competing interests: The authors declare no competing interests., (© 2024. The Author(s).)

Background: Rectal cancer is a common cancer worldwide and lacks effective prognostic markers. The development of prognostic markers by computational pathology methods has attracted increasing attention. This paper aims to construct a prognostic signature from whole slide images for predicting progression-free survival (PFS) of rectal cancer through an unsupervised artificial intelligence algorithm., Methods: A total of 238 patients with rectal cancer from two datasets were collected for the development and validation of the prognostic signature. A tumor detection model was built by transfer learning. Then, on the basis of the tumor patches recognized by the tumor detection model, a convolutional autoencoder model was built for decoding the tumor patches into deep latent features. Next, on the basis of the deep latent features, the tumor patches were divided into different clusters. The cluster number and other hyperparameters were optimized by a nested cross-validation method. The percentage of each cluster from the patient's tumor patches, which is hereafter called PCF, was calculated for prognostic signature construction. The prognostic signature was constructed by Cox proportional hazard regression with L2 regularization. Finally, bioinformatic analysis was performed to explore the underlying biological mechanisms of the PCFs., Results: The accuracy of the tumor detection model in distinguishing tumor patches from non-tumor patches achieved 99.3%. The optimal cluster number was determined to be 9. Therfore, 9 PCFs were calculated to construct the prognostic signature. The prognostic signature achieved a concordance index of 0.701 in the validation cohort. The Kaplan-Meier survival curves showed the prognostic signature had good risk stratification ability. Through the bioinformatic analysis, several PCF-associated genes were identified. These genes were enriched in various gene ontology terms., Conclusion: The developed prognostic signature can effectively predict PFS in patients with rectal cancer and exploration of the underlying biological mechanisms may help to promote its clinical translation., Competing Interests: Declarations. Ethics approval and consent to participate: This study was approved by the hospital’s ethics committee, and informed consent was waived. Consent for publication: Not applicable. Competing interests: The authors declare no competing interests., (© 2024. The Author(s).)

Background: Traditional supervised learning methods applied to DNA sequence taxonomic classification rely on the labor-intensive and time-consuming step of labelling the primary DNA sequences. Additionally, standard DNA classification/clustering methods involve time-intensive multiple sequence alignments, which impacts their applicability to large genomic datasets or distantly related organisms. These limitations indicate a need for robust, efficient, and scalable unsupervised DNA sequence clustering methods that do not depend on sequence labels or alignment., Results: This study proposes CGRclust, a novel combination of unsupervised twin contrastive clustering of Chaos Game Representations (CGR) of DNA sequences, with convolutional neural networks (CNNs). To the best of our knowledge, CGRclust is the first method to use unsupervised learning for image classification (herein applied to two-dimensional CGR images) for clustering datasets of DNA sequences. CGRclust overcomes the limitations of traditional sequence classification methods by leveraging unsupervised twin contrastive learning to detect distinctive sequence patterns, without requiring DNA sequence alignment or biological/taxonomic labels. CGRclust accurately clustered twenty-five diverse datasets, with sequence lengths ranging from 664 bp to 100 kbp, including mitochondrial genomes of fish, fungi, and protists, as well as viral whole genome assemblies and synthetic DNA sequences. Compared with three recent clustering methods for DNA sequences (DeLUCS, iDeLUCS, and MeShClust v3.0.), CGRclust is the only method that surpasses 81.70% accuracy across all four taxonomic levels tested for mitochondrial DNA genomes of fish. Moreover, CGRclust also consistently demonstrates superior performance across all the viral genomic datasets. The high clustering accuracy of CGRclust on these twenty-five datasets, which vary significantly in terms of sequence length, number of genomes, number of clusters, and level of taxonomy, demonstrates its robustness, scalability, and versatility., Conclusion: CGRclust is a novel, scalable, alignment-free DNA sequence clustering method that uses CGR images of DNA sequences and CNNs for twin contrastive clustering of unlabelled primary DNA sequences, achieving superior or comparable accuracy and performance over current approaches. CGRclust demonstrated enhanced reliability, by consistently achieving over 80% accuracy in more than 90% of the datasets analyzed. In particular, CGRclust performed especially well in clustering viral DNA datasets, where it consistently outperformed all competing methods., Competing Interests: Declarations. Ethics approval and consent to participate: Not applicable. Consent for publication: Not applicable. Competing interests: The authors declare no competing interests., (© 2024. The Author(s).)

Background: Anomaly detection is crucial in healthcare data due to challenges associated with the integration of smart technologies and healthcare. Anomaly in electronic health record can be associated with an insider trying to access and manipulate the data. This article focuses around the anomalies under different contexts., Methodology: This research has proposed methodology to secure Electronic Health Records (EHRs) within a complex environment. We have employed a systematic approach encompassing data preprocessing, labeling, modeling, and evaluation. Anomalies are not labelled thus a mechanism is required that predicts them with greater accuracy and less false positive results. This research utilized unsupervised machine learning algorithms that includes Isolation Forest and Local Outlier Factor clustering algorithms. By calculating anomaly scores and validating clustering through metrics like the Silhouette Score and Dunn Score, we enhanced the capacity to secure sensitive healthcare data evolving digital threats. Three variations of Isolation Forest (IForest)models (SVM, Decision Tree, and Random Forest) and three variations of Local Outlier Factor (LOF) models (SVM, Decision Tree, and Random Forest) are evaluated based on accuracy, sensitivity, specificity, and F1 Score., Results: Isolation Forest SVM achieves the highest accuracy of 99.21%, high sensitivity (99.75%) and specificity (99.32%), and a commendable F1 Score of 98.72%. The Isolation Forest Decision Tree also performs well with an accuracy of 98.92% and an F1 Score of 99.35%. However, the Isolation Forest Random Forest exhibits lower specificity (72.84%) than the other models., Conclusion: The experimental results reveal that Isolation Forest SVM emerges as the top performer showcasing the effectiveness of these models in anomaly detection tasks. The proposed methodology utilizing isolation forest and SVM produced better results by detecting anomalies with less false positives in this specific EHR of a hospital in North England. Furthermore the proposal is also able to identify new contextual anomalies that were not identified in the baseline methodology., Competing Interests: Declarations Ethics approval and consent to participate Not applicable. Consent for publication All authors are agreed to publish this work. Competing interests The authors declare no competing interests., (© 2024. The Author(s).)

Background: Evaluating patients' experiences is essential when incorporating the patients' perspective in improving healthcare. Experiences are mainly collected using closed-ended questions, although the value of open-ended questions is widely recognized. Natural language processing (NLP) can automate the analysis of open-ended questions for an efficient approach to patient-centeredness.Methods: We developed the Artificial Intelligence Patient-Reported Experience Measures (AI-PREM) tool, consisting of a new, open-ended questionnaire, an NLP pipeline to analyze the answers using sentiment analysis and topic modeling, and a visualization to guide physicians through the results. The questionnaire and NLP pipeline were iteratively developed and validated in a clinical context.Results: The final AI-PREM consisted of five open-ended questions about the provided information, personal approach, collaboration between healthcare professionals, organization of care, and other experiences. The AI-PREM was sent to 867 vestibular schwannoma patients, 534 of which responded. The sentiment analysis model attained an F1 score of 0.97 for positive texts and 0.63 for negative texts. There was a 90% overlap between automatically and manually extracted topics. The visualization was hierarchically structured into three stages: the sentiment per question, the topics per sentiment and question, and the original patient responses per topic.Conclusions: The AI-PREM tool is a comprehensive method that combines a validated, open-ended questionnaire with a well-performing NLP pipeline and visualization. Thematically organizing and quantifying patient feedback reduces the time invested by healthcare professionals to evaluate and prioritize patient experiences without being confined to the limited answer options of closed-ended questions. [ABSTRACT FROM AUTHOR]

Background: Population heterogeneity and the lack of clinical and sociodemographic information in transgender individuals with gender dysphoria (GD) remains a challenge for specialized services in mental health and surgical procedures. It aimed to identify and describe profiles in a sample waiting for gender-affirming surgery. Methods: A sample of 100 outpatients with GD was assessed through a structured interview, Emotion Regulation Difficulty Scale (DERS), Ruminative Response Scale (RRS), Depression, Anxiety and Stress Scale (DASS-21) and Life Satisfaction scale (SWLS). Cluster analysis was used to identify different profile categories. Results: Two subgroups with different profiles were identified: with less clinical severity (LCS) and with high clinical severity (HCS) on emotional dysregulation, acute symptoms of depression, anxiety, stress and association with mental rumination. The HCS cluster had greater vulnerability in terms of psychiatric history, use of psychotropic drugs, HIV positive, child abuse and suicidal behavior. Conclusion: Different profiles were found regarding the vulnerability to mental health in a sample of transgender people with GD who seek a public hospital service for the same clinical-surgical objective. Longitudinal studies are essential to monitor the impact of these contrasts and to target personalized therapeutic approaches in the prevention of psychiatric disorders. [ABSTRACT FROM AUTHOR]

Background: Recent studies have demonstrated that individuals hospitalized due to COVID-19 can be affected by "long-COVID" symptoms for as long as one year after discharge., Objectives: Our study objective is to identify data-driven clusters of patients using a novel, unsupervised machine learning technique., Methods: The study uses data from 437 patients hospitalized in New York City between March 3rd and May 15th of 2020. The data used was abstracted from medical records and collected from a follow-up survey for up to one-year post-hospitalization. Hospitalization data included demographics, comorbidities, and in-hospital complications. The survey collected long-COVID symptoms, and information on general health, social isolation, and loneliness. To perform the analysis, we created a graph by projecting the data onto eight principal components (PCs) and running the K-nearest neighbors algorithm. We then used Louvain's algorithm to partition this graph into non-overlapping clusters., Results: The cluster analysis produced four clusters with distinct health and social connectivity patterns. The first cluster (n = 141) consisted of patients with both long-COVID neurological symptoms (74%) and social isolation/loneliness. The second cluster (n = 137) consisted of healthy patients who were also more socially connected and not lonely. The third cluster (n = 96) contained patients with neurological symptoms who were socially connected but lonely, and the fourth cluster (n = 63) consisted entirely of patients who had traumatic COVID hospitalization, were intubated, suffered symptoms, but were socially connected and experienced recovery., Conclusion: The cluster analysis identified social isolation and loneliness as important features associated with long-COVID symptoms and recovery after hospitalization. It also confirms that social isolation and loneliness, though connected, are not necessarily the same. Physicians need to be aware of how social characteristics relate to long-COVID and patient's ability to cope with the resulting symptoms., (© 2024. The Author(s).)

Background: Heart failure (HF) with preserved or mildly reduced ejection fraction includes a heterogenous group of patients. Reclassification into distinct phenogroups to enable targeted interventions is a priority. This study aimed to identify distinct phenogroups, and compare phenogroup characteristics and outcomes, from electronic health record data., Methods: 2,187 patients admitted to five UK hospitals with a diagnosis of HF and a left ventricular ejection fraction ≥ 40% were identified from the NIHR Health Informatics Collaborative database. Partition-based, model-based, and density-based machine learning clustering techniques were applied. Cox Proportional Hazards and Fine-Gray competing risks models were used to compare outcomes (all-cause mortality and hospitalisation for HF) across phenogroups., Results: Three phenogroups were identified: (1) Younger, predominantly female patients with high prevalence of cardiometabolic and coronary disease; (2) More frail patients, with higher rates of lung disease and atrial fibrillation; (3) Patients characterised by systemic inflammation and high rates of diabetes and renal dysfunction. Survival profiles were distinct, with an increasing risk of all-cause mortality from phenogroups 1 to 3 (p < 0.001). Phenogroup membership significantly improved survival prediction compared to conventional factors. Phenogroups were not predictive of hospitalisation for HF., Conclusions: Applying unsupervised machine learning to routinely collected electronic health record data identified phenogroups with distinct clinical characteristics and unique survival profiles., (© 2024. The Author(s).)

Background: Arteriosclerosis obliterans (ASO) is a chronic arterial disease that can lead to critical limb ischemia. Endovascular therapy is increasingly used for limb salvage in ASO patients, but the outcomes vary. The development of prediction models using unsupervised machine learning may lead to the identification of novel subtypes to guide patient prognosis and treatment., Methods: This retrospective study analyzed clinical data from 448 patients with ASOs who underwent endovascular therapy. Unsupervised machine learning algorithms were employed to identify subgroups. To validate the precision of the clustering outcomes, an analysis of the postoperative results of the clusters was conducted. A prediction model was constructed using binary logistic regression., Results: Two distinct subgroups were identified by unsupervised machine learning and characterized by differing patterns of clinical features. Patients in Cluster 2 had significantly worse conditions and prognoses than those in Cluster 1. For the novel ASO subtypes, a nomogram was developed using six predictive factors, namely, platelet count, ankle brachial index, Rutherford category, operation method, hypertension, and diabetes status. The nomogram achieved excellent discrimination for predicting membership in the two identified clusters, with an area under the curve of 0.96 and 0.95 in training cohort and internal test cohort., Conclusion: This study demonstrated that unsupervised machine learning can reveal novel phenotypic subgroups of patients with varying prognostic risk who underwent endovascular therapy. The prediction model developed could support clinical decision-making and risk counseling for this complex patient population. Further external validation is warranted to assess the generalizability of the findings., (© 2024. The Author(s).)

Background: Machine learning (ML) has emerged as the predominant computational paradigm for analyzing large-scale datasets across diverse domains. The assessment of dataset quality stands as a pivotal precursor to the successful deployment of ML models. In this study, we introduce DREAMER (Data REAdiness for MachinE learning Research), an algorithmic framework leveraging supervised and unsupervised machine learning techniques to autonomously evaluate the suitability of tabular datasets for ML model development. DREAMER is openly accessible as a tool on GitHub and Docker, facilitating its adoption and further refinement within the research community.., Results: The proposed model in this study was applied to three distinct tabular datasets, resulting in notable enhancements in their quality with respect to readiness for ML tasks, as assessed through established data quality metrics. Our findings demonstrate the efficacy of the framework in substantially augmenting the original dataset quality, achieved through the elimination of extraneous features and rows. This refinement yielded improved accuracy across both supervised and unsupervised learning methodologies., Conclusion: Our software presents an automated framework for data readiness, aimed at enhancing the integrity of raw datasets to facilitate robust utilization within ML pipelines. Through our proposed framework, we streamline the original dataset, resulting in enhanced accuracy and efficiency within the associated ML algorithms., (© 2024. The Author(s).)

Purpose: Accurate deformable registration of magnetic resonance imaging (MRI) scans containing pathologies is challenging due to changes in tissue appearance. In this paper, we developed a novel automated three-dimensional (3D) convolutional U-Net based deformable image registration (ConvUNet-DIR) method using unsupervised learning to establish correspondence between baseline pre-operative and follow-up MRI scans of patients with brain glioma., Methods: This study involved multi-parametric brain MRI scans (T1, T1-contrast enhanced, T2, FLAIR) acquired at pre-operative and follow-up time for 160 patients diagnosed with glioma, representing the BraTS-Reg 2022 challenge dataset. ConvUNet-DIR, a deep learning-based deformable registration workflow using 3D U-Net style architecture as a core, was developed to establish correspondence between the MRI scans. The workflow consists of three components: (1) the U-Net learns features from pairs of MRI scans and estimates a mapping between them, (2) the grid generator computes the sampling grid based on the derived transformation parameters, and (3) the spatial transformation layer generates a warped image by applying the sampling operation using interpolation. A similarity measure was used as a loss function for the network with a regularization parameter limiting the deformation. The model was trained via unsupervised learning using pairs of MRI scans on a training data set (n = 102) and validated on a validation data set (n = 26) to assess its generalizability. Its performance was evaluated on a test set (n = 32) by computing the Dice score and structural similarity index (SSIM) quantitative metrics. The model's performance also was compared with the baseline state-of-the-art VoxelMorph (VM1 and VM2) learning-based algorithms., Results: The ConvUNet-DIR model showed promising competency in performing accurate 3D deformable registration. It achieved a mean Dice score of 0.975 ± 0.003 and SSIM of 0.908 ± 0.011 on the test set (n = 32). Experimental results also demonstrated that ConvUNet-DIR outperformed the VoxelMorph algorithms concerning Dice (VM1: 0.969 ± 0.006 and VM2: 0.957 ± 0.008) and SSIM (VM1: 0.893 ± 0.012 and VM2: 0.857 ± 0.017) metrics. The time required to perform a registration for a pair of MRI scans is about 1 s on the CPU., Conclusions: The developed deep learning-based model can perform an end-to-end deformable registration of a pair of 3D MRI scans for glioma patients without human intervention. The model could provide accurate, efficient, and robust deformable registration without needing pre-alignment and labeling. It outperformed the state-of-the-art VoxelMorph learning-based deformable registration algorithms and other supervised/unsupervised deep learning-based methods reported in the literature., (© 2024. The Author(s).)

Background: Subclinical diastolic dysfunction is a precursor for developing heart failure with preserved ejection fraction (HFpEF); yet not all patients progress to HFpEF. Our objective was to evaluate clinical and echocardiographic variables to identify patients who develop HFpEF.Methods: Clinical, laboratory, and echocardiographic data were retrospectively collected for 81 patients without HF and 81 matched patients with HFpEF at the time of first documentation of subclinical diastolic dysfunction. Density-based clustering or hierarchical clustering to group patients was based on 65 total variables including 19 categorical and 46 numerical variables. Logistic regression analysis was conducted on the entire study population as well as each individual cluster to identify independent predictors of HFpEF.Results: Unsupervised clustering identified 3 subgroups which differed in gender composition, severity of cardiac hypertrophy and aortic stenosis, NT-proBNP, percentage of patients who progressed to HFpEF, and timing of disease progression from diastolic dysfunction to HFpEF to death. Clusters that had higher percentages of women had progressively milder cardiac hypertrophy, less severe aortic stenosis, lower NT-proBNP, were diagnosed at an older age with HFpEF, and survived to an older age. Independent predictors of HFpEF for the entire cohort included diabetes, chronic kidney disease, atrial fibrillation, and diuretic use, with additional predictive variables found for each cluster.Conclusions: Cluster analysis can identify phenotypically distinct subgroups of patients with diastolic dysfunction. Clusters differ in HFpEF and mortality outcome. In addition, the variables that correlate with and predict HFpEF outcome differ among clusters. [ABSTRACT FROM AUTHOR]

Background: Persistent critical illness is common in critically ill patients and is associated with vast medical resource use and poor clinical outcomes. This study aimed to define when patients with sepsis would be stabilized and transitioned to persistent critical illness, and whether such transition time varies between latent classes of patients.Methods: This was a retrospective cohort study involving sepsis patients in the eICU Collaborative Research Database. Persistent critical illness was defined at the time when acute physiological characteristics were no longer more predictive of in-hospital mortality (i.e., vital status at hospital discharge) than antecedent characteristics. Latent growth mixture modeling was used to identify distinct trajectory classes by using Sequential Organ Failure Assessment score measured during intensive care unit stay as the outcome, and persistent critical illness transition time was explored in each latent class.Results: The mortality was 16.7% (3828/22,868) in the study cohort. Acute physiological model was no longer more predictive of in-hospital mortality than antecedent characteristics at 15 days after intensive care unit admission in the overall population. Only a minority of the study subjects (n = 643, 2.8%) developed persistent critical illness, but they accounted for 19% (15,834/83,125) and 10% (19,975/198,833) of the total intensive care unit and hospital bed-days, respectively. Five latent classes were identified. Classes 1 and 2 showed increasing Sequential Organ Failure Assessment score over time and transition to persistent critical illness occurred at 16 and 27 days, respectively. The remaining classes showed a steady decline in Sequential Organ Failure Assessment scores and the transition to persistent critical illness occurred between 6 and 8 days. Elevated urea-to-creatinine ratio was a good biochemical signature of persistent critical illness.Conclusions: While persistent critical illness occurred in a minority of patients with sepsis, it consumed vast medical resources. The transition time differs substantially across latent classes, indicating that the allocation of medical resources should be tailored to different classes of patients. [ABSTRACT FROM AUTHOR]

Background: Modern medicine is rapidly moving towards a data-driven paradigm based on comprehensive multimodal health assessments. Integrated analysis of data from different modalities has the potential of uncovering novel biomarkers and disease signatures. Methods: We collected 1385 data features from diverse modalities, including metabolome, microbiome, genetics, and advanced imaging, from 1253 individuals and from a longitudinal validation cohort of 1083 individuals. We utilized a combination of unsupervised machine learning methods to identify multimodal biomarker signatures of health and disease risk. Results: Our method identified a set of cardiometabolic biomarkers that goes beyond standard clinical biomarkers. Stratification of individuals based on the signatures of these biomarkers identified distinct subsets of individuals with similar health statuses. Subset membership was a better predictor for diabetes than established clinical biomarkers such as glucose, insulin resistance, and body mass index. The novel biomarkers in the diabetes signature included 1-stearoyl-2-dihomo-linolenoyl-GPC and 1-(1-enyl-palmitoyl)-2-oleoyl-GPC. Another metabolite, cinnamoylglycine, was identified as a potential biomarker for both gut microbiome health and lean mass percentage. We identified potential early signatures for hypertension and a poor metabolic health outcome. Additionally, we found novel associations between a uremic toxin, p-cresol sulfate, and the abundance of the microbiome genera Intestinimonas and an unclassified genus in the Erysipelotrichaceae family. Conclusions: Our methodology and results demonstrate the potential of multimodal data integration, from the identification of novel biomarker signatures to a data-driven stratification of individuals into disease subtypes and stages—an essential step towards personalized, preventative health risk assessment. [ABSTRACT FROM AUTHOR]

Amyotrophic lateral sclerosis (ALS) displays considerable clinical and genetic heterogeneity. Machine learning approaches have previously been utilised for patient stratification in ALS as they can disentangle complex disease landscapes. However, lack of independent validation in different populations and tissue samples have greatly limited their use in clinical and research settings. We overcame these issues by performing hierarchical clustering on the 5000 most variably expressed autosomal genes from motor cortex expression data of people with sporadic ALS from the KCL BrainBank (N = 112). Three molecular phenotypes linked to ALS pathogenesis were identified: synaptic and neuropeptide signalling, oxidative stress and apoptosis, and neuroinflammation. Cluster validation was achieved by applying linear discriminant analysis models to cases from TargetALS US motor cortex (N = 93), as well as Italian (N = 15) and Dutch (N = 397) blood expression datasets, for which there was a high assignment probability (80-90%) for each molecular subtype. The ALS and motor cortex specificity of the expression signatures were tested by mapping KCL BrainBank controls (N = 59), and occipital cortex (N = 45) and cerebellum (N = 123) samples from TargetALS to each cluster, before constructing case-control and motor cortex-region logistic regression classifiers. We found that the signatures were not only able to distinguish people with ALS from controls (AUC 0.88 ± 0.10), but also reflect the motor cortex-based disease process, as there was perfect discrimination between motor cortex and the other brain regions. Cell types known to be involved in the biological processes of each molecular phenotype were found in higher proportions, reinforcing their biological interpretation. Phenotype analysis revealed distinct cluster-related outcomes in both motor cortex datasets, relating to disease onset and progression-related measures. Our results support the hypothesis that different mechanisms underpin ALS pathogenesis in subgroups of patients and demonstrate potential for the development of personalised treatment approaches. Our method is available for the scientific and clinical community at https://alsgeclustering.er.kcl.ac.uk ., (© 2023. The Author(s).)

Background: Fraud, Waste, and Abuse (FWA) in medical claims have a negative impact on the quality and cost of healthcare. A major component of FWA in claims is procedure code overutilization, where one or more prescribed procedures may not be relevant to a given diagnosis and patient profile, resulting in unnecessary and unwarranted treatments and medical payments. This study aims to identify such unwarranted procedures from millions of healthcare claims. In the absence of labeled examples of unwarranted procedures, the study focused on the application of unsupervised machine learning techniques., Methods: Experiments were conducted with deep autoencoders to find claims containing anomalous procedure codes indicative of FWA, and were compared against a baseline density-based clustering model. Diagnoses, procedures, and demographic data associated with healthcare claims were used as features for the models. A dataset of one hundred thousand claims sampled from a larger claims database is used to initially train and tune the models, followed by experimentations on a dataset with thirty-three million claims. Experimental results show that the autoencoder model, when trained with a novel feature-weighted loss function, outperforms the density-based clustering approach in finding potential outlier procedure codes., Results: Given the unsupervised nature of our experiments, model performance was evaluated using a synthetic outlier test dataset, and a manually annotated outlier test dataset. Precision, recall and F1-scores on the synthetic outlier test dataset for the autoencoder model trained on one hundred thousand claims were 0.87, 1.0 and 0.93, respectively, while the results for these metrics on the manually annotated outlier test dataset were 0.36, 0.86 and 0.51, respectively. The model performance on the manually annotated outlier test dataset improved further when trained on the larger thirty-three million claims dataset with precision, recall and F1-scores of 0.48, 0.90 and 0.63, respectively., Conclusions: This study demonstrates the feasibility of leveraging unsupervised, deep-learning methods to identify potential procedure overutilization from healthcare claims., (© 2023. BioMed Central Ltd., part of Springer Nature.)

Mass-Suite (MSS) is a Python-based, open-source software package designed to analyze high-resolution mass spectrometry (HRMS)-based non-targeted analysis (NTA) data, particularly for water quality assessment and other environmental applications. MSS provides flexible, user-defined workflows for HRMS data processing and analysis, including both basic functions (e.g., feature extraction, data reduction, feature annotation, data visualization, and statistical analyses) and advanced exploratory data mining and predictive modeling capabilities that are not provided by currently available open-source software (e.g., unsupervised clustering analyses, a machine learning-based source tracking and apportionment tool). As a key advance, most core MSS functions are supported by machine learning algorithms (e.g., clustering algorithms and predictive modeling algorithms) to facilitate function accuracy and/or efficiency. MSS reliability was validated with mixed chemical standards of known composition, with 99.5% feature extraction accuracy and ~ 52% overlap of extracted features relative to other open-source software tools. Example user cases of laboratory data evaluation are provided to illustrate MSS functionalities and demonstrate reliability. MSS expands available HRMS data analysis workflows for water quality evaluation and environmental forensics, and is readily integrated with existing capabilities. As an open-source package, we anticipate further development of improved data analysis capabilities in collaboration with interested users., (© 2023. The Author(s).)

Background: Cardiogenic shock (CS) is a complex state with many underlying causes and associated outcomes. It is still difficult to differentiate between various CS phenotypes. We investigated if the CS phenotypes with distinctive clinical profiles and prognoses might be found using the machine learning (ML) consensus clustering approach., Methods: The current study included patients who were diagnosed with CS at the time of admission from the electronic ICU (eICU) Collaborative Research Database. Among 21,925 patients with CS, an unsupervised ML consensus clustering analysis was conducted. The optimal number of clusters was identified by means of the consensus matrix (CM) heat map, cumulative distribution function (CDF), cluster-consensus plots, and the proportion of ambiguously clustered pairs (PAC) analysis. We calculated the standardized mean difference (SMD) of each variable and used the cutoff of ± 0.3 to identify each cluster's key features. We examined the relationship between the phenotypes and several clinical endpoints utilizing logistic regression (LR) analysis., Results: The consensus cluster analysis identified two clusters (Cluster 1: n = 9,848; Cluster 2: n = 12,077). The key features of patients in Cluster 1, compared with Cluster 2, included: lower blood pressure, lower eGFR (estimated glomerular filtration rate), higher BUN (blood urea nitrogen), higher creatinine, lower albumin, higher potassium, lower bicarbonate, lower red blood cell (RBC), higher red blood cell distribution width (RDW), higher SOFA score, higher APS III score, and higher APACHE IV score on admission. The results of LR analysis showed that the Cluster 2 was associated with lower in-hospital mortality (odds ratio [OR]: 0.374; 95% confidence interval [CI]: 0.347-0.402; P < 0.001), ICU mortality (OR: 0.349; 95% CI: 0.318-0.382; P < 0.001), and the incidence of acute kidney injury (AKI) after admission (OR: 0.478; 95% CI: 0.452-0.505; P < 0.001)., Conclusions: ML consensus clustering analysis synthesized the pattern of clinical and laboratory data to reveal distinct CS phenotypes with different clinical outcomes., (© 2023. BioMed Central Ltd., part of Springer Nature.)

Background: Acute Kidney Injury (AKI) is a multifactorial condition which presents a substantial burden to healthcare systems. There is limited evidence on whether it is seasonal. We sought to investigate the seasonality of AKI hospitalisations in England and use unsupervised machine learning to explore clustering of underlying comorbidities, to gain insights for future intervention., Methods: We used Hospital Episodes Statistics linked to the Clinical Practice Research Datalink to describe the overall incidence of AKI admissions between 2015 and 2019 weekly by demographic and admission characteristics. We carried out dimension reduction on 850 diagnosis codes using multiple correspondence analysis and applied k-means clustering to classify patients. We phenotype each group based on the dominant characteristics and describe the seasonality of AKI admissions by these different phenotypes., Results: Between 2015 and 2019, weekly AKI admissions peaked in winter, with additional summer peaks related to periods of extreme heat. Winter seasonality was more evident in those diagnosed with AKI on admission. From the cluster classification we describe six phenotypes of people admitted to hospital with AKI. Among these, seasonality of AKI admissions was observed among people who we described as having a multimorbid phenotype, established risk factor phenotype, and general AKI phenotype., Conclusion: We demonstrate winter seasonality of AKI admissions in England, particularly among those with AKI diagnosed on admission, suggestive of community triggers. Differences in seasonality between phenotypes suggests some groups may be more likely to develop AKI as a result of these factors. This may be driven by underlying comorbidity profiles or reflect differences in uptake of seasonal interventions such as vaccines., (© 2023. BioMed Central Ltd., part of Springer Nature.)

Introduction: Significant regional variations in the HIV epidemic hurt effective common interventions in sub-Saharan Africa. It is crucial to analyze HIV positivity distributions within clusters and assess the homogeneity of countries. We aim at identifying clusters of countries based on socio-behavioural predictors of HIV for screening., Method: We used an agglomerative hierarchical, unsupervised machine learning, approach for clustering to analyse data for 146,733 male and 155,622 female respondents from 13 sub-Saharan African countries with 20 and 26 features, respectively, using Population-based HIV Impact Assessment (PHIA) data from the survey years 2015-2019. We employed agglomerative hierarchical clustering and optimal silhouette index criterion to identify clusters of countries based on the similarity of socio-behavioural characteristics. We analyse the distribution of HIV positivity with socio-behavioural predictors of HIV within each cluster., Results: Two principal components were obtained, with the first describing 62.3% and 70.1% and the second explaining 18.3% and 20.6% variance of the total socio-behavioural variation in females and males, respectively. Two clusters per sex were identified, and the most predictor features in both sexes were: relationship with family head, enrolled in school, circumcision status for males, delayed pregnancy, work for payment in last 12 months, Urban area indicator, known HIV status and delayed pregnancy. The HIV positivity distribution with these variables was significant within each cluster., Conclusions /findings: The findings provide a potential use of unsupervised machine learning approaches for substantially identifying clustered countries based on the underlying socio-behavioural characteristics., (© 2023. The Author(s).)

Background: Accurate risk stratification in pulmonary arterial hypertension (PAH), a devastating cardiopulmonary disease, is essential to guide successful therapy. Machine learning may improve risk management and harness clinical variability in PAH., Methods: We conducted a long-term retrospective observational study (median follow-up: 67 months) including 183 PAH patients from three Austrian PAH expert centers. Clinical, cardiopulmonary function, laboratory, imaging, and hemodynamic parameters were assessed. Cox proportional hazard Elastic Net and partitioning around medoid clustering were applied to establish a multi-parameter PAH mortality risk signature and investigate PAH phenotypes., Results: Seven parameters identified by Elastic Net modeling, namely age, six-minute walking distance, red blood cell distribution width, cardiac index, pulmonary vascular resistance, N-terminal pro-brain natriuretic peptide and right atrial area, constituted a highly predictive mortality risk signature (training cohort: concordance index = 0.82 [95%CI: 0.75 - 0.89], test cohort: 0.77 [0.66 - 0.88]). The Elastic Net signature demonstrated superior prognostic accuracy as compared with five established risk scores. The signature factors defined two clusters of PAH patients with distinct risk profiles. The high-risk/poor prognosis cluster was characterized by advanced age at diagnosis, poor cardiac output, increased red cell distribution width, higher pulmonary vascular resistance, and a poor six-minute walking test performance., Conclusion: Supervised and unsupervised learning algorithms such as Elastic Net regression and medoid clustering are powerful tools for automated mortality risk prediction and clinical phenotyping in PAH., (© 2023. The Author(s).)

Background: Structural variation (SV), which ranges from 50 bp to [Formula: see text] 3 Mb in size, is an important type of genetic variations. Deletion is a type of SV in which a part of a chromosome or a sequence of DNA is lost during DNA replication. Three types of signals, including discordant read-pairs, reads depth and split reads, are commonly used for SV detection from high-throughput sequence data. Many tools have been developed for detecting SVs by using one or multiple of these signals., Results: In this paper, we develop a new method called EigenDel for detecting the germline submicroscopic genomic deletions. EigenDel first takes advantage of discordant read-pairs and clipped reads to get initial deletion candidates, and then it clusters similar candidates by using unsupervised learning methods. After that, EigenDel uses a carefully designed approach for calling true deletions from each cluster. We conduct various experiments to evaluate the performance of EigenDel on low coverage sequence data., Conclusions: Our results show that EigenDel outperforms other major methods in terms of improving capability of balancing accuracy and sensitivity as well as reducing bias. EigenDel can be downloaded from https://github.com/lxwgcool/EigenDel ., (© 2023. The Author(s).)

Radiation dermatitis is a major concern in intensity modulated proton therapy (IMPT) for head and neck cancer (HNC) despite its demonstrated superiority over contemporary photon radiotherapy. In this study, dose surface histogram data extracted from forty-four patients of HNC treated with IMPT was used to predict the normal tissue complication probability (NTCP) of skin. Grades of NTCP-skin were clustered using the K-means clustering unsupervised machine learning (ML) algorithm. A new skin-sparing IMPT (IMPT-SS) planning strategy was developed with three major changes and prospectively implemented in twenty HNC patients. Across skin surfaces exposed from 10 (S10) to 70 (S70) GyRBE, the skin's NTCP demonstrated the strongest associations with S50 and S40 GyRBE (0.95 and 0.94). The increase in the NTCP of skin per unit GyRBE is 0.568 for skin exposed to 50 GyRBE as compared to 0.418 for 40 GyRBE. Three distinct clusters were formed, with 41% of patients in G1, 32% in G2, and 27% in G3. The average (± SD) generalised equivalent uniform dose for G1, G2, and G3 clusters was 26.54 ± 6.75, 38.73 ± 1.80, and 45.67 ± 2.20 GyRBE. The corresponding NTCP (%) were 4.97 ± 5.12, 48.12 ± 12.72 and 87.28 ± 7.73 respectively. In comparison to IMPT, new IMPT-SS plans significantly (P < 0.01) reduced SX GyRBE, gEUD, and associated NTCP-skin while maintaining identical dose volume indices for target and other organs at risk. The mean NTCP-skin value for IMPT-SS was 34% lower than that of IMPT. The dose to skin in patients treated prospectively for HNC was reduced by including gEUD for an acceptable radiation dermatitis determined from the local patient population using an unsupervised MLA in the spot map optimization of a new IMPT planning technique. However, the clinical finding of acute skin toxicity must also be related to the observed reduction in skin dose., (© 2023. The Author(s).)

Background: Coronary atherosclerosis can lead to serious cardiovascular events. In type 2 diabetes (T2DM) patients, the effects of clinical factors on coronary atherosclerosis have not been fully elucidated. We used a clustering method to distinguish the population heterogeneity of T2DM and the differences in coronary atherosclerosis evaluated on coronary computed tomography angiography (CCTA) among groups and to facilitate clinical management., Methods: Clinical data from 1157 T2DM patients with coronary atherosclerosis who underwent CCTA in our hospital from January 2018 to September 2021 were retrospectively collected. The coronary artery segment plaque type and stenosis, the number of involved vessels, the segment involvement score (SIS) and the segment stenosis score (SSS) were evaluated and calculated. Unsupervised clustering analysis based on clinical information was used (cluster 1: n = 463; cluster 2: n = 341; cluster 3: n = 353). The association of coronary plaque characteristics with cluster groups was evaluated., Results: The clinical data among the three groups were different in several aspects: (1) Cluster 1 had the least male patients (41.7%), the lowest proportion of patients with smoking (0%) or alcohol history (0.9%), and the lowest level of serum creatinine (74.46 ± 22.18 µmol/L); (2) Cluster 2 had the shortest duration of diabetes (7.90 ± 8.20 years) and was less likely to be treated with diabetes (42.2%) or statins (17.6%) and (3) Cluster 3 was the youngest (65.89 ± 10.15 years old) and had the highest proportion of male patients (96.6%), the highest proportion of patients with smoking (91.2%) and alcohol (59.8%) history, the highest level of eGFR (83.81 ± 19.06 ml/min/1.73m 2 ), and the lowest level of HDL-C (1.07 ± 0.28 mmol/L). The CCTA characteristics varied with different clusters: (1) Cluster 1 had the largest number of segments with calcified plaques (2.43 ± 2.46) and the least number of segments with mixed plaques (2.24 ± 2.59) and obstructive stenosis (0.98 ± 2.00); (2) Cluster 1 had the lowest proportion of patients with mixed plaques (68%) and obstructive stenosis (32.2%); (3) Cluster 3 had more segments with noncalcified plaques than cluster 1 (0.63 ± 1.02 vs 0.40 ± 0.78, P < 0.05) and the highest proportion of patients with noncalcified plaques (39.9%) and (4) There was no significant difference in the extent of coronary plaques among the three clusters., Conclusions: The unsupervised clustering method could address T2DM patients with heterogeneous clinical indicators and identify groups with different types of coronary plaque and degrees of coronary stenosis. This method has the potential for patient stratification, which is essential for the clinical management of T2DM patients with coronary atherosclerosis., (© 2022. The Author(s).)

Background: Emerging infectious disease-causing RNA viruses, such as the SARS-CoV-2 and Ebola viruses, are thought to rely on bats as natural reservoir hosts. Since these zoonotic viruses pose a great threat to humans, it is important to characterize the bat genome from multiple perspectives. Unsupervised machine learning methods for extracting novel information from big sequence data without prior knowledge or particular models are highly desirable for obtaining unexpected insights. We previously established a batch-learning self-organizing map (BLSOM) of the oligonucleotide composition that reveals novel genome characteristics from big sequence data., Results: In this study, using the oligonucleotide BLSOM, we conducted a comparative genomic study of humans and six bat species. BLSOM is an explainable-type machine learning algorithm that reveals the diagnostic oligonucleotides contributing to sequence clustering (self-organization). When unsupervised machine learning reveals unexpected and/or characteristic features, these features can be studied in more detail via the much simpler and more direct standard distribution map method. Based on this combined strategy, we identified the Mb-level enrichment of CG dinucleotide (Mb-level CpG islands) around the termini of bat long-scaffold sequences. In addition, a class of CG-containing oligonucleotides were enriched in the centromeric and pericentromeric regions of human chromosomes. Oligonucleotides longer than tetranucleotides often represent binding motifs for a wide variety of proteins (e.g., transcription factor binding sequences (TFBSs)). By analyzing the penta- and hexanucleotide composition, we observed the evident enrichment of a wide range of hexanucleotide TFBSs in centromeric and pericentromeric heterochromatin regions on all human chromosomes., Conclusion: Function of transcription factors (TFs) beyond their known regulation of gene expression (e.g., TF-mediated looping interactions between two different genomic regions) has received wide attention. The Mb-level TFBS and CpG islands are thought to be involved in the large-scale nuclear organization, such as centromere and telomere clustering. TFBSs, which are enriched in centromeric and pericentromeric heterochromatin regions, are thought to play an important role in the formation of nuclear 3D structures. Our machine learning-based analysis will help us to understand the differential features of nuclear 3D structures in the human and bat genomes., (© 2022. The Author(s).)