Classification of patients with bipolar disorder using k-means clustering

Who benefits from digital interventions for bipolar disorder? Stage of illness characteristics as predictors of changes in quality of life

OBJECTIVES

This study explored the potential role of stage-related variables in intervention outcomes in bipolar disorder (BD). Specifically, we aimed to identify which subgroups of individuals were most likely to experience improved quality of life following digitally delivered psychosocial interventions for BD.

METHODS

The study involved a secondary analysis of combined data from two randomised control trials (RCTs). Each trial assessed the effectiveness of digitally delivered interventions for improving quality of life, in late-stage (ORBIT RCT) or early-stage (BETTER RCT) BD. Three iterations of cluster analyses were performed, identifying subgroups of individuals based on (i) current phenomenology, (ii) course of illness and (iii) medication response. The resultant subgroups were compared with regard to changes in quality of life pre-post intervention, via repeated measures ANOVAs.

RESULTS

In each cluster analysis, two clusters were found. The current phenomenology clusters reflected two impairment levels, 'moderate impairment' and 'low impairment'. The course of illness clusters reflected 'more chronicity' and 'less chronicity' and the medication response clusters reflected 'good medication response' and 'poor medication response'. Differences in changes in quality of life over time were observed between the two current phenomenology clusters and between the medication response clusters, while the course of illness subgroups did not respond differently.

CONCLUSIONS

There are at least two distinct groups of treatment-seeking individuals with established BD, based on illness features with previously established links to different illness stages. Clusters within the current phenomenology and medication response domains demonstrated significantly different trajectories of QoL change over time in the context of our interventions, highlighting potential implications for treatment selection aligned with precision psychiatry.

Bi-level gene selection of cancer by combining clustering and sparse learning

The diagnosis of cancer based on gene expression profile data has attracted extensive attention in the field of biomedical science. This type of data usually has the characteristics of high dimensionality and noise. In this paper, a hybrid gene selection method based on clustering and sparse learning is proposed to choose the key genes with high precision. We first propose a filter method, which combines the k-means clustering algorithm and signal-to-noise ratio ranking method, and then, a weighted gene co-expression network has been applied to the reduced data set to identify modules corresponding to biological pathways. Moreover, we choose the key genes by using group bridge and sparse group lasso as wrapper methods. Finally, we conduct some numerical experiments on six cancer datasets. The numerical results show that our proposed method has achieved good performance in gene selection and cancer classification.

Functional recovery prediction during rehabilitation after rotator cuff tears by decision support system

BACKGROUND

Today's rehabilitation decision-making still relies on conventional methods and different specific targeted rehabilitation protocols. Our study focuses on the decision support system for early rehabilitation after rotator cuff (RC) tears repair, where a multicriteria decision-making framework (MCDM) is applied for the prediction of successful functional recovery and selection of a rehabilitation protocol.

OBJECTIVE

To identify factors that affect recovery outcomes and to develop a decision support system methodology for predicting functional recovery outcomes at early rehabilitation after RC repair.

METHODS

Twelve rehabilitation experts were involved in the design, calibration, and evaluation of a rehabilitation protocol based on the proposed decision support system constructed using the MCDM framework. For the development of a decision support system, 20 patients after RC surgery undergoing outpatient rehabilitation were enrolled in a prospective cohort clinical trial.

RESULTS

The MCDM framework (SWARA method) sensitively assesses different criteria and determines the corresponding criteria weights that were similar to criteria weights assessed subjectively by rehabilitation experts. The assignment of patients into the classes, according to the heuristic evaluation method based on expert opinion and the standard qualitative evaluation methods showed the validity of MCDM methods remain the best new alternative in predicting recovery during rehabilitation.

CONCLUSIONS

The results of this paper show that sustainable rehabilitation is an area that is quite suitable for the use of MCDM. The most of rehabilitation protocols are based on traditional methods and approaches, but the sensitive results showed the validity of MCDM methods and remains the best new alternative in prediction recovery protocols during rehabilitation.

LRFMV: An efficient customer segmentation model for superstores

The Recency, Frequency, and Monetary model, also known as the RFM model, is a popular and widely used business model for determining beneficial client segments and analyzing profit. It is also recommended and frequently used in superstores to identify customer segments and increase profit margins. Later, the Length, Recency, Frequency, and Monetary model, also known as the LRFM model, was introduced as an improved version of the RFM model to identify more relevant and exact consumer groups for profit maximization. Superstores have a varying number of different products. In RFM and LRFM models, the relationship between profit and purchased quantity has never been investigated. Therefore, this paper proposed an efficient customer segmentation model, namely LRFMV (Length, Recency, Frequency, Monetary and Volume) and studied the profit-quantity relationship. A new dimension V (volume) has been added to the existing LRFM model to show a direct profit-quantity relationship in customer segmentation. The V stands for volume, which is derived by calculating the average number of products purchased by a frequent superstore client in a single day. The data obtained from feature extraction of the LRMFV model is then clustered by using conventional K-means, K-Medoids, and Mini Batch K-means methods. The results obtained from the three algorithms are compared, and the K-means algorithm is chosen for the superstore dataset of the proposed LRFMV model. All clusters created using these three algorithms are evaluated in the LRFMV model, and a close relationship between profit and volume is observed. A clear profit-quantity relationship of items has yet not been seen in any prior study on the RFM and LRFM models. Grouping customers aiming at profit maximization existed previously, but there was no clear and direct depiction of profit and quantity of sold items. This study applied unsupervised machine learning to investigate the patterns, trends, and correlations between volume and profit. The traits of all the clusters are analyzed by the Customer-Classification Matrix. The LRFMV values, larger or less than the overall average for each cluster, are identified as their traits. The performance of the proposed LRFMV model is compared with the legacy RFM and LRFM customer segmentation models. The outcome shows that the LRFMV model creates precise customer segments with the same number of customers while maintaining a greater profit.

Patterns of patients with polypharmacy in adult population from Korea

Polypharmacy and its rising global prevalence is a growing public health burden. Using a large representative nationwide Korean cohort (N = 761,145), we conducted a retrospective cross-sectional study aiming to identify subpopulations of patients with polypharmacy and characterize their unique patterns through cluster analysis. Patients aged ≥ 30 years who were prescribed at least one medication between 2014 and 2018 were included in our study. Six clusters were identified: cluster 1 mostly included patients who were hospitalized for a long time (4.3 ± 5.3 days); cluster 2 consisted of patients with disabilities (100.0%) and had the highest mean number of prescription drugs (7.7 ± 2.8 medications); cluster 3 was a group of low-income patients (99.9%); cluster 4 was a group of high-income patients (80.2%) who frequently (46.4 ± 25.9 days) visited hospitals/clinics (7.3 ± 2.7 places); cluster 5 was mostly elderly (74.9 ± 9.8 years) females (80.3%); and cluster 6 comprised mostly middle-aged (56.4 ± 1.5 years) males (88.6%) (all P < 0.001). Patients in clusters 1-5 had more prescribed medications and outpatient visit days than those in cluster 6 (all P < 0.001). Given limited health care resources, individuals with any of the identified phenotypes may be preferential candidates for participation in intervention programs for optimal medication use.

Association between neutrophil-lymphocyte ratio, platelet-lymphocyte ratio, and C-reactive protein levels and metabolic status in patients with a bipolar disorder

OBJECTIVES

Neutrophil-lymphocyte ratio (NLR), platelet-lymphocyte ratio (PLR), and C-Reactive Protein (CRP) are markers of inflammation that are elevated in bipolar disorder (BD) and are also related to a higher risk of metabolic syndrome (MetS). This study aimed at investigating for the first time the association between NLR, PLR, and CRP and the metabolic status in BD.

METHODS

We assessed the association between biomarkers and the metabolic status: number of metabolic risk factors, presence of MetS, insulin sensitivity (Quantitative Insulin Sensitivity Check Index, QUICKI) and insulin resistance (Homeostatic Model Assessment for Insulin Resistance, HOMA-IR index), in a sample of 219 outpatients with BD.

RESULTS

25.9% of the sample met the criteria for MetS. High levels of CRP were found in 12% of the sample. Older age, low PLR, high NLR, and high CRP levels significantly predicted a higher number of MetS risk factors (p < 0.001). Older age and low PLR were associated with a greater likelihood of developing MetS (p = 0.007).

CONCLUSIONS

Although further studies are needed to replicate and validate these findings, inflammatory biomarkers as CRP, PLR and NLR could be useful tools to identify patients with a BD at risk for a metabolic adverse outcome.

An Improved K-means Clustering Algorithm Towards an Efficient Data-Driven Modeling

K-means algorithm is one of the well-known unsupervised machine learning algorithms. The algorithm typically finds out distinct non-overlapping clusters in which each point is assigned to a group. The minimum squared distance technique distributes each point to the nearest clusters or subgroups. One of the K-means algorithm’s main concerns is to find out the initial optimal centroids of clusters. It is the most challenging task to determine the optimum position of the initial clusters’ centroids at the very first iteration. This paper proposes an approach to find the optimal initial centroids efficiently to reduce the number of iterations and execution time. To analyze the effectiveness of our proposed method, we have utilized different real-world datasets to conduct experiments. We have first analyzed COVID-19 and patient datasets to show our proposed method’s efficiency. A synthetic dataset of 10M instances with 8 dimensions is also used to estimate the performance of the proposed algorithm. Experimental results show that our proposed method outperforms traditional kmeans++ and random centroids initialization methods regarding the computation time and the number of iterations.

STI epidemic re-emergence, socio-epidemiological clusters characterisation and HIV coinfection in Catalonia, Spain, during 2017-2019: a retrospective population-based cohort study

OBJECTIVES

To describe the epidemiology of sexually transmitted infections (STIs), identify and characterise socio-epidemiological clusters and determine factors associated with HIV coinfection.

DESIGN

Retrospective population-based cohort.

SETTING

Catalonia, Spain.

PARTICIPANTS

42 283 confirmed syphilis, gonorrhoea, chlamydia and lymphogranuloma venereum cases, among 34 600 individuals, reported to the Catalan HIV/STI Registry in 2017-2019.

PRIMARY AND SECONDARY OUTCOMES

Descriptive analysis of confirmed STI cases and incidence rates. Factors associated with HIV coinfection were determined using logistic regression. We identified and characterized socio-epidemiological STI clusters by Basic Health Area (BHA) using K-means clustering.

RESULTS

The incidence rate of STIs increased by 91.3% from 128.2 to 248.9 cases per 100 000 population between 2017 and 2019 (p<0.001), primarily driven by increase among women (132%) and individuals below 30 years old (125%). During 2017-2019, 50.1% of STIs were chlamydia and 31.6% gonorrhoea. Reinfections accounted for 10.8% of all cases and 6% of cases affected HIV-positive individuals. Factors associated with the greatest likelihood of HIV coinfection were male sex (adjusted OR (aOR) 23.69; 95% CI 16.67 to 35.13), age 30-39 years (versus <20 years, aOR 18.58; 95% CI 8.56 to 52.13), having 5-7 STI episodes (vs 1 episode, aOR 5.96; 95% CI 4.26 to 8.24) and living in urban areas (aOR 1.32; 95% CI 1.04 to 1.69). Living in the most deprived BHAs (aOR 0.60; 95% CI 0.50 to 0.72) was associated with the least likelihood of HIV coinfection. K-means clustering identified three distinct clusters, showing that young women in rural and more deprived areas were more affected by chlamydia, while men who have sex with men in urban and less deprived areas showed higher rates of STI incidence, multiple STI episodes and HIV coinfection.

CONCLUSIONS

We recommend socio-epidemiological identification and characterisation of STI clusters and factors associated with HIV coinfection to identify at-risk populations at a small health area level to design effective interventions.

Treatment selection using prototyping in latent-space with application to depression treatment

Machine-assisted treatment selection commonly follows one of two paradigms: a fully personalized paradigm which ignores any possible clustering of patients; or a sub-grouping paradigm which ignores personal differences within the identified groups. While both paradigms have shown promising results, each of them suffers from important limitations. In this article, we propose a novel deep learning-based treatment selection approach that is shown to strike a balance between the two paradigms using latent-space prototyping. Our approach is specifically tailored for domains in which effective prototypes and sub-groups of patients are assumed to exist, but groupings relevant to the training objective are not observable in the non-latent space. In an extensive evaluation, using both synthetic and Major Depressive Disorder (MDD) real-world clinical data describing 4754 MDD patients from clinical trials for depression treatment, we show that our approach favorably compares with state-of-the-art approaches. Specifically, the model produced an 8% absolute and 23% relative improvement over random treatment allocation. This is potentially clinically significant, given the large number of patients with MDD. Therefore, the model can bring about a much desired leap forward in the way depression is treated today.

Pharmacological treatment profiles in the FACE-BD cohort: An unsupervised machine learning study, applied to a nationwide bipolar cohort✰

BACKGROUND

Despite thorough and validated clinical guidelines based on bipolar disorders subtypes, large pharmacological treatment heterogeneity remains in these patients. There is limited knowledge about the different treatment combinations used and their influence on patient outcomes. We attempted to determine profiles of patients based on their treatments and to understand the clinical characteristics associated with these treatment profiles.

METHODS

This multicentre longitudinal study was performed on a French nationwide bipolar cohort database. We performed hierarchical agglomerative clustering to search for clusters of individuals based on their treatments during the first year following inclusion. We then compared patient clinical characteristics according to these clusters.

RESULTS

Four groups were identified among the 1795 included patients: group 1 ("heterogeneous" n = 1099), group 2 ("lithium" n = 265), group 3 ("valproate" n = 268), and group 4 ("lamotrigine" n = 163). Proportion of bipolar 1 disorder, in groups 1 to 4 were: 48.2%, 57.0%, 48.9% and 32.5%. Groups 1 and 4 had greater functional impact at baseline and a less favorable clinical and functioning evolution at one-year follow-up, especially on GAF and FAST scales.

LIMITATIONS

The one-year period used for the analysis of mood stabilizing treatments remains short in the evolution of bipolar disorder.

CONCLUSIONS

Treatment profiles are associated with functional evolution of patients and were not clearly determined by bipolar subtypes. These profiles seem to group together common patient phenotypes. These findings do not seem to be influenced by the duration of disease prior to inclusion and neither by the number of treatments used during the follow-up period.

Conceptualizing the later stage of bipolar disorder: Descriptive analyses from the ORBIT trial

OBJECTIVES

This study aimed to investigate the features of later stage bipolar disorder (BD) and specifically, factors underlying later stage BD and potential subgroups within this stage, to understand more about the later stage group and contribute to the measurement of stage.

METHODS

An exploratory factor analysis was conducted using variables relating to current phenomenological aspects of illness, followed by cluster analyses based on the identified factors. Finally, the resultant clusters were compared based on course of illness variables.

RESULTS

Fourteen extracted factors explained 57 percent of the variance. Latent structures aligned with current depressive symptoms, energy and interest, independence, occupational functioning, symptoms of anxiety, pain, elevated symptoms, interpersonal functioning, anger, perceptions of social connections, and perceptions of current medication effectiveness, cognitive issues, sleep issues, and sense of isolation. Two clusters were identified which differed significantly on each of these factors, and on a range of course of illness features including lifetime number of episodes, duration of illness and number of depressive hospitalizations.

CONCLUSIONS

Latent phenomenological features relevant to individuals in the later stage of BD were identified. Two clusters of individuals in later stage BD differ based on these features as well as course of illness, suggesting that there are distinct subgroups of individuals in the later stage of BD, distinguishable based on current phenomenology and illness history. However, findings are exploratory and therefore require confirmation before they can be applied clinically.

Demystifying machine learning: a primer for physicians

Machine learning is a tool for analysing digitised data sets and formulating predictions that can optimise clinical decision-making. It aims to identify complex patterns in large data sets and encode them into models that can then classify new unseen cases or make predictions on new data. Machine learning methods take several forms and individual models can be of many different types. More than 50 models have been approved for use in routine healthcare, and the numbers continue to grow exponentially. The reliability and robustness of any model depends on multiple factors, including the quality and quantity of the data used to develop the models, and the selection of features in the data considered most important to maximising accuracy. In ensuring models are safe, effective and reproducible in routine care, physicians need to have some understanding of how these models are developed and evaluated, and to collaborate with data and computer scientists in their design and validation. This narrative review introduces principles, methods and examples of machine learning in a way that does not require mastery of highly complex statistical and computational concepts.

Discovering prescription patterns in pediatric acute-onset neuropsychiatric syndrome patients

OBJECTIVE

Pediatric acute-onset neuropsychiatric syndrome (PANS) is a complex neuropsychiatric syndrome characterized by an abrupt onset of obsessive-compulsive symptoms and/or severe eating restrictions, along with at least two concomitant debilitating cognitive, behavioral, or neurological symptoms. A wide range of pharmacological interventions along with behavioral and environmental modifications, and psychotherapies have been adopted to treat symptoms and underlying etiologies. Our goal was to develop a data-driven approach to identify treatment patterns in this cohort.

MATERIALS AND METHODS

In this cohort study, we extracted medical prescription histories from electronic health records. We developed a modified dynamic programming approach to perform global alignment of those medication histories. Our approach is unique since it considers time gaps in prescription patterns as part of the similarity strategy.

RESULTS

This study included 43 consecutive new-onset pre-pubertal patients who had at least 3 clinic visits. Our algorithm identified six clusters with distinct medication usage history which may represent clinician's practice of treating PANS of different severities and etiologies i.e., two most severe groups requiring high dose intravenous steroids; two arthritic or inflammatory groups requiring prolonged nonsteroidal anti-inflammatory drug (NSAID); and two mild relapsing/remitting group treated with a short course of NSAID. The psychometric scores as outcomes in each cluster generally improved within the first two years.

DISCUSSION AND CONCLUSION

Our algorithm shows potential to improve our knowledge of treatment patterns in the PANS cohort, while helping clinicians understand how patients respond to a combination of drugs.

Unsupervised color image segmentation: A case of RGB histogram based K-means clustering initialization

Color-based image segmentation classifies pixels of digital images in numerous groups for further analysis in computer vision, pattern recognition, image understanding, and image processing applications. Various algorithms have been developed for image segmentation, but clustering algorithms play an important role in the segmentation of digital images. This paper presents a novel and adaptive initialization approach to determine the number of clusters and find the initial central points of clusters for the standard K-means algorithm to solve the segmentation problem of color images. The presented scheme uses a scanning procedure of the paired Red, Green, and Blue (RGB) color-channel histograms for determining the most salient modes in every histogram. Next, the histogram thresholding is applied and a search in every histogram mode is performed to accomplish RGB pairs. These RGB pairs are used as the initial cluster centers and cluster numbers that clustered each pixel into the appropriate region for generating the homogeneous regions. The proposed technique determines the best initialization parameters for the conventional K-means clustering technique. In this paper, the proposed approach was compared with various unsupervised image segmentation techniques on various image segmentation benchmarks. Furthermore, we made use of a ranking approach inspired by the Evaluation Based on Distance from Average Solution (EDAS) method to account for segmentation integrity. The experimental results show that the proposed technique outperforms the other existing clustering techniques by optimizing the segmentation quality and possibly reducing the classification error.

Identifying GPCR-drug interaction based on wordbook learning from sequences

BACKGROUND

G protein-coupled receptors (GPCRs) mediate a variety of important physiological functions, are closely related to many diseases, and constitute the most important target family of modern drugs. Therefore, the research of GPCR analysis and GPCR ligand screening is the hotspot of new drug development. Accurately identifying the GPCR-drug interaction is one of the key steps for designing GPCR-targeted drugs. However, it is prohibitively expensive to experimentally ascertain the interaction of GPCR-drug pairs on a large scale. Therefore, it is of great significance to predict the interaction of GPCR-drug pairs directly from the molecular sequences. With the accumulation of known GPCR-drug interaction data, it is feasible to develop sequence-based machine learning models for query GPCR-drug pairs.

RESULTS

In this paper, a new sequence-based method is proposed to identify GPCR-drug interactions. For GPCRs, we use a novel bag-of-words (BoW) model to extract sequence features, which can extract more pattern information from low-order to high-order and limit the feature space dimension. For drug molecules, we use discrete Fourier transform (DFT) to extract higher-order pattern information from the original molecular fingerprints. The feature vectors of two kinds of molecules are concatenated and input into a simple prediction engine distance-weighted K-nearest-neighbor (DWKNN). This basic method is easy to be enhanced through ensemble learning. Through testing on recently constructed GPCR-drug interaction datasets, it is found that the proposed methods are better than the existing sequence-based machine learning methods in generalization ability, even an unconventional method in which the prediction performance was further improved by post-processing procedure (PPP).

CONCLUSIONS

The proposed methods are effective for GPCR-drug interaction prediction, and may also be potential methods for other target-drug interaction prediction, or protein-protein interaction prediction. In addition, the new proposed feature extraction method for GPCR sequences is the modified version of the traditional BoW model and may be useful to solve problems of protein classification or attribute prediction. The source code of the proposed methods is freely available for academic research at https://github.com/wp3751/GPCR-Drug-Interaction.

Internet Gaming Disorder Clustering Based on Personality Traits in Adolescents, and Its Relation with Comorbid Psychological Symptoms

In recent years, the evidence regarding Internet Gaming Disorder (IGD) suggests that some personality traits are important risk factors for developing this problem. The heterogeneity involved in problematic online gaming and differences found in the literature regarding the comorbid psychopathology associated with the problem could be explained through different types of gamers. Clustering analysis can allow organization of a collection of personality traits into clusters based on similarity. The objectives of this study were: (1) to obtain an empirical classification of IGD patients according to personality variables and (2) to describe the resultant groups in terms of clinical and sociodemographic variables. The sample included 66 IGD adolescent patients who were consecutive referrals at a mental health center in Barcelona, Spain. A Gaussian mixture model cluster analysis was used in order to classify the subjects based on their personality. Two clusters based on personality traits were detected: type I "higher comorbid symptoms" (n = 24), and type II "lower comorbid symptoms" (n = 42). The type I included higher scores in introversive, inhibited, doleful, unruly, forceful, oppositional, self-demeaning and borderline tendency traits, and lower scores in histrionic, egotistic and conforming traits. The type I obtained higher scores on all the Symptom Check List-90 items-Revised, all the State-Trait Anxiety Index scales, and on the DSM-5 IGD criteria. Differences in personality can be useful in determining clusters with different types of dysfunctionality.

Machine learning methods in psychiatry: a brief introduction

Machine learning (ML) techniques have been widely used to address mental health questions. We discuss two main aspects of ML in psychiatry in this paper, that is, supervised learning and unsupervised learning. Examples are used to illustrate how ML has been implemented in recent mental health research.

A clinical staging model for bipolar disorder: longitudinal approach

Bipolar disorder (BD) has been identified as a life-course illness with different clinical manifestations from an at-risk to a late stage, supporting the assumption that it would benefit from a staging model. In a previous study, we used a clustering approach to stratify 224 patients with a diagnosis of BD into five clusters based on clinical characteristics, functioning, cognition, general health, and health-related quality of life. This study was design to test the construct validity of our previously developed k-means clustering model and to confirm its longitudinal validity over a span of 3 years. Of the 224 patients included at baseline who were used to develop our model, 129 (57.6%) reached the 3-year follow-up. All life domains except mental health-related quality of life (QoL) showed significant worsening in stages (p < 0.001), suggesting construct validity. Furthermore, as patients progressed through stages, functional decline (p < 0.001) and more complex treatment patterns (p = 0.002) were observed. As expected, at 3 years, the majority of patients remained at the same stage (49.6%), or progressed (20.9%) or regressed (23.3%) one stage. Furthermore, 85% of patients who stayed euthymic during that period remained at the same stage or regressed to previous stages, supporting its longitudinal validity. For that reason, this study provides evidence of the construct and longitudinal validity of an empirically developed, comprehensive staging model for patients with BD. Thus, it may help clinicians and researchers to better understand the disorder and, at the same time, to design more accurate and personalized treatment plans.