Predict Cluster Membership Research Articles

Trajectory Analysis of Healthcare Use Before and After Gastrointestinal Cancer Surgery.

Frailty correlates with worse post-operative outcomes and higher surgical costs, but the long-term impact on healthcare utilization remains ill-defined. We sought to evaluate patterns of healthcare utilization pre- and post-surgery among patients with gastrointestinal cancer and characterize the association with frailty. Data on patients who underwent surgical resection for liver, biliary, pancreatic, colon and rectal cancer were obtained from the 2005-2020 SEER-Medicare database. Frailty was assessed using the claims-based frailty index. Group-based trajectory modelling identified clusters of patients with discrete patterns of healthcare utilization. Multivariable regression was performed to predict cluster membership based on preoperative factors, including frailty. Among 66,684 beneficiaries, four distinct utilization trajectories based on data from 12 months before and after surgery were identified. Following a surge in utilization during the month of surgery, most patients reverted to pre-surgery baseline utilization (low: n=6588, 9.9%; moderate: n=17,627, 26.4%; high: n=29,850, 44.8%). However, a notable trajectory involving 12,619 (18.9%) patients was identified, wherein surgery precipitated a transition from a "low" pre-surgery utilization state to a "high" utilization state post-surgery. Frail patients were more likely to be among those individuals who transitioned to high utilizers (low: 4.2% vs. vs. transition: 12.6% vs. high: 7.5%; p<0.001). On multivariable analysis incorporating preoperative variables, frailty was associated with high group trajectory membership (ref: least and moderate; highest: OR 4.90, 95%CI 4.49-5.35; p<0.001). Patients with gastrointestinal cancer demonstrated distinct clusters of healthcare utilization after surgical resection. Preoperative predictive models may help differentiate different health care utilization trajectories to help tailor care for patients in the postoperative period.

Social inequality in admission chances for prestigious higher education programs in Germany: do application patterns matter?

Abstract Research has shown that admission to prestigious higher education programs varies by students' socio-economic status (SES). Access to these programs is characterized by high competition and often rather complex admission procedures. Thus, access may depend not only on students’ performance and decisions to apply but also on their application patterns: Where and how they apply, which may vary by social background due to differences in educational achievement, aspirations, and constraints. Using applications to highly prestigious medical programs in Germany, we examine whether admission chances are socially selective even among the positively selected group of applicants, and whether this is due to SES differences in application patterns or performance. Based on complete application register data, we identify application patterns through cluster analysis. We then used the resulting cluster model to predict cluster membership in the 2018 applicant cohort, for which we collected survey data with information on applicants’ SES, preferences, and motivations. We find that application patterns vary primarily by applicants’ performance (grades and test scores) and SES-specific geographic constraints. However, our multivariate analyses on admission chances show that application patterns do not mediate SES differences in admission chances. Instead, these differences are entirely due to SES differences in applicants’ performance.

Employee Technostress: Analyzing the Influence of Sociodemographic Characteristics on Techno-Stressors and Their Consequences

Technostress has garnered increasing interest in recent times due to the pervasive role of technology and its impact on well-being and overall quality of life. Understanding and addressing technostress is crucial to promoting healthier relationships with technology, mitigating its negative effects, and developing strategies to manage the challenges it poses. This article presents a set of sociodemographic characteristics, explores their relationships with different dimensions of technostress (i.e., techno-stressors), and analyzes the consequences of these techno-stressors. The aim of this research is threefold: first, we examine how the sociodemographic characteristics of gender, age, living arrangement, education level, work experience, tenure, organization size, and organization type relate to techno-stressors, namely techno-overload, techno-invasion, techno-complexity, techno-insecurity, and techno-uncertainty. Second, we identify technostress profiles based on their levels of technostress across these five dimensions. Third, we analyze the impact of the selected sociodemographic variables in predicting cluster membership and examine differences in role overload, role conflict, life satisfaction, and work–family conflict across the identified clusters. Data were collected cross-sectionally from a convenience sample of 1187 Chilean workers and analyzed using R Studio for Windows. Descriptive statistics, cluster analysis, and multinomial logistic regression were performed. Our results show that certain sociodemographic variables have a significant impact on specific dimensions of technostress. We also identified three distinct clusters based on their technostress levels (high, moderate, and low). Cluster 1 emerged as the profile with the lowest levels across all dimensions, cluster 2 displayed moderate levels, and cluster 3 exhibited the highest levels. Multinomial logistic regression also revealed that living alone, education level, tenure, and organization size were all significant predictors of technostress cluster membership. There were also significant differences among the three clusters in terms of the consequences of technostress. Specifically, individuals in the high technostress cluster reported higher levels of role overload and role conflict, lower life satisfaction, and greater work–family conflict. Future research avenues and implications for mental health and well-being are discussed at the end of the article.

Towards stratified treatment of JIA: machine learning identifies subtypes in response to methotrexate from four UK cohorts

Methotrexate (MTX) is the gold-standard first-line disease-modifying anti-rheumatic drug for juvenile idiopathic arthritis (JIA), despite only being either effective or tolerated in half of children and young people (CYP). To facilitate stratified treatment of early JIA, novel methods in machine learning were used to i) identify clusters with distinct disease patterns following MTX initiation; ii) predict cluster membership; and iii) compare clusters to existing treatment response measures. Discovery and verification cohorts included CYP who first initiated MTX before January 2018 in one of four UK multicentre prospective cohorts of JIA within the CLUSTER consortium. JADAS components (active joint count, physician (PGA) and parental (PGE) global assessments, ESR) were recorded at MTX start and over the following year. Clusters of MTX 'response' were uncovered using multivariate group-based trajectory modelling separately in discovery and verification cohorts. Clusters were compared descriptively to ACR Pedi 30/90 scores, and multivariate logistic regression models predicted cluster-group assignment. The discovery cohorts included 657 CYP and verification cohorts 1241 CYP. Six clusters were identified: Fast improvers (11%), Slow Improvers (16%), Improve-Relapse (7%), Persistent Disease (44%), Persistent PGA (8%) and Persistent PGE (13%), the latter two characterised by improvement in all features except one. Factors associated with clusters included ethnicity, ILAR category, age, PGE, and ESR scores at MTX start, with predictive model area under the curve values of 0.65-0.71. Singular ACR Pedi 30/90 scores at 6 and 12 months could not capture speeds of improvement, relapsing courses or diverging disease patterns. Six distinct patterns following initiation of MTX have been identified using methods in artificial intelligence. These clusters demonstrate the limitations in traditional yes/no treatment response assessment (e.g., ACRPedi30) and can form the basis of a stratified medicine programme in early JIA. Medical Research Council, Versus Arthritis, Great Ormond Street Hospital Children's Charity, Olivia's Vision, and the National Institute for Health Research.

Utility of Baseline Pathological, Neuroimaging and Clinical Markers for Progression Prognosis and Subtype Classification in Early Parkinson’s Disease: Implications for Cognitive Impairment

AbstractBackgroundCognitive dysfunction is a well‐recognised PD symptom, but it is highly heterogenous and under‐assessed clinically. Interestingly, PD motor subtypes are associated with different risk of cognitive decline. However, subtype classification is currently reliant on subjective clinician assessment and may consequently be misdiagnosed. Therefore, objective biomarkers of PD progression are critically needed, but must be validated prior to clinical deployment. We assessed the utility of various clinical, neuroimaging and pathological measures as predictors of cognitive outcomes up to 5‐years follow‐up in early PD.MethodData was extracted from the Parkinson’s Progression Markers Initiative. Cognitive assessments were compiled into an overall cognition score via principal component analysis. Distinct subgroups were identified via Fuzzy C‐Means clustering on Year‐5 Unified Parkinson’s Disease Rating Scale (UPDRS) motor scores. Models predicting cluster membership were created via logistic regression and a stratified cross‐validation machine learning pipeline, comparing two predictor inputs: 1) baseline motor assessments; and 2) incorporating additional measures, including prodromal assessments (sleep, olfactory and autonomic function), neuroimaging (proxy SN volume, striatal DaT binding) and biofluid markers (CSF alpha‐syn, p‐tau, a‐beta, and serum IGF‐1) at baseline. Finally, cluster membership was used for retroactive assessment of baseline and follow‐up cognitive data.ResultsTwo clusters were identified, with cluster two (n = 117) demonstrating higher rigidity, lower DaT binding, worsened motor outcomes and increased mood dysfunction at follow‐up compared to tremor‐dominant cluster one (n = 183). Additionally, rigid‐dominant membership was predicted by male sex, higher p‐tau, along with lower a‐beta, predicting 40% of variance (n = 111). This was significantly higher (P&lt;0.001) than baseline motor scores alone, only accounting for 18% of variance. This was supported by machine learning, whereby incorporating additional assessments corresponded to a classification accuracy of 62%, compared to 26% when omitted. Finally, higher probability of belonging to the rigid‐dominant cluster yielded significant negative correlations cognition at baseline and follow‐up.ConclusionIncreased likelihood of rigid‐dominance corresponded to more profound cognitive decline, which may in part result from neuropathological marker concentrations as highlighted by predictive models. Therefore, further tracking of this cohort, specifically conversion into MCI or dementia, could provide insight into pathological contribution to differences in cognitive decline between PD subtypes.

Hepatic Involvement across the Metabolic Syndrome Spectrum: Non-Invasive Assessment and Risk Prediction Using Machine Learning.

Metabolic-dysfunction-associated steatotic liver disease (MASLD) and metabolic syndrome (MetS) are inextricably linked conditions, both of which are experiencing an upward trend in prevalence, thereby exerting a substantial clinical and economic burden. The presence of MetS should prompt the search for metabolic-associated liver disease. Liver fibrosis is the main predictor of liver-related morbidity and mortality. Non-invasive tests (NIT) such as the Fibrosis-4 index (FIB4), aspartate aminotransferase-to-platelet ratio index (APRI), aspartate aminotransferase-to-alanine aminotransferase ratio (AAR), hepatic steatosis index (HIS), transient elastography (TE), and combined scores (AGILE3+, AGILE4) facilitate the detection of liver fibrosis or steatosis. Our study enrolled 217 patients with suspected MASLD, 109 of whom were diagnosed with MetS. We implemented clinical and biological evaluations complemented by transient elastography (TE) to discern the most robust predictors for liver disease manifestation patterns. Patients with MetS had significantly higher values of FIB4, APRI, HSI, liver stiffness, and steatosis parameters measured by TE, as well as AGILE3+ and AGILE4 scores. Machine-learning algorithms enhanced our evaluation. A two-step cluster algorithm yielded three clusters with reliable model quality. Cluster 1 contained patients without significant fibrosis or steatosis, while clusters 2 and 3 showed a higher prevalence of significant liver fibrosis or at least moderate steatosis as measured by TE. A decision tree algorithm identified age, BMI, liver enzyme levels, and metabolic syndrome characteristics as significant factors in predicting cluster membership with an overall accuracy of 89.4%. Combining NITs improves the accuracy of detecting patterns of liver involvement in patients with suspected MASLD.

Cognitive subgroups and the relationships with symptoms, psychosocial functioning and quality of life in first-episode non-affective psychosis: a cluster-analysis approach.

Prior research examining cognitive heterogeneity in psychotic disorders primarily focused on chronic schizophrenia, with limited data on first-episode psychosis (FEP). We aimed to identify distinct cognitive subgroups in adult FEP patients using data-driven cluster-analytic approach, and examine relationships between cognitive subgroups and a comprehensive array of illness-related variables. Two-hundred-eighty-nine Chinese patients aged 26-55 years presenting with FEP to an early intervention program in Hong Kong were recruited. Assessments encompassing premorbid adjustment, illness-onset profile, symptom severity, psychosocial functioning, subjective quality-of-life, and a battery of cognitive tests were conducted. Hierarchical cluster-analysis was employed, optimized with k-means clustering and internally-validated by discriminant-functional analysis. Cognitive subgroup comparisons in illness-related variables, followed by multivariable multinominal-regression analyzes were performed to identify factors independently predictive of cluster membership. Three clusters were identified including patients with globally-impaired (n = 101, 34.9%), intermediately-impaired (n = 112, 38.8%) and relatively-intact (n = 76, 26.3%) cognition (GIC, IIC and RIC subgroups) compared to demographically-matched healthy-controls' performance (n = 50). GIC-subgroup was older, had lower educational attainment, greater positive, negative and disorganization symptom severity, poorer insight and quality-of-life than IIC- and RIC-subgroups, and higher antipsychotic-dose than RIC-subgroup. IIC-subgroup had lower education levels and more severe negative symptoms than RIC-subgroup, which had better psychosocial functioning than two cognitively-impaired subgroups. Educational attainment and disorganization symptoms were found to independently predict cluster membership. Our results affirmed cognitive heterogeneity in FEP and identified three subgroups, which were differentially associated with demographic and illness-related variables. Further research should clarify longitudinal relationships of cognitive subgroups with clinical and functional outcomes in FEP.

Physical activity and sedentary behaviour changes during and after cardiac rehabilitation: Can patients be clustered?

Objective: To identify clusters of patients according to changes in their physical behaviour during and after cardiac rehabilitation, and to predict cluster membership.Methods: The study included 533 patients (mean age 57.9 years; 18.2% females) with a recent acute coronary syndrome who participated in a 12-week multi-disciplinary cardiac rehabilitation programme, within a cohort study design. Physical behaviour (light physical activity, moderate-to vigorous physical activity, step count, and sedentary behaviour) was measured using accelerometry at 4 timepoints. To identify clusters of patients according to changes in physical behaviour during and after cardiac rehabilitation, latent class trajectory modelling was applied. Baseline factors to predict cluster membership were assessed using multinomial logistic regression.Results: During and after cardiac rehabilitation, 3 separate clusters were identified for all 4 physical behaviour outcomes: patients with steady levels (comprising 68–83% of the patients), and improving (6–21%) or deteriorating (4–23%) levels. Main predictor for membership to a specific cluster was baseline physical behaviour. Patients with higher starting physical behaviour were more likely to be a member of clusters with deteriorating levels.Conclusion: Separate clusters of physical behaviour changes during and after cardiac rehabilitation could be identified. Clusters were mainly distinguished by baseline physical behaviour level. LAY ABSTRACTPhysical behaviour is a construct including both physical activity and sedentary behaviour. Healthy levels of physical behaviour are important for cardiac patients. Cardiac rehabilitation programs are designed to promote a heart-healthy lifestyle for this group. Nevertheless, not all patients perform sufficient physical activity after cardiac rehabilitation. It is important to identify patients at risk for disappointing physical behaviour outcomes at an early stage to provide additional care. Outcomes of the current study show that cardiac patients can be clustered according to their change in physical behavior during and after cardiac rehabilitation. The majority showed steady levels and no improvements, but we could also identify groups of patients with improving and deteriorating levels. Patients with higher starting physical activity levels or low sedentary behaviour levels were more likely to be a member of clusters with deteriorating levels. These patients could benefit of additional interventions.

PATTERNS IN THE USE OF DIGITAL TECHNOLOGY AMONG SCHOOL COUNSELLORS IN ROMANIA

Digital technology has become ubiquitous in the practice of school counselling in Romania, but currently there is little evidence and critical reflection on how it is employed in daily activities. The aim of the study was to gather information regarding the practices and resources associated with school counselling, especially regarding the use of digital technology. A total of 528 valid answers were collected from school counsellors across Romania’s 42 administrative divisions, using an online questionnaire. A cluster analysis was used to identify patterns of use of digital technology in professional practices, with three groups emerging: a) all-around use of digital technology, b) less frequent use of digital technology and c) regular use for specific professional tasks. Several socio-demographic factors were used to attempt to predict cluster membership. We discuss the implications of this exploratory study and chart potential avenues of action and research.

P159. Unsupervised machine learning driven stratification of spinal cord injury patients for prediction of long-term outcomes

BACKGROUND CONTEXT Several studies have attempted to predict outcomes of spinal cord injury (SCI) patients, emphasizing the American Spinal Association (ASA) Scale or Functional Independence Measure (FIM). However, no studies to date have attempted to create a unified metric to define patient outcomes that span multiple domains. In this study, we attempt to create a novel machine learning driven classification system of long-term outcomes and attempt to observe natural clusters in combinations of known outcomes. METHODS We utilized the multi-institutional SCI Model Systems Public Access dataset to identify patients. We selected 7 outcomes consistently reported at 1-year and performed dimensionality reduction using t-distributed stochastic network embedding (tSNE) and identified clusters using unsupervised k-mean clustering. We used patient features collected at admission and during the initial hospitalization to predict cluster membership. RESULTS A total of 4,830 patients were identified. Four natural clusters emerged from our analysis of outcomes. Cluster C was associated with improved outcomes in multiple outcome metrics. Cluster A and B were able to differentiate between independence and mobility. We sought to predict membership in Cluster C from data only collected during the initial hospitalization and rehabilitation of the patient. We tested six classification algorithms to predict membership into Cluster C with 5-fold cross-validation. Adaptive Boosted (AdaBoost) clustering proved to have the highest accuracy of 78.7%, and an AUC of 0.74. CONCLUSIONS We demonstrate the capacity of unsupervised machine learning models to identify new patterns to better prognosticate long-term outcomes for SCI patients. We identify membership to a unique cluster of SCI patients associated with significantly better outcomes at 1-year followup. FDA DEVICE/DRUG STATUS This abstract does not discuss or include any applicable devices or drugs. Several studies have attempted to predict outcomes of spinal cord injury (SCI) patients, emphasizing the American Spinal Association (ASA) Scale or Functional Independence Measure (FIM). However, no studies to date have attempted to create a unified metric to define patient outcomes that span multiple domains. In this study, we attempt to create a novel machine learning driven classification system of long-term outcomes and attempt to observe natural clusters in combinations of known outcomes. We utilized the multi-institutional SCI Model Systems Public Access dataset to identify patients. We selected 7 outcomes consistently reported at 1-year and performed dimensionality reduction using t-distributed stochastic network embedding (tSNE) and identified clusters using unsupervised k-mean clustering. We used patient features collected at admission and during the initial hospitalization to predict cluster membership. A total of 4,830 patients were identified. Four natural clusters emerged from our analysis of outcomes. Cluster C was associated with improved outcomes in multiple outcome metrics. Cluster A and B were able to differentiate between independence and mobility. We sought to predict membership in Cluster C from data only collected during the initial hospitalization and rehabilitation of the patient. We tested six classification algorithms to predict membership into Cluster C with 5-fold cross-validation. Adaptive Boosted (AdaBoost) clustering proved to have the highest accuracy of 78.7%, and an AUC of 0.74. We demonstrate the capacity of unsupervised machine learning models to identify new patterns to better prognosticate long-term outcomes for SCI patients. We identify membership to a unique cluster of SCI patients associated with significantly better outcomes at 1-year followup.

Identifying Novel Clusters of Patients With Prolonged Mechanical Ventilation Using Trajectories of Rapid Shallow Breathing Index

ObjectivePatients with prolonged mechanical ventilation (PMV) are comprised of a heterogeneous population, creating great challenges for clinical management and study design. The study aimed to identify subclusters of PMV patients based on trajectories of rapid shallow breathing index (RSBI), and to develop a machine learning model to predict the cluster membership based on baseline variables.MethodsThis was a retrospective cohort study conducted in respiratory care center (RCC) at a tertiary academic medical center. The RCC referral criteria were patients with mechanical ventilation for at least 21 days with stable hemodynamic and oxygenation status. Patients admitted to the RCC from April 2009 to December 2020 were screened. Two-step clustering through linear regression modeling and k-means was employed to find clusters of the trajectories of RSBI. The number of clusters was chosen by statistical metrics and domain expertise. A gradient boosting machine (GBM) was trained, exploiting variables on RCC admission, to predict cluster membership.ResultsA total of 1371 subjects were included in the study. Four clusters were identified: cluster A showed persistently high RSBI; cluster B was characterized by a constant low RSBI over time; Cluster C was characterized by increasing RSBI; and cluster D showed a declining RSBI. Cluster A showed the highest mortality rate (72%), followed by cluster D (63%), C (62%) and B (61%; p = 0.005 for comparison between 4 clusters). GBM was able to predict cluster membership with an accuracy of > 0.95 in ten-fold cross validation. Highly ranked variables for the prediction of clusters included thyroid-stimulating hormone (TSH), cortisol, platelet, free thyroxine (T4) and serum magnesium.ConclusionsPatients with PMV are composed of a heterogeneous population that can be classified into four clusters by using trajectories of RSBI. These clusters can be easily predicted with baseline clinical variables.

Osteoarthritis endotype discovery via clustering of biochemical marker data

ObjectivesOsteoarthritis (OA) patient stratification is an important challenge to design tailored treatments and drive drug development. Biochemical markers reflecting joint tissue turnover were measured in the IMI-APPROACH cohort at baseline...

Caregiver Task Difficulties at the End of Life: A Latent Cluster Analysis

Caregivers engage in myriad tasks from household help to complex medical care. However, little information is available on how caregivers experience individual tasks – particularly key end-of-life tasks such as managing breathing problems or patients' sadness and anxiety. The purpose of this study was therefore to assess task difficulty. Using data from the National Health and Aging Trends Survey and the National Survey on Caregivers (2015-2017), we assessed eleven caregiving tasks in 241 primary caregivers of care recipients in their last month of life. A latent cluster analysis revealed three key clusters: 1) pervasive difficulties, in which caregivers reported difficulty across most or all of the tasks; 2) minimal difficulties; and 3) emotional management difficulties, in which caregivers reported difficulty with managing sadness and anxiety and lower levels of difficulty on the other tasks. Weighted frequency analyses revealed that caregivers in the pervasive difficulties cluster were most likely to be filial caregivers (85% versus 63% of the full sample, p<0.05) or co-residing with the care recipient (49% versus 37% of the full sample, p<0.05). Caregivers identified as having pervasive difficulties were also more likely to report providing intensive care, more than 100 hours per week (54% versus 36% of the full sample, p<0.05). Care recipient condition was not associated with cluster membership. The findings highlight the need to consider caregiver coping at the task-level and have implications for understanding unmet needs. Future research will assess predictors of cluster membership and how task difficulties are associated with symptoms and well-being outcomes.

Substance consumption and intoxication patterns in a medically supervised overdose prevention program for people experiencing homelessness

Background: Opioid overdose is a leading cause of death among homeless individuals. Combining psychoactive substances with opioids increases overdose risk. This study aimed to describe intoxication patterns at a drop-in space offering medical monitoring and harm reduction services to individuals who arrive intoxicated and at risk of overdose. Methods: We examined data from visits to the Supportive Place for Observation and Treatment at Boston Health Care for the Homeless Program between January 1, 2017 and December 31, 2017. We used k-means cluster analysis to characterize intoxication patterns based on clinically assessed sedation levels and vital sign parameters. Multinomial logistic regression analysis assessed demographic and substance consumption predictors of cluster membership. Linear and logistic regression models examined associations between cluster membership and care outcomes. Results: Across 305 care episodes involving 156 unique patients, cluster analysis revealed 3 distinct intoxication patterns. Cluster A (26.6%) had mild sedation and normal vital signs. Cluster B (44.5%) featured greater sedation with bradycardia and/or hypotension. Cluster C (28.9%) was comparable to cluster B but with the addition of hypoxia. Self-reported consumption of non-opioid sedatives prior to arrival was common (63.3% of episodes) and predicted membership in cluster B (aOR 2.75, 95% CI 1.40, 5.40) and cluster C (aOR 3.38, 95% CI 1.48, 7.70). In comparison to cluster A episodes, cluster C episodes were longer (mean 4.8 vs. 2.3 hours, p < 0.001) and more likely to require supplemental oxygen (27.3% vs. 2.5%, p < 0.001). Few episodes required hospital transfer (4.7%) or naloxone (1.0%). No deaths occurred. Conclusions: In a medically supervised overdose monitoring program, reported use of non-opioid sedatives strongly predicted more complex clinical courses and should be factored into overdose prevention efforts. Low-threshold medical monitoring in an ambulatory setting was sufficient for most episodes, suggesting a role for such programs in reducing harm and averting costly emergency services.

A data-driven approach to post-stroke aphasia classification and lesion-based prediction.

Aphasia is an acquired impairment in the production or comprehension of language, typically caused by left hemisphere stroke. The subtyping framework used in clinical aphasiology today is based on the Wernicke-Lichtheim model of aphasia formulated in the late 19th century, which emphasizes the distinction between language production and comprehension. The current study used a data-driven approach that combined modern statistical, machine learning, and neuroimaging tools to examine behavioural deficit profiles and their lesion correlates and predictors in a large cohort of individuals with post-stroke aphasia. First, individuals with aphasia were clustered based on their behavioural deficit profiles using community detection analysis (CDA) and these clusters were compared with the traditional aphasia subtypes. Random forest classifiers were built to evaluate how well individual lesion profiles predict cluster membership. The results of the CDA analyses did not align with the traditional model of aphasia in either behavioural or neuroanatomical patterns. Instead, the results suggested that the primary distinction in aphasia (after severity) is between phonological and semantic processing rather than between production and comprehension. Further, lesion-based classification reached 75% accuracy for the CDA-based categories and only 60% for categories based on the traditional fluent/non-fluent aphasia distinction. The results of this study provide a data-driven basis for a new approach to classification of post-stroke aphasia subtypes in both research and clinical settings.

Disentangling Cognitive Heterogeneity in Psychotic Spectrum Disorders

Neuropsychological impairments represent a central feature of psychosis-spectrum disorders. It is characterized by a great both within- and between-subjects variability (i.e. cognitive heterogeneity), which needs to be better disentangled. The present study aimed to describe the distribution of performance on the Brief Assessment of Cognition in Schizophrenia (BACS) by using the Equivalent Scores, in order to balance statistical methodological problems. To do so, cognitive performance groups were branded, identifying the main factors contributing to cognitive heterogeneity. A sample of 583 patients with a diagnosis of Schizophrenia or Psychotic Disorder Not Otherwise Specified was enrolled and assessed for neurocognition and intellectual level. K-means cluster analysis was performed based on BACS Equivalent Scores. Differences among clusters were analyzed throughout Analysis of Variance and Discriminant Function Analysis in order to identify the most significant predictors of cluster membership. For each cognitive task, roughly 40% of patients displayed poor performance, while up to 63% displayed a symbol-coding deficit. K-means cluster analysis depicted three profiles characterized by "near-normal" cognition, widespread impairment, and "borderline" profile. Discriminant analysis selected Verbal IQ and diagnosis as predictors of cluster membership. Our findings support the usefulness of Equivalent Scores and cluster analysis to explain cognitive heterogeneity, and tailor better interventions.

Newly walking infants' night sleep impacts next day learning and problem solving.

Sleep is part of the process that prepares children and adults for next day cognitive activity. Insufficient or fragmented sleep has a detrimental impact on subsequent encoding (Rouleau et al., 2002) and cognitive functioning (Joo et al., 2012). However, fragmented sleep early in life is a developmental norm, limiting the extent to which conclusions derived from older populations can be generalized. To directly test the continuity of this relationship, newly-walking infants' (N=58) sleep was monitored overnight using actigraphy. The next morning they were taught a motor problem-solving task. The task required infants to navigate through a tunnel to reach a goal at the other end. We coded infants' exploratory behaviors and the extent of training required to solve the task. Using a cluster analysis that accounted for exploratory behaviors and number of training prompts, infants were sorted into three profiles: those who found the task Easy to solve, those who found it Difficult, and those who Never solved it. Wake episodes and sleep efficiency were entered as predictors of cluster membership in a multinomial logistic regression. Of the infants who ultimately solved the task, those with more wake episodes and lower sleep efficiency had more difficulty. Specifically, fragmentation appeared to negatively impact preparedness to learn. Contrary to our expectations, infants who Never solved the task had the least fragmented sleep, indicating that an optimal level of fragmentation is needed for efficient problem-solving. For infants, some level of sleep fragmentation is needed the night before learning in order to solve a task efficiently. These findings highlight the interaction between developmental domains, from sleep quality to motor experience, and their impact on infant learning in real time.

A-092 Behavioral Phenotypes of Childhood Idiopathic Epilepsies

Abstract Objective To characterize the presence and nature of discrete behavioral phenotypes and their correlates in a cohort of youth with new and recent onset focal and generalized epilepsies (YWE). Methods The parents of 297 YWE (age 8–18; n = 183) and typically developing participants (n = 107), completed the Child Behavior Checklist. The 8 behavior problem scales were subjected to hierarchical clustering analytics to identify behavioral subgroups. Factors examined included: demographics, neuropsychological data, history of academic problems, epilepsy characteristics, parental IQ and education, cortical thickness, daily executive functioning, and number of lifetime-to-date DSM-IV diagnoses. Results Hierarchical clustering identified three behavioral phenotypes which included: no behavioral complications (Cluster 1, 67% of YWE), internalizing problems (Cluster 2, 11% of YWE), and combined internalizing and externalizing problems (Cluster 3, 22% of YWE). These behavioral phenotypes were characterized by orderly differences in the factors analyzed. Cluster 1 was most similar to controls across most metrics while Cluster 3 was the most abnormal. Epilepsy syndrome was not a predictor of cluster membership. Conclusion YWE fall into 3 distinct behavioral phenotypes associated with a variety of co-occurring features and comorbidities. This approach identifies important phenotypes of behavior problem presentations and their accompanying factors that serve to advance clinical and theoretical understanding of the behavioral complications of children with epilepsy and the complex conditions with which they co-occur.

Cross-Validation Approach to Evaluate Clustering Algorithms: An Experimental Study Using Multi-Label Datasets

Clustering validation is one of the most important and challenging parts of clustering analysis, as there is no ground truth knowledge to compare the results with. Up till now, the evaluation methods for clustering algorithms have been used for determining the optimal number of clusters in the data, assessing the quality of clustering results through various validity criteria, comparison of results with other clustering schemes, etc. It is also often practically important to build a model on a large amount of training data and then apply the model repeatedly to smaller amounts of new data. This is similar to assigning new data points to existing clusters which are constructed on the training set. However, very little practical guidance is available to measure the prediction strength of the constructed model to predict cluster labels for new samples. In this study, we proposed an extension of the cross-validation procedure to evaluate the quality of the clustering model in predicting cluster membership for new data points. The performance score was measured in terms of the root mean squared error based on the information from multiple labels of the training and testing samples. The principal component analysis (PCA) followed by k-means clustering algorithm was used to evaluate the proposed method. The clustering model was tested using three benchmark multi-label datasets and has shown promising results with overall RMSE of less than 0.075 and MAPE of less than 12.5% in three datasets.

Adoption and Performance of Complementary Clinical Information Technologies: Analysis of a Survey of General Practitioners.

BackgroundThe benefits from the combination of 4 clinical information systems (CISs)—electronic health records (EHRs), health information exchange (HIE), personal health records (PHRs), and telehealth—in primary care depend on the configuration of their functional capabilities available to clinicians. However, our empirical knowledge of these configurations and their associated performance implications is very limited because they have mostly been studied in isolation.ObjectiveThis study aims to pursue 3 objectives: (1) characterize general practitioners (GPs) by uncovering the typical profiles of combinations of 4 major CIS capabilities, (2) identify physician and practice characteristics that predict cluster membership, and (3) assess the variation in the levels of performance associated with each configuration.MethodsWe used data from a survey of GPs conducted throughout the European Union (N=5793). First, 4 factors, that is, EHRs, HIE, PHRs, and Telehealth, were created. Second, a cluster analysis helps uncover clusters of GPs based on the 4 factors. Third, we compared the clusters according to five performance outcomes using an analysis of variance (ANOVA) and a Tamhane T2 post hoc test. Fourth, univariate and multivariate multinomial logistic regressions were used to identify predictors of the clusters. Finally, with a multivariate multinomial logistic regression, among the clusters, we compared performance in terms of the number of patients treated (3 levels) over the last 2 years.ResultsWe unveiled 3 clusters of GPs with different levels of CIS capability profiles: strong (1956/5793, 37.36%), medium (2764/5793, 47.71%), and weak (524/5793, 9.04%). The logistic regression analysis indicates that physicians (younger, female, and less experienced) and practice (solo) characteristics are significantly associated with a weak profile. The ANOVAs revealed a strong cluster associated with significantly high practice performance outcomes in terms of the quality of care, efficiency, productivity, and improvement of working processes, and two noncomprehensive medium and weak profiles associated with medium (equifinal) practice performance outcomes. The logistic regression analysis also revealed that physicians in the weak profile are associated with a decrease in the number of patients treated over the last 2 years.ConclusionsDifferent CIS capability profiles may lead to similar equifinal performance outcomes. This underlines the importance of looking beyond the adoption of 1 CIS capability versus a cluster of capabilities when studying CISs. GPs in the strong cluster exhibit a comprehensive CIS capability profile and outperform the other two clusters with noncomprehensive profiles, leading to significantly high performance in terms of the quality of care provided to patients, efficiency of the practice, productivity of the practice, and improvement of working processes. Our findings indicate that medical practices should develop high capabilities in all 4 CISs if they have to maximize their performance outcomes because efforts to develop high capabilities selectively may only be in vain.