Predicting Urgent Dialysis at Ambulance Transport to the Emergency Department Using Machine Learning Methods.

Hemodialysis patients frequently require ambulance transport to the hospital for dialysis. Some patients require urgent dialysis (UD) within 24 hours of transport to hospital to avoid morbidity and mortality. UD is not available in all hospitals; therefore, predicting patients who need UD prior to hospital transport can help paramedics with destination planning. In this paper, we developed machine learning models for paramedics to predict whether a patient needs UD based on patient characteristics available at the time of ambulance transport. This paper presented a study based on ambulance data collected in Halifax, Canada. Given that relatively few patients need UD, a class imbalance problem is addressed by up-sampling methods and prediction models are developed using multiple machine learning methods. The achieved prediction scores are F1-score=0.76, sensitivity=0.76, and specificity=0.97, confirming that models can predict UD with limited patient characteristics.

Characterizing Cluster-Based Frailty Phenotypes in a Multicenter Prospective Cohort of Kidney Transplant Candidates.

Frailty is associated with a higher risk of death among kidney transplant candidates. Currently available frailty indices are often based on clinical impression, physical exam or an accumulation of deficits across domains of health. In this paper we investigate a clustering based approach that partitions the data based on similarities between individuals to generate phenotypes of kidney transplant candidates. We analyzed a multicenter cohort that included several features typically used to determine an individual's level of frailty. We present a clustering based phenotyping approach, where we investigated two clustering approaches-i.e. neural network based Self-Organizing Maps (SOM) with hierarchical clustering, and KAMILA (KAy-means for MIxed LArge data sets). Our clustering results partition the individuals across 3 distinct clusters. Clusters were used to generate and study feature-level phenotypes of each group.

Ensemble Clustering to Generate Phenotypes of Kidney Transplant Donors and Recipients.

In this paper we investigate the generation of phenotypes for kidney transplant donors and recipients to assist with decision making around organ allocation. We present an ensemble clustering approach for multi-type data (numerical and categorical) using two different clustering approaches-i.e., model based and vector quantization based clustering. These clustering approaches were applied to a large, US national deceased donor kidney transplant recipient database to characterize members of each cluster (in an unsupervised fashion) and to determine whether the subsequent risk of graft failure differed for each cluster. We generated three distinct clusters of recipients, which were subsequently used to generate phenotypes. Each cluster phenotype had recipients with varying clinical features, and the risk of kidney transplant graft failure and mortality differed across clusters. Importantly, the clustering results by both approaches demonstrated a significant overlap. Utilization of two distinct clustering approaches may be a novel way to validate unsupervised clustering techniques and clustering can be used for organ allocation decision making on the basis of differential outcomes.

Digital Therapeutics for COPD Patient Self-Management: Needs Analysis and Design Study.

Timely management of Chronic Obstructive Pulmonary Disease (COPD) exacerbations can improve recovery and reduce the risk of hospitalization. Digital therapeutics are digital interventions, based on best evidence, designed to provide home-based, patient-centered and pervasive self-management support to patients. Digital therapeutics can be effectively used to offer personalized and explainable self-management and behaviour modification resources to patients to reduce the burden of COPD, especially the prevention of acute COPD exacerbations. The functionalities of COPD specific digital therapeutics for self-management need to be grounded in clinical evidence and behavioral theories, in keeping with the self-management needs of COPD patients and their care providers. In this paper, we report the functionalities of a COPD digital therapeutic mobile application based on a needs analysis qualitative study involving both COPD patients and physicians, and, based on the study's finding, we present a knowledge-driven digital therapeutic for COPD self-management.

Spatial Hotspots and Sociodemographic Profiles Associated With Traumatic Brain Injury in Nova Scotia.

Traumatic brain injury (TBI) is a leading cause of death and disability, primarily caused by falls and motor vehicle collisions (MVCs). Although many TBIs are preventable, there is a notable lack of studies exploring the association of geographically defined TBI hotspots with social deprivation. Geographic information systems (GIS) can be used to identify at-risk neighborhoods (hotspots) for targeted interventions. This study aims to determine the spatial distribution of TBI by major causes and to explore the sociodemographic and economic characteristics of TBI hotspots and cold spots in Nova Scotia. Patient data for TBIs from 2003 to 2019 were obtained from the Nova Scotia Trauma Registry. Residential postal codes were geocoded and assigned to dissemination areas (DA). Area-based risk factors and deprivation status (residential instability [RI], economic dependency [ED], ethnocultural composition [EC], and situational vulnerability [SV]) from the national census data were linked to DAs. Spatial autocorrelation was assessed using Moran's I, and hotspot analysis was performed using Getis-Ord Gi* statistic. Differences in risk factors between hot and cold spots were evaluated using the Mann-Whitney U test for numerical variables and the χ2 test or Fisher's exact test for categorical variables. A total of 5394 TBI patients were eligible for inclusion in the study. The distribution of hotspots for falls exhibited no significant difference between urban and rural areas (p = 0.71). Conversely, hotspots related to violence were predominantly urban (p = 0.001), whereas hotspots for MVCs were mostly rural (p < 0.001). Distinct dimensions of deprivation were associated with falls, MVCs, and violent hotspots. Fall hotspots were significantly associated with areas characterized by higher RI (p < 0.001) and greater ethnocultural diversity (p < 0.001). Conversely, the same domains exhibited an inverse relationship with MVC hotspots; areas with low RI and ethnic homogeneity displayed a higher proportion of MVC hotspots. ED and SV exhibited a strong gradient with MVC hotspots; the most deprived quintiles displayed the highest proportion of MVC hotspots compared with cold spots (ED; p = 0.002, SV; p < 0.001). Areas with the highest levels of ethnocultural diversity were found to have a significantly higher proportion of violence-related hotspots than cold spots (p = 0.005). This study offers two significant contributions to spatial epidemiology. First, it demonstrates the distribution of TBI hotspots by major injury causes using the smallest available geographical unit. Second, we disentangle the various pathways through which deprivation impacts the risk of main mechanisms of TBI. These findings provide valuable insights for public health officials to design targeted injury prevention strategies in high-risk areas.

Needs and expectations for artificial intelligence in emergency medicine according to Canadian physicians.

BACKGROUND: Artificial Intelligence (AI) is recognized by emergency physicians (EPs) as an important technology that will affect clinical practice. Several AI-tools have already been developed to aid care delivery in emergency medicine (EM). However, many EM tools appear to have been developed without a cross-disciplinary needs assessment, making it difficult to understand their broader importance to general-practice. Clinician surveys about AI tools have been conducted within other medical specialties to help guide future design. This study aims to understand the needs of Canadian EPs for the apt use of AI-based tools. METHODS: A national cross-sectional, two-stage, mixed-method electronic survey of Canadian EPs was conducted from January-May 2022. The survey includes demographic and physician practice-pattern data, clinicians' current use and perceptions of AI, and individual rankings of which EM work-activities most benefit from AI. RESULTS: The primary outcome is a ranked list of high-priority AI-tools for EM that physicians want translated into general use within the next 10 years. When ranking specific AI examples, 'automated charting/report generation', 'clinical prediction rules' and 'monitoring vitals with early-warning detection' were the top items. When ranking by physician work-activities, 'AI-tools for documentation', 'AI-tools for computer use' and 'AI-tools for triaging patients' were the top items. For secondary outcomes, EPs indicated AI was 'likely' (43.1%) or 'extremely likely' (43.7%) to be able to complete the task of 'documentation' and indicated either 'a-great-deal' (32.8%) or 'quite-a-bit' (39.7%) of potential for AI in EM. Further, EPs were either 'strongly' (48.5%) or 'somewhat' (39.8%) interested in AI for EM. CONCLUSIONS: Physician input on the design of AI is essential to ensure the uptake of this technology. Translation of AI-tools to facilitate documentation is considered a high-priority, and respondents had high confidence that AI could facilitate this task. This study will guide future directions regarding the use of AI for EM and help direct efforts to address prevailing technology-translation barriers such as access to high-quality application-specific data and developing reporting guidelines for specific AI-applications. With a prioritized list of high-need AI applications, decision-makers can develop focused strategies to address these larger obstacles.

Improving Mental Health Literacy and Reducing Psychological Problems Among Teachers in Zambia: Protocol for Implementation and Evaluation of a Wellness4Teachers Email Messaging Program.

BACKGROUND: Primary, basic, secondary, and high school teachers are constantly faced with increased work stressors that can result in psychological health challenges such as burnout, anxiety, and depression, and in some cases, physical health problems. It is presently unknown what the mental health literacy levels are or the prevalence and correlates of psychological issues among teachers in Zambia. It is also unknown if an email mental messaging program (Wellness4Teachers) would effectively reduce burnout and associated psychological problems and improve mental health literacy among teachers. OBJECTIVE: The primary objectives of this study are to determine if daily supportive email messages plus weekly mental health literacy information delivered via email can help improve mental health literacy and reduce the prevalence of moderate to high stress symptoms, burnout, moderate to high anxiety symptoms, moderate to high depression symptoms, and low resilience among school teachers in Zambia. The secondary objectives of this study are to evaluate the baseline prevalence and correlates of moderate to high stress, burnout, moderate to high anxiety, moderate to high depression, and low resilience among school teachers in Zambia. METHODS: This is a quantitative longitudinal and cross-sessional study. Data will be collected at the baseline (the onset of the program), 6 weeks, 3 months, 6 months (the program midpoint), and 12 months (the end point) using web-based surveys. Individual teachers will subscribe by accepting an invitation to do so from the Lusaka Apex Medical University organizational account on the ResilienceNHope web-based application. Data will be analyzed using SPSS version 25 with descriptive and inferential statistics. Outcome measures will be evaluated using standardized rating scales. RESULTS: The Wellness4Teachers email program is expected to improve the participating teachers' mental health literacy and well-being. It is anticipated that the prevalence of stress, burnout, anxiety, depression, and low resilience among teachers in Zambia will be similar to those reported in other jurisdictions. In addition, it is expected that demographic, socioeconomic, and organizational factors, class size, and grade teaching will be associated with burnout and other psychological disorders among teachers, as indicated in the literature. Results are expected 2 years after the program's launch. CONCLUSIONS: The Wellness4Teachers email program will provide essential insight into the prevalence and correlates of psychological problems among teachers in Zambia and the program's impact on subscribers' mental health literacy and well-being. The outcome of this study will help inform policy and decision-making regarding psychological interventions for teachers in Zambia. INTERNATIONAL REGISTERED REPORT IDENTIFIER (IRRID): PRR1-10.2196/44370.

Validation of a flow-cytometry-based red blood cell antigen phenotyping method.

BACKGROUND AND OBJECTIVES: Current manual and automated phenotyping methods are based on visual detection of the antigen-antibody interaction. This approach has several limitations including the use of large volumes of patient and reagent red blood cells (RBCs) and antisera to produce a visually detectable reaction. We sought to determine whether the flow cytometry could be developed and validated to perform RBC phenotyping to enable a high-throughput method of phenotyping using comparatively miniscule reagent volumes via fluorescence-based detection of antibody binding. MATERIALS AND METHODS: RBC phenotyping by flow cytometry was performed using monoclonal direct typing antisera (human IgM): anti-C, -E, -c, -e, -K, -Jka , -Jkb and indirect typing antisera (human IgG): anti-k, -Fya , -Fyb , -S, -s that are commercially available and currently utilized in our blood transfusion services (BTS) for agglutination-based phenotyping assays. RESULTS: Seventy samples were tested using both flow-cytometry-based-phenotyping and a manual tube standard agglutination assay. For all the antigens tested, 100% concordance was achieved. The flow-cytometry-based method used minimal reagent volume (0.5-1 µl per antigen) compared with the volumes required for manual tube standard agglutination (50 µl per antigen) CONCLUSION: This study demonstrates the successful validation of flow-cytometry-based RBC phenotyping. Flow cytometry offers many benefits compared to common conventional RBC phenotyping methods including high degrees of automation, quantitative assessment with automated interpretation of results and extremely low volumes of reagents. This method could be used for high-throughput, low-cost phenotyping for both blood suppliers and hospital BTS.

Towards an Adaptive Clinical Transcription System for In-Situ Transcribing of Patient Encounter Information.

Electronic patient charts are essential for follow-up and multi-disciplinary care, but either take up an exorbitant amount of time during the patient encounter using a key-stroke entry system, or suffer from poor recall when made long after the encounter. Transcribing in-situ, natural dictations by the clinician, recorded during the encounter, with minimal workflow impact, is a promising solution. However, human transcription requires significant manual resources, whereas automated transcription currently lacks the accuracy for specialized clinical language. Our ultimate goal is to automate clinical transcription, particularly for Emergency Departments, with as an end-result a structured SOAP report. Towards this goal, we present the Adaptive Clinical Transcription System (ACTS). We compare the accuracy and processing times of state-of-the-art speech recognition tools, studying the feasibility of streaming-style dynamic transcription and opportunities of incremental learning.

A Knowledge Graph of Mechanistic Associations Between COVID-19, Diabetes Mellitus, and Chronic Kidney Disease.

We present an automated knowledge synthesis and discovery framework to analyze published literature to identify and represent underlying mechanistic associations that aggravate chronic conditions due to COVID-19. Our literature-based discovery approach integrates text mining, knowledge graphs and medical ontologies to discover hidden and previously unknown pathophysiologic relations, dispersed across multiple public literature databases, between COVID-19 and chronic disease mechanisms. We applied our approach to discover mechanistic associations between COVID-19 and chronic conditions-i.e. diabetes mellitus and chronic kidney disease-to understand the long-term impact of COVID-19 on patients with chronic diseases. We found several gene-disease associations that could help identify mechanisms driving poor outcomes for COVID-19 patients with underlying conditions.

Using Interactive Visual Analytics to Optimize Blood Products Inventory at a Blood Bank.

Blood products and their derivatives are perishable commodities that require an efficient inventory management to ensure both a low wastage rate and a high product availability rate. To optimize blood product inventory, Blood Transfusion Services (BTS) need to reduce wastage by avoiding outdates and improving availability of different blood products. We took a blood product lifecycle approach and used advanced visualization techniques to design and develop a highly interactive web-based dashboard to audit retrospective data and consequently, to identify and learn from procedural inefficiencies based on analysis of transactional data. We present pertinent scenarios to show how the blood transfusion staff can use the dashboard to investigate blood product lifecycles so as to probe transition sequence patterns that led to wastage as a means to discover causes of procedural inefficiencies in the BTS.

Applying Machine Learning to Arsenic Species and Metallomics Profiles of Toenails to Evaluate Associations of Environmental Arsenic with Incident Cancer Cases.

Chronic exposure to environmental arsenic has been linked to a number of human diseases affecting multiple organ systems, including cancer. The greatest concern for chronic exposure to arsenic is contaminated groundwater used for drinking as it is the main contributor to the amount of arsenic present in the body. An estimated 40% of households in Nova Scotia (Canada) use water from private wells, and there is a concern that exposure to arsenic may be linked to/associated with cancer. In this preliminary study, we are aiming to gain insights into the association of environmental metal's pathogenicity and carcinogenicity with prostate cancer. We use toenails as a novel biomarker for capturing long-term exposure to arsenic, and have performed toxicological analysis to generate data about differential profiles of arsenic species and the metallome (entirety of metals) for both healthy and individuals with a history cancer. We have applied feature selection and machine learning algorithms to arsenic species and metallomics profiles of toenails to investigate the complex association between environmental arsenic (as a carcinogen) and prostate cancer. We present machine learning based models to ultimately predict the association of environmental arsenic exposure in cancer cases.

Predicting Kidney Graft Survival Using Machine Learning Methods: Prediction Model Development and Feature Significance Analysis Study.

BACKGROUND: Kidney transplantation is the optimal treatment for patients with end-stage renal disease. Short- and long-term kidney graft survival is influenced by a number of donor and recipient factors. Predicting the success of kidney transplantation is important for optimizing kidney allocation. OBJECTIVE: The aim of this study was to predict the risk of kidney graft failure across three temporal cohorts (within 1 year, within 5 years, and after 5 years following a transplant) based on donor and recipient characteristics. We analyzed a large data set comprising over 50,000 kidney transplants covering an approximate 20-year period. METHODS: We applied machine learning-based classification algorithms to develop prediction models for the risk of graft failure for three different temporal cohorts. Deep learning-based autoencoders were applied for data dimensionality reduction, which improved the prediction performance. The influence of features on graft survival for each cohort was studied by investigating a new nonoverlapping patient stratification approach. RESULTS: Our models predicted graft survival with area under the curve scores of 82% within 1 year, 69% within 5 years, and 81% within 17 years. The feature importance analysis elucidated the varying influence of clinical features on graft survival across the three different temporal cohorts. CONCLUSIONS: In this study, we applied machine learning to develop risk prediction models for graft failure that demonstrated a high level of prediction performance. Acknowledging that these models performed better than those reported in the literature for existing risk prediction tools, future studies will focus on how best to incorporate these prediction models into clinical care algorithms to optimize the long-term health of kidney recipients.

Decision support for comorbid conditions via execution-time integration of clinical guidelines using transaction-based semantics and temporal planning.

In case of comorbidity, i.e., multiple medical conditions, Clinical Decision Support Systems (CDSS) should issue recommendations based on all relevant disease-related Clinical Practice Guidelines (CPG). However, treatments from multiple comorbid CPG often interact adversely (e.g., drug-drug interactions) or introduce operational inefficiencies (e.g., redundant scans). A common solution is the a-priori integration of computerized CPG, which involves integration decisions such as discarding, replacing or delaying clinical tasks (e.g., treatments) to avoid adverse interactions or inefficiencies. We argue this insufficiently deals with execution-time events: as the patient's health profile evolves, acute conditions occur, and real-time delays take place, new CPG integration decisions will often be needed, and prior ones may need to be reverted or undone. Any realistic CPG integration effort needs to further consider temporal aspects of clinical tasks-these are not only restricted by temporal constraints from CPGs (e.g., sequential relations, task durations) but also by CPG integration efforts (e.g., avoid treatment overlap). This poses a complex execution-time challenge and makes it difficult to determine an up-to-date, optimal comorbid care plan. We present a solution for dynamic integration of CPG in response to evolving health profiles and execution-time events. CPG integration policies are formulated by clinical experts for coping with comorbidity at execution-time, with clearly defined integration semantics that build on Description and Transaction Logics. A dynamic planning approach reconciles temporal constraints of CPG tasks at execution-time based on their importance, and continuously updates an optimal task schedule.

Staged reflexive artificial intelligence driven testing algorithms for early diagnosis of pituitary disorders.

BACKGROUND: Sellar masses (SM) frequently present with insidious hormonal dysfunction. We previously showed that, by utilizing a combined reflex/reflecting approach involving a laboratory clinician (LC) on common endocrine test results requested by non-specialists, and subsequently adding further warranted tests, previously undiagnosed pituitary disorders can be identified. However, manually employing these strategies by an LC is not feasible for wider screening of pituitary disorders. OBJECTIVE: The aim of this study was to compare the accuracy and financial impact of an Artificial Intelligence (AI) based, fully computerized reflex protocol with manual reflex/reflective intervention protocol led by an LC. METHODS: We developed a proof-of-concept AI-based framework to fully computerize multi-stage reflex testing protocols for pituitary dysfunction using automated reasoning methods. We compared the efficacy of this AI-based protocol with a reflex/reflective protocol based on manually curated retrospective data in identifying pituitary dysfunction based on 12 months of laboratory testing. RESULTS: The AI-based reflex protocol, as compared with the manual protocol, would have identified laboratory tests for add-on that either directly matched or included all manual add-on tests in 92% of cases, and recommended a similar specialist referral in 90% of the cases. The AI-based protocol would have issued 2.8 times the total number of manual add-on laboratory tests at an 85% lower operation cost than the manual protocol when considering marginal test costs, technical staff and specialist salary. CONCLUSION/DISCUSSION: Our AI-based reflex protocol can successfully identify patients with pituitary dysfunction, with lower estimated laboratory cost. Future research will focus on enhancing the protocol's accuracy and incorporating the AI-based reflex protocol into institutional laboratory and hospital information systems for the detection of undiagnosed pituitary disorders.

Building a Knowledge Graph Representing Causal Associations Between Risk Factors and Incidence of Breast Cancer.

This paper explores the use of semantic- and evidence-based biomedical knowledge to build the RiskExplorer knowledge graph that outlines causal associations between risk factors and chronic disease or cancers. The intent of this work is to offer an interactive knowledge synthesis platform to empower health-information-seeking individuals to learn about and mitigate modifiable risk factors. Our approach analyzes biomedical text (from PubMed abstracts), Semantic Medline database, evidence-based semantic associations, literature-based discovery, and graph database to discover associations between risk factors and breast cancer. Our methodological framework involves (a) identifying relevant literature on specified chronic diseases or cancers, (b) extracting semantic associations via knowledge mining tool, (c) building rich semantic graph by transforming semantic associations to nodes and edges, (d) applying frequency-based methods and using semantic edge properties to traverse the graph and identify meaningful multi-node NCD risk paths. Generated multi-node risk paths consist of a source node (representing the source risk factor), one or more intermediate nodes (representing biomedical phenotypes), a target node (representing a chronic disease or cancer), and edges between nodes representing meaningful semantic associations. The results demonstrate that our methodology is capable of generating biomedically valid knowledge related to causal risk and protective factors related to breast cancer.

Ontology-Based Personalized Cognitive Behavioural Plans for Patients with Mild Depression.

Cognitive Behavioural Therapy (CBT) is an action-oriented psychotherapy that combines cognitive and behavioural techniques for psychosocial treatment for depression, and is considered by many to be the golden standard in psychotherapy. More recently, computerized CBT (CCBT) has been deployed to help increase availability and access to this evidence-based therapy. In this vein, a CBT ontology, as a shared common understanding of the domain, can facilitate the aggregation, verification, and operationalization of computerized CBT knowledge. Moreover, as opposed to black-box applications, ontology-enabled systems allow recommended, evidence-based treatment interventions to be traced back to the corresponding psychological concepts. We used a Knowledge Management approach to synthesize and computerize CBT knowledge from multiple sources into a CBT ontology, which allows generating personalized action plans for treating mild depression, using the Web Ontology Language (OWL) and Semantic Web Rule Language (SWRL). We performed a formative evaluation of the CBT ontology in terms of its completeness, consistency, and conciseness.

Analyzing Association Rules for Graft Failure Following Deceased and Live Donor Kidney Transplantation.

This paper investigates the clinical attributes that contribute to kidney graft failure following live and deceased donor transplantation using an association rule mining approach. The generated rules are used to analyze the distinctive co-occurrence of attributes for those with or without all-cause graft failure. Analysis of a kidney transplantation dataset acquired from the Scientific Registry of Transplant Recipients that included over 95000 deceased and live donor recipients over 5-years was performed. Using an association rule mining approach, we were able to confirm established risk factors for graft loss after live and deceased donor transplantation and identify novel combinations of factors that may have implications for clinical care and risk prediction post kidney transplantation. Using lift as the metric to evaluate association rules, our results indicate that advanced recipient age (i.e. over 60 years), end stage kidney disease due to diabetes, the presence of recipient peripheral vascular disease and recipient coronary artery disease have a high likelihood of graft failure within 5 years after transplantation.

Using Interactive Visual Analytics to Optimize in Real-Time Blood Products Inventory at a Blood Bank.

Blood products and their derivatives are perishable commodities that require an efficient inventory management to ensure both a low wastage rate and a high product availability rate. To optimize blood product inventory, blood transfusion services need to reduce wastage by avoiding outdates and improve availability of different blood products. We used advance visualization techniques to design and develop a highly interactive real-time web-based dashboard to monitor the blood product inventory and the on-going blood unit transactions in near-real-time based on analysis of transactional data. Blood transfusion staff use the dashboard to locate units with specific characteristics, investigate the lifecycle of the units, and efficiently transfer units between facilities to minimize outdates.

A Knowledge Graph of Mechanistic Associations Between COVID-19, Diabetes Mellitus and Kidney Diseases.

This paper proposes an automated knowledge synthesis and discovery framework to analyze published literature to identify and represent underlying mechanistic associations that aggravate chronic conditions due to COVID-19. We present a literature-based discovery approach that integrates text mining, knowledge graphs and ontologies to discover semantic associations between COVID-19 and chronic disease concepts that were represented as a complex disease knowledge network that can be queried to extract plausible mechanisms by which COVID-19 may be exacerbated by underlying chronic conditions.