Readily accessible risk model to predict in-hospital major adverse cardiac events in patients with acute myocardial infarction: a retrospective study of Chinese patients.

OBJECTIVE: Rapid, accurate identification of patients with acute myocardial infarction (AMI) at high risk of in-hospital major adverse cardiac events (MACE) is critical for risk stratification and prompt management. This study aimed to develop a simple, accessible tool for predicting in-hospital MACE in Chinese patients with AMI. DESIGN: Retrospective review of deidentified medical records. SETTING: 38 urban and rural hospitals across diverse economic and geographic areas in China (Beijing, Henan Province and Jilin Province). PARTICIPANTS: 15 009 patients discharged from hospital with a diagnosis of AMI. MAIN OUTCOME MEASURE: The primary outcome was MACE occurrence during index hospitalisation. A multivariate logistic regression model (China AMI Risk Model, CHARM) derived using patient data from Beijing (n=7329) and validated with data from Henan (n=4247) and Jilin (n=3433) was constructed to predict the primary outcome using variables of age, white cell count (WCC) and Killip class. C-statistics evaluated discrimination in the derivation and validation cohorts, with goodness-of-fit assessed using Hosmer-Lemeshow statistics. RESULTS: The CHARM model included age (OR: 1.06 per 1-year increment, 95% CI 1.05 to 1.07, p<0.001), WCC (OR per 109/L increment: 1.10 (95% CI 1.07 to 1.13), p<0.001) and Killip class (class II vs class I: OR 1.34 (95% CI 0.99 to 1.83), p=0.06; class III vs class I: OR 2.74 (95% CI 1.86 to 3.97), p<0.001; class IV vs class I: OR 14.12 (95% CI 10.35 to 19.29), p<0.001). C-statistics were similar between the derivation and validation datasets. CHARM had a higher true positive rate than the Thrombolysis In Myocardial Infarction score and similar to the Global Registry of Acute Coronary Events (GRACE). Hosmer-Lemeshow statistics were 5.5 (p=0.703) for derivation, 41.1 (p<0.001) for Henan, and 103.2 for Jilin (p<0.001) validation sets with CHARM, compared with 119.6, 34.0 and 459.1 with GRACE (all p<0.001). CONCLUSIONS: The CHARM model provides an inexpensive, accurate and readily accessible tool for predicting in-hospital MACE in Chinese patients with AMI.

A Word Graph-based Method for Disease Topic Identification in Biomedical Literature.

An important task in biomedical literature precise search is to identify paper describing a certain disease. The tradi- tional topic identification approaches based on neural network can be used to recognize the disease topic of literature. To achieve better performance, we propose a novel word graph-based method for disease topic identification in this paper. Word graphs are constructed from literature title and abstract. Graph features are extracted and used for disease topic classification using a logistic regression or random forest classifier. Experiment results showed the word graph features outperformed disease mention frequency by a large margin. Our approach achieved better perfor- mance in identifying disease topic compared to hierarchical attention networks, which is a deep learning approach for document classification. We also demonstrated the use of the proposed method in identifying the disease topic in an application scenario.

Identification and external validation of IgA nephropathy patients benefiting from immunosuppression therapy.

BACKGROUND: Although IgA nephropathy (IgAN), an immune-mediated disease with heterogeneous clinical and pathological phenotypes, is the most common glomerulonephritis worldwide, it remains unclear which IgAN patients benefit from immunosuppression (IS) therapy. METHODS: Clinical and pathological data from 4047 biopsy-proven IgAN patients from 24 renal centres in China were included. The derivation and validation cohorts were composed of 2058 and 1989 patients, respectively. Model-based recursive partitioning, a machine learning approach, was performed to partition patients in the derivation cohort into subgroups with different IS long-term benefits, associated with time to end-stage kidney disease, measured by adjusted Kaplan-Meier estimator and adjusted hazard ratio (HR) using Cox regression. FINDINGS: Three identified subgroups obtained a significant IS benefits with HRs ≤ 1. In patients with serum creatinine ≤ 1·437 mg/dl, the benefits of IS were observed in those with proteinuria > 1·525 g/24h (node 6; HR = 0·50; 95% CI, 0·29 to 0·89; P = 0·02), especially in those with proteinuria > 2·480 g/24h (node 8; HR =  0·23; 95% CI, 0·11 to 0·50; P <0·001). In patients with serum creatinine > 1·437 mg/dl, those with high proteinuria and crescents benefitted from IS (node 12; HR = 0·29; 95% CI, 0·09 to 0·94; P = 0·04). The treatment benefits were externally validated in the validation cohort. INTERPRETATION: Machine learning could be employed to identify subgroups with different IS benefits. These efforts promote decision-making, assist targeted clinical trial design, and shed light on individualised treatment in IgAN patients. FUNDING: National Key Research and Development Program of China (2016YFC0904103), National Key Technology R&D Program (2015BAI12B02).

Decentralized Privacy-Preserving Platform for Clinical Data Sharing and Analysis.

Collection and management of clinical data for administration and analysis is a time-consuming and complex task, especially when multiple data providers been involved. Even if people are willing to take on the burden for it, there is still no mature solution to protect data privacy for distributed data providers. Distributed ledger is an emerging technology that supports decentralized data sharing and management. Based on this, we present a platform which enables distributed and truthful data collection and serves privacy-preserving needs in clinical data management. Our system, built on Hyperledger Fabric, used smart contract to execute data aggregation and provide basic analysis methods. The system used ledger and world status to record data access history and other metadata. This decentralized platform enables data providers to proactively share and protect their data, Thus can simplify clinical data collection procedure and promote efficient collaboration between providers.

A Method to Accelerate and Visualize Iterative Clinical Paper Searching.

Clinical paper searching is a major task for clinical researchers to collect authoritative and up-to-date evidences to support their research works and clinical practices. Currently, this task needs huge amount of labor work. Researchers usually spend a lot of time searching on the online repository and iterating many times to get the final paper list. Systematic review is a special case, in which the paper searching process is a critical step. To address this challenge, this paper introduces a method to streamline the iterative paper searching process. It automatically selects the most probably matched papers, and then generates new search strategy. All the intermediate results are visualized based on the paper citation graph. It assembles technologies such as PageRank and Topic-based clustering to accelerate the paper searching tasks. The precision, recall, and execution time of the proposed method are then evaluated by comparing with published systematic reviews.

Facilitating Clinical Trial Recruitment by Recommending Cost-Efficient Medical Exams.

Clinical trials are key and essential processes for researchers to develop new treatments as well as evaluate their effectiveness and safety, whilst more than half of all clinical trials experience delays, which leads to a considerable amount of cost. In this paper, we present a cost-effective framework to reduce the time and monetary cost in the stage of recruiting and screening eligible clinical trial participants. By leveraging patients' observed conditions and the cost of medical examinations, the proposed framework uses collaborative filtering techniques to predict the utilized cost for the to-do medical examinations and then rank patients and medical examinations. The preliminary experiment results indicate that the framework is promising to reduce the cost spent on medical examinations by three quarters or even more and accelerate the recruitment process in the screening stage.

Network-Based Prediction of Major Adverse Cardiac Events in Acute Coronary Syndromes from Imbalanced EMR Data.

The low proportion and the rapid evolvement of major adverse cardiac events (MACE) present challenges for predicting MACE by machine learning models. In this paper, we propose a method to predict MACE from large-scale imbalanced EMR data by using a network-based one-class classifier. It only used the reliably known MACE samples to establish the hyperspherical model. Experiments show that our model outperforms the state-of-the-art models.

Knowledge Learning Symbiosis for Developing Risk Prediction Models from Regional EHR Repositories.

Secondary use of regional EHR data suffers several problems, including data selection bias and limited data size caused by data incompleteness. Here, we propose knowledge learning symbiosis (KLS) as a framework to incorporate domain knowledge to address the problems and make better secondary use of EHR data. Under the framework, we introduce three main categories of methods: knowledge injection to input features, objective functions, and output labels, where knowledge-enhanced neural network (KENN) was first introduced to inject knowledge into objective functions. A case study was conducted to build a cardiovascular disease risk prediction model on the type 2 diabetes patient cohort using regional EHR repositories. By incorporating a well-established knowledge risk model as domain knowledge under our KLS framework, we increased risk prediction performance both on small and biased data, where KENN showed the best performance among all methods.

Outcome-Driven Clustering of Acute Coronary Syndrome Patients Using Multi-Task Neural Network with Attention.

Cluster analysis aims at separating patients into phenotypically heterogenous groups and defining therapeutically homogeneous patient subclasses. It is an important approach in data-driven disease classification and subtyping. Acute coronary syndrome (ACS) is a syndrome due to sudden decrease of coronary artery blood flow, where disease classification would help to inform therapeutic strategies and provide prognostic insights. Here we conducted an outcome-driven cluster analysis of ACS patients, which jointly considers treatment and patient outcome as indicators for patient state. Multi-task neural network with attention was used as a modeling framework, including learning of the patient state, cluster analysis, and feature importance profiling. Seven patient clusters were discovered. The clusters have different characteristics, as well as different risk profiles to the outcome of in-hospital major adverse cardiac events. The results demonstrate cluster analysis using outcome-driven multi-task neural network as promising for patient classification and subtyping.

Prediction and Risk Stratification of Kidney Outcomes in IgA Nephropathy.

RATIONALE & OBJECTIVE: Immunoglobulin A nephropathy (IgAN) is common worldwide and has heterogeneous phenotypes. Predicting long-term outcomes and stratifying risk are important for clinical decision making and designing future clinical trials. STUDY DESIGN: Multicenter retrospective cohort study of 2,047 patients with IgAN. SETTING & PARTICIPANTS: Derivation and validation cohorts composed of 1,022 Chinese patients with IgAN from a single center and 1,025 patients with IgAN from 18 renal centers, respectively. PREDICTORS: 36 characteristics, including demographic, clinical, and pathologic variables. OUTCOMES: Combined event of end-stage kidney disease or 50% reduction in estimated glomerular filtration rate within 5 years after diagnostic kidney biopsy. ANALYTICAL APPROACH: A gradient tree boosting method implemented in the eXtreme Gradient Boosting (XGBoost) system was used to select the 10 most important variables from 36 candidate variables. Stepwise Cox regression analysis was used to derive a simplified scoring scale model (SSM) based on these 10 variables. Model discrimination and calibration were assessed using the C statistic and Hosmer-Lemeshow test. Risk stratification of the SSM was evaluated using Kaplan-Meier analysis. RESULTS: In the derivation and validation cohorts, 74 and 114 patients reached the outcome, respectively. XGBoost predicted the outcome with a C statistic of 0.84 (95% CI, 0.80-0.88) for the validation cohort. The SSM included 3 variables: urine protein excretion, global sclerosis, and tubular atrophy/interstitial fibrosis. Using Kaplan-Meier analysis, the SSM identified significant risk stratification (P < 0.001). LIMITATIONS: Retrospective study design, application for other ethnic groups needs to be verified. CONCLUSIONS: A prediction model using routinely available characteristics and based on the combination of a machine learning algorithm and survival analysis can stratify risk for kidney disease progression in the setting of IgAN. An online calculator, the Nanjing IgAN Risk Stratification System, permits easy implementation of this model.

Using Machine Learning Approaches for Emergency Room Visit Prediction Based on Electronic Health Record Data.

Emergency room(ER) visit prediction, especially whether visit ER or not and ER visit count, is crucial for hospitals to reasonably adapt resource allocation and` for patients to know future health state. Some existing studies have explored to use machine learning methods especially kinds of general linear model to settle down the task. But, in the clinical problems, there exist complex correlation between targets and features. Generally, liner model is difficult to model complex correlation to make better prediction. Hence, in this paper, we propose to use two non-linear models to settle the problem, which are XGBoost and Recurrent Neural Network. Experimental results show both methods have better performance.

Comparative whole-genome sequence analysis of Mycobacterium tuberculosis isolated from tuberculous meningitis and pulmonary tuberculosis patients.

Tuberculous meningitis (TBM) is a severe form of tuberculosis with a high mortality rate. The factors associated with TBM pathogenesis are still unclear. Using comparative whole-genome sequence analysis we compared Mycobacterium tuberculosis (Mtb) isolates from cerebrospinal fluid of TBM cases (n = 73) with those from sputum of pulmonary tuberculosis (PulTB) patients (n = 220) from Thailand. The aim of this study was to seek genetic variants of Mtb associated with TBM. Regardless of Mtb lineage, we found 242 variants that were common to all TBM isolates. Among these variants, 28 were missense SNPs occurring mainly in the pks genes (involving polyketide synthesis) and the PE/PPE gene. Six lineage-independent SNPs were commonly found in TBM isolates, two of which were missense SNPs in Rv0532 (PE_PGRS6). Structural variant analysis revealed that PulTB isolates had 14 genomic regions containing 2-3-fold greater read depth, indicating higher copy number variants and half of these genes belonged to the PE/PPE gene family. Phylogenetic analysis revealed only two small clusters of TBM clonal isolates without support from epidemiological data. This study reported genetic variants of Mtb commonly found in TBM patients compared to PulTB patients. Variants associated with TBM disease warrant further investigation.

Mining Disease-Symptom Relation from Massive Biomedical Literature and Its Application in Severe Disease Diagnosis.

Disease-symptom relation is an important biomedical relation that can be used for clinical decision support including building medical diagnostic systems. Here we present a study on mining disease-symptom relation from massive biomedical literature and constructing biomedical knowledge graph from the relation. From 15,970,134 MEDLINE/PubMed citation records, occurrences of 8,514 disease concepts from the Human Disease Ontology and 842 symptom concepts from the Symptom Ontology and their relation were analyzed and characterized. We improve previous disease-symptom relation mining work by: (1) leveraging the hierarchy information of concepts in medical entity association discovery; and (2) including more exquisite relationship with weights between entities for knowledge graph construction. A medical diagnostic system for severe disease diagnosis was implemented based on the constructed knowledge graph and achieved the best performance compared to all other methods.

Learning Doctors' Medicine Prescription Pattern for Chronic Disease Treatment by Mining Electronic Health Records: A Multi-Task Learning Approach.

Increasing learning ability from massive medical data and building learning methods robust to data quality issues are key factors toward building data-driven clinical decision support systems for medicine prescription decision support. Here, we attempted accordingly to address the factors using a multi-task neural network approach, benefiting from multi-task learning's advantage in modeling commonalities to increase learning performance and neural network's robustness to imprecise data. By mining electronic health record data, we learned medicine prescription patterns of multiple correlated antidiabetic agents in blood glucose control and antihypertensive drugs in blood pressure control scenarios. We achieved AUC increases of 0.02 to 0.06 in single drug prescription and an accuracy increase of 0.05 in prescription pattern prediction compared to logistic regression, demonstrating the efficacy of multi-task neural network approach in learning medicine prescription patterns.

Applying Risk Models on Patients with Unknown Predictor Values: An Incremental Learning Approach.

In clinical practice, many patients may have unknown or missing values for some predictors, causing that the developed risk models cannot be directly applied on these patients. In this paper, we propose an incremental learning approach to apply a developed risk model on new patients with unknown predictor values, which imputes a patient's unknown values based on his/her k-nearest neighbors (k-NN) from the incremental population. We perform a real world case study by developing a risk prediction model of stroke for patients with Type 2 diabetes mellitus from EHR data, and incrementally applying the risk model on a sequence of new patients. The experimental results show that our risk prediction model of stroke has good prediction performance. And the k-nearest neighbors based incremental learning approach for data imputation can gradually increase the prediction performance when the model is applied on new patients.

Gathering Real World Evidence with Cluster Analysis for Clinical Decision Support.

Clinical decision support systems are information technology systems that assist clinical decision-making tasks, which have been shown to enhance clinical performance. Cluster analysis, which groups similar patients together, aims to separate patient cases into phenotypically heterogenous groups and defining therapeutically homogeneous patient subclasses. Useful as it is, the application of cluster analysis in clinical decision support systems is less reported. Here, we describe the usage of cluster analysis in clinical decision support systems, by first dividing patient cases into similar groups and then providing diagnosis or treatment suggestions based on the group profiles. This integration provides data for clinical decisions and compiles a wide range of clinical practices to inform the performance of individual clinicians. We also include an example usage of the system under the scenario of blood lipid management in type 2 diabetes. These efforts represent a step toward promoting patient-centered care and enabling precision medicine.

In silico region of difference (RD) analysis of Mycobacterium tuberculosis complex from sequence reads using RD-Analyzer.

BACKGROUND: Whole-genome sequencing is increasingly used in clinical diagnosis of tuberculosis and study of Mycobacterium tuberculosis complex (MTC). MTC consists of several genetically homogenous mycobacteria species which can cause tuberculosis in humans and animals. Regions of difference (RDs) are commonly regarded as gold standard genetic markers for MTC classification. RESULTS: We develop RD-Analyzer, a tool that can accurately infer the species and lineage of MTC isolates from sequence reads based on the presence and absence of a set of 31 RDs. Applied on a publicly available diverse set of 377 sequenced MTC isolates from known major species and lineages, RD-Analyzer achieved an accuracy of 98.14 % (370/377) in species prediction and a concordance of 98.47 % (257/261) in Mycobacterium tuberculosis lineage prediction compared to predictions based on single nucleotide polymorphism markers. By comparing respective sequencing read depths on each genomic position between isolates of different sublineages, we were able to identify the known RD markers in different sublineages of Lineage 4 and provide support for six potential delineating markers having high sensitivities and specificities for sublineage prediction. An extended version of RD-Analyzer was thus developed to allow user-defined RDs for lineage prediction. CONCLUSIONS: RD-Analyzer is a useful and accurate tool for species, lineage and sublineage prediction using known RDs of MTC from sequence reads and is extendable to accepting user-defined RDs for analysis. RD-Analyzer is written in Python and is freely available at https://github.com/xiaeryu/RD-Analyzer .

Tracking inter-institutional spread of NDM and identification of a novel NDM-positive plasmid, pSg1-NDM, using next-generation sequencing approaches.

OBJECTIVES: Owing to gene transposition and plasmid conjugation, New Delhi metallo-ß-lactamase (NDM) is typically identified among varied Enterobacteriaceae species and STs. We used WGS to characterize the chromosomal and plasmid molecular epidemiology of NDM transmission involving four institutions in Singapore. METHODS: Thirty-three Enterobacteriaceae isolates (collection years 2010-14) were sequenced using short-read sequencing-by-synthesis and analysed. Long-read single molecule, real-time sequencing (SMRTS) was used to characterize genetically a novel plasmid pSg1-NDM carried on Klebsiella pneumoniae ST147. RESULTS: In 20 (61%) isolates, blaNDM was located on the pNDM-ECS01 plasmid in the background of multiple bacterial STs, including eight K. pneumoniae STs and five Escherichia coli STs. In six (18%) isolates, a novel blaNDM-positive plasmid, pSg1-NDM, was found only in K. pneumoniae ST147. The pSg1-NDM-K. pneumoniae ST147 clone (Sg1-NDM) was fully sequenced using SMRTS. pSg1-NDM, a 90 103 bp IncR plasmid, carried genes responsible for resistance to six classes of antimicrobials. A large portion of pSg1-NDM had no significant homology to any known plasmids in GenBank. pSg1-NDM had no conjugative transfer region. Combined chromosomal-plasmid phylogenetic analysis revealed five clusters of clonal bacterial NDM-positive plasmid transmission, of which two were inter-institution clusters. The largest inter-institution cluster involved six K. pneumoniae ST147-pSg1-NDM isolates. Fifteen patients were involved in transmission clusters, of which four had ward contact, six had hospital contact and five had an unknown transmission link. CONCLUSIONS: A combined sequencing-by-synthesis and SMRTS approach can determine effectively the transmission clusters of blaNDM and genetically characterize novel plasmids. Plasmid molecular epidemiology is important to understanding NDM spread as blaNDM-positive plasmids can conjugate extensively across species and STs.

Whole-Genome Sequencing Analysis of Serially Isolated Multi-Drug and Extensively Drug Resistant Mycobacterium tuberculosis from Thai Patients.

Multi-drug and extensively drug-resistant tuberculosis (MDR and XDR-TB) are problems that threaten public health worldwide. Only some genetic markers associated with drug-resistant TB are known. Whole-genome sequencing (WGS) is a promising tool for distinguishing between re-infection and persistent infection in isolates taken at different times from a single patient, but has not yet been applied in MDR and XDR-TB. We aim to detect genetic markers associated with drug resistance and distinguish between reinfection and persistent infection from MDR and XDR-TB patients based on WGS analysis. Samples of Mycobacterium tuberculosis (n = 7), serially isolated from 2 MDR cases and 1 XDR-TB case, were retrieved from Siriraj Hospital, Bangkok. The WGS analysis used an Illumina Miseq sequencer. In cases of persistent infection, MDR-TB isolates differed at an average of 2 SNPs across the span of 2-9 months whereas in the case of reinfection, isolates differed at 61 SNPs across 2 years. Known genetic markers associated with resistance were detected from strains susceptible to streptomycin (2/7 isolates), p-aminosalicylic acid (3/7 isolates) and fluoroquinolone drugs. Among fluoroquinolone drugs, ofloxacin had the highest phenotype-genotype concordance (6/7 isolates), whereas gatifloxcain had the lowest (3/7 isolates). A putative candidate SNP in Rv2477c associated with kanamycin and amikacin resistance was suggested for further validation. WGS provided comprehensive results regarding molecular epidemiology, distinguishing between persistent infection and reinfection in M/XDR-TB and potentially can be used for detection of novel mutations associated with drug resistance.