MultiNEP: a multi-omics network enhancement framework for prioritizing disease genes and metabolites simultaneously.

MOTIVATION: Many studies have successfully used network information to prioritize candidate omics profiles associated with diseases. The metabolome, as the link between genotypes and phenotypes, has accumulated growing attention. Using a "multi-omics" network constructed with a gene-gene network, a metabolite-metabolite network, and a gene-metabolite network to simultaneously prioritize candidate disease-associated metabolites and gene expressions could further utilize gene-metabolite interactions that are not used when prioritizing them separately. However, the number of metabolites is usually 100 times fewer than that of genes. Without accounting for this imbalance issue, we cannot effectively use gene-metabolite interactions when simultaneously prioritizing disease-associated metabolites and genes. RESULTS: Here, we developed a Multi-omics Network Enhancement Prioritization (MultiNEP) framework with a weighting scheme to reweight contributions of different sub-networks in a multi-omics network to effectively prioritize candidate disease-associated metabolites and genes simultaneously. In simulation studies, MultiNEP outperforms competing methods that do not address network imbalances and identifies more true signal genes and metabolites simultaneously when we down-weight relative contributions of the gene-gene network and up-weight that of the metabolite-metabolite network to the gene-metabolite network. Applications to two human cancer cohorts show that MultiNEP prioritizes more cancer-related genes by effectively using both within- and between-omics interactions after handling network imbalance. AVAILABILITY AND IMPLEMENTATION: The developed MultiNEP framework is implemented in an R package and available at: https://github.com/Karenxzr/MultiNep.

Diagnosis and Management of Idiopathic Hypothalamic Dysfunction: A Case Series and Systematic Review.

INTRODUCTION: Idiopathic hypothalamic dysfunction (IHD) is a rare syndrome with heterogeneous clinical symptoms. This study aimed to systematically review the clinical features and potential treatment of IHD based on our case series and literature. METHODS: We analysed six recently diagnosed cases of IHD in Peking Union Medical College Hospital and conducted a systematic review of IHD case studies published before August 25, 2021, using the PubMed/Medline database. All 12 articles that met the definition of IHD and provided individual clinical data were reviewed. RESULTS: Of the 19 cases reviewed (13 from the literature and 6 from our centre), the median age at onset was 6 years. Obesity/weight gain (n = 14, 73.7%) and electrolyte abnormalities (n = 14, 73.7%) were the most common hypothalamic physiological dysfunction, followed by autonomic dysregulation (n = 13, 68.4%) and adipsia (n = 13, 68.4%). The most common initial symptom of young patients was obesity/weight gain, whereas the initial symptoms of the three adult patients were hypothalamic amenorrhoea, delayed sexual development, and polydipsia. 11 (57.9%) patients had obesity, and three of our patients were diagnosed with metabolic syndrome in late adolescence or early adulthood. Three of our cases diagnosed with growth hormone deficiency received growth hormone therapy, which exerted positive effects on growth promotion and weight stabilization. CONCLUSION: Although obesity/weight gain was the most common symptom of IHD, uncommon initial symptoms such as electrolyte abnormalities and sexual disorders also require attention, especially in patients with late childhood- or adult-onset IHD. Consistent monitoring of metabolic profiles is recommended. Positive effects of growth hormone replacement therapy on growth and weight were observed, but more extensive cohort studies are required to confirm its efficacy and safety.

Multi-Sensor Fault Diagnosis Based on Time Series in an Intelligent Mechanical System.

Intelligent mechanical systems are a focused area nowadays. One of the requirements of intelligent mechanical systems is to achieve intelligent fault diagnosis through the real-time acquisition and analysis of data from various sensors installed on mechanical components. In this paper, a new fault diagnosis method is proposed to solve the problems of difficulty in integrating the fault diagnosis algorithm and locating fault parts due to the complexity of modern mechanical systems. The complexity of modern industrial intelligent systems is due to the fact that the systems are composed of multiple components and there are various connections between them. Common fault diagnosis is to design specialized fault identification algorithms for the physical characteristics of each component, and the integration of different algorithms is a major challenge for system performance. Therefore, this paper investigates a general algorithm for the fault diagnosis of complex systems using the timing characteristics of sensors and transfer entropy. The fault diagnosis algorithm is based on the prediction of multi-dimensional long time series using Autoformer, and fault identification is performed based on the deviation of the predicted value from the actual value. After fault identification, a root cause analysis method of faults based on transfer entropy is proposed. The method can locate the component where the fault occurs more accurately based on the analysis of the cause-effect relationship of each component and help maintenance personnel to troubleshoot the fault.

Identification of QTNs and their candidate genes for flowering time and plant height in soybean using multi-locus genome-wide association studies.

Flowering time (FT) and plant height (PH) are important agronomic traits in soybean. However, their genetic foundations are not fully understood. Thus, in this study, a total of 106,013 single nucleotide polymorphisms in 286 soybean accessions were used to associate with the first and full FT (FT1 and FT2) and PH in 4 environments and their BLUP values using 6 multi-locus genome-wide association study methods. As a result, 38, 43, and 27 stable quantitative trait nucleotides (QTNs) were identified, respectively, for FT1, FT2, and PH across at least 3 methods and/or environments. Among these QTNs for FT1, FT2, and PH, 31, 36, and 21 were found to have significant phenotype differences across 2 alleles; 22, 18, and 13 were consistent with the corresponding loci in previous studies; 13 and 8 genes, with more than average expression level, around 64 FT and 27 PH QTNs were predicted as their corresponding candidate genes. Among these candidate genes, GmPRR3b, and GmGIa for FT, and GmTFL1b for PH were known, while some were new, e.g., GmPHYA4, GmVRN5, GmFPA, and GmSPA1 for FT, and Glyma.02g300200, GmFPA, and Glyma.13g339800 for PH. All the validated QTNs were used to design the best cross-combinations in 2 FT directions. In each FT direction, the best 5 cross-combinations were predicted, such as Heihe 54 × Qincha 1 for early FT, and Yingdejiadou × Wuhuabayuehuang for late FT. This study provides solid foundations for genetic basis, molecular biology, and breeding by design of soybean FT and PH. Supplementary Information: The online version contains supplementary material available at 10.1007/s11032-021-01230-3.

Machine learning of clinical variables and coronary artery calcium scoring for the prediction of obstructive coronary artery disease on coronary computed tomography angiography: analysis from the CONFIRM registry.

AIMS: Symptom-based pretest probability scores that estimate the likelihood of obstructive coronary artery disease (CAD) in stable chest pain have moderate accuracy. We sought to develop a machine learning (ML) model, utilizing clinical factors and the coronary artery calcium score (CACS), to predict the presence of obstructive CAD on coronary computed tomography angiography (CCTA). METHODS AND RESULTS: The study screened 35 281 participants enrolled in the CONFIRM registry, who underwent ≥64 detector row CCTA evaluation because of either suspected or previously established CAD. A boosted ensemble algorithm (XGBoost) was used, with data split into a training set (80%) on which 10-fold cross-validation was done and a test set (20%). Performance was assessed of the (1) ML model (using 25 clinical and demographic features), (2) ML + CACS, (3) CAD consortium clinical score, (4) CAD consortium clinical score + CACS, and (5) updated Diamond-Forrester (UDF) score. The study population comprised of 13 054 patients, of whom 2380 (18.2%) had obstructive CAD (≥50% stenosis). Machine learning with CACS produced the best performance [area under the curve (AUC) of 0.881] compared with ML alone (AUC of 0.773), CAD consortium clinical score (AUC of 0.734), and with CACS (AUC of 0.866) and UDF (AUC of 0.682), P < 0.05 for all comparisons. CACS, age, and gender were the highest ranking features. CONCLUSION: A ML model incorporating clinical features in addition to CACS can accurately estimate the pretest likelihood of obstructive CAD on CCTA. In clinical practice, the utilization of such an approach could improve risk stratification and help guide downstream management.

Single-Cell Omics for Transcriptome CHaracterization (SCOTCH): isoform-level characterization of gene expression through long-read single-cell RNA sequencing.

The advent of long-read single-cell transcriptome sequencing (lr-scRNA-Seq) represents a significant leap forward in single-cell genomics. With the recent introduction of R10 flowcells by Oxford Nanopore, we propose that previous computational methods designed to handle high sequencing error rates are no longer relevant, and that the prevailing approach using short reads to compile "barcode space" (candidate barcode list) to de-multiplex long reads are no longer necessary. Instead, computational methods should now shift focus on harnessing the unique benefits of long reads to analyze transcriptome complexity. In this context, we introduce a comprehensive suite of computational methods named Single-Cell Omics for Transcriptome CHaracterization (SCOTCH). Our method is compatible with the single-cell library preparation platform from both 10X Genomics and Parse Biosciences, facilitating the analysis of special cell populations, such as neurons, hepatocytes and developing cardiomyocytes. We specifically re-formulated the transcript mapping problem with a compatibility matrix and addressed the multiple-mapping issue using probabilistic inference, which allows the discovery of novel isoforms as well as the detection of differential isoform usage between cell populations. We evaluated SCOTCH through analysis of real data across different combinations of single-cell libraries and sequencing technologies (10X + Illumina, Parse + Illumina, 10X + Nanopore_R9, 10X + Nanopore_R10, Parse + Nanopore_R10), and showed its ability to infer novel biological insights on cell type-specific isoform expression. These datasets enhance the availability of publicly available data for continued development of computational approaches. In summary, SCOTCH allows extraction of more biological insights from the new advancements in single-cell library construction and sequencing technologies, facilitating the examination of transcriptome complexity at the single-cell level.

Oxidized Dextran/Chitosan Hydrogel Engineered with Tetrasulfide-Bridged Silica Nanoparticles for Postsurgical Treatment.

Tumor recurrence and wound microbial infection after tumor excision are serious threats to patients. Thus, the strategy to supply a sufficient and sustained release of cancer drugs and simultaneously engineer antibacterial properties and satisfactory mechanical strength is highly desired for tumor postsurgical treatment. Herein, A novel double-sensitive composite hydrogel embedded with tetrasulfide-bridged mesoporous silica (4S-MSNs) is developed. The incorporation of 4S-MSNs into oxidized dextran/chitosan hydrogel network, not only enhances the mechanical properties of hydrogels, but also can increase the specificity of drug with dual pH/redox sensitivity, thereby allowing more efficient and safer therapy. Besides, 4S-MSNs hydrogel preserves the favorable physicochemical properties of polysaccharide hydrogel, such as high hydrophilicity, satisfactory antibacterial activity, and excellent biocompatibility. Thus, the prepared 4S-MSNs hydrogel can be served as an efficient strategy for postsurgical bacterial infection and inhibition of tumor recurrence.

Semi-Supervised, Attention-Based Deep Learning for Predicting TMPRSS2:ERG Fusion Status in Prostate Cancer Using Whole Slide Images.

IMPLICATIONS: Our study illuminates the potential of deep learning in effectively inferring key prostate cancer genetic alterations from the tissue morphology depicted in routinely available histology slides, offering a cost-effective method that could revolutionize diagnostic strategies in oncology.

PhenoSV: interpretable phenotype-aware model for the prioritization of genes affected by structural variants.

Structural variants (SVs) represent a major source of genetic variation associated with phenotypic diversity and disease susceptibility. While long-read sequencing can discover over 20,000 SVs per human genome, interpreting their functional consequences remains challenging. Existing methods for identifying disease-related SVs focus on deletion/duplication only and cannot prioritize individual genes affected by SVs, especially for noncoding SVs. Here, we introduce PhenoSV, a phenotype-aware machine-learning model that interprets all major types of SVs and genes affected. PhenoSV segments and annotates SVs with diverse genomic features and employs a transformer-based architecture to predict their impacts under a multiple-instance learning framework. With phenotype information, PhenoSV further utilizes gene-phenotype associations to prioritize phenotype-related SVs. Evaluation on extensive human SV datasets covering all SV types demonstrates PhenoSV's superior performance over competing methods. Applications in diseases suggest that PhenoSV can determine disease-related genes from SVs. A web server and a command-line tool for PhenoSV are available at https://phenosv.wglab.org .

Intensity-modulated radiotherapy for cushing's disease: single-center experience in 70 patients.

Context: Intensity-modulated radiotherapy (IMRT) is a modern precision radiotherapy technique for the treatment of the pituitary adenoma. Objective: Aim to investigate the efficacy and toxicity of IMRT in treating Cushing's Disease (CD). Methods: 70 of 115 patients with CD treated with IMRT at our institute from April 2012 to August 2021 were included in the study. The radiation doses were usually 45-50 Gy in 25 fractions. After IMRT, endocrine evaluations were performed every 6 months and magnetic resonance imaging (MRI) annually. Endocrine remission was defined as suppression of 1 mg dexamethasone test (DST) or normal 24-hour urinary free cortisol level (24hUFC). The outcome of endocrine remission, endocrine recurrence, tumor control and complications were retrieved from medical record. Results: At a median follow-up time of 36.8 months, the endocrine remission rate at 1, 2, 3 and 5 years were 28.5%, 50.2%, 62.5% and 74.0%, respectively. The median time to remission was 24 months (95%CI: 14.0-34.0). Endocrine recurrence was found in 5 patients (13.5%) till the last follow-up. The recurrence-free rate at 1, 2, 3 and 5 years after endocrine remission was 98.2%, 93.9%, 88.7% and 88.7%, respectively. The tumor control rate was 98%. The overall incidence of new onset hypopituitarism was 22.9%, with hypothyroidism serving as the most common individual axis deficiency. Univariate analysis indicated that only higher Ki-67 index (P=0.044) was significant favorable factors for endocrine remission. Conclusion: IMRT was a highly effective second-line therapy with low side effect profile for CD patients. Endocrine remission, tumor control and recurrence rates were comparable to previous reports on FRT and SRS.

Machine learning can aid in prediction of IDH mutation from H&E-stained histology slides in infiltrating gliomas.

While Machine Learning (ML) models have been increasingly applied to a range of histopathology tasks, there has been little emphasis on characterizing these models and contrasting them with human experts. We present a detailed empirical analysis comparing expert neuropathologists and ML models at predicting IDH mutation status in H&E-stained histology slides of infiltrating gliomas, both independently and synergistically. We find that errors made by neuropathologists and ML models trained using the TCGA dataset are distinct, representing modest agreement between predictions (human-vs.-human κ = 0.656; human-vs.-ML model κ = 0.598). While no ML model surpassed human performance on an independent institutional test dataset (human AUC = 0.901, max ML AUC = 0.881), a hybrid model aggregating human and ML predictions demonstrates predictive performance comparable to the consensus of two expert neuropathologists (hybrid classifier AUC = 0.921 vs. two-neuropathologist consensus AUC = 0.920). We also show that models trained at different levels of magnification exhibit different types of errors, supporting the value of aggregation across spatial scales in the ML approach. Finally, we present a detailed interpretation of our multi-scale ML ensemble model which reveals that predictions are driven by human-identifiable features at the patch-level.

Metabolic syndrome as a common comorbidity in adults with hypothalamic dysfunction.

Objective: Hypothalamic dysfunction (HD) results in various endocrine disorders and is associated with an increased risk of metabolic comorbidities. This study aimed to analyze the clinical characteristics and metabolic abnormalities of adults with HD of various causes. Methods: This study retrospectively reviewed adults with HD treated at our center between August 1989 and October 2020. Metabolic characteristics of patients were compared to those of age-, sex-matched lean, and body mass index (BMI)-matched controls. Results: Temperature dysregulation (61.0%) was the most common hypothalamic physiological dysfunction. At least one anterior pituitary hormone deficiency was observed in 50 patients (84.7%), with hypogonadotropic hypogonadism being the most frequently observed. Metabolic syndrome was confirmed in 31 patients (52.5%) and was significantly more prevalent in those with panhypopituitarism or overweight/obesity. Metabolic syndrome (MetS) was significantly more common in patients with HD than in both lean and BMI-matched controls (P < 0.001 and P = 0.030, respectively). Considering the components of MetS, elevated fasting glucose levels were significantly more common in patients with HD than in BMI-matched controls (P = 0.029). Overweight/obesity and panhypopituitarism were significant risk factors for MetS in the multivariate analysis on patients with HD. Moreover, in the multivariate analysis on patients and BMI-matched control, HD was a significant risk factor of MetS (P=0.035, OR 2.919) after adjusted for age, sex and BMI. Conclusions: Temperature dysregulation and hypogonadotropic hypogonadism are the most common physiological and endocrine dysfunctions, respectively. MetS and unfavorable metabolic profiles were prevalent in adults with HD. HD was a significant risk factor of MetS after adjusted for BMI.

Maximizing noble metal utilization in solid catalysts by control of nanoparticle location.

Maximizing the utilization of noble metals is crucial for applications such as catalysis. We found that the minimum loading of platinum for optimal performance in the hydroconversion of n-alkanes for industrially relevant bifunctional catalysts could be reduced by a factor of 10 or more through the rational arranging of functional sites at the nanoscale. Intentionally depositing traces of platinum nanoparticles on the alumina binder or the outer surface of zeolite crystals, instead of inside the zeolite crystals, enhanced isomer selectivity without compromising activity. Separation between platinum and zeolite acid sites preserved the metal and acid functions by limiting micropore blockage by metal clusters and enhancing access to metal sites. Reduced platinum nanoparticles were more active than platinum single atoms strongly bonded to the alumina binder.

Mapping theme trends and recognizing research hot spots in the use of ultrasound in orthopaedics: a bibliometric analysis of global research.

BACKGROUND: In the past decade, ultrasound has been increasingly used in the field of orthopaedics. The purpose of this study is to inspire future research in this field by analyzing the publications relating to ultrasound research in orthopaedics. METHODS: All relevant articles published between 2009 and 2020 were retrieved from Web of Science. Statistical Package for Social Science and GraphPad Prism 8 software were used to generate and analyse diagrams. VOSviewer software and CiteSpace were employed to visualize the research trends based on co-occurring keywords. Finally, we obtained information about relevant clinical randomized controlled trials (http://clinicaltrials.gov.com/). RESULTS: The United States had the most publications in this field and the most citations and the highest H-index. Furthermore, Skeletal Radiology published the most papers related to the use of ultrasound in orthopaedics, Ozcakar L published the most papers, and a study by Kwon, YM had the highest citation frequency. The keywords "MRI", "complication", "female" and "male" were identified as being indicative of emerging topics. CONCLUSIONS: While the contribution of United States to publications in this field has been substantial, the future contributions of China cannot be ignored. Moreover, it is hypothesized that diagnostic and epidemiological aspects may become hotspots.

Deep learning predicts chromosomal instability from histopathology images.

Chromosomal instability (CIN) is a hallmark of human cancer yet not readily testable for patients with cancer in routine clinical setting. In this study, we sought to explore whether CIN status can be predicted using ubiquitously available hematoxylin and eosin histology through a deep learning-based model. When applied to a cohort of 1,010 patients with breast cancer (Training set: n = 858, Test set: n = 152) from The Cancer Genome Atlas where 485 patients have high CIN status, our model accurately classified CIN status, achieving an area under the curve of 0.822 with 81.2% sensitivity and 68.7% specificity in the test set. Patch-level predictions of CIN status suggested intra-tumor heterogeneity within slides. Moreover, presence of patches with high predicted CIN score within an entire slide was more predictive of clinical outcome than the average CIN score of the slide, thus underscoring the clinical importance of intra-tumor heterogeneity.

Preventive Effects of Bacillus licheniformis on Heat Stroke in Rats by Sustaining Intestinal Barrier Function and Modulating Gut Microbiota.

Heat stroke (HS) models in rats are associated with severe intestinal injury, which is often considered as the key event at the onset of HS. Probiotics can regulate the gut microbiota by inhibiting the colonization of harmful bacteria and promoting the proliferation of beneficial bacteria. Here, we investigated the preventive effects of a probiotic Bacillus licheniformis strain (BL, CMCC 63516) on HS rats as well as its effects on intestinal barrier function and gut microbiota. All rats were randomly divided into four groups: control (Con) + PBS (pre-administration with 1 ml PBS twice a day for 7 days, without HS induction), Con + BL group (pre-administration with 1 ml 1 × 108 CFU/ml BL twice a day for 7 days, without HS induction), HS + PBS (PBS, with HS induction), and HS + BL (BL, with HS induction). Before the study, the BL strain was identified by genomic DNA analysis. Experimental HS was induced by placing rats in a hot and humid chamber for 60 min until meeting the diagnostic criterion of HS onset. Body weight, core body temperature, survival rate, biochemical markers, inflammatory cytokines, and histopathology were investigated to evaluate the preventive effects of BL on HS. D-Lactate, I-FABP, endotoxin, and tight-junction proteins were investigated, and the fluorescein isothiocyanate-dextran (FD-4) test administered, to assess the degree of intestinal injury and integrity. Gut microbiota of rats in each group were analyzed by 16S rRNA sequencing. The results showed that pre-administration with BL significantly attenuated hyperthermia, reduced HS-induced death, alleviated multiple-organ injury, and decreased the levels of serum inflammatory cytokines. Furthermore, BL sustained the intestinal barrier integrity of HS rats by alleviating intestinal injury and improving tight junctions. We also found that BL significantly increased the ratios of two probiotic bacteria, Lactobacillus and Lactococcus. In addition, Romboutsia, a candidate biomarker for HS diagnosis, was unexpectedly detected. In summary, BL pre-administration for 7 days has preventative effects on HS that may be mediated by sustaining intestinal barrier function and modulating gut microbiota.

Efficient electrochemical production of glucaric acid and H2 via glucose electrolysis.

Glucose electrolysis offers a prospect of value-added glucaric acid synthesis and energy-saving hydrogen production from the biomass-based platform molecules. Here we report that nanostructured NiFe oxide (NiFeOx) and nitride (NiFeNx) catalysts, synthesized from NiFe layered double hydroxide nanosheet arrays on three-dimensional Ni foams, demonstrate a high activity and selectivity towards anodic glucose oxidation. The electrolytic cell assembled with these two catalysts can deliver 100 mA cm-2 at 1.39 V. A faradaic efficiency of 87% and glucaric acid yield of 83% are obtained from the glucose electrolysis, which takes place via a guluronic acid pathway evidenced by in-situ infrared spectroscopy. A rigorous process model combined with a techno-economic analysis shows that the electrochemical reduction of glucose produces glucaric acid at a 54% lower cost than the current chemical approach. This work suggests that glucose electrolysis is an energy-saving and cost-effective approach for H2 production and biomass valorization.

Comparing a novel machine learning method to the Friedewald formula and Martin-Hopkins equation for low-density lipoprotein estimation.

BACKGROUND: Low-density lipoprotein cholesterol (LDL-C) is a target for cardiovascular prevention. Contemporary equations for LDL-C estimation have limited accuracy in certain scenarios (high triglycerides [TG], very low LDL-C). OBJECTIVES: We derived a novel method for LDL-C estimation from the standard lipid profile using a machine learning (ML) approach utilizing random forests (the Weill Cornell model). We compared its correlation to direct LDL-C with the Friedewald and Martin-Hopkins equations for LDL-C estimation. METHODS: The study cohort comprised a convenience sample of standard lipid profile measurements (with the directly measured components of total cholesterol [TC], high-density lipoprotein cholesterol [HDL-C], and TG) as well as chemical-based direct LDL-C performed on the same day at the New York-Presbyterian Hospital/Weill Cornell Medicine (NYP-WCM). Subsequently, an ML algorithm was used to construct a model for LDL-C estimation. Results are reported on the held-out test set, with correlation coefficients and absolute residuals used to assess model performance. RESULTS: Between 2005 and 2019, there were 17,500 lipid profiles performed on 10,936 unique individuals (4,456 females; 40.8%) aged 1 to 103. Correlation coefficients between estimated and measured LDL-C values were 0.982 for the Weill Cornell model, compared to 0.950 for Friedewald and 0.962 for the Martin-Hopkins method. The Weill Cornell model was consistently better across subgroups stratified by LDL-C and TG values, including TG >500 and LDL-C <70. CONCLUSIONS: An ML model was found to have a better correlation with direct LDL-C than either the Friedewald formula or Martin-Hopkins equation, including in the setting of elevated TG and very low LDL-C.

Extraction of radiographic findings from unstructured thoracoabdominal computed tomography reports using convolutional neural network based natural language processing.

BACKGROUND: Heart failure (HF) is a major cause of morbidity and mortality. However, much of the clinical data is unstructured in the form of radiology reports, while the process of data collection and curation is arduous and time-consuming. PURPOSE: We utilized a machine learning (ML)-based natural language processing (NLP) approach to extract clinical terms from unstructured radiology reports. Additionally, we investigate the prognostic value of the extracted data in predicting all-cause mortality (ACM) in HF patients. MATERIALS AND METHODS: This observational cohort study utilized 122,025 thoracoabdominal computed tomography (CT) reports from 11,808 HF patients obtained between 2008 and 2018. 1,560 CT reports were manually annotated for the presence or absence of 14 radiographic findings, in addition to age and gender. Thereafter, a Convolutional Neural Network (CNN) was trained, validated and tested to determine the presence or absence of these features. Further, the ability of CNN to predict ACM was evaluated using Cox regression analysis on the extracted features. RESULTS: 11,808 CT reports were analyzed from 11,808 patients (mean age 72.8 ± 14.8 years; 52.7% (6,217/11,808) male) from whom 3,107 died during the 10.6-year follow-up. The CNN demonstrated excellent accuracy for retrieval of the 14 radiographic findings with area-under-the-curve (AUC) ranging between 0.83-1.00 (F1 score 0.84-0.97). Cox model showed the time-dependent AUC for predicting ACM was 0.747 (95% confidence interval [CI] of 0.704-0.790) at 30 days. CONCLUSION: An ML-based NLP approach to unstructured CT reports demonstrates excellent accuracy for the extraction of predetermined radiographic findings, and provides prognostic value in HF patients.

Utilizing electronic health data and machine learning for the prediction of 30-day unplanned readmission or all-cause mortality in heart failure.

Background: Existing risk assessment tools for heart failure (HF) outcomes use structured databases with static, single-timepoint clinical data and have limited accuracy. Objective: The purpose of this study was to develop a comprehensive approach for accurate prediction of 30-day unplanned readmission and all-cause mortality (ACM) that integrates clinical and physiological data available in the electronic health record system. Methods: Three predictive models for 30-day unplanned readmissions or ACM were created using an extreme gradient boosting approach: (1) index admission model; (2) index discharge model; and (3) feature-aggregated model. Performance was assessed by the area under the curve (AUC) metric and compared with that of the HOSPITAL score, a widely used predictive model for hospital readmission. Results: A total of 3774 patients with a primary billing diagnosis of HF were included (614 experienced the primary outcome), with 796 variables used in the admission and discharge models, and 2032 in the feature-aggregated model. The index admission model had AUC = 0.723, the index discharge model had AUC = 0.754, and the feature-aggregated model had AUC = 0.756 for prediction of 30-day unplanned readmission or ACM. For comparison, the HOSPITAL score had AUC = 0.666 (admission model: P = .093; discharge model: P = .022; feature aggregated: P = .012). Conclusion: These models predict risk of HF hospitalizations and ACM in patients admitted with HF and emphasize the importance of incorporating large numbers of variables in machine learning models to identify predictors for future investigation.