Abstract: Clinical trials are essential for drug development but often suffer from expensive, inaccurate and insufficient patient recruitment. The core problem of patient-trial matching is to find qualified patients for a trial, where patient information is stored in electronic health records (EHR) while trial eligibility criteria (EC) are described in text documents available on the web. How to represent longitudinal patient EHR? How to extract complex logical rules from EC? Most existing works rely on manual rule-based extraction, which is time consuming and inflexible for complex inference. To address these challenges, we proposed DeepEnroll, a cross-modal inference learning model to jointly encode enrollment criteria (text) and patients records (tabular data) into a shared latent space for matching inference. DeepEnroll applies a pre-trained Bidirectional Encoder Representations from Transformers(BERT) model to encode clinical trial information into sentence embedding. And uses a hierarchical embedding model to represent patient longitudinal EHR. In addition, DeepEnroll is augmented by a numerical information embedding and entailment module to reason over numerical information in both EC and EHR. These encoders are trained jointly to optimize patient-trial matching score. We evaluated DeepEnroll on the trial-patient matching task with demonstrated on real world datasets. DeepEnroll outperformed the best baseline by up to 12.4% in average F1.

Abstract: Deep learning has demonstrated success in health risk prediction especially for patients with chronic and progressing conditions. Most existing works focus on learning disease patterns from longitudinal patient data, but pay little attention to the disease progression stage itself. To fill the gap, we propose a Stage-aware neural Network (StageNet) model to extract disease stage information from patient data and integrate it into risk prediction. StageNet is enabled by (1) a stage-aware long short-term memory (LSTM) module that extracts health stage variations unsupervisedly; (2) a stage-adaptive convolutional module that incorporates stage-related progression patterns into risk prediction. We evaluate StageNet on two real-world datasets and show that StageNet outperforms state-of-the-art models in risk prediction task and patient subtyping task. Compared to the best baseline model, StageNet achieves up to 12% higher AUPRC for risk prediction task on two real-world patient datasets. StageNet also achieves over 58% higher Calinski-Harabasz score (a cluster quality metric) for a patient subtyping task.

Abstract: Molecule optimization is a fundamental task for accelerating drug discovery, with the goal of generating new valid molecules that maximize multiple drug properties while maintaining similarity to the input molecule. Existing generative models and reinforcement learning approaches made initial success, but still face difficulties in simultaneously optimizing multiple drug properties. To address such challenges, we propose the MultI-constraint MOlecule SAmpling (MIMOSA) approach, a sampling framework to use input molecule as an initial guess and sample molecules from the target distribution. MIMOSA first pretrains two property agnostic graph neural networks (GNNs) for molecule topology and substructure-type prediction, where a substructure can be either atom or single ring. For each iteration, MIMOSA uses the GNNs' prediction and employs three basic substructure operations (add, replace, delete) to generate new molecules and associated weights. The weights can encode multiple constraints including similarity and drug property constraints, upon which we select promising molecules for next iteration. MIMOSA enables flexible encoding of multiple property- and similarity-constraints and can efficiently generate new molecules that satisfy various property constraints and achieved up to 49.6% relative improvement over the best baseline in terms of success rate.

Abstract: Clinical trials are essential for drug development but often suffer from expensive, inaccurate and insufficient patient recruitment. The core problem of patient-trial matching is to find qualified patients for a trial, where patient information is stored in electronic health records (EHR) while trial eligibility criteria (EC) are described in text documents available on the web. How to represent longitudinal patient EHR? How to extract complex logical rules from EC? Most existing works rely on manual rule-based extraction, which is time consuming and inflexible for complex inference. To address these challenges, we proposed a cross-modal inference learning model to jointly encode enrollment criteria (text) and patients records (tabular data) into a shared latent space for matching inference. pplies a pre-trained Bidirectional Encoder Representations from Transformers(BERT) model to encode clinical trial information into sentence embedding. And uses a hierarchical embedding model to represent patient longitudinal EHR. In addition, s augmented by a numerical information embedding and entailment module to reason over numerical information in both EC and EHR. These encoders are trained jointly to optimize patient-trial matching score. We evaluated n the trial-patient matching task with demonstrated on real world datasets. utperformed the best baseline by up to 12.4% in average F1.

Abstract: Molecule optimization is about generating molecule Y with more desirable properties based on an input molecule X. The state-of-the-art approaches partition the molecules into a large set of substructures S and grow the new molecule structure by iteratively predicting which substructure from S to add. However, since the set of available substructures S is large, such an iterative prediction task is often inaccurate especially for substructures that are infrequent in the training data. To address this challenge, we propose a new generating strategy called “Copy&Refine” (CORE), where at each step the generator first decides whether to copy an existing substructure from input X or to generate a new substructure, then the most promising substructure will be added to the new molecule. Combining together with scaffolding tree generation and adversarial training, CORE can significantly improve several latest molecule optimization methods in various measures including drug likeness (QED), dopamine receptor (DRD2) and penalized LogP. We tested CORE and baselines using the ZINC database and CORE obtained up to 11% and 21% relatively improvement over the baselines on success rate on the complete test set and the subset with infrequent substructures, respectively.

Abstract: CONTEXT Intermediate-term glycemic control metrics fulfill a need for measures beyond hemoglobin A1C. OBJECTIVE Compare glycated albumin (GA), a 14-day blood glucose measure, with other glycemic indices. DESIGN 24-week prospective study of assay performance. SETTING 8 US clinics. PARTICIPANTS Subjects with type 1 (n = 73) and type 2 diabetes (n = 77) undergoing changes to improve glycemic control (n = 98) or with stable diabetes therapy (n = 52). INTERVENTIONS GA, fructosamine, and A1C measured at prespecified intervals. Mean blood glucose (MBG) calculated using weekly self-monitored blood glucose profiles. MAIN OUTCOME MEASURES Primary: Pearson correlation between GA and fructosamine. Secondary: magnitude (Spearman correlation) and direction (Kendall correlation) of change of glycemic indices in the first 3 months after a change in diabetes management. RESULTS GA was more concordant (60.8%) with changes in MBG than fructosamine (55.5%) or A1C (45.5%). Across all subjects and visits, the GA Pearson correlation with fructosamine was 0.920. Pearson correlations with A1C were 0.655 for GA and 0.515 for fructosamine (P < .001) and with MBG were 0.590 and 0.454, respectively (P < .001). At the individual subject level, Pearson correlations with both A1C and MBG were higher for GA than for fructosamine in 56% of subjects; only 4% of subjects had higher fructosamine correlations with A1C and MBG. GA had a higher Pearson correlation with A1C and MBG in 82% and 70% of subjects, respectively. CONCLUSIONS Compared with fructosamine, GA correlates significantly better with both short-term MBG and long-term A1C and may be more useful than fructosamine in clinical situations requiring monitoring of intermediate-term glycemic control (NCT02489773).

Abstract: Molecule optimization is a fundamental task for accelerating drug discovery, with the goal of generating new valid molecules that maximize multiple drug properties while maintaining similarity to the input molecule. Existing generative models and reinforcement learning approaches made initial success, but still face difficulties in simultaneously optimizing multiple drug properties. To address such challenges, we propose the MultI-constraint MOlecule SAmpling (MIMOSA) approach, a sampling framework to use input molecule as an initial guess and sample molecules from the target distribution. MIMOSA first pretrains two property agnostic graph neural networks (GNNs) for molecule topology and substructure-type prediction, where a substructure can be either atom or single ring. For each iteration, MIMOSA uses the GNNs' prediction and employs three basic substructure operations (add, replace, delete) to generate new molecules and associated weights. The weights can encode multiple constraints including similarity and drug property constraints, upon which we select promising molecules for next iteration. MIMOSA enables flexible encoding of multiple property- and similarity-constraints and can efficiently generate new molecules that satisfy various property constraints and achieved up to 49.6% relative improvement over the best baseline in terms of success rate.

Abstract: Accurate prediction of the transmission of epidemic diseases such as COVID-19 is crucial for implementing effective mitigation measures. In this work, we develop a tensor method to predict the evolution of epidemic trends for many regions simultaneously. We construct a 3-way spatio-temporal tensor (location, attribute, time) of case counts and propose a nonnegative tensor factorization with latent epidemiological model regularization named STELAR. Unlike standard tensor factorization methods which cannot predict slabs ahead, STELAR enables long-term prediction by incorporating latent temporal regularization through a system of discrete-time difference equations of a widely adopted epidemiological model. We use latent instead of location/attribute-level epidemiological dynamics to capture common epidemic profile sub-types and improve collaborative learning and prediction. We conduct experiments using both county- and state-level COVID-19 data and show that our model can identify interesting latent patterns of the epidemic. Finally, we evaluate the predictive ability of our method and show superior performance compared to the baselines, achieving up to 21% lower root mean square error and 25% lower mean absolute error for county-level prediction.

Abstract: Deep learning has demonstrated success in health risk prediction especially for patients with chronic and progressing conditions. Most existing works focus on learning disease Network (StageNet) model to extract disease stage information from patient data and integrate it into risk prediction. StageNet is enabled by (1) a stage-aware long short-term memory (LSTM) module that extracts health stage variations unsupervisedly; (2) a stage-adaptive convolutional module that incorporates stage-related progression patterns into risk prediction. We evaluate StageNet on two real-world datasets and show that StageNet outperforms state-of-the-art models in risk prediction task and patient subtyping task. Compared to the best baseline model, StageNet achieves up to 12% higher AUPRC for risk prediction task on two real-world patient datasets. StageNet also achieves over 58% higher Calinski-Harabasz score (a cluster quality metric) for a patient subtyping task.

Abstract: Accurate prediction of drug-target interactions (DTI) is crucial for drug discovery. Recently, deep learning (DL) models for show promising performance for DTI prediction. However, these models can be difficult to use for both computer scientists entering the biomedical field and bioinformaticians with limited DL experience. We present DeepPurpose, a comprehensive and easy-to-use deep learning library for DTI prediction. DeepPurpose supports training of customized DTI prediction models by implementing 15 compound and protein encoders and over 50 neural architectures, along with providing many other useful features. We demonstrate state-of-the-art performance of DeepPurpose on several benchmark datasets.

Abstract: Venous thromboembolism (VTE), which includes deep vein thrombosis (DVT) and pulmonary embolism (PE), is an important cause of post-surgery morbidity and mortality. However, it is unclear whether thromboprophylaxis with the low-molecular-weight heparin (LMWH) enoxaparin after non-orthopedic surgery could balance the cost and clinical outcomes or not. The purpose of this research was to evaluate the cost-effectiveness of enoxaparin for the universal prophylaxis of VTE and associated long-term complications in patients after non-orthopedic surgery compared with no prevention in a Chinese healthcare setting. A decision model, which included both acute VTE and long-term complications, was developed to assess the economic outcomes of the two strategies for patients after non-orthopedic surgery. Quality-adjusted life years (QALYs) and direct medical costs were measured over a 5-year horizon. Incremental cost-effectiveness ratios (ICERs) were calculated. Compared with no prevention, patients under enoxaparin treatment with Caprini risk scores of 3–4, 5–6, 7–8, and ≥ 9 increased by 0.012, 0.017, 0.034, and 0.102 in QALYs, respectively. The results were either that ICERs of thromboprophylaxis with enoxaparin over no prevention were lower than the thresholds or that thromboprophylaxis with enoxaparin was dominant. For patients with a Caprini risk score ≥ 9, thromboprophylaxis with enoxaparin is dominant across the whole drug use duration range. The sensitivity analysis confirmed the results. As the first analysis evaluating the economic outcomes of enoxaparin in patients undergoing general non-orthopedic surgery, this study suggests that thromboprophylaxis with enoxaparin is highly cost-effective compared with no prevention in patients with Caprini risk score ≥ 3.

Abstract: There is a long-standing controversy about the role of serotonin in sleep/wake control, with competing theories that it either promotes sleep or causes arousal. Here, we show that there is a marked increase in wakefulness when all serotonin neurons are genetically deleted in mice hemizygous for ePet1-Cre and homozygous for floxed Lmx1b (Lmx1bf/f/p). However, this only occurs at cool ambient temperatures and can be explained by a thermoregulatory defect that leads to an increase in motor activity to generate heat. Because some serotonin neurons are stimulated by CO2, and serotonin activates thalamocortical networks, we hypothesized that serotonin neurons cause arousal in response to hypercapnia. We found that Lmx1bf/f/p mice completely lacked any arousal response to inhalation of 10% CO2 (with 21% O2 in balance N2) but had normal arousal responses to hypoxia, sound, and air puff. We propose that serotonin neurons mediate the potentially life-saving arousal response to hypercapnia. Impairment of this response may contribute to sudden unexpected death in epilepsy, sudden infant death syndrome, and sleep apnea.

Abstract: Coronavirus disease 2019 (COVID-19) is a new infectious respiratory disease that has caused the ongoing global pandemic. The primary purpose of this article is to describe evolving clinical epidemiology of COVID-19, including 1) infection and testing, 2) clinical spectrum including classification of clinical type, asymptomatic cases, severe cases and comorbidity, and clinical and immunological response, 3) regional variation in clinical presentation, 4) population distribution by age, sex, and occupation, and finally, 5) case-fatality. This content may provide important information on detailed clinical type and presentation of the disease, in which appropriate clinical outcomes can be derived for developing prevention strategies and clinical studies or trials that aim to test potential therapeutics or products for different patient populations.

Abstract: This chapter provides an introduction into the diverse field of data mining, as viewed from the perspective of a clinical statistician. We start with a discussion of data mining and its relationship with machine learning and classical statistics. To facilitate the presentation of material, we map some common problems occurring in analysis of clinical data onto general machine learning tasks, such as supervised, unsupervised, and semi-supervised learning. We then review key concepts of data mining and machine learning with emphasis on methods that are most relevant for analyses of clinical data. We also present our view of the key elements of a statistical analysis plan that ensure principled data mining of randomized clinical trials. This topic is rarely addressed, yet of interest for many clinical statisticians who are routinely using data mining to gain insights and knowledge from the available data beyond the “pre-specified analyses.”

Abstract: Objective: To examine real-world changes in acute migraine-specific medication (AMSMs) use among patients prescribed erenumab in the United States (US). Background: Overuse of AMSMs can potentially complicate rather than support migraine management. Erenumab is a calcitonin gene-related peptide (CGRP) antagonist indicated for the preventive treatment of migraine in adults that has been shown to significantly reduce use of AMSMs. Design/Methods: We conducted a retrospective cohort study using data from IQVIA’s open source pharmacy (LRx) and medical claims (Dx) databases. Patients 18 years or older were included if they had completed an adequate trial of erenumab (≥3 claims) between 5/1/18–4/30/19 (first claim was index date) with data continuity in the 12 months prior to and ≥6 months following the index date. Post-index change in AMSMs use (triptans and ergotamine derivatives used both pre-and-post index) included discontinuation (no refills for ≥60 days after last post-index fill) and change (post-pre index) in units (tablets/pills) filled. Results: We identified 43,185 patients who received ≥3 doses of erenumab; female (85.8%), average (SD) age 47 (12.9) years. After initiation of Erenumab, AMSMs were discontinued in 8,556 of 23,222 patients with both pre- and post-index use (36.8%); triptans 8,021 of 22,338 patients (35.9%) and ergotamines 535 of 884 patients (60.5%). AMSM units changed in 18,571 of 23,222 patients (80.0%); triptans 18,034 of 22,338 patients (80.7%) and ergotamines 537 of 884 patients (60.7%) with an overall mean (SD) change of −1.2 (6.6) units for triptans and −0.4 (6.9) units for ergotamines. Conclusions: In this US focused real-world study, a proportion of patients completing an adequate trial of erenumab discontinued and/or reduced consumption of their AMSMs. There is evidence that the benefits of erenumab seen in clinical trials, in terms of reductions in acute migraine-specific medications are being realized in real-world clinical practice. Disclosure: Dr. Hines has received personal compensation for consulting, serving on a scientific advisory board, speaking, or other activities with IQVIA. Dr. Shah has received personal compensation for consulting, serving on a scientific advisory board, speaking, or other activities with Amgen. Dr. Multani has received personal compensation for consulting, serving on a scientific advisory board, speaking, or other activities with IQVIA. Dr. Wade has received personal compensation for consulting, serving on a scientific advisory board, speaking, or other activities with IQVIA. Dr. Buse has received personal compensation for consulting, serving on a scientific advisory board, speaking, or other activities with Dawn C. Buse, PhD, has received grant support and honoraria from Allergan, Avanir, Amgen, Biohaven, Eli Lilly and Company, and Promius and for work on the editorial board of Current Pain and Headache Reports.. Dr. Buse has received personal compensation in an editorial capacity for Dawn C. Buse, PhD, has received grant support and honoraria from Allergan, Avanir, Amgen, Biohaven, Eli Lilly and Company, and Promius and for work on the editorial board of Current Pain and Headache Reports.. Dr. Buse has received research support from Dawn C. Buse, PhD, has received grant support and honoraria from Allergan, Avanir, Amgen, Biohaven, Eli Lilly and Company, and Promius and for work on the editorial board of Current Pain and Headache Reports.. Dr. Bensink has received personal compensation for consulting, serving on a scientific advisory board, speaking, or other activities with Amgen. Dr. Bensink holds stock and/or stock options in Amgen which sponsored research in which Dr. Bensink was involved as an investigator.

Abstract: There have been a series of recent discussions and debates on the p-value and statistical significance. These discussions, including publications of more than 40 papers in a special issue of the American Statistician, provide an excellent opportunity to think about some technical measures for practical implementation in grant applications and publications. While several factors have been discussed, it may be the rigor of a study that determines the p-value for reporting study results and judging a consistent replication of research. Both p-values and power, which integrate Fisherian and Neyman-Pearson methods, should be used for hypothesis testing. We propose new criteria, which can be implemented without fundamental changes in existing statistics, to reduce false positives and irreplicability of studies that are either inadequately powered or overpowered.

Abstract: In late 2019, a novel coronavirus, SARS-CoV-2, which causes Coronavirus disease 2019 (COVID-19), appeared in Wuhan, China On January 30, 2020, the World Health

Abstract: Clinical trials are essential for drug development but often suffer from expensive, inaccurate and insufficient patient recruitment. The core problem of patient-trial matching is to find qualified patients for a trial, where patient information is stored in electronic health records (EHR) while trial eligibility criteria (EC) are described in text documents available on the web. How to represent longitudinal patient EHR? How to extract complex logical rules from EC? Most existing works rely on manual rule-based extraction, which is time consuming and inflexible for complex inference. To address these challenges, we proposed a cross-modal inference learning model to jointly encode enrollment criteria (text) and patients records (tabular data) into a shared latent space for matching inference. pplies a pre-trained Bidirectional Encoder Representations from Transformers(BERT) model to encode clinical trial information into sentence embedding. And uses a hierarchical embedding model to represent patient longitudinal EHR. In addition, s augmented by a numerical information embedding and entailment module to reason over numerical information in both EC and EHR. These encoders are trained jointly to optimize patient-trial matching score. We evaluated n the trial-patient matching task with demonstrated on real world datasets. utperformed the best baseline by up to 12.4% in average F1.

Abstract: Accurate prediction of the transmission of epidemic diseases such as COVID-19 is crucial for implementing effective mitigation measures. In this work, we develop a tensor method to predict the evolution of epidemic trends for many regions simultaneously. We construct a 3-way spatio-temporal tensor (location, attribute, time) of case counts and propose a nonnegative tensor factorization with latent epidemiological model regularization named STELAR. Unlike standard tensor factorization methods which cannot predict slabs ahead, STELAR enables long-term prediction by incorporating latent temporal regularization through a system of discrete-time difference equations of a widely adopted epidemiological model. We use latent instead of location/attribute-level epidemiological dynamics to capture common epidemic profile sub-types and improve collaborative learning and prediction. We conduct experiments using both county- and state-level COVID-19 data and show that our model can identify interesting latent patterns of the epidemic. Finally, we evaluate the predictive ability of our method and show superior performance compared to the baselines, achieving up to 21% lower root mean square error and 25% lower mean absolute error for county-level prediction.

Abstract: Objective: Evaluate changes in measures of depression and anxiety after 3 months of galcanezumab treatment in patients with prior migraine preventive treatment failures. Background: While depression and anxiety are well-described migraine comorbidities, little is known about the effects of migraine prevention on these comorbidities. Design/Methods: Adults with episodic or chronic migraine and multiple preventive treatment failures received 3-month treatment with galcanezumab 120mg/month (n=232) or placebo (n=230). Patients rated symptoms and severity of depression and anxiety using Patient Health Questionnaire-9 (PHQ-9) and 7-item Generalized Anxiety Disorder Scale (GAD-7) at baseline and endpoint. The PHQ-9 algorithm for possible major depressive disorder (MDD) is based on the Diagnostic and Statistical Manual (DSM-IV) criteria for MDD. The criterion for possible anxiety disorder is a GAD-7 total score ≥10. Results: The percent of patients with “minimal” depressive symptoms (PHQ-9 total score=0–4) increased from 30.6% baseline to 54.3% after 3 months galcanezumab treatment (difference=23.7%), compared with a change from 28.7% baseline to 42.3% (difference=13.6%) with placebo. Percent of patients with possible MDD decreased from 12.9% at baseline to 6.3% after 3 months galcanezumab treatment and from 15.7% baseline to 9.9% with placebo. Mean decrease in PHQ-9 (mean baseline score=7.7) was greater in galcanezumab-treated patients vs placebo (−2.1 vs −1.2; p=0.009), and a greater proportion of galcanezumab-treated patients reported “not at all” on individual PHQ items. Percent of patients with possible anxiety disorder decreased from 13.8% at baseline to 8.5% after 3 months with galcanezumab treatment and from 15.7% baseline to 14.4% with placebo. The mean decrease in GAD-7 (mean baseline score=4.9) was numerically greater in galcanezumab-treated patients vs placebo, without statistical significance (−0.9 vs −0.4; p=0.069). Conclusions: More galcanezumab-treated patients achieved a state of “minimal” depressive symptomatology and experienced significantly greater reductions in depressive symptoms compared with placebo. Anxiety ratings decreased at endpoint but did not differ significantly between treatment groups. Disclosure: Dr. Maizels has received personal compensation for consulting, serving on a scientific advisory board, speaking, or other activities with Amgen, Eli Lilly and Company, Teva. Dr. Maizels has received research support from Eli Lilly and Company. Dr. Buse has received personal compensation for consulting, serving on a scientific advisory board, speaking, or other activities with Dawn C. Buse, PhD, has received grant support and honoraria from Allergan, Avanir, Amgen, Biohaven, Eli Lilly and Company, and Promius and for work on the editorial board of Current Pain and Headache Reports.. Dr. Buse has received personal compensation in an editorial capacity for Dawn C. Buse, PhD, has received grant support and honoraria from Allergan, Avanir, Amgen, Biohaven, Eli Lilly and Company, and Promius and for work on the editorial board of Current Pain and Headache Reports.. Dr. Buse has received research support from Dawn C. Buse, PhD, has received grant support and honoraria from Allergan, Avanir, Amgen, Biohaven, Eli Lilly and Company, and Promius and for work on the editorial board of Current Pain and Headache Reports.. Dr. Jedynak has received personal compensation for consulting, serving on a scientific advisory board, speaking, or other activities with Eli Lilly and Company. Dr. Jedynak holds stock and/or stock options in Eli Lilly and Company which sponsored research in which Dr. Jedynak was involved as an investigator. Dr. Hand has received personal compensation for consulting, serving on a scientific advisory board, speaking, or other activities with IQVIA.Dr. Ford has received personal compensation for consulting, serving on a scientific advisory board, speaking, or other activities with Eli Lilly and Company (Dr. Ford is an employee). Dr. Ford holds stock and/or stock options in Eli Lilly and Company. Dr. Ford has received research support from Eli Lilly and Company. Dr. Detke has received personal compensation for consulting, serving on a scientific advisory board, speaking, or other activities with Eli Lilly and Company.

Abstract: In recent years, significant attention has been devoted towards integrating deep learning technologies in the healthcare domain. However, to safely and practically deploy deep learning models for home health monitoring, two significant challenges must be addressed: the models should be (1) robust against noise; and (2) compact and energy-efficient. We propose REST, a new method that simultaneously tackles both issues via 1) adversarial training and controlling the Lipschitz constant of the neural network through spectral regularization while 2) enabling neural network compression through sparsity regularization. We demonstrate that REST produces highly-robust and efficient models that substantially outperform the original full-sized models in the presence of noise. For the sleep staging task over single-channel electroencephalogram (EEG), the REST model achieves a macro-F1 score of 0.67 vs. 0.39 achieved by a state-of-the-art model in the presence of Gaussian noise while obtaining 19x parameter reduction and 15x MFLOPS reduction on two large, real-world EEG datasets. By deploying these models to an Android application on a smartphone, we quantitatively observe that REST allows models to achieve up to 17x energy reduction and 9x faster inference. We open-source the code repository with this paper: this https URL.

Abstract: The International Classification of Disease (ICD) is a widely used diagnostic ontology for the classification of health disorders and a valuable resource for healthcare analytics. However, ICD is an evolving ontology and subject to periodic revisions (e.g. ICD-9-CM to ICD-10-CM) resulting in the absence of complete cross-walks between versions. While clinical experts can create custom mappings across ICD versions, this process is both time-consuming and costly. We propose an automated solution that facilitates interoperability without sacrificing accuracy. Our solution leverages the SNOMED-CT ontology whereby medical concepts are organised in a directed acyclic graph. We use this to map ICD-9-CM to ICD-10-CM by associating codes to clinical concepts in the SNOMED graph using a nearest neighbors search in combination with natural language processing. To assess the impact of our method, the performance of a gradient boosted tree (XGBoost) developed to classify patients with Exocrine Pancreatic Insufficiency (EPI) disorder, was compared when using features constructed by our solution versus clinically-driven methods. This dataset comprised of 23, 204 EPI patients and 277, 324 non-EPI patients with data spanning from October 2011 to April 2017. Our algorithm generated clinical predictors with comparable stability across the ICD-9-CM to ICD-10-CM transition point when compared to ICD-9-CM/ICD-10-CM mappings generated by clinical experts. Preliminary modeling results showed highly similar performance for models based on the SNOMED mapping vs clinically defined mapping (71% precision at 20% recall for both models). Overall, the framework does not compromise on accuracy at the individual code level or at the model-level while obviating the need for time-consuming manual mapping.

Abstract: Background: Advice for cancer patients on the impact of COVID 19 in terms of their therapy is available from governments and professional societies, but is open to interpretation by clinicians and their patients. We hypothesise that different healthcare systems may respond in different ways leading to variation in the delivery of chemotherapy. To this end we have charted chemotherapy usage before and during the COVID pandemic in 15 hospitals in France and Spain. Methods: We analysed de-identified data of patients on systemic anti-cancer IV therapies that was automatically and continuously transmitted by 15 hospitals in Spain and France up to and including May 2020. The data are bridged to a common data model containing information about demographics, diagnosis, and treatments.We compared the number of patients treated and number of IV administrations in April and May with hospital specific historic means up until and including February 2020. Numbers are broken down by hospital, cancer type, therapy class, new vs. existing patients, gender, and age. Findings: Total monthly patient numbers in hospitals within the Oncology Data Network (ODN) fell from a historic mean of 2643·5 (SD=54·4) to 2443·5 (SD=7·8) in the peri-COVID period (mean of April and May) (p=0·04) while number of administrations dropped from 21954·5 (SD=1730·3) to 20570·0 (SD=1115·8) (p=0·44) compared to the pre-COVID period. Spanish hospitals registered a steeper decline in both patient (1558·5 (SD=47·4) to 1416·5 (SD=0·7) (p=0·05)) and administration numbers (12868·0 (SD=1138·4) to 12054·0 (SD=770·7) (p=0·49)) than French hospitals, patients (1056·7 (SD=49·3) to 1027 (SD=7·1) (p=0·48) and administrations 8852·0 (SD=583·2) to 8516·0 (SD=245·1) (p=0·53).For most cancer types, i.e. chest and respiratory (330·5 (SD=3·5) to 308·5 (SD=6·4) (p=0·05)), gastrointestinal (606 (SD=4·2) to 566 SD=0·0) (p=0·0056)), haematology (233·5 (SD=26·2) to 175·5 (SD=17·7) (p=0·12)), urology (140·5 (SD=9·2) to 122·0 (SD=5·7) (p=0·14)), patient numbers receiving chemotherapy decreased overall, while there was an increase in patients for breast cancer (491·5 (SD=2·1) to 534·5 (SD=6·4) (p=0·01)) and gynecological cancer (196·0 (SD=5·7) to 219·5 (SD=7·8) (p=0·07)).The decrease in patient numbers was mainly driven by reduction in new patients (412·5 (SD=137·9) to 219·5 (SD=40·3) (p=0·20), initiating chemotherapy. Numbers of male patients fell from 1133 to 1013 (-11%), while numbers of female patients fell from 1377 to 1332 (-3%). We could not identify a notable reduction in elderly patients. Interpretation: Delivery of chemotherapy to patients who had already initiated treatment was maintained at broadly similar pre-COVID levels, but there was a reduction in the number of new cancer patients starting therapy during this time. Funding: The Oncology Data Network is funded by IQVIA Holdings Inc. Declaration of Interests: Nicolas Niklas is fully paid and employed by IQVIA. Ashley Woolmore is fully paid and employed by IQVIA. Lisa Chin-A-Young is fully paid and employed by IQVIA. Nathalie Bazylewicz during data analysis fully paid and employed by IQVIA. David Kerr acted as consultant to IQVIA. Ethics Approval Statement: Longi.tudinal, de-identified diagnosis and treatment data was automatically extracted from existing clinical systems and transmitted in full compliance with GDPR through the privacy protecting technology infrastructure in near real-time to IQVIA. The study was approved by the Steering Committee (https://odn-cancer.com/the-collaboration-for-oncology-data-in-europe-code-establishes-clinical-and-analytical-steering-committee/).). Beyond that, given the nature of the data and its collection, with no patient facing impact, this study is exempt from ethics.

Abstract: Matrix factorization (MF) plays an important role in a wide range of machine learning and data mining models. MF is commonly used to obtain item embeddings and feature representations due to its ability to capture correlations and higher-order statistical dependencies across dimensions. In many applications, the categories of items exhibit a hierarchical tree structure. For instance, human diseases can be divided into coarse categories, e.g., bacterial, and viral. These categories can be further divided into finer categories, e.g., viral infections can be respiratory, gastrointestinal, and exanthematous viral diseases. In e-commerce, products, movies, books, etc., are grouped into hierarchical categories, e.g., clothing items are divided by gender, then by type (formal, casual, etc.). While the tree structure and the categories of the different items may be known in some applications, they have to be learned together with the embeddings in many others. In this work, we propose eTREE, a model that incorporates the (usually ignored) tree structure to enhance the quality of the embeddings. We leverage the special uniqueness properties of Nonnegative MF (NMF) to prove identifiability of eTREE. The proposed model not only exploits the tree structure prior, but also learns the hierarchical clustering in an unsupervised data-driven fashion. We derive an efficient algorithmic solution and a scalable implementation of eTREE that exploits parallel computing, computation caching, and warm start strategies. We showcase the effectiveness of eTREE on real data from various application domains: healthcare, recommender systems, and education. We also demonstrate the meaningfulness of the tree obtained from eTREE by means of domain experts interpretation.

Abstract: The long-established anti-malaria medicines, Chloroquine (CQ), and Hydroxychloroquine (HCQ) have been widely mentioned as potential treatments for patients with COVID-19 caused by the novel coronavirus SARS-CoV-2. This is partly based on preclinical studies that have demonstrated the antiviral potential of these two drugs. The potential advantage of CQ and HCQ is based on decades of experience with these drugs, their known safety profiles, and widespread global access. Regulatory authorities recommended their use in COVID-19 patients mainly in the context of clinical trials or following national protocols at the initiation of the pandemic in US. In this paper, we review the up-to-date information on the efficacy and safety of CQ and HCQ for practitioners in the treatment of COVID-19 patents.

Abstract: Clinical narratives that describe complex medical events are often accompanied by meta-information such as a patient’s demographics, diagnoses and medications. This structured information implicitly relates to the logical and semantic structure of the entire narrative, and thus affects vocabulary choices for the narrative composition. To leverage this meta-information, we propose a supervised topic compositional neural language model, called MeTRNN, that integrates the strength of supervised topic modeling in capturing global semantics with the capacity of contextual recurrent neural networks (RNN) in modeling local word dependencies. MeTRNN generates interpretable topics from global meta-information and uses them to facilitate contextual RNNs in modeling local dependencies of text. For efficient training of MeTRNN, we develop an autoencoding variational Bayes inference method. We evaluate MeTRNN on the word prediction tasks using public text datasets. MeTRNN consistently outperforms all baselines across all datasets in perplexity ranging from 5% to 40%. Our case studies on real world electronic health records (EHR) data show that MeTRNN can learn and benefit from meaningful topics.