Data and systems for medication-related text classification and concept normalization from Twitter: insights from the Social Media Mining for Health (SMM4H)-2017 shared task

Objective

We executed the Social Media Mining for Health (SMM4H) 2017 shared tasks to enable the community-driven development and large-scale evaluation of automatic text processing methods for the classification and normalization of health-related text from social media. An additional objective was to publicly release manually annotated data.

Materials and Methods

We organized 3 independent subtasks: automatic classification of self-reports of 1) adverse drug reactions (ADRs) and 2) medication consumption, from medication-mentioning tweets, and 3) normalization of ADR expressions. Training data consisted of 15 717 annotated tweets for (1), 10 260 for (2), and 6650 ADR phrases and identifiers for (3); and exhibited typical properties of social-media-based health-related texts. Systems were evaluated using 9961, 7513, and 2500 instances for the 3 subtasks, respectively. We evaluated performances of classes of methods and ensembles of system combinations following the shared tasks.

Results

Among 55 system runs, the best system scores for the 3 subtasks were 0.435 (ADR class F₁-score) for subtask-1, 0.693 (micro-averaged F₁-score over two classes) for subtask-2, and 88.5% (accuracy) for subtask-3. Ensembles of system combinations obtained best scores of 0.476, 0.702, and 88.7%, outperforming individual systems.

Discussion

Among individual systems, support vector machines and convolutional neural networks showed high performance. Performance gains achieved by ensembles of system combinations suggest that such strategies may be suitable for operational systems relying on difficult text classification tasks (eg, subtask-1).

Conclusions

Data imbalance and lack of context remain challenges for natural language processing of social media text. Annotated data from the shared task have been made available as reference standards for future studies (http://dx.doi.org/10.17632/rxwfb3tysd.1).

P1217Zirconium-89 labelled probe for molecular imaging of inflammation in experimental atherosclerosis

Background

Early detection of inflamed atherosclerotic lesions by molecular imaging might improve risk assessment beyond that of vascular stenosis and plaque morphology imaging, and improve the clinical management of high-risk patients.

Purpose

To target the key features of unstable atherosclerotic lesions, we studied the feasibility of our radiotracer, based on modified human serum albumin (HSA), to identify inflamed atherosclerotic lesions by in vivo molecular imaging.

Methods

We applied a maleylated HSA (Mal-HSA) probe, recognised by scavenger receptors on macrophages, in an experimental in vivo imaging study of atherosclerosis. Mal-HSA was coupled with a positron-emittingmetal ion, Zirconium-89 (89Zr). The targeting potential of this probe was evaluated and compared with unspecific 89Zr-HSA and 18F-FDG in a mouse model of atherosclerosis (Apoe−/−, n=22) and compared with wild-type (WT) mice (C57BL/6, n=21) as controls. Radiotracer accumulation in the aortic arch was analysed in vivo by the fusion of positron emission tomography–magnetic resonance imaging (PET-MRI), radiotracer bio-distribution was measured ex vivo by gamma counter, and plaque uptake was evaluated by phosphor imaging (PI) autoradiography (ARG).

Results

PET-MRI, gamma counter measurements, and PI-ARG showed the accumulation of 89Zr-Mal-HSA in the atherosclerotic lesions of Apoe−/− mice. The maximum standardised uptake value (SUVmax) for 89Zr-Mal-HSA at 16 and 20 weeks were 26% and 20% higher (P<0.05) in Apoe−/− mice than control WT mice, whereas no difference in SUVmax was found for 18F-FDG in the same animals. 89Zr-Mal-HSA uptake in the aorta as evaluated by gamma counter 48 h post-injection was 32% higher (P<0.01) for Apoe−/− mice compared to WT mice, and the aorta-to-blood ratio was 10-fold higher (P<0.001) for 89Zr-Mal-HSA compared with unspecific 89Zr-HSA. HSA probes were mainly distributed to the liver, spleen, kidneys, bone and lymph nodes. The PI-ARG results corroborated the PET and gamma counter measurements, showing higher accumulation of 89Zr-Mal-HSA in the aortas of Apoe−/− mice compared to WT mice; 9.4±1.4 vs 0.8±0.3% (P<0.001).

Conclusions

The modified HSA-based radiotracer showed in vivo targeting of inflamed atherosclerotic lesions of mouse aorta, which could also be verified ex vivo. 89Zr-Mal-HSA seems to be a promising diagnostic tool for the identification of vascular inflammation. Further methodological studies are needed to verify its applicability for detecting rupture-prone plaques.

Acknowledgement/Funding

Swedish Research Council (22036); the Swedish HLF (20150423, 20170669); ALF (20150517, 447561, 726481); Söderberg Foundations, VINNOVA and KI

Distant supervision for treatment relation extraction by leveraging MeSH subheadings

The growing body of knowledge in biomedicine is too vast for human consumption. Hence there is a need for automated systems able to navigate and distill the emerging wealth of information. One fundamental task to that end is relation extraction, whereby linguistic expressions of semantic relationships between biomedical entities are recognized and extracted. In this study, we propose a novel distant supervision approach for relation extraction of binary treatment relationships such that high quality positive/negative training examples are generated from PubMed abstracts by leveraging associated MeSH subheadings. The quality of generated examples is assessed based on the quality of supervised models they induce; that is, the mean performance of trained models (derived via bootstrapped ensembling) on a gold standard test set is used as a proxy for data quality. We show that our approach is preferable to traditional distant supervision for treatment relations and is closer to human crowd annotations in terms of annotation quality. For treatment relations, our generated training data performs at 81.38%, compared to traditional distant supervision at 64.33% and crowd-sourced annotations at 90.57% on the model-wide PR-AUC metric. We also demonstrate that examples generated using our method can be used to augment crowd-sourced datasets. Augmented models improve over non-augmented models by more than two absolute points on the more established F1 metric. We lastly demonstrate that performance can be further improved by implementing a classification loss that is resistant to label noise.

[Robot-assisted Whipple resection; results of the first 100 procedures in the Netherlands]

OBJECTIVE

Investigation into the results of robot-assisted Whipple resection in the Netherlands. These results were compared with those of open Whipple procedures on the basis of recent large case series of patients.

DESIGN

Case series of patients and systematic literature review.

METHOD

We carried out a post hoc analysis of prospectively collected data on the first 100 consecutive patients who underwent robot-assisted Whipple procedures in the period from March 2016 until March 2018 at the Erasmus MC, the Maasstad hospital or the Regional Academic Cancer Centre Utrecht. We were mainly interested in surgery characteristics and postoperative outcomes. We compared our results with those of case series of patients with more than 500 open Whipple procedures carried out in a single hospital, published in the last 5 years.

RESULTS

There were one or more serious complications in 22 patients (22%) and 2 patients (2%) developed multiple organ failure. 7 patients (7%) underwent reoperation. There was no postoperative mortality. In 14 case series (n = 12,708), complications occurred in 38% of patients and 7% of patients underwent reoperation. Mean mortality rate was 3%.

CONCLUSION

Our findings show that robotic Whipple procedures can be carried out safely in the Netherlands. The number of complications and mortality rates are comparable with results of large case series of patients who underwent open Whipple procedures in a centre of expertise.

Restoration of susceptibility to amikacin by 8-hydroxyquinoline analogs complexed to zinc

Gram-negative pathogens resistant to amikacin and other aminoglycosides of clinical relevance usually harbor the 6’-N-acetyltransferase type Ib [AAC(6')-Ib], an enzyme that catalyzes inactivation of the antibiotic by acetylation using acetyl-CoA as donor substrate. Inhibition of the acetylating reaction could be a way to induce phenotypic conversion to susceptibility in these bacteria. We have previously observed that Zn²⁺ acts as an inhibitor of the enzymatic acetylation of aminoglycosides by AAC(6')-Ib, and in complex with ionophores it effectively reduced the levels of resistance in cellulo. We compared the activity of 8-hydroxyquinoline, three halogenated derivatives, and 5-[N-Methyl-N-Propargylaminomethyl]-8-Hydroxyquinoline in complex with Zn²⁺ to inhibit growth of amikacin-resistant Acinetobacter baumannii in the presence of the antibiotic. Two of the compounds, clioquinol (5-chloro-7-iodo-8-hydroxyquinoline) and 5,7-diiodo-8-hydroxyquinoline, showed robust inhibition of growth of the two A. baumannii clinical isolates that produce AAC(6')-Ib. However, none of the combinations had any activity on another amikacin-resistant A. baumannii strain that possesses a different, still unknown mechanism of resistance. Time-kill assays showed that the combination of clioquinol or 5,7-diiodo-8-hydroxyquinoline with Zn²⁺ and amikacin was bactericidal. Addition of 8-hydroxyquinoline, clioquinol, or 5,7-diiodo-8-hydroxyquinoline, alone or in combination with Zn²⁺, and amikacin to HEK293 cells did not result in significant toxicity. These results indicate that ionophores in complex with Zn²⁺ could be developed into potent adjuvants to be used in combination with aminoglycosides to treat Gram-negative pathogens in which resistance is mediated by AAC(6')-Ib and most probably other related aminoglycoside modifying enzymes.

Effects of mass loading on the viability of assessing the state of healing of a fixated fractured long bone

Introduction

This paper aims to evaluate the effects of mass loading on the healing assessment of an internally fixated femur by vibrational means. The presence of soft tissue surrounding a femur increases damping and mass of a system, and hence affects the vibrational response of a mechanical structure by obscuring the coherent modes. This may compromise vibration-based monitoring strategies in identifying modes associated with fracture healing.

Methods

This paper presents a series of experimental works to address this issue. Two osteotomised composite femurs were internally fixated using a plate-screw system and an intramedullary nail. Soft tissue is approximated by surrounding an artificial Sawbone femur with modelling clay. The femur is excited by an instrumented impact hammer and instrumented with two accelerometers to record bending and torsion modes between 0 and 600 Hz. A 30-min epoxy was applied to simulate the healing of the fractured femur in the osteotomised region. The resonant frequencies and its modes are monitored while union is being formed and a healing index is calculated at various times to quantify the degree of healing.

Results

The results demonstrate that the effect of modelling clay compressed the natural modes along the frequency axis. It is observed that frequency bandwidth in the vicinity of 150 Hz and 500 Hz is sensitive to the state of healing of the fixated femurs, which is due to the increase in stiffness of the osteotomised region. These findings were used to formulate the healing index which assists in identifying the initial, later and complete healing stages in conjunction with the index derivative.

Conclusion

In this study, a two-sensor measurement strategy to quantify fixated femur healing is investigated. It is shown that the mass loading effect did not affect this vibrational analysis method ability to assess the state of healing, and both coherent bending and twisting modes associated with healing were easily identified. The proposed healing index, its derivative, and the cross-spectra are a viable tool for quantitative healing assessment.

Mortality risk reduction differs according to bisphosphonate class: a 15-year observational study

In this prospective cohort of 6120 participants aged 50+, nitrogen-bisphosphonates but not non-nitrogen bisphosphonates were associated with a significant 34% mortality risk reduction compared to non-treated propensity score matched controls. These findings open new avenues for research into mechanistic pathways.

INTRODUCTION

Emerging evidence suggests that bisphosphonates (BP), first-line treatment of osteoporosis, are associated with reduced risks for all-cause mortality. This study aimed to determine the association between different BP types and mortality risk in participants with or without a fracture.

METHODS

A prospective cohort study of users of different BPs matched to non-users by propensity score (age, gender, co-morbidities, fragility fracture status) and time to starting the BP medication from the population-based Canadian Multicentre Osteoporosis Study from nine Canadian centres followed from 1995 to 2013. Mortality risk for bisphosphonate users vs matched non-users was assessed using pairwise multivariable Cox proportional hazards models.

RESULTS

There were 2048 women and 308 men on BP and 1970 women and 1794 men who did not receive medication for osteoporosis. The relationship between BP and mortality risk was explored in three separate 1:1 propensity score-matched cohorts of BP users and no treatment (etidronate, n = 599, alendronate, n = 498, and risedronate n = 213). Nitrogen BP (n-BP) (alendronate and risedronate) was associated with lower mortality risks [pairwise HR, 0.66 (95% CI, 0.48-0.91)] while the less potent non-n-BP, etidronate, was not [pairwise HR: 0.89 (95% CI, 0.66-1.20)]. A direct comparison between n-BP and etidronate (n = 340 pairs) also suggested a better survival for n-BP [paired HR, 0.47 (95%CI, (95% CI, 031-0.70)] for n-BP vs. etidronate].

CONCLUSION

Compared to no treatment, nitrogen but not non-nitrogen bisphosphonates appear to be associated with better survival.

Confirm or refute?: A comparative study on citation sentiment classification in clinical research publications

Quantifying scientific impact of researchers and journals relies largely on citation counts, despite the acknowledged limitations of this approach. The need for more suitable alternatives has prompted research into developing advanced metrics, such as h-index and Relative Citation Ratio (RCR), as well as better citation categorization schemes to capture the various functions that citations serve in a publication. One such scheme involves citation sentiment: whether a reference paper is cited positively (agreement with the findings of the reference paper), negatively (disagreement), or neutrally. The ability to classify citation function in this manner can be viewed as a first step toward a more fine-grained bibliometrics. In this study, we compared several approaches, varying in complexity, for classification of citation sentiment in clinical trial publications. Using a corpus of 285 discussion sections from as many publications (a total of 4,182 citations), we developed a rule-based method as well as supervised machine learning models based on support vector machines (SVM) and two variants of deep neural networks; namely, convolutional neural network (CNN) and bidirectional long short-term memory (BiLSTM). A CNN model augmented with hand-crafted features yielded the best performance (0.882 accuracy and 0.721 macro-F₁ on held-out set). Our results show that baseline performances of traditional supervised learning algorithms and deep neural network architectures are similar and that hand-crafted features based on sentiment dictionaries and rhetorical structure allow neural network approaches to outperform traditional machine learning approaches for this task. We make the rule-based method and the best-performing neural network model publicly available at: https://github.com/kilicogluh/clinical-citation-sentiment.

Overview of the BioCreative VI Precision Medicine Track: mining protein interactions and mutations for precision medicine

The Precision Medicine Initiative is a multicenter effort aiming at formulating personalized treatments leveraging on individual patient data (clinical, genome sequence and functional genomic data) together with the information in large knowledge bases (KBs) that integrate genome annotation, disease association studies, electronic health records and other data types. The biomedical literature provides a rich foundation for populating these KBs, reporting genetic and molecular interactions that provide the scaffold for the cellular regulatory systems and detailing the influence of genetic variants in these interactions. The goal of BioCreative VI Precision Medicine Track was to extract this particular type of information and was organized in two tasks: (i) document triage task, focused on identifying scientific literature containing experimentally verified protein-protein interactions (PPIs) affected by genetic mutations and (ii) relation extraction task, focused on extracting the affected interactions (protein pairs). To assist system developers and task participants, a large-scale corpus of PubMed documents was manually annotated for this task. Ten teams worldwide contributed 22 distinct text-mining models for the document triage task, and six teams worldwide contributed 14 different text-mining systems for the relation extraction task. When comparing the text-mining system predictions with human annotations, for the triage task, the best F-score was 69.06%, the best precision was 62.89%, the best recall was 98.0% and the best average precision was 72.5%. For the relation extraction task, when taking homologous genes into account, the best F-score was 37.73%, the best precision was 46.5% and the best recall was 54.1%. Submitted systems explored a wide range of methods, from traditional rule-based, statistical and machine learning systems to state-of-the-art deep learning methods. Given the level of participation and the individual team results we find the precision medicine track to be successful in engaging the text-mining research community. In the meantime, the track produced a manually annotated corpus of 5509 PubMed documents developed by BioGRID curators and relevant for precision medicine. The data set is freely available to the community, and the specific interactions have been integrated into the BioGRID data set. In addition, this challenge provided the first results of automatically identifying PubMed articles that describe PPI affected by mutations, as well as extracting the affected relations from those articles. Still, much progress is needed for computer-assisted precision medicine text mining to become mainstream. Future work should focus on addressing the remaining technical challenges and incorporating the practical benefits of text-mining tools into real-world precision medicine information-related curation.

Bacterial population dynamics after foliar fertilization of almond leaves

AIMS

To describe the effects of foliar fertilizer application on the bacterial populations of almond tree leaves.

METHODS AND RESULTS

We applied a commercially available foliar fertilizer or a water control onto the leaves of almond trees and collected leaves after 1, 7, 14 and 56 days and examined their bacterial populations by 16S rRNA gene sequence analysis. After 1 day, we observed significant differences in 3 of the 4 predominant bacterial phyla, and 5 of the 13 predominant bacterial families. After 7 days, we observed significant differences in all of the predominant phyla, and 8 of the 13 predominant families. After 14 days, the number of significant differences decreased, and after 56 days only 2 of the 13 predominant families differed significantly.

CONCLUSIONS

Foliar fertilization significantly altered the bacterial population structure of almond leaves as compared to the water control. While most of the observed perturbation was transient, significant differences remained after 56 days.

SIGNIFICANCE AND IMPACT OF THE STUDY

This is the first report describing the effects of foliar fertilization on the bacterial populations of almond leaves and provides new insights as to how this process alters the leaf bacterial population structure.

Solar Sources of Interplanetary Magnetic Clouds Leading to Helicity Prediction

This study identifies the solar origins of magnetic clouds that are observed at 1 AU and predicts the helical handedness of these clouds from the solar surface magnetic fields. We started with the magnetic clouds listed by the Magnetic Field Investigation (MFI) team supporting NASA's Wind spacecraft in what is known as the MFI table and worked backward in time to identify solar events that produced these clouds. Our methods utilize magnetograms from the Helioseismic and Magnetic Imager instrument on the Solar Dynamics Observatory spacecraft so that we could only analyze MFI entries after the beginning of 2011. This start date and the end date of the MFI table gave us 37 cases to study. Of these we were able to associate only eight surface events with clouds detected by Wind at 1 AU. We developed a simple algorithm for predicting the cloud helicity that gave the correct handedness in all eight cases. The algorithm is based on the conceptual model that an ejected flux tube has two magnetic origination points at the positions of the strongest radial magnetic field regions of opposite polarity near the places where the ejected arches end at the solar surface. We were unable to find events for the remaining 29 cases: lack of a halo or partial halo coronal mass ejection in an appropriate time window, lack of magnetic and/or filament activity in the proper part of the solar disk, or the event was too far from disk center. The occurrence of a flare was not a requirement for making the identification but in fact flares, often weak, did occur for seven of the eight cases.