Active learning strives to reduce annotation costs by choosing the most critical examples to label. Typically, the active learning strategy is contingent on the classification model. For instance, uncertainty sampling depends on poorly calibrated model confidence scores. In the cold-start setting, active learning is impractical because of model instability and data scarcity. Fortunately, modern NLP provides an additional source of information: pretrained language models. The pre-training loss can find examples that surprise the model and should be labeled for efficient fine-tuning. Therefore, we treat the language modeling loss as a proxy for classification uncertainty. With BERT, we develop a simple strategy based on the masked language modeling loss that minimizes labeling costs for text classification. Compared to other baselines, our approach reaches higher accuracy within less sampling iterations and computation time.

Background: Emerging data suggest that direct oral anticoagulants may be a suitable choice for anticoagulation for cerebral venous thrombosis (CVT). However, conducting high-quality trials in CVT is challenging as it is a rare disease with low rates of adverse outcomes such as major bleeding and functional dependence. To facilitate the design of future CVT trials, SECRET (Study of Rivaroxaban for Cerebral Venous Thrombosis) assessed (1) the feasibility of recruitment, (2) the safety of rivaroxaban compared with standard-of-care anticoagulation, and (3) patient-centered functional outcomes. Methods: This was a phase II, prospective, open-label blinded-end point 1:1 randomized trial conducted at 12 Canadian centers. Participants were aged ≥18 years, within 14 days of a new diagnosis of symptomatic CVT, and suitable for oral anticoagulation; they were randomized to receive rivaroxaban 20 mg daily, or standard-of-care anticoagulation (warfarin, target international normalized ratio, 2.0–3.0, or low-molecular-weight heparin) for 180 days, with optional extension up to 365 days. Primary outcomes were annual rate of recruitment (feasibility); and a composite of symptomatic intracranial hemorrhage, major extracranial hemorrhage, or mortality at 180 days (safety). Secondary outcomes included recurrent venous thromboembolism, recanalization, clinically relevant nonmajor bleeding, and functional and patient-reported outcomes (modified Rankin Scale, quality of life, headache, mood, fatigue, and cognition) at days 180 and 365. Results: Fifty-five participants were randomized. The rate of recruitment was 21.3 participants/year; 57% of eligible candidates consented. Median age was 48.0 years (interquartile range, 38.5–73.2); 66% were female. There was 1 primary event (symptomatic intracranial hemorrhage), 2 clinically relevant nonmajor bleeding events, and 1 recurrent CVT by day 180, all in the rivaroxaban group. All participants in both arms had at least partial recanalization by day 180. At enrollment, both groups on average reported reduced quality of life, low mood, fatigue, and headache with impaired cognitive performance. All metrics improved markedly by day 180. Conclusions: Recruitment targets were reached, but many eligible participants declined randomization. There were numerically more bleeding events in patients taking rivaroxaban compared with control, but rates of bleeding and recurrent venous thromboembolism were low overall and in keeping with previous studies. Participants had symptoms affecting their well-being at enrollment but improved over time. REGISTRATION: URL: https://www.clinicaltrials.gov ; Unique identifier: NCT03178864.

Objective: Friedreich ataxia (FRDA), an autosomal recessive neurodegenerative disease caused by mutations in the gene encoding for the mitochondrial protein frataxin, is characterized by ataxia and gait instability, immobility, and eventual death. We evaluated corneal confocal microscopy (CCM) quantification of corneal nerve morphology as a novel, noninvasive, in vivo quantitative imaging biomarker for the severity of neurological manifestations in FRDA. Methods: Corneal nerve fiber density, branch density, and fiber length were quantified in individuals with FRDA (n = 23) and healthy age-matched controls (n = 14). All individuals underwent genetic testing and a detailed neurological assessment with the Scale for the Assessment and Rating of Ataxia (SARA) and Friedreich's Ataxia Rating Scale (FARS). A subset of individuals with FRDA who were ambulatory underwent quantitative gait assessment. Results: CCM demonstrated a significant reduction in nerve fiber density and length in FRDA compared to healthy controls. Importantly, CCM parameters correlated with genotype, SARA and FARS neurological scales, and linear regression modeling of CCM nerve parameter-generated equations that predict the neurologic severity of FRDA. Interpretation: Together, the data suggest that CCM quantification of corneal nerve morphology is a rapid, sensitive imaging biomarker for quantifying the severity of neurologic disease in individuals with FRDA. ANN NEUROL 2018;84:893-904 F riedreich ataxia (FRDA), an inherited autosomal recessive disorder with an incidence of 1:50,000 in the Caucasian population, is characterized by progressive ataxia, gait instability, immobility, and eventual death. 1,2 FRDA is caused by a guanine-adenine-adenine (GAA) trinucleotide repeat expansion in the first intron of the frataxin (FXN) gene, which causes epigenetic silencing of the FXN gene and decreases levels of FXN mRNA. 2 The resulting deficiency of FXN protein causes an imbalance of iron homeostasis in the mitochondria, with increased oxygen radical production and oxidative stress. 3 FRDA is characterized by spinocerebellar ataxia, dysarthria, pyramidal weakness, sensory loss, optic atrophy, hypertrophic cardiomyopathy, and diabetes mellitus. 1,4 The onset of FRDA is usually between the ages of 10 and 15 years, leading to wheelchair dependence and premature death by the 4th decade. 4,5 The recent focus on disease-modifying therapies for FRDA has highlighted the need for validated biomarkers of disease progression. 6,7 The current standards to assess FRDA disease severity are the two clinical scales Friedreich's Ataxia Rating Scale (FARS) 8,9 and Scale for the Assessment and Rating of Ataxia (SARA). 10 Based on a body of literature demonstrating the presence of a distal axonopathy and abnormal quantitative sensory thresholds in FRDA, indicative of unmyelinated nerve fiber pathology, 11 we hypothesized that quantification of View this article online at wileyonlinelibrary.com.

Background SFTSV, an emerging tick-borne pathogen that can cause fatal severe fever with thrombocytopenia syndrome (SFTS), was first identified in China in 2009. Limited evidence suggests that SFTSV can be transmitted between humans via blood contact raising concerns over transfusion safety. A study of donor samples from three Chinese blood centers was conducted to investigate the seroprevalence and rate of SFTSV viremia among Chinese blood donors. Materials and Methods From April 16 to Oct 31, 2012, 17,208 plasma samples were collected from donors at Xinyang (located in an SFTSV endemic area), Mianyang and Luoyang blood centers. Assessment of anti-SFTSV total antibody was performed on all samples using Enzyme Linked Immunoassay (ELISA). Repeat-reactive samples were tested for SFTSV RNA using reverse transcription (RT) Real-time-PCR assay with Taqman probes. In addition, 9960 of the Xinyang samples were tested in pools of 4 by the same PCR method and each of samples in a reactive pool was tested individually. Results Donor seroreactivity rates were: Xinyang, 0.54% (80/14,752); Mianyang, 0.27% (3/1,130); and Luoyang, 0.28% (3/1,326). All seroreactive samples were negative on RT-PCR single-sample testing. Two RT-PCR reactive donor samples were identified, both with estimated viral load <20pfu/ml. The RNA prevalence rate for SFTSV among donors in Xinyang was 0.02%. Conclusion This was the first multi-region study of SFTSV sero- and viral-prevalence among Chinese blood donors. Viral prevalence was low and no seroreactive sample was viremic, suggesting limited impact of SFTSV on blood safety in China.

Cross-lingual word embeddings transfer knowledge between languages: models trained on high-resource languages can predict in low-resource languages. We introduce CLIME, an interactive system to quickly refine cross-lingual word embeddings for a given classification problem. First, CLIME ranks words by their salience to the downstream task. Then, users mark similarity between keywords and their nearest neighbors in the embedding space. Finally, CLIME updates the embeddings using the annotations. We evaluate CLIME on identifying health-related text in four low-resource languages: Ilocano, Sinhalese, Tigrinya, and Uyghur. Embeddings refined by CLIME capture more nuanced word semantics and have higher test accuracy than the original embeddings. CLIME often improves accuracy faster than an active learning baseline and can be easily combined with active learning to improve results.