• Digital Humanities and Cultural Heritage
• Publications close
• Research data close
• Other research products close
• AU close
• FR close
• Europe PubMed Central close

Previous work on clinical relation extraction from free-text sentences leveraged information about semantic types from clinical knowledge bases as a part of entity representations. In this paper, we exploit additional evidence by also making use of domain-specific semantic type dependencies. We encode the relation between a span of tokens matching a Unified Medical Language System (UMLS) concept and other tokens in the sentence. We implement our method and compare against different named entity recognition (NER) architectures (i.e., BiLSTM-CRF and BiLSTM-GCN-CRF) using different pre-trained clinical embeddings (i.e., BERT, BioBERT, UMLSBert). Our experimental results on clinical datasets show that in some cases NER effectiveness can be significantly improved by making use of domain-specific semantic type dependencies. Our work is also the first study generating a matrix encoding to make use of more than three dependencies in one pass for the NER task.

Abstract Background Dementia is defined as a cognitive decline that affects functional status. Longitudinal ageing surveys often lack a clinical diagnosis of dementia though measure cognition and daily function over time. We used unsupervised machine learning and longitudinal data to identify transition to probable dementia. Methods Multiple Factor Analysis was applied to longitudinal function and cognitive data of 15,278 baseline participants (aged 50 years and more) from the Survey of Health, Ageing, and Retirement in Europe (SHARE) (waves 1, 2 and 4–7, between 2004 and 2017). Hierarchical Clustering on Principal Components discriminated three clusters at each wave. We estimated probable or “Likely Dementia” prevalence by sex and age, and assessed whether dementia risk factors increased the risk of being assigned probable dementia status using multistate models. Next, we compared the “Likely Dementia” cluster with self-reported dementia status and replicated our findings in the English Longitudinal Study of Ageing (ELSA) cohort (waves 1–9, between 2002 and 2019, 7840 participants at baseline). Results Our algorithm identified a higher number of probable dementia cases compared with self-reported cases and showed good discriminative power across all waves (AUC ranged from 0.754 [0.722–0.787] to 0.830 [0.800–0.861]). “Likely Dementia” status was more prevalent in older people, displayed a 2:1 female/male ratio, and was associated with nine factors that increased risk of transition to dementia: low education, hearing loss, hypertension, drinking, smoking, depression, social isolation, physical inactivity, diabetes, and obesity. Results were replicated in ELSA cohort with good accuracy. Conclusions Machine learning clustering can be used to study dementia determinants and outcomes in longitudinal population ageing surveys in which dementia clinical diagnosis is lacking. International audience

PURPOSE Histopathologic features are critical for studying risk factors of colorectal polyps, but remain deeply embedded within unstructured pathology reports, requiring costly and time-consuming manual abstraction for research. In this study, we developed and evaluated a natural language processing (NLP) pipeline to automatically extract histopathologic features of colorectal polyps from pathology reports, with an emphasis on individual polyp size. These data were then linked with structured electronic health record (EHR) data, creating an analysis-ready epidemiologic data set. METHODS We obtained 24,584 pathology reports from colonoscopies performed at the University of Utah’s Gastroenterology Clinic. Two investigators annotated 350 reports to determine inter-rater agreement, develop an annotation scheme, and create a reference standard for performance evaluation. The pipeline was then developed, and performance was compared against the reference for extracting polyp location, histology, size, shape, dysplasia, and the number of polyps. Finally, the pipeline was applied to 24,225 unseen reports and NLP-extracted data were linked with structured EHR data. RESULTS Across all features, our pipeline achieved a precision of 98.9%, a recall of 98.0%, and an F1-score of 98.4%. In patients with polyps, the pipeline correctly extracted 95.6% of sizes, 97.2% of polyp locations, 97.8% of histology, 98.3% of shapes, and 98.3% of dysplasia levels. When applied to unseen data, the pipeline classified 12,889 patients as having polyps, 4,907 patients without polyps, and extracted the features of 28,387 polyps. Tubular adenomas were the most common subtype (55.9%), 8.1% of polyps were advanced adenomas, and the mean polyp size was 0.57 (±0.4) cm. CONCLUSION Our pipeline extracted histopathologic features of colorectal polyps from colonoscopy pathology reports, most notably individual polyp sizes, with considerable accuracy. This study demonstrates the utility of NLP for extracting polyp features and linking these data with EHR data to create an epidemiologic data set to study colorectal polyp risk factors and outcomes.

AbstractSeveral teams have been publishing global estimates of excess deaths during the COVID‐19 pandemic. Here, we examine potential flaws and underappreciated sources of uncertainty in global excess death calculations. Adjusting for changing population age structure is essential. Otherwise, excess deaths are markedly overestimated in countries with increasingly aging populations. Adjusting for changes in other high‐risk indicators, such as residence in long‐term facilities, may also make a difference. Death registration is highly incomplete in most countries; completeness corrections should allow for substantial uncertainty and consider that completeness may have changed during pandemic years. Excess death estimates have high sensitivity to modelling choice. Therefore different options should be considered and the full range of results should be shown for different choices of pre‐pandemic reference periods and imposed models. Any post‐modelling corrections in specific countries should be guided by pre‐specified rules. Modelling of all‐cause mortality (ACM) in countries that have ACM data and extrapolating these models to other countries is precarious; models may lack transportability. Existing global excess death estimates underestimate the overall uncertainty that is multiplicative across diverse sources of uncertainty. Informative excess death estimates require risk stratification, including age groups and ethnic/racial strata. Data to‐date suggest a death deficit among children during the pandemic and marked socioeconomic differences in deaths, widening inequalities. Finally, causal explanations require great caution in disentangling SARS‐CoV‐2 deaths, indirect pandemic effects and effects from measures taken. We conclude that excess deaths have many uncertainties, but globally deaths from SARS‐CoV‐2 may be the minority of calculated excess deaths.

Abstract Objective To develop an automated deidentification pipeline for radiology reports that detect protected health information (PHI) entities and replaces them with realistic surrogates “hiding in plain sight.” Materials and Methods In this retrospective study, 999 chest X-ray and CT reports collected between November 2019 and November 2020 were annotated for PHI at the token level and combined with 3001 X-rays and 2193 medical notes previously labeled, forming a large multi-institutional and cross-domain dataset of 6193 documents. Two radiology test sets, from a known and a new institution, as well as i2b2 2006 and 2014 test sets, served as an evaluation set to estimate model performance and to compare it with previously released deidentification tools. Several PHI detection models were developed based on different training datasets, fine-tuning approaches and data augmentation techniques, and a synthetic PHI generation algorithm. These models were compared using metrics such as precision, recall and F1 score, as well as paired samples Wilcoxon tests. Results Our best PHI detection model achieves 97.9 F1 score on radiology reports from a known institution, 99.6 from a new institution, 99.5 on i2b2 2006, and 98.9 on i2b2 2014. On reports from a known institution, it achieves 99.1 recall of detecting the core of each PHI span. Discussion Our model outperforms all deidentifiers it was compared to on all test sets as well as human labelers on i2b2 2014 data. It enables accurate and automatic deidentification of radiology reports. Conclusions A transformer-based deidentification pipeline can achieve state-of-the-art performance for deidentifying radiology reports and other medical documents.

The emergence of the ability to think about future possibilities must have played an influential role in human evolution, driving a range of foresightful behaviours, including preparation, communication and technological innovation. Here we review the archeological evidence for such behavioural indicators of foresight. We find the earliest signs of hominins retaining tools and transporting materials for repeated future use emerging from around 1.8 Ma. From about 0.5 Ma onwards, there are indications of technical and social changes reflecting advances in foresight. And in a third period, starting from around 140 000 years ago, hominins appear to have increasingly relied on material culture to shape the future and to exchange their ideas about possibilities. Visible signs of storytelling, even about entirely fictional scenarios, appear over the last 50 000 years. Although the current evidence suggests that there were distinct transitions in the evolution of our capacity to think about the future, we warn that issues of taphonomy and archaeological sampling are likely to skew our picture of human cognitive evolution.This article is part of the theme issue ‘Thinking about possibilities: mechanisms, ontogeny, functions and phylogeny’.

We present the first ancient DNA data from the Pre-Pottery Neolithic of Mesopotamia (Southeastern Turkey and Northern Iraq), Cyprus, and the Northwestern Zagros, along with the first data from Neolithic Armenia. We show that these and neighboring populations were formed through admixture of pre-Neolithic sources related to Anatolian, Caucasus, and Levantine hunter-gatherers, forming a Neolithic continuum of ancestry mirroring the geography of West Asia. By analyzing Pre-Pottery and Pottery Neolithic populations of Anatolia, we show that the former were derived from admixture between Mesopotamian-related and local Epipaleolithic-related sources, but the latter experienced additional Levantine-related gene flow, thus documenting at least two pulses of migration from the Fertile Crescent heartland to the early farmers of Anatolia. The newly reported dataset is described in detail in an accompanying Research Article, where we also acknowledge the funders who supported dataset generation (12). Analysis of data was supported by the National Institutes of Health (GM100233 and HG012287), the John Templeton Foundation (grant 61220), a private gift from Jean-Francois Clin, the Allen Discovery Center program, a Paul G. Allen Frontiers Group advised program of the Paul G. Allen Family Foundation, and the Howard Hughes Medical Institute (D.R.). National Institutes of Health [GM100233, HG012287]; John Templeton Foundation [61220]; Allen Discovery Center program; Paul G. Allen Frontiers Group advised program of the Paul G. Allen Family Foundation; Howard Hughes Medical Institute