[P4–341]: Levels of Blood Pressure, Body Mass Index and Total Serum Cholesterol at Different Time Points Prior to Dementia Diagnosis: A Case Control Study of Over 28 Million Electronic Health Records From the Emif Ehr Data Resource

Perera, Gayan; Gungabissoon, Usha; Alexander, Myriam; Ansel, David; Avillach, Paul; Salles, Talita Duarte; Gordon, Mark Forrest; Nevado‐Holgado, Alejo J.; Novak, Gerald; Pasqua, Alessandro; Pedersen, Lars Jacob Tynes; Ponjoan, Anna; Lei, Johan van der; Stewart, Robert

doi:10.1016/j.jalz.2017.06.2211

Cited by 1 publication

(2 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…medication, dosage, mode, frequency, duration, reason; see section 2.1) from the rest, but again without a method clearly outperforming the other. We also explored with embedding sizes of 2 4 to 2 10 and noticed diminishing improvements in performance at values above 2 7 , ultimately settling at an embedding size of 100.…”

Section: Intrinsic Evaluation Of Embeddingsmentioning

confidence: 99%

“…In number of subjects (for example, 50 million patients in the case of European Medical Information Framework (EMIF)), EHRs are the largest source of empirical data in biomedical research [4,8], making them ideal for studying disease (e.g. Alzheimer's [9], cardiovascular disease [10], or associated risk factors [11][12][13]) and evaluating service (e.g. monitoring adverse drug reactions [14]).…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Named entity recognition in electronic health records using transfer learning bootstrapped Neural Networks

et al. 2020

View full text Add to dashboard Cite

Neural networks (NNs) have become the state of the art in many machine learning applications, such as image, sound [1] and natural language processing [2,3]. However, the success of NNs remains dependent on the availability of large labelled datasets, such as in the case of electronic health records (EHRs). With scarce data, NNs are unlikely to be able to extract this hidden information with practical accuracy. In this study, we develop an approach that solves these problems for named entity recognition, obtaining 94.6 F1 score in I2B2 2009 Medical Extraction Challenge [6], 4.3 above the architecture that won the competition. To achieve this, we bootstrap our NN models through transfer learning by pretraining word embeddings on a secondary task performed on a large pool of unannotated EHRs and using the output embeddings as a foundation of a range of NN architectures. Beyond the official I2B2 challenge, we further achieve 82.4 F1 on extracting relationships between medical terms using attention-based seq2seq models bootstrapped in the same manner.

show abstract

Section: Intrinsic Evaluation Of Embeddingsmentioning

confidence: 99%