Dataset Information

Deep learning for pharmacovigilance: recurrent neural network architectures for labeling adverse drug reactions in Twitter posts.

ABSTRACT: Objective:Social media is an important pharmacovigilance data source for adverse drug reaction (ADR) identification. Human review of social media data is infeasible due to data quantity, thus natural language processing techniques are necessary. Social media includes informal vocabulary and irregular grammar, which challenge natural language processing methods. Our objective is to develop a scalable, deep-learning approach that exceeds state-of-the-art ADR detection performance in social media. Materials and Methods:We developed a recurrent neural network (RNN) model that labels words in an input sequence with ADR membership tags. The only input features are word-embedding vectors, which can be formed through task-independent pretraining or during ADR detection training. Results:Our best-performing RNN model used pretrained word embeddings created from a large, non-domain-specific Twitter dataset. It achieved an approximate match F-measure of 0.755 for ADR identification on the dataset, compared to 0.631 for a baseline lexicon system and 0.65 for the state-of-the-art conditional random field model. Feature analysis indicated that semantic information in pretrained word embeddings boosted sensitivity and, combined with contextual awareness captured in the RNN, precision. Discussion:Our model required no task-specific feature engineering, suggesting generalizability to additional sequence-labeling tasks. Learning curve analysis showed that our model reached optimal performance with fewer training examples than the other models. Conclusion:ADR detection performance in social media is significantly improved by using a contextually aware model and word embeddings formed from large, unlabeled datasets. The approach reduces manual data-labeling requirements and is scalable to large social media datasets.

SUBMITTER: Cocos A

PROVIDER: S-EPMC7651964 | biostudies-literature | 2017 Jul

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Deep learning for pharmacovigilance: recurrent neural network architectures for labeling adverse drug reactions in Twitter posts.

Cocos Anne A Fiks Alexander G AG Masino Aaron J AJ

Journal of the American Medical Informatics Association : JAMIA 20170701 4

<h4>Objective</h4>Social media is an important pharmacovigilance data source for adverse drug reaction (ADR) identification. Human review of social media data is infeasible due to data quantity, thus natural language processing techniques are necessary. Social media includes informal vocabulary and irregular grammar, which challenge natural language processing methods. Our objective is to develop a scalable, deep-learning approach that exceeds state-of-the-art ADR detection performance in social ...[more]

PMID: 28339747

Similar Datasets

Project description:INTRODUCTION:Key to pharmacovigilance is spontaneously reporting all Adverse Drug Reactions (ADR) during post-market surveillance. This facilitates the identification and evaluation of previously unreported ADR's, acknowledging the trade-off between benefits and potential harm of medications. Only 41% Antiretroviral (ARV) ADR's documented in Harare city clinical records for January to December 2016 were reported to Medicines Control Authority of Zimbabwe (MCAZ). We investigated reasons contributing to underreporting of ARV ADR's in Harare city. METHODS:A descriptive cross-sectional study and the Centers for Disease Control (CDC) guided surveillance evaluation was conducted. Two hospitals were purposively included. Seventeen health facilities and 52 health workers were randomly selected. Interviewer-administered questionnaires, key informant interviews and WHO pharmacovigilance checklists were used to collect data. Likert scales were applied to draw inferences and Epi info 7 used to generate frequencies and proportions. RESULTS:Of the 52 participants, 32 (61.5%) distinguished the ARV ADR defining criteria. Twenty-nine (55.8%) knew system's purpose whilst 28 (53.8%) knew the reporting process. Knowledge scored average on the 5-point-Likert scale. Thirty-eight (73.1%) participants identified ARV ADR's following client complaints and nine (1.3%) enquired clients' medication response. Forty-six (88.5%) cited non-feedback from MCAZ for underreporting. Inadequate ARV ADR identification skills were cited by 21 (40.4%) participants. Reporting forms were available in five (26.3%) facilities and reports were generated from hospitals only. Forty-two (90.6%) clinicians made therapeutic decisions from ARV ADR's. Averaged usefulness score was 4, on the 5-point-Likert scale. All 642 generated signals were committed to Vigiflow by MCAZ, reflecting a case detection rate of 4/ 100 000. Data quality was 0.75-1.0 (WHO) and all reports were causally assessed. CONCLUSION:The pharmacovigilance system was useful, simple, and acceptable despite being unstable, not representative and not sensitive. It was threatened by suboptimal health worker knowledge, weak detection strategies and referral policy preventing ARV ADR identification by person place and time. Revisiting local policy, advocacy, communication and health worker orientation might improve pharmacovigilance performance in Harare city.

Project description:BackgroundThe introduction of immune checkpoint inhibitors (ICIs) in clinical practice has brought significant benefits for patients. Seven ICIs are available in Europe: nivolumab, pembrolizumab, atezolizumab, avelumab, durvalumab, cemiplimab, and ipilimumab. Despite their proven clinical efficacy, these innovative drugs may cause serious immune-related adverse drugs reactions (irADRs). Given the significance of these ADRs for patients' health, we analyzed individual case safety reports (ICSRs) related to ICIs, focusing on those reporting irADRs, collected in the Italian spontaneous reporting database.MethodsWe analyzed ICI-induced irADRs collected in the Italian Pharmacovigilance database (Rete Nazionale di Farmacovigilanza [RNF]) from January 1, 2002, to February 28, 2019, focusing on those reported in the Campania Region. We retrieved from an open-access Italian pharmacovigilance system, the RAM system (for national safety data), and from the RNF (for Campania safety data) all ICSRs reporting ADRs related to ICIs authorized until the analysis date. Focusing on irADRs, we performed descriptive and disproportionality analyses through the reporting odds ratio (ROR) with 95% confidence interval.ResultsNational results. Among 2,088 ICI-related ICSRs, 801 reported irADRs. The majority of such ADRs occurred in male patients reporting gastrointestinal and skin toxicities. Nivolumab and pembrolizumab were drugs most commonly reported as suspect drugs. Compared to other ICIs, ROR was statistically significant for pembrolizumab and ipilimumab.Campania Region results. Out of 253 ICI-related ICSRs sent to Regional Pharmacovigilance Center of Campania Region, 121 reported at least one ICI-induced irADR. These were serious in 37.2% of cases and had an unfavorable outcome in 32.2% of cases. Overall, out of 8 ICSRs reported ADR with a fatal outcome, four reported irADRs. From disproportionality analyses on Campania Region ICSRs, statistically significant ROR emerged only for ipilimumab.ConclusionsOur results showed that during the study period several serious irADRs were reported, some of which had fatal outcome. Given the clinical relevance of irADRs, further investigations in real-life context are necessary for a better characterization of ICIs safety profiles. Oncologists should be trained to early recognize and adequately manage irADRs. Patients should also be educated to immediately report any new symptom or worsening of pre-existed ones during the ICI treatment.

Dataset Information

Deep learning for pharmacovigilance: recurrent neural network architectures for labeling adverse drug reactions in Twitter posts.

Publications

Deep learning for pharmacovigilance: recurrent neural network architectures for labeling adverse drug reactions in Twitter posts.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets