Dataset Information

Filtering Entities to Optimize Identification of Adverse Drug Reaction From Social Media: How Can the Number of Words Between Entities in the Messages Help?

ABSTRACT:

Background

With the increasing popularity of Web 2.0 applications, social media has made it possible for individuals to post messages on adverse drug reactions. In such online conversations, patients discuss their symptoms, medical history, and diseases. These disorders may correspond to adverse drug reactions (ADRs) or any other medical condition. Therefore, methods must be developed to distinguish between false positives and true ADR declarations.

Objective

The aim of this study was to investigate a method for filtering out disorder terms that did not correspond to adverse events by using the distance (as number of words) between the drug term and the disorder or symptom term in the post. We hypothesized that the shorter the distance between the disorder name and the drug, the higher the probability to be an ADR.

Methods

We analyzed a corpus of 648 messages corresponding to a total of 1654 (drug and disorder) pairs from 5 French forums using Gaussian mixture models and an expectation-maximization (EM) algorithm .

Results

The distribution of the distances between the drug term and the disorder term enabled the filtering of 50.03% (733/1465) of the disorders that were not ADRs. Our filtering strategy achieved a precision of 95.8% and a recall of 50.0%.

Conclusions

This study suggests that such distance between terms can be used for identifying false positives, thereby improving ADR detection in social media.

SUBMITTER: Abdellaoui R

PROVIDER: S-EPMC5500778 | biostudies-literature | 2017 Jun

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Filtering Entities to Optimize Identification of Adverse Drug Reaction From Social Media: How Can the Number of Words Between Entities in the Messages Help?

Abdellaoui Redhouane R Schück Stéphane S Texier Nathalie N Burgun Anita A

JMIR public health and surveillance 20170622 2

<h4>Background</h4>With the increasing popularity of Web 2.0 applications, social media has made it possible for individuals to post messages on adverse drug reactions. In such online conversations, patients discuss their symptoms, medical history, and diseases. These disorders may correspond to adverse drug reactions (ADRs) or any other medical condition. Therefore, methods must be developed to distinguish between false positives and true ADR declarations.<h4>Objective</h4>The aim of this study ...[more]

PMID: 28642212

Similar Datasets

Project description:IntroductionPhysical distancing (PD) is an important public health strategy to reduce the transmission of COVID-19 and has been promoted by public health authorities through social media. Although youth have a tendency to engage in high-risk behaviors that could facilitate COVID-19 transmission, there is limited research on the characteristics of PD messaging targeting this population on social media platforms with which youth frequently engage. This study examined social media posts created by Canadian public health entities (PHEs) with PD messaging aimed at youth and young adults aged 16-29 years and reported behavioral change techniques (BCTs) used in these posts.MethodsA content analysis of all social media posts of Canadian PHEs from Facebook, Twitter, Instagram and YouTube were conducted from April 1st to May 31st, 2020. Posts were classified as either implicitly or explicitly targeting youth and young adults. BCTs in social media posts were identified and classified based on Behavior Change Technique Taxonomy version 1 (BCTTv1). Frequency counts and proportions were used to describe the data.ResultsIn total, 319 youth-targeted PD posts were identified. Over 43% of the posts originated from Ontario Regional public health units, and 36.4 and 32.6% of them were extracted from Twitter and Facebook, respectively. Only 5.3% of the total posts explicitly targeted youth. Explicit posts were most frequent from federal PHEs and posted on YouTube. Implicit posts elicited more interactions than explicit posts regardless of jurisdiction level or social media format. Three-quarters of the posts contained at least one BCT, with a greater portion of BCTs found within implicit posts (75%) than explicit posts (52.9%). The most common BCTs from explicit posts were instructions on how to perform a behavior (25.0%) and restructuring the social environment (18.8%).ConclusionsThere is a need for more PD messaging that explicitly targets youth. BCTs should be used when designing posts to deliver public health messages and social media platforms should be selected depending on the target population.

Project description:BACKGROUND:Incidence rate of cancer is increasing worldwide, with longer life expectancy being one of the main causes. Yet, between 30% and 50% of cancer cases are preventable, and early detection contributes to a better prognosis. This makes health communication strategies essential. Facebook, the world's most used social networking site in 2017 and 2018, can be a useful tool for disseminating powerful messages on health promotion, prevention, and early detection. OBJECTIVE:We aimed to (1) offer ways of optimizing health messages about cancer on Facebook, focusing on topics, such as risk factors, prevention, treatment, early diagnosis, and cure, and (2) investigate which aspects of these messages generate greater engagement. METHODS:To verify what generates greater engagement in topics related to cancer on Facebook, we analyzed 16 Brazilian pages with the main theme of cancer. We performed a manual analysis of texts, content, and engagement rates. Finally, we developed a software program to operationalize the analysis of Facebook posts. The tool we devised aims to automate the analysis of any Facebook page with cancer as the main theme. RESULTS:We analyzed 712 posts over a 1-month period. We divided the posts into the following 8 categories: "Testimonies or real-life stories," "Solidarity," "Anniversaries," "Science and health," "Events," "Institutional," "Risk factors," and "Beauty." The pages were also organized into groups according to the type of profile to which they belonged (ie, hospitals or foundations, informative, nongovernmental organizations, and personal pages).The results showed that the categories generating greater engagement in Brazil were not those with the highest percentage of cancer-related content. For instance, in the "Informative" group the "Testimonies or real-life stories" category generated an engagement of 79.5%. However, only 9.5% (25/261) of the content within the relevant time period dealt with such topics. Another example concerns the category "Science and health." Despite being the one with the highest number of posts (129/261, 49.4%), it scored 5th in terms of engagement. This investigation served as the basis for the development of a tool designed to automate the analysis of Facebook pages. The list of categories and keywords generated by this analysis was employed to feed the system, which was then able to categorize posts appearing on a Facebook page. We tested the system on 163 posts and only 34 were classified incorrectly, which amounts to a 20.8% error rate (79.2% accuracy). CONCLUSIONS:The analysis we conducted by categorizing posts and calculating engagement rates shows that the potential of Facebook pages is often underutilized. This occurs because the categories that generate the greatest engagement are often not those most frequently used. The software developed in this research may help administrators of cancer-related pages analyze their posts more easily and increase public interest as a result.

Project description:Background: The importance of print health literacy is widely recognized, but oral literacy has been largely ignored. Because much health information is conveyed via spoken word, an inaccurate or incomplete comprehension of spoken health messages may have important consequences. This study explored the extent to which listeners understood critical concepts in spoken messages about cancer prevention and screening.Methods: Forty-four adults from three health plans took part in a 1-hour interview. Participants viewed six brief media clips from TV or the Web about cancer prevention and screening. Each clip contained multiple messages. Participants paraphrased the clips? main points and key concepts. We coded the accuracy of participants? responses with respect to the message content.Results: Of 44 participants, aged 41 to 70, 52% were female; 48% were non-white; and 5% had a high school education or less. Messages were generally understood by most participants but some participants misunderstood critical cancer prevention concepts. Nine of 44 could not define the term ?at risk.? Others provided approximately accurate synonyms, such as ?susceptible to,? or ?inclined to,? or gave examples of risk factors (e.g. fair skin for skin cancer) that indicated a partial understanding of the concept. In response to a news report about a comprehensive cancer study synthesizing the results of 7,000 clinical trials, 10% of the participants viewed the study size as small, or mistook the number of trials for number of patients. Some participants had trouble distinguishing ?screening? and ?prevention,? apparently believing that screening is inherently preventive.Conclusions: Most participants in this sample of moderately to well-educated adults understood the main points contained in spoken messages about cancer screening and prevention. However, important concepts such as ?at risk? (applicable to conditions besides cancer), were sometimes misunderstood. Similarly, some participants had difficulty understanding the strength of research evidence and the value of multiple studies. Comprehension depends on foundational knowledge, which even educated lay people may be missing. Speakers, whether news anchors or providers?cannot assume that listeners understand critical health concepts even if the words themselves seem simple.

Project description:Background: The UK print and online media is an important channel by which scientific research is communicated to the public. Media risk messages relating to pregnancy or fertility contribute to the context of reproductive decision making, but their fidelity to the underlying science has been questioned. Method: We measured the volume, distribution and content of science-based risk headlines relating to pregnancy or fertility in the UK media over four months. We grouped headlines into unique stories and categorised them by exposure and outcome of interest. We selected four unique stories for closer content analysis and assessed their fidelity to the underlying science, with attention to the role of press releases. Results: We identified 171 headlines over four months (average 43 per month), comprising 56 unique stories. The unique stories most commonly concerned maternal risk factors (n=46) and child health outcomes (n=46). Maternal health outcomes were less frequently the focus (n=20). The most common risk factors in the media coverage were maternal food and drink (n=15), maternal medication and medical interventions (n=9), and maternal health factors (n=6). Media reports were largely faithful to press releases. Where substantive deviations from the underlying scientific study were identified, these could mostly be traced back to press releases or quotes from the study's authors. Press releases often omitted caveats which were reinstated at the media reporting stage, alongside additional expert criticism. Conclusions: Frequent science-based risk messages in the UK media frame mothers as vectors of potential harm to children, who are the focus of health outcomes. Largely, the media does not introduce misinformation, but reports press releases faithfully with additional caveats and expert commentary. Press releases fulfil an interpretative role, often omitting caveats and introducing new elements and advice to women. Their role as a bridge between scientific and lay audiences is discussed.

Dataset Information

Filtering Entities to Optimize Identification of Adverse Drug Reaction From Social Media: How Can the Number of Words Between Entities in the Messages Help?

Background

Objective

Methods

Results

Conclusions

Publications

Filtering Entities to Optimize Identification of Adverse Drug Reaction From Social Media: How Can the Number of Words Between Entities in the Messages Help?

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets