The total number of users of social media continues to grow worldwide, resulting in the generation of vast amounts of data. Popular social networking sites such as Facebook, Twitter and Instagram dominate this sphere. According to estimates, 500 million tweets and 4.3 billion Facebook messages are posted every day.1 A Pew Research Report on Social Media2 estimates that nearly half of adults worldwide and two-thirds of all American adults (65%) use social networking. The report states that of the total users, 26% have discussed health information, and, of those, 30% changed behavior based on this information and 42% discussed current medical conditions. Advances in automated data processing, machine learning and NLP present the possibility of utilizing this massive data source for biomedical and public health applications, if researchers adequately address the methodological challenges unique to this media.
Despite numerous published studies, however, myths and misconceptions persist about the suitability and adequate use of these data, impacting the perception of researchers, institutional review boards, and the general public on the validity of the studies for health research. In this talk, we will discuss (and hopefully debunk!) some of the more poignant myths and misconceptions.

This talk will explore the wealth of textual ‘big data’ about our health in social media and the role of Natural Language Processing (NLP) in extracting useful patterns to help inform new scientific discoveries. I will exemplify the technical challenge by focussing on a range of approaches to automatically link mentions of everyday language to large-scale scientific vocabularies such as SNOMED CT.

Text mining of scientific libraries and social media has already proven itself as a reliable tool for drug repurposing and pharmacovigilance. Recent attempts have been made to utilize social media resources, such as askapatient.com, to provide monitoring of a population's health status. In this talk, I overview a task of medical concept recognition discussing in particular social media messages. This task aims to extract medical concepts in real conditions: given a set of user messages, the system has to find health-related entity mentions in a free-form text and map them to a certain medical concept (disease, adverse drug reaction, etc.). In other words, the system translates a text written in 'social media language' to 'formal medical language'. The latter can be used for identification of mentions of certain terms, extraction of drug-disease, drug-effect, or disease-symptom relations, to name a few. The objective is challenging due to the differences in medical terminology between health care professionals and social media texts coming from the lay public. Simple keyword matching approaches may not capture the semantics of user messages. I will describe the medical concept normalization task, the challenges it poses and its relation to current research in natural language processing (NLP).

We will present tools developed by the OntoGene group (University of Zurich and Swiss Institute of Bioinformatics) for biomedical text mining over the scientific literature and social media.

The OntoGene Entity Recognizer (OGER) is a fast, efficient, accurate Named Entity Recognition application which is based on terminologies derived from reference knowledge repositories (e.g. SwissProt, NCBI taxonomy, EntrezGene, Gene Ontology, UMLS, etc). OGER has been shown to be among the best tools for biomedical NER through participation in several scientific competitive evaluation.

The OntoGene group has recently participated in a shared task on social media mining (SMM4H) where it has achieved best results in one of the challenges. Detailed results and methods will be presented.

Source format variability and noise are major challenges when harvesting content from social media. This presentation discusses methods and abstractions for gathering user generated content from Web pages and social media platforms covering (i) structured content, (ii) platforms that leverage Semantic Web standard such as Microformats, RDFa and JSON-LD, and (iii) semi-structured or even unstructured content that is typically found in Web forums.
We then discuss pre-processing and anonymization tasks and outline how the collected content is annotated, aggregated and summarized in a so called contextualized information space. An interactive dashboard provides efficient means for analyzing, browsing and visualizing this information space. The dashboard supports analysts in identifying emerging trends and topics, exploring the lexical, geospatial and relational context of topics and entities such as health conditions, diseases, symptoms and drugs, and performing drill-down analysis to shed light on individual posts and statements that cause the observed effects.

The identification of fit-for-purpose outcome measures for clinical trials relies on an authentic and detailed understanding of the lived experience of disease from the patient perspective. The Food and Drug Administration patient-reported outcome (PRO) guidance provides standards for clinical outcome assessment (COA) development, but these standards bring scientific and logistical challenges which can result in a lengthy and expensive instrument development process. Thus, more pragmatic methods are needed alongside traditional approaches. Traditional approaches to eliciting concepts of interest include in-depth qualitative interviews with patients and families, observational research and surveys. However, while these approaches benefit from the rich insights provided, they can be expensive and time-consuming to conduct. Social media data provide a novel approach to understanding the lived experience of a disease and are increasingly utilized by researchers seeking rapid yet reliable insights. Although these methods are gaining traction, there remain outstanding challenges related to the reliability and specificity of insights obtained. Through real-world examples, this presentation will explore the benefits and limitations of traditional versus novel concept elicitation approaches with a specific emphasis on social media data. The presentation will focus on a number of key areas including practical and ethical considerations, regulatory acceptance and scalability.

Background: Online forums are commonly used by patients to share disease information, grievances and for seeking mutual support. However, social media analysis has seldom been applied to evaluate patients’ perspective on COPD.
Aim: To use social media data for understanding the patients’ perspective on COPD with respect to symptoms, unmet needs, and impact on their health-related quality of life (HRQoL).
Methods: A social media data aggregator tool was used to download relevant records posted between July 2016 and Jan 2018 using predefined keywords. Data sources were forums, blogs, Twitter and newswires. Post data anonymization, text algorithms and manual curation techniques were used to analyse and describe the impact of COPD symptoms on patient’s HRQoL.
Results: 695 unique patient records were considered for the analysis. Patients’ discussions were primarily on symptoms, diagnosis, treatments and HRQoL. The most commonly reported symptoms were cough (25%), mucous (23%), and shortness of breath (23%). These symptoms impacted patients’ HRQoL by affecting their sleep, mobility, work, inducing panic attacks and emotional wellbeing.
Conclusions: Social media sourced data (unfiltered, uninfluenced) can generate useful insights on patients’ experiences with COPD and identify unmet needs. This can inform decision-making in early drug development and guide discussions with regulatory bodies.

Background: In drug development patient-reported outcomes (PROs) are now routinely collected in clinical trials. Semantic analysis based on natural language processing (NLP) offers the opportunity to generate insights from unstructured patient conversations in online forums and social media to inform the design of clinical trials regarding PROs.
We aimed to investigate whether NLP would allow us to:
1) identify a small sub-population of patients with distinct features,
2) describe and quantify the symptom burden.
Methods: We studied 180,000 patient conversations in 2 online forums and > 2mn tweets about Systemic Lupus Erythematosus, a severe autoimmune disease. The algorithm was refined to identify conversations of Lupus Nephritis, a rare sub-population.
Results: We identified 7,562 forum conversations and 354 tweets relating to Lupus Nephritis. The five most frequently mentioned symptoms were pain, swelling of body parts, rash, arthritis and fatigue. The symptoms were linked to body-parts. Qualitative analysis of randomly chosen conversations showed that some conversations were false-positive.
Discussion: NLP is a powerful tool to identify relevant PROs and thereby inform the design of clinical trials, even for rare diseases. Further analyses are required to understand the sensitivity and specificity of the approach.