gms | German Medical Science

21. Deutscher Kongress für Versorgungsforschung

Deutsches Netzwerk Versorgungsforschung e. V.

05.10. - 07.10.2022, Potsdam

On the potential of social media data in health services research – using the example of a Psoriasis patient forum

Meeting Abstract

Search Medline for

  • Lukas Westphal - Institute for Health Services Research in Dermatology and Nursing (IVDP), University Medical Center Hamburg-Eppendorf, Hamburg, Deutschland
  • Rachel Sommer - Institute for Health Services Research in Dermatology and Nursing (IVDP), University Medical Center Hamburg-Eppendorf, Hamburg, Deutschland
  • Juliane Traxler - Institute for Health Services Research in Dermatology and Nursing (IVDP), University Medical Center Hamburg-Eppendorf, Hamburg, Deutschland

21. Deutscher Kongress für Versorgungsforschung (DKVF). Potsdam, 05.-07.10.2022. Düsseldorf: German Medical Science GMS Publishing House; 2022. Doc22dkvf260

doi: 10.3205/22dkvf260, urn:nbn:de:0183-22dkvf2601

Published: September 30, 2022

© 2022 Westphal et al.
This is an Open Access article distributed under the terms of the Creative Commons Attribution 4.0 License. See license information at http://creativecommons.org/licenses/by/4.0/.


Outline

Text

Background and state of (inter)national research: In our digitized world, people increasingly share and search for health-related information on the internet. These activities leave digital traces that open up new ways for scientists to obtain large amounts of timely data that are valuable resources for studies on health behavior, epidemiology, or health services research. A frequently used data source is social media, were many patients share their illness and treatment experiences in online health communities. Consideration of these first-person perspectives fits well with a modern understanding of person-centered care, which is oriented toward the needs of the individual. For this study, an online forum for people with psoriasis was scraped and posts that relate to the psychosocial burden of the disease were subsetted to examine the most prevalent topics discussed within this domain. The emerging topics represent an inductive and data-driven account of patient reports and are potentially less susceptible for socially desirable responding or existing preconceptions.

Research question and objective: What are the most common topics related to psychosocial burden among people with psoriasis, and to what extent do they match those identified in previous research?

Method or hypothesis: An online forum for people with psoriasis was scraped with explicit consent of the patients’ association that runs the forum. Over 53.000 posts from between January 2015 and April 2022 were collected. After removing punctuation, digits, and URLs, the posts were tokenized, lemmatized, and stopwords removed. Subsequently, posts related to psychosocial burden were selected using a semi-automated keyword retrieval and document selection approach developed by King, Lam, and Roberts [1]. In a final step that is yet to be completed, topic models will be computed and qualitatively compared to existing themes in the literature of psychosocial burden in patients with psoriasis.

Results: The analyses have not yet been completed at this point in time. Either Latent Dirichlet Allocation (LDA) or Structural Topic Models (STM) will be used. Both fall into the realm of unsupervised machine learning.

Discussion: Based on our results, the applicability of social media data to assess psychosocial burden will be discussed. Subsequently, the perspective will be broadened and the potential of using online data in health services research will be evaluated with a special focus on common pitfalls, ethical concerns, and its potential to be used in hypothesis-generating research.

Practical implications: Social media can be a valuable, timely, and rich source of data for addressing public health issues. The potential of which is only starting to be acknowledged.


References

1.
King G, Lam P, Roberts ME. Computer‐Assisted Keyword and Document Set Discovery from Unstructured Text. Am J Pol Sci. 2017;61(4):971-88.