Identity Signaling Strategies Among Medical Professionals on Twitter: Mixed-Methods Study (Preprint)

Abstract

BACKGROUND The increased use of social media by physicians has created new opportunities and challenges for projecting medical expertise and engaging with the public. Twitter's massive reach and information diffusion-oriented architecture make it a key space for understanding how physicians construct their online identities and establish credibility. OBJECTIVE This study aimed to investigate how physicians use language and platform features to construct their online identities and establish themselves as part of the medical community on Twitter using content analysis. It also explored how computational methods, such as sentiment analysis and topic modeling, can provide further insights into these practices. Additionally, it sought to establish a baseline understanding of legitimate communication among credible physicians on Twitter METHODS A dataset of 600 tweets from 30 physicians with "MD" credentials and over 10,000 followers was collected. Content analysis was conducted to identify prevalent themes in the tweets, resulting in eight main categories. Inter-rater reliability was assessed using Cohen's Kappa (κ = 0.6350, substantial agreement). Computational methods, including code co-occurrence analysis using phi correlation coefficients, sentiment analysis using Valence Aware Dictionary and sEntiment Reasoner (VADER), and topic modeling using Latent Dirichlet Allocation (LDA), were employed to further explore the dataset. RESULTS Qualitative analysis revealed eight primary codes in the tweets: networking (36.1%, n = 217), sharing news/articles (35.8%, n = 215), professional topics (33.6%, n = 202), personal updates (25.0%, n = 150), commentary on current events (16.3%, n = 98), job/academic achievements (13.3%, n = 80), advocacy (11.2%, n = 67), and patient care/medical advice (5.0%, n = 30). Analysis of platform features showed physicians actively leveraging Twitter's affordances to construct and communicate their professional identities. Computational methods provided insights into code co-occurrences, with the strongest associations between "Personal" and "News/Articles" (correlation = 1.060), "Personal" and "Professional" (correlation = 1.046), and "CE Commentary" and "Networking" (correlation = 1.037). Sentiment analysis revealed variations in emotional tone across codes, with the most positive sentiment associated with "Networking" (0.604) and its combinations with "Achievements" (0.422) and "Personal" (0.419). LDA topic modeling identified eight latent topics within the corpus, capturing specific focus areas such as healthcare leadership and achievements, medical education, and information sharing. CONCLUSIONS Physicians employ many strategies to construct their online identities on Twitter while navigating the challenges of context collapse and the blurring of personal and professional boundaries. The study highlights the value of combining qualitative and computational approaches to gain a more thorough understanding of physician behavior and communication patterns online. By establishing a baseline understanding of legitimate communication among credible physicians on Twitter, the findings lay the groundwork for future research on medical misinformation and have implications for developing evidence-based strategies to support physicians in maintaining a professional presence on social media.