A Machine Learning Pipeline to Analyze Global Sentiment and Factors Influencing Retinoblastoma Treatment Hesitancy: Observational Infodemiology Study.

Authors

Emily S Wong, Richard W Choy, Esther W Tang, Yuzhou Zhang, Xiu Juan Zhang, Linbin Zhou, Wai Kit Chu, Li Jia Chen, Clement C Tham, Chi Pui Pang, Jason C Yam

Published in

Journal of medical Internet research. Volume 28. Pages e73364. Aug 11, 2026. Epub Aug 11, 2026.

Abstract

The use of social media in cancer research, patient support, and information sharing has been well documented.
Using retinoblastoma as a model, we use the information provided from Twitter (subsequently rebranded X) to understand patients' treatment-seeking behavior and barriers, as well as investigate its application in research and epidemiology for rare diseases.
Posts on retinoblastoma were extracted from Twitter. We trained BERT (Bidirectional Encoder Representations from Transformers) models to identify relevance and conducted sentiment analysis. The hierarchical Dirichlet process was trained to identify topics with BERTopic used as a sensitivity analysis. We enriched user profiles with OpenStreetMap-based geotagging and CoreNLP-based occupation tagging. Retinoblastoma outcomes were obtained from a systematic review and meta-analysis, which covered articles published between January 1, 1981, and October 8, 2021.
The dataset covered 2,382,511 posts from 797,870 Twitter users. Most of the information dissemination and discussion originated from North America and Western Europe. A lack of disease understanding and the need for more support and counseling remained the most significant barriers to receiving treatment worldwide, as reflected by both the intensity and number of posts. The number of new posts per year related to treatment barriers and enucleation hesitancy showed exponential growth after 2016 (βlog-linear=0.957, P=.002). For the perceived barriers to treatment, sentiment was the strongest over time for worries over treatment failure (βlinear=-0.003, P=.79, estimate2022=0.679). Posts with higher negative sentiment intensity related to enucleation were concentrated in Central and Southern America, Asia, and Africa. Stronger negative sentiment toward enucleation (β=-0.726, 95% CI -1.224 to -0.228) was associated with worse overall survival outcomes. The association between lower overall survival rates and enucleation hesitancy was observed in Asia (β=-1.518, 95% CI -2.602 to -0.434) and Africa (β=-0.812, 95% CI -1.412 to -0.021) in the subgroup analysis. The active participation of clinical staff (β=-0.105, 95% CI -0.186 to -0.024; P=.01) and academia (β=-0.116, 95% CI -0.208 to -0.024; P=.01) on retinoblastoma topics on Twitter correlated with lower enucleation hesitancy.
Computational social media analysis can generate actionable insights for public health interventions for retinoblastoma. Negative sentiment toward enucleation is associated with poorer survival. The active participation of clinical staff and academia on Twitter is correlated with lower enucleation hesitancy. However, they remain underrepresented in social media discussions, suggesting a significant opportunity for greater engagement from stakeholders and targeted information dissemination to improve acceptance and outcomes in vulnerable zones.

PMID:
42579819
Bibliographic data and abstract were imported from PubMed on 12 Aug 2026.

Read full publication at:
Please sign in to see all details.

Stats

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.