Large scale social media analytics on urinary incontinence: insights from Chinese online user generated contents

Fan Y1, Huang W2, Dong Q1, Luo D1

Research Type

Clinical

Abstract Category

Female Stress Urinary Incontinence (SUI)

Abstract 191
Personal and Social Dimensions of Incontinence
Scientific Podium Short Oral Session 24
Thursday 28th September 2023
16:15 - 16:22
Room 103
Female Stress Urinary Incontinence Urgency Urinary Incontinence
1. Department of Urology, Institute of Urology, West China Hospital, Sichuan University, No. 37 Guo Xue Xiang, Chengdu, Sichuan, 610041, P.R. China, 2. School of Computer Science and Engineering, University of Electronic Science and Technology of China, Chengdu, Sichuan, China
Presenter
Links

Abstract

Hypothesis / aims of study
Urinary incontinence (UI) is a prevalent condition globally that negatively impacts quality of life (QoL). The exact prevalence of UI is difficult to quantify due to patients’ stigma and people’s perception of UI as either a disease or just annoyance. Social media data is instant, conversational, and large in scale and thus an alternative resource for real-time public health surveillance. This study aimed to evaluate public opinions, experiences and sentiment on UI based on Chinese social media data, with the goal of identifying unmet needs in dealing with UI symptoms that require further medical resource investments.
Study design, materials and methods
We extracted Chinese social media posts from mainstream Chinese platforms (Weibo, Xiaohongshu, and Baidu Tieba) between 2019 and 2022 using a python web crawler and manual screening. We developed search terms that aimed to exhaustively include posts about UI while exclude ads, official accounts of organizations, and entities other than adults. Latent Dirichlet Allocation (LDA), a generative machine learning model, was used to identify hidden topics and themes in the posts, and sentiment analysis was conducted for each post based on the emotion dictionary. We assigned positive or negative emotion to each theme category using the Wilcoxon signed rank test. We assessed the quality of posts containing medical information using a self-designed information quality scale (ranging from -6 to 10 with higher points indicating better quality in terms of reliability, readability, accuracy, time effectiveness and usefulness) and compared them to posts of the same topics generated by ChatGPT, an AI language model. We also evaluated the temporal and spatial distribution of posts using a stacked line graph and a geographical map. The number of posts for each province was adjusted by local resident population to generate the hotspot indexes that measured the social attention to UI. All statistical analyses were conducted using R program (version 4.2.2).
Results
A total of 24,770 Chinese social media posts were extracted, and six themes of UI were identified: 1. impact on the quality of life (5,908); 2. traditional Chinese treatments (4,793); 3. female stress urinary incontinence (4,651); 4. neurogenic lower urinary tract dysfunction (4,408); 5. etiology of UI (3,348); and 6. senile incontinence (1,662). The sentiments related to themes 1, 2, 4 and 6 were all negative (P < 0.001), and 95% confidence intervals (CI) for the negative score were (-0.50, -0.00), (-1.50, -1.00), (-7.50, -6.50), and (-2.50, -1.50) respectively. Twenty representative posts (user generated content, UGC) from theme 3 and twenty topic matched ChatGPT posts (artificial intelligence generated content, AIGC) were assessed for their quality. The mean information quality score for AIGC and UGC is 6.6 and 3.1, respectively [95% CI of the mean difference: (1.68, 5.32)]. 
There was no significant trend regarding UI among social media platforms since 2019 (Figure 1). Traditional media channels like TV and newspaper reports were the main drivers for temporary social media spikes. The high spike during December 2022 was mainly due to a surge of posts belonging to theme 6, positively correlated with the rapid growth of COVID-19 patients in China under the loosened quarantine policy. Spatial distribution of the posts and hotspot indexes showed discrepancies among developed and underdeveloped regions, with underdeveloped regions showing lower attention to UI (Figure 2). Regions that previously had no report on the prevalence of UI also displayed hotspot indexes larger than 1, indicating a nationwide attention to and prevalence of UI.
Interpretation of results
This is to our knowledge the first comprehensive analysis of mainstream Chinese social media data on UI. Topic analysis revealed a new theme that haven’t been reported before, i.e. the Chinese traditional therapies, especially acupuncture, as promising alternative treatments for UI. Sentiment analysis showed that compared with female and senile incontinence, neurogenic lower urinary tract dysfunction was described with more negative emotion and more devastating for patients. Quality of posts containing medical information about UI were poor, particularly in terms of accuracy and time effectiveness. Meanwhile, ChatGPT was good at generating readable, accurate and useful information, showing its potential in future patient education. 
Visualization of the number of UI posts against time and space showed that the real-life prevalence of UI may well exceed the one drawn upon from national epidemiological surveys. Abnormal spikes in the timeline identified “outliers”, hinting that UI might be a comorbidity of COVID-19 infection among elderly patients, which was a meaningful hypothesis that warrants further investigation. With the rapid economic development of western China, the demand for medical resources to manage UI may increase. Further cost benefit analysis could be done to determine the optimal supply of pelvic reconstructive surgery training for urologists in western China.
Concluding message
Social media data can be a valuable resource for monitoring public perceptions of UI. Our findings suggest a need for improved public education and more targeted medical resource allocation. ChatGPT has the potential for better patient education, and urology professionals should consider using it to enhance public understanding of UI and improve patients’ compliance and QoL.
Figure 1 The stacked line graph of the number of posts against the posted time, with different colors indicating content themes 1 to 6 identified by LDA.
Figure 2 The geographical map showed the hotspot indexes of each province, with higher index indicating higher social media attention on UI.
Disclosures
Funding This study was supported by natural science foundation of Sichuan Province, No. NSFSC1308. Clinical Trial No Subjects Human Ethics Committee The current study was approved by the Institutional Review Board of West China Hospital and was determined exempt from review. Helsinki Yes Informed Consent No
Citation

Continence 7S1 (2023) 100909
DOI: 10.1016/j.cont.2023.100909

14/11/2024 03:22:51