Quality and reliability of ground-glass nodule-related short Chinese videos on TikTok and Bilibili: a cross-sectional content analysis

Abstract

Background

The widespread use of chest computed tomography (CT) has substantially increased the detection of ground-glass nodules (GGNs). This often causes significant patient anxiety. While most GGNs are slow-growing, misinformation or incomplete guidance on social media can worsen “scan anxiety”. This may lead to demands for unnecessary overtreatment or result in poor adherence to surveillance protocols. This study evaluated the content, quality, and reliability of GGN-related short videos on TikTok and Bilibili to determine their utility for patient education.

Methods

We searched both platforms using the keyword “ground-glass nodules” (“GGNs”) between September 30–October 8, 2025. We analyzed the top 130 videos per platform. We classified uploaders as professionals (surgeons, radiologists, internists including traditional Chinese medicine physicians) or patients. Content was coded for etiology, imaging, diagnosis, treatment, and follow-up. Video quality and reliability were assessed using the Global Quality Score (GQS, 1–5) and modified DISCERN (mDISCERN). Two physicians rated all videos independently, with adjudication by a senior clinician. Nonparametric tests and Spearman correlations were applied (two-sided P<0.05).

Results

A total of 237 videos were included (TikTok, n=125; Bilibili, n=112). Content analysis revealed significant information gaps: while 92.83% of videos discussed treatment options (often emphasizing surgery), only 16.88% explained GGN etiology, and systematic guidance on risk stratification was frequently lacking. Professionally produced videos (surgeons/radiologists) scored significantly higher than patient-generated content. Although Bilibili had higher median GQS scores (3.00 vs. 2.00, P<0.001) than TikTok, the overall reliability (mDISCERN) across both platforms was modest, with no significant difference. Engagement metrics (likes/shares) did not correlate with medical quality.

Conclusions

Current short-video algorithms prioritize engagement over clinical accuracy, resulting in fragmented health information that may distort patients’ risk perception of GGNs. While professionals produce higher-quality content, the overall reliability remains suboptimal. Clinicians must be aware of these online information deficits to proactively address patient anxiety and correct misconceptions during consultations, ensuring adherence to evidence-based surveillance pathways.

Highlight box.

Key findings

• This analysis of 237 short videos revealed that while healthcare professionals produce higher-quality content than non-professionals, the overall reliability of information regarding ground-glass nodules (GGNs) on TikTok and Bilibili remains suboptimal.

• Bilibili videos demonstrated higher Global Quality Scores (GQS) compared to TikTok, but user engagement metrics (such as likes and shares) showed no correlation with medical accuracy on either platform.

What is known and what is new?

• The widespread detection of GGNs via computed tomography often triggers significant patient anxiety, prompting individuals to seek health information on social media platforms where misinformation is a known concern.

• This study is the first to systematically evaluate GGN-related content on China’s dominant video platforms. It identifies a critical disconnect where algorithms prioritize engagement over accuracy and highlights significant information gaps, specifically a lack of guidance on etiology and risk stratification despite a heavy emphasis on treatment.

What is the implication, and what should change now?

• Clinicians must proactively address the information deficits and alarmist narratives patients encounter online during consultations. Additionally, platform developers should refine algorithms to better balance user engagement with professional medical accuracy to prevent the spread of fragmented health information.

Introduction

Ground-glass nodules (GGNs) are frequently identified on high-resolution computed tomography (HRCT). With advances in imaging, detection rates have increased, and GGNs now serve as key markers for early lung cancer screening (1). Most GGNs are benign, yet a subset carries malignant potential. This creates diagnostic and management challenges for clinicians (2). Optimizing risk assessment and care pathways for GGNs remains a priority in thoracic medicine.

Social media has become a major channel for health communication (3,4). Short videos offer strong visual clarity and rapid dissemination. Consequently, they are increasingly used for medical information (5). High-resolution visuals can improve the understanding of disease status and treatment options. Interactive functions enable communication regardless of time or place, and they may support better adherence among viewers exposed to structured education (6,7). Algorithm-driven recommendations can increase delivery efficiency to defined patient groups. However, they may also raise concerns about “information cocoons” (8). In the context of GGNs, these algorithms may inadvertently trap users in a feedback loop of alarmist content or anecdotal experiences, thereby amplifying health anxiety (“scan anxiety”) rather than improving literacy. Understanding how these platforms filter and present medical information is critical for clinicians attempting to counteract misinformation.

Content quality and reliability vary widely and may influence understanding and decisions (9). Rigorous evaluation of GGN-related short videos is needed to inform public health literacy. This study systematically analyzes GGN-related videos on Bilibili and TikTok, assesses content coverage, quality, and reliability, and examines differences based on the uploader background. The goal is to provide patients and clinicians with accurate, trustworthy information and to inform future strategies for digital health education.

Methods

Ethics

This study used only publicly accessible videos from TikTok and Bilibili. No clinical data, human specimens, or animal experiments were involved. No personal identifiers were collected and no user interaction occurred. Under these conditions, formal ethics review was not required.

Video identification and selection

We conducted a cross-sectional search for “ground-glass nodules” (“GGNs”) on Bilibili and TikTok between September 30 and October 8, 2025. To reduce personalization bias, searches were performed while logged out. This approach ensured that the retrieved videos represented the default content pushed by platform algorithms to a general user rather than results tailored to a specific viewing history. We retrieved the top 130 videos to exceed the typical user attention span (usually limited to the top 50–100 results), thereby capturing the most visible and influential content. We specifically chose the keyword GGNs rather than broader terms (e.g., “pulmonary nodule”) to ensure specificity. Since the study aimed to evaluate information relevant to patients who had already received a radiological diagnosis, using the precise diagnostic term minimized the inclusion of irrelevant content related to purely solid or calcified nodules, which fall outside the scope of GGNs management. Exclusion criteria were: (I) content unrelated to GGN diagnosis or management (e.g., pure pathology demonstrations or operative footage); (II) overt self-promotional content by physicians; (III) missing authorship or title. Figure 1 outlines the screening workflow. For included videos, we extracted Uniform Resource Locators (URLs), uploader identity, likes, saves/collections, shares, comments, and video length, and recorded these in Excel (Microsoft Corp).

Figure 1.

Flowchart of video selection for GGN analysis on TikTok and Bilibili. GGN, ground-glass nodule.

Uploader classification and content coding

Uploaders were classified as professional or non-professional. Professionals included internists (pulmonologists and traditional Chinese medicine physicians), surgeons, and radiologists. Non-professionals were primarily patients. We coded content elements covering etiology, representative imaging, imaging manifestations, nodule size, diagnosis, treatment, and follow-up (Table 1).

Table 1. General information, quality and reliability scores, and content of GGN videos on TikTok and Bilibili.

Characteristics	Total	Bilibili	TikTok	Statistic	P
General information
Number	237	112	125
Video length	130.00 (72.00, 228.00)	158.50 (102.25, 291.00)	101.00 (47.00, 193.00)	Z=−4.46	<0.001
Likes	267.00 (39.00, 2,556.00)	37.50 (9.75, 85.25)	2,237.00 (646.00, 8,499.00)	Z=−12.03	<0.001
Collections	76.00 (13.00, 712.00)	16.00 (5.00, 50.75)	636.00 (90.00, 2,307.00)	Z=−9.84	<0.001
Comments	62.00 (3.00, 393.00)	3.00 (1.00, 12.00)	310.00 (127.00, 1,071.00)	Z=−11.86	<0.001
Shares	31.00 (6.00, 355.00)	7.00 (1.00, 27.25)	273.00 (31.00, 1,014.00)	Z=−9.19	<0.001
Quality and reliability
GQS	3.00 (2.00, 3.00)	3.00 (3.00, 3.00)	2.00 (2.00, 3.00)	Z=−6.41	<0.001
	2.62±0.72	2.91±0.62	2.35±0.70	t=6.51	<0.001
mDISCERN	2.00 (2.00, 3.00)	2.00 (2.00, 3.00)	2.00 (2.00, 3.00)	Z=−1.02	0.31
	2.40±0.93	2.48±0.84	2.32±1.00	t=1.35	0.18
Content analysis
Etiology	40 (16.88)	19 (16.96)	21 (16.80)	–	–
Image pictures display	101 (42.62)	61 (54.46)	40 (32.00)	–	–
Imaging description	139 (58.65)	78 (69.64)	61 (48.80)	–	–
Size	212 (89.45)	100 (89.29)	112 (89.60)	–	–
Diagnosis	219 (92.41)	103 (91.96)	116 (92.80)	–	–
Treatment	220 (92.83)	106 (94.64)	114 (91.20)	–	–
Follow-up	202 (85.23)	102 (91.07)	100 (80.00)	–	–

Data are presented as median (Q1, Q3), n (%) or mean ± standard deviation. GGN, ground-glass nodule; GQS, Global Quality Score; mDISCERN, modified DISCERN.

Quality and reliability appraisal

We assessed overall quality using the Global Quality Score (GQS; 1= poor to 5= excellent) (10) and reliability using a modified DISCERN (mDISCERN) instrument tailored for video content (five items; higher scores indicate greater reliability) (11). Two physicians rated all videos independently. Prior to rating, all reviewers underwent standardized training based on current clinical guidelines. It was explicitly established that ‘Clinical Accuracy’ was a fundamental component of the GQS. Videos containing factually incorrect medical advice were automatically penalized and assigned low GQS scores (≤2), regardless of their technical production quality. Discrepant ratings were adjudicated by a chief physician. All raters received standardized training before scoring to enhance consistency and reduce bias.

Statistical analysis

Normality was evaluated with the Shapiro-Wilk test. Parametric data are reported as mean ± standard deviation (SD), using analysis of variance (ANOVA) for multiple-group comparisons, while nonparametric data are presented as median and interquartile range (IQR). Between-group comparisons used the Mann-Whitney U test for two groups and the Kruskal-Wallis H test for three or more groups. Spearman’s rank correlation (ρ) quantified associations between engagement metrics (likes, saves, shares, comments, video length) and GQS/mDISCERN. Statistical significance was defined as a two-sided P<0.05 (adjusted where applicable). Analyses were performed in R version 4.4.0.

Results

Uploader characteristics

After applying the exclusion criteria, 237 videos were included (TikTok, n=125; Bilibili, n=112). Most uploaders were professionals. Specifically 16.46% were internal medicine physicians (including traditional Chinese medicine practitioners and respiratory specialists), 52.32% were surgeons, and 26.58% were radiologists. Patients accounted for 4.64% of uploaders (Figure 2A). Professionals were the main content providers on both platforms, and surgeons constituted the largest proportion, especially on TikTok (Figure 2B).

Figure 2.

Distribution of video uploaders. (A) Overall distribution. (B) Distribution by platform.

Video characteristics

We compared video duration and engagement metrics across platforms. As shown in Table 1, TikTok and Bilibili differed significantly in these features (P<0.001). TikTok videos were substantially shorter. TikTok also demonstrated higher engagement, with greater counts of likes (median 2,237; IQR, 646–8,499), comments (median 636; IQR, 90–2,307), collections (median 310; IQR, 127–1,071), and shares (median 273; IQR, 31–1,014). Bilibili showed consistently lower values across all engagement measures. Stratification by uploader identity (Table 2) indicated that patient-uploaded videos achieved higher engagement than other groups, whereas videos from internal medicine physicians performed less well. Patient videos were also longer in duration, although the difference was not statistically significant.

Table 2. Characteristics of video uploaders about GGNs on TikTok and Bilibili.

Characteristics	Total	Patient	Internist	Surgeon	Radiologist	Statistic	P
General information
Number	237	11	39	124	63		–
Video length, (seconds)	130.00 (72.00, 228.00)	175.00 (74.00, 452.00)	100.00 (73.00, 146.00)	136.50 (62.75, 262.00)	134.00 (88.00, 191.00)	χ2=4.41	0.22
Likes	267.00 (39.00, 2,556.00)	1,053.00 (226.00, 1,697.50)	23.00 (6.00, 163.00)	466.00 (44.00, 3,310.50)	267.00 (76.00, 5,139.00)	χ2=22.23	<0.001
Collections	76.00 (13.00, 712.00)	293.00 (83.00, 747.00)	10.00 (3.00, 167.50)	88.00 (13.75, 1,135.00)	84.00 (31.50, 710.50)	χ2=16.76	<0.001
Comments	62.00 (3.00, 393.00)	353.00 (41.00, 1021.50)	2.00 (0.00, 55.00)	98.00 (7.50, 478.00)	64.00 (5.00, 441.50)	χ2=20.68	<0.001
Shares	31.00 (6.00, 355.00)	317.00 (32.50, 548.50)	7.00 (0.00, 108.50)	51.00 (6.00, 552.25)	35.00 (11.00, 188.00)	χ2=12.98	0.005
Quality and reliability
GQS	3.00 (2.00, 3.00)	1.00 (1.00, 2.50)	2.00 (2.00, 3.00)	3.00 (2.00, 3.00)	3.00 (2.00, 3.00)	χ2=16.14	0.001
	2.62±0.72	1.73±0.90	2.49±0.51	2.65±0.77	2.78±0.58	F=7.81	<0.001
mDISCERN	2.00 (2.00, 3.00)	1.00 (1.00, 1.50)	2.00 (2.00, 2.00)	3.00 (2.00, 3.00)	2.00 (2.00, 3.00)	χ2=25.08	<0.001
	2.40±0.93	1.36±0.67	2.05±0.60	2.55±1.01	2.49±0.78	F=8.42	<0.001

Data are presented as median (Q1, Q3) or mean ± standard deviation unless otherwise specified. χ², Kruskal-Wallis test; F, analysis of variance. GGNs, ground-glass nodules; GQS, Global Quality Score; mDISCERN, modified DISCERN.

Video content

As summarized in Table 1 and Figure 3, among the 237 videos, 16.88% mentioned causes of GGNs, 42.62% displayed representative imaging, 58.65% analyzed imaging manifestations, 89.45% reported lesion size, 92.41% provided diagnostic opinions, 92.83% discussed treatment strategies, and 85.23% emphasized follow-up. Qualitatively, treatment discussions in surgeon-uploaded videos heavily favored surgical intervention (e.g., segmentectomy), while radiologist-led content placed greater emphasis on active surveillance protocols for pure GGNs. The distinction between pure and part-solid GGNs was frequently omitted in patient-generated content.

Figure 3.

Comparative content coverage of GGN videos between TikTok and Bilibili. GGN, ground-glass nodule.

Video quality

According to Table 1, the median GQS on TikTok was 2.00 (Q1 =2.00, Q3 =3.00; mean 2.35±0.70), whereas Bilibili had a median of 3.00 (Q1 =3.00, Q3 =3.00; mean 2.91±0.62). For mDISCERN, TikTok had a median of 2.00 (Q1 =2.00, Q3 =3.00; mean 2.32±1.00) and Bilibili a median of 2.00 (Q1 =2.00, Q3 =3.00; mean 2.48±0.84). Figure 4A,4B show significantly higher GQS on Bilibili, indicating superior completeness and educational value. Differences in mDISCERN between platforms were not statistically significant, and overall reliability was modest. Comparisons by uploader identity (Table 2; Figure 4C,4D) revealed significant variation in both GQS and mDISCERN, with surgeons and radiologists outperforming patients and internal medicine physicians.

Figure 4.

Quality and reliability comparisons. (A) GQS. (B) mDISCERN. (C) GQS by uploader. (D) mDISCERN by uploader. ns, no significance; *, P<0.05; **, P<0.01; ***, P<0.001; ****, P<0.0001. GQS, Global Quality Score; mDISCERN, modified DISCERN.

Correlation analysis

Spearman correlations examined associations between engagement metrics and quality/reliability scores. Engagement measures were strongly intercorrelated (e.g., likes with collections r=0.94; likes with comments r=0.95; likes with shares r=0.91; Figure 5). By contrast, correlations between engagement and GQS or mDISCERN were very weak and near zero. Both GQS and mDISCERN showed moderate positive correlations with video length (GQS r=0.55; mDISCERN r=0.44), suggesting that longer videos may achieve higher quality and reliability, likely by accommodating more comprehensive and accurate content.

Figure 5.

Spearman correlation matrix between engagement metrics and quality scores. GQS, Global Quality Score; mDISCERN, modified DISCERN.

Discussion

Background

Pulmonary GGNs have heterogeneous etiologies, including early lung cancer (particularly adenocarcinoma), infection, inflammation, and focal fibrosis. Evidence suggests these nodules may represent a principal manifestation of early-stage lung cancer, especially where smoking prevalence is declining (12). The pathogenesis remains incompletely understood and may involve driver gene mutations and immune microenvironment changes (13). GGNs are subsolid nodules categorized as pure GGNs (no solid component) and part-solid nodules (both ground-glass and solid components). The presence of a solid component increases malignant risk. Persistent nodules ≥10 mm that remain for more than three months carry an estimated malignancy probability of 10–50% (12). The correlation between LUAD subtypes and imaging features is well-established. Preinvasive lesions, including atypical adenomatous hyperplasia (AAH) and adenocarcinoma in situ (AIS), typically present as pure GGNs. AAH is generally small (<5 mm) and low-density, whereas AIS is larger (5–30 mm) with higher attenuation and frequent vacuole signs. Minimally invasive adenocarcinoma (MIA) often appears as pGGNs or mixed GGNs (mGGNs) with a solid component under 5 mm. In contrast, invasive adenocarcinoma is characterized by mGGNs, where a solid component >9 mm is highly specific for diagnosis. Malignant features such as spiculation, lobulation, and air bronchograms are significantly more common in IAC than in precursor lesions (12). One large study identified 6,725 subsolid nodules among 4,545 participants. The incidence of newly detected subsolid nodules of 1.9%. The overall incidence was higher in never-smokers (10.7%) than in smokers (7.7%) (14). Diagnosis relies primarily on CT, yet distinguishing benign from malignant lesions remains challenging (15). Surgical resection remains the mainstay for malignant or high-risk lesions (16). Microwave ablation is a safe and effective alternative with low complication rates (17,18). Follow-up should consider nodule size, solid component proportion, and individual risk factors (19), pGGNs, particularly those <30 mm, may be appropriate for surveillance, Conversely, part-solid nodules, especially with a solid component of 4 mm or greater, may warrant more proactive intervention (20,21). A growth-risk threshold of 5% has been suggested to guide follow-up intervals (22). As public health awareness increases, short-video platforms have popularized knowledge about pulmonary nodules. Notably, 70–80% of internet users search for health-related information online (23), though accuracy varies.

Quality and reliability of short GGN videos

To our knowledge, this is the first study to systematically evaluate the quality and reliability of GGN-related short videos on China’s two major platforms, TikTok and Bilibili. This provides foundational evidence to enhance public awareness. TikTok content was more popular, while Bilibili hosted longer videos. Using GQS and mDISCERN, we found overall modest reliability on both platforms. Possible reasons include low entry barriers and a lack of pre-publication academic review. Furthermore, predominantly nonprofessional audiences incentivize rapid, case-based delivery over systematic coverage and transparent sourcing (24,25). Uploader identity significantly influenced quality and reliability (24). Professionals, primarily surgeons, dominated content creation. Videos from surgeons and radiologists scored higher than those from patients and internists. This likely reflects their integrated understanding of imaging interpretation, surgical pathways, and pathological follow-up. Short duration and single-mode presentation further constrained depth (24). We observed moderate positive correlations between video length and GQS/mDISCERN scores. This is consistent with prior research showing higher-quality videos tend to be longer (9,26).

Narrative styles

Looking beyond the raw scores, we noticed a sharp contrast in how content was presented. Given that misinformation on social media can significantly influence health attitudes (27). We frequently saw shock-value titles and emotional stories that focused on rare, worst-case scenarios. These narratives tended to twist standard medical observation into ‘passively waiting for death’, which only fuels patient anxiety.

On the flip side, while trusted messengers, including healthcare, providers play a vital role in public health messaging (27). They stuck to the facts, explaining the slow-growing nature of most GGNs and why active surveillance is safe. However, these videos often felt like classroom lectures. They lacked the emotional pull or storytelling hooks found in the alarmist content. This suggests that while professionals are providing the correct data, they might be losing the audience to less accurate—but more emotionally gripping—video.

Video content gaps

Across platforms, short videos tended to focus on conclusions rather than evidence. Common emphases included nodule size, treatment decisions, and the importance of follow-up. There was less systematic discussion of etiology, malignancy risk stratification, or guideline-based recommendations on surveillance intervals. Some creators used fewer imaging figures and limited interpretation of key radiologic features, focusing instead on familiar therapeutic modalities such as ablation or surgery. This imbalance carries risks (28), including skewed risk perception with anxiety or delayed care, departures from evidence-based pathways and resource misallocation, and reduced quality of shared decision-making (24,29). Cross-platform studies in other conditions similarly show missing core elements in short-form health content, which relates to misunderstanding or inappropriate practices—highlighting the need for stronger guideline concordance and clearer evidence presentation (30-32).

Quality vs. engagement

We did not find meaningful positive correlations between quality and engagement (likes, comments, shares, saves). This aligns with prior social media health research suggesting platform algorithms amplify reach without ensuring accuracy or completeness (24,33). The nature of short-form content, fragmented storytelling, and engagement-oriented tactics (e.g., emotive titles, anecdotal narratives) may drive interaction while weakening guideline-based evidence and transparent uncertainty communication (24). Patient-uploaded videos drew higher engagement but lower professional quality than clinician-produced content, suggesting platforms should better balance experiential empathy with evidentiary reliability. Platform-level differences also stood out (34). Bilibili, with longer videos and a more knowledge-oriented community, demonstrated superior GQS performance. TikTok achieved higher engagement through broader distribution and shorter formats but lagged in content quality.

Practical implications

GGNs have diverse causes and imaging patterns, complicating diagnosis and management. With widespread internet access and mobile media, short videos have become a major channel for health education. However, content quality varies and misinformation risks persist (24). High-quality videos should be scientifically rigorous, transparently sourced, and clearly communicated—avoiding inaccurate or misleading statements. Continuous evaluation can help the public find reliable sources and inform platform governance and content development. These priorities deserve emphasis moving forward.

Limitations

This study has several limitations. First, the cross-sectional design captured videos within a restricted timeframe; therefore, causal relationships cannot be inferred, and future trends in content quality remain unknown. Second, our search strategy relied on a single specific keyword (‘ground-glass nodule’). While this ensured specificity for the radiologic entity and excluded irrelevant solid nodule content, it may have missed videos labeled exclusively with lay terms such as “lung shadow”, potentially affecting the retrieval of some patient-generated narratives. Third, although two physicians performed independent ratings with consensus adjudication, subjective elements persist, and inter-rater reliability statistics were not computed. Fourth, the recommendation algorithms on TikTok and Bilibili personalize results based on user history and geo-location (Internet protocol address), introducing potential variability across different users and regions. Fifth, while our physician raters incorporated medical accuracy into the overall GQS assessment, we did not employ a separate, itemized “Clinical Accuracy Checklist” to quantify the exact rate of guideline concordance for each recommendation. Furthermore, existing assessment instruments contain subjective components, and there is currently no widely validated objective standard for evaluating online medical videos. Finally, our platform selection focused exclusively on video-centric apps (TikTok and Bilibili) and excluded Xiaohongshu (RedNote). We acknowledge this as a significant limitation given that GGNs are epidemiologically prevalent in young, non-smoking Asian females—a demographic that heavily utilizes Xiaohongshu for peer support and medical experience sharing. Consequently, our findings may not fully reflect the information landscape accessible to this specific patient subgroup. Future studies should specifically evaluate community-based platforms like Xiaohongshu to complement these findings.

Conclusions

We evaluated the information quality of 237 GGN-related videos on TikTok and Bilibili. TikTok demonstrated greater user engagement, whereas Bilibili provided higher information quality; overall reliability on both platforms was suboptimal. Videos produced by healthcare professionals tended to score higher. Short-video platforms transcend temporal and spatial limits, offering clinicians a path to deliver accurate advice and disease management recommendations and to improve public health literacy. At the same time, platforms should strengthen oversight and quality control to reduce misinformation and potential delays in care. Users should remain cautious when relying on short-video platforms for medical management information.

Supplementary

The article’s supplementary files as

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.