JMIR Publications

ABSTRACT

Background:

YouTube is a widely used source of health information for people considering glucagon-like peptide-1 (GLP-1) receptor agonists. Existing studies have assessed the quality of small, purposively sampled sets of videos, typically 100 to 240 videos. However, none has characterized the structure of the information environment itself, including the distribution of content across channels, the concentration of attention, and whether commercial discourse is concentrated around particular drug vocabularies.

Objective:

This study aimed to characterize the information supply ecology of GLP-1 receptor agonist content on YouTube by quantifying attention concentration across channels, measuring the prevalence of content signals, and comparing videos that used brand names with those that used molecule names.

Methods:

We conducted a cross-sectional infoveillance study of publicly available YouTube video metadata retrieved through the YouTube Data API v3 for 8 GLP-1-related search terms. We deduplicated records at the video level and assigned each video to 1 or more drugs by matching brand and molecule names against the title and description, rather than according to the search term that returned it. We excluded videos that mentioned no GLP-1 product or drug class as algorithmic false positives and restricted the primary analysis to English-language videos. We measured channel-level attention concentration using the Herfindahl-Hirschman Index (HHI), Gini coefficient, and top-k channel share. We classified 6 content signal categories using a prespecified regular expression lexicon and compared brand-only and molecule-only videos using chi-square tests with Bonferroni correction.

Results:

Of 7793 unique videos retrieved, 577 (7.4%) mentioned no GLP-1 product or drug class yet accounted for 37.8% of all views. Nine of the 10 most-viewed retrieved videos were off-topic. The primary corpus comprised 4260 English-language, on-topic videos from 2015 channels, with 137,251,720 views. Content production was fragmented (HHI 0.0095), whereas attention was highly unequal (Gini coefficient 0.917). The 20 highest-reach channels, representing 0.99% of all channels, accounted for 35.9% of views while contributing 4.8% of videos. In total, 832 videos (19.5%) had fewer than 100 views, and 50 had no views. Compounding or shortage signals were more prevalent in molecule-only than brand-only videos (7.8% vs 1.7%; adjusted P<.001), whereas safety signals (30.6% vs 20.6%; adjusted P<.001) and personal narrative signals (15.2% vs 9.2%; adjusted P=.003) were more prevalent in brand-only videos. Search-term classification disagreed with content-derived classification for 31.7% of videos.

Conclusions:

The GLP-1 information environment on YouTube is fragmented in content production but highly concentrated in the distribution of attention. A small number of channels accounted for a disproportionate share of views, whereas many videos received little attention. Brand-name and molecule-name vocabularies indexed distinct discourses, with safety and personal narrative content concentrated in brand-name video environments and compounding and shortage content concentrated in molecule-name video environments. Exposure-weighted surveillance that incorporates molecule-level keywords may provide an efficient approach to monitoring direct-to-consumer pharmaceutical discourse on social media.