PO.PS01.06 · 人群科学
AI生成的图片式酒精警示通过增强注意力、学习与情绪加工强烈提高减少饮酒的动机：一项中介分析
AI-generated pictorial alcohol warnings strongly increase motivation to reduce drinking through enhanced attention, learning, and emotional processing: A mediation analysis
该海报暂无可下载的资料
AACR 官方页面
作者与单位 Authors & Affiliations
摘要 Abstract
中文摘要
背景：酒精是美国第3大癌症病因，与七种癌症相关，但只有30%的美国人对此知晓。健康警示信息可以有效传达这一风险，但其机制仍不明确。
目的：本研究考察AI生成的图片式信息（图形化、中性）是否优于文字信息，并探索影响减少饮酒动机的中介因素。
方法：2025年3月，一项全国性在线交叉实验纳入596名美国成年饮酒者（49.3%为女性；年龄21岁以上），评估8条酒精-癌症风险信息（4条为一般性，4条为癌症特异性），每条设3种强度：文字、中性图片和图形化图片（n = 24）。每位参与者从8条信息中随机分配观看两条，每条含3种强度（n = 6），顺序采用平衡设计。主要结局为未来30天内减少饮酒的动机。中介变量为注意力、感知学习、可信度、恐惧和回避。我们采用多层结构方程模型估计并行和序列中介路径，比较图形化 vs. 文字、中性 vs. 文字和图形化 vs. 中性，同时校正年龄、性别、饮酒水平（中度、重度）、饮酒社会规范以及信息顺序和特异性（一般性 vs. 癌症特异性）。
结果：图形化和中性图片比文字信息引起更多的注意力、感知学习、可信度和恐惧（所有p<0.01）。图形化图片相比中性和文字信息还增加了回避。在并行中介模型中，图形化图片对动机有显著的总效应（beta = 0.91，SE = 0.12），其中93.6%的效应通过间接路径中介，主要通过注意力（47.2%）和恐惧（24.3%）。相比文字信息，中性图片也增加了动机（beta = 0.47，SE = 0.10），其中99.3%的效应被中介，主要通过注意力（56.8%）和可信度（16.7%）。图形化图片优于中性图片（beta = 0.439，SE = 0.092，所有p < 0.001），其中87.6%的效应主要通过注意力和恐惧中介。序列中介分析确认注意力为主要上游驱动因素。对于图形化 vs. 文字的比较，经由注意力和可信度（31.8%）、注意力和恐惧（22.9%）以及注意力和学习（17.6%）的序列路径显著预测动机。在中性 vs. 文字和图形化 vs. 中性的比较中观察到类似模式。在所有模型中，信息强度对动机的直接效应均不显著。
结论：AI生成的图片式酒精警示，尤其是图形化图片，通过增强注意力、恐惧、感知学习和可信度，大幅提高减少饮酒的动机。这些发现凸显了AI在改善酒精-癌症风险传播中的作用。
查看英文原文 English abstract
Background: Alcohol is the 3rd leading cause of cancer in the US, linked to seven types, but only 30% of Americans are aware. Health warning messages can effectively communicate this risk, though the mechanism remains unclear.
Objective: This study examined if Al-generated pictorial messages (graphic, neutral) surpass text messages and explored mediators affecting motivation to reduce alcohol use.
Methods: In March 2025, a national online crossover experiment with 596 U.S. adult alcohol users (49.3% female; aged 21 +) evaluated 8 alcohol-cancer risk messages (4 general, 4 cancer-specific), each at 3 intensities: text, neutral pictorial, and graphic pictorial (n = 24). Each participant viewed two randomly assigned messages out of the 8, each at 3 intensities (n = 6), in a counterbalanced order. The primary outcome was motivation to reduce alcohol in the next 30 days. Mediators were attention, perceived learning, believability, fear, and avoidance. We used multilevel structural equation modeling to estimate parallel and serial mediation pathways, comparing graphic vs. text, neutral vs. text, and graphic vs. neutral, while adjusting for age, sex, alcohol level consumption (moderate, heavy), alcohol social norms, and message order and specificity (general vs. cancer-specific).
Results: Graphic and neutral pictorials drew more attention, perceived learning, believability, and fear than text messages (p<0.01 for all). Graphic pictorials also increased avoidance relative to both neutral and text messages. In parallel mediation models, graphic pictorials had a significant total effect on motivation (beta = 0.91, SE = 0.12), with 93.6% of the effect mediated through indirect paths, primarily through attention (47.2%) and fear (24.3%). Neutral pictorials also increased motivation compared to text messages (beta= 0.47, SE= 0.10), with 99.3% of the effect mediated, mainly by attention (56.8%) and believability (16.7%). Graphic pictorials outperformed neutral ones (beta= 0.439, SE = 0.092, p < 0.001 for all), with 87.6% of the effect mediated mainly through attention and fear. Serial mediation analyses confirmed attention as the primary upstream driver. For graphic vs. text comparisons, sequential paths through attention and believability (31.8%), attention and fear (22.9%), and attention and learning (17.6%) significantly predicted motivation. Similar patterns were observed in comparing neutral vs. text and graphic vs. neutral. Across all models, the direct effects of message intensity on motivation were not significant.
Conclusions: Al-generated pictorial alcohol warnings, particularly graphics, substantially increase motivation to reduce alcohol consumption by enhancing attention, fear, perceived learning, and believability. These findings highlight Al's role in improving alcohol-cancer risk communication.
利益披露 Disclosure
T. Asfar, None..
F. J. Penedo, None..
O. J. Oluwole, None..
E. Brown, None..
E. N. Kobetz, None..
C. Calfa, None.