Background: Digital health interventions offer scalable opportunities for primary cancer prevention, yet concerns persist that they may exacerbate the “digital divide” due to socioeconomic disparities in access, literacy, and engagement. Evidence on whether personalized, risk-based digital tools promote cancer-preventive behaviors equitably across socioeconomic status (SES) groups remains limited, particularly in rapidly digitalizing, non-Western settings.
 
Objectives: To evaluate the effectiveness of a personalized digital cancer risk communication platform in promoting adherence to cancer-preventive lifestyle behaviors and to assess whether its impact differs across socioeconomic strata.
 
Methods: We analyzed longitudinal data from 49,207 Chinese adults enrolled in a digital cancer prevention platform embedded within a widely used social media application and supported by optional offline assistance from local Centers for Disease Control and Prevention, community health centers, and hospitals. Using brief self-reported information, the platform applied Rothman–Keller prediction models to generate visual cancer risk alerts identifying cancer types for which predicted risk exceeded population averages, accompanied by actionable lifestyle recommendations. Data were collected between 2021 and 2025. Risk-alert burden was defined as the number of high-risk cancer types flagged (0, 1–3, or >3). The primary outcome was change in adherence to the 2018 World Cancer Research Fund/American Institute for Cancer Research (WCRF/AICR) Cancer Prevention Recommendations (score range 0–8, covering body mass index, physical activity, diet, alcohol consumption, and smoking) between baseline and follow-up. Changes were categorized as worsened (decrease >0.5), stable (±0.5), or improved (increase >0.5). Multivariable logistic regression models estimated odds ratios (ORs) and 95% confidence intervals (CIs). Effect modification of SES indicators (education, occupational class, and household income) was tested using likelihood ratio tests.
 
Results: Median follow-up was 7.9 months (Q1–Q3: 7.2–8.7). Participants had a median age of 37 years, and 65.8% were women. At follow-up, 23.5% of participants improved, 58.4% remained stable, and 18.1% worsened their WCRF/AICR adherence. Risk-alert burden showed a clear dose–response relationship with behavioral improvement. Compared with participants receiving no alerts, the ORs for improvement versus worsening were 1.32 (95% CI 1.19–1.47) for 1–3 alerts and 1.94 (1.75–2.16) for >3 alerts, while associations with behavioural stability were modest. Equity-focused analyses indicated that the positive association between receiving >3 alerts and behavioral improvement was consistently observed across SES strata, with no strong evidence of effect modification (interaction P>0.05). Across educational, occupational and income groups, higher alert burden was associated with increased odds of improvement. Findings were robust after stratification by sex and urban–rural residence. Age-stratified analyses suggested some SES-related modification for behavioral stability, particularly among younger adults with lower education and older unemployed adults.
 
Conclusions/Implications: In this large real-world digital cancer prevention program, personalized cancer risk alerts were associated with meaningful improvements in adherence to cancer-prevention lifestyle behaviors, with broadly similar benefits across SES groups. Integrating personalized risk communication into ubiquitous digital platforms, complemented by optional offline support, may help advance equitable primary cancer prevention. Further research is needed to clarify mechanisms and to identify strategies for sustaining long-term behavioral change.