Your Social Media Footprint as an Early Signal for Mental Health
LDiHK turns the data exports every user can request under privacy law (YouTube via Google Takeout, Instagram and TikTok) into an anonymous picture of their digital habits: how much, how fragmented and how late at night they use social media. Our concept: together with insurers, build models that recognise depression risk early, so prevention can start before treatment becomes necessary, rewarding healthier habits along the way.
Built at the EuroTech x Hong Kong Talent Exchange Hackathon 2026 in Munich
LDiHK (Life Data Intelligence Hong Kong) was created by a team of four for the EuroTech x HKTE Hackathon 2026. The challenge we chose: Hong Kong's mental health care is overloaded, with long waits for psychiatric referrals, while decades of research link heavy, fragmented and late-night social media use to depression risk.
This website is the working prototype from the hackathon. You can upload your own data exports and see your usage patterns, a comparison with a synthetic population and a study-based risk indicator. The insurance partnerships, roadmap and business model below are the concept we pitched to the jury. No insurer is involved, and no data is shared with anyone.
Read the story behind the projectStrategic Fit: Hong Kong & Global Expansion
Mental health is a critical concern in high-density markets like Hong Kong. Figures compiled by the Legislative Council Research Office put the share of people with a mental disorder at 13% of working adults, 24% of youth and 12.5% of the elderly, with depression, bipolar disorder and anxiety the most common. Most cases are recognised late, when treatment is long and expensive.
Hong Kong is the ideal launchpad due to two unique advantages:
- ✓Hybrid Insurance System: The split between private and government-funded care rewards insurers who manage chronic risks early.
- ✓East-Meets-West Digital Intersection: Access to both Western (YouTube, Instagram, Spotify) and Eastern ecosystems enables training models for seamless future expansion into China.
How the Data Flows
Data Export by the User
Users download their own data package from YouTube, Instagram or TikTok, a right every platform grants under privacy law (GDPR and equivalents).
Anonymous Upload & Digital Biomarkers
Videos and photos are removed in the browser before upload. Under an anonymous LDiHK-ID we extract only timing signals: daily volume, fragmentation and late-night use. No names, handles or messages are stored.
Risk Signal & Prevention
The user sees their own patterns compared with people of the same age and sex. With consent, insurers use the risk signal for rewards and early, low-threshold support offers.
The Science Behind the Score
A neural network that links social media use to medical conditions needs a lot of training data. Partnering insurers provide it in Phase 1. Until then, a mathematical model built from independent, peer-reviewed studies serves as the starting point.
Built on Published Evidence
We combined findings of peer-reviewed studies on social media use and depression (e.g. Kelly et al. 2018, Lin et al. 2016, Primack et al. 2017, Riehm et al. 2019 and a dose-response meta-analysis by Liu et al. 2022), including their adjusted odds ratios, into one database and fit an initial linear model to them. It validates the pipeline and scoring logic and is the basis for a machine-learning model as the dataset grows.
Three Models by Age
Digital stressors affect age groups very differently, so one model would not suffice. We use separate models for adolescents, young adults and older adults, with sex as a further baseline. It adds complexity in the backend but delivers more reliable results, a trade-off we consider essential for the standards of HealthTech.
Focused on Depression First
We deliberately start with depression: the most prevalent and best-studied mental health disorder. It offers the highest achievable accuracy and the largest potential impact, for public health and for insurers' costs.
Digital Biomarkers Today
- •Daily volume: watch and usage time per day across the uploaded platforms.
- •Fragmentation: how many hours of the day are interrupted by usage.
- •Late-night use: activity between 23:00 and 05:00 that disrupts the circadian rhythm.
- •Demographic baselines: age range and sex, because usage and risk differ strongly between groups.
All of them are tracked over time, to recognise personal patterns and recent changes rather than single days.
Coming Next
The format and content of what is consumed (short-form video or public forums, Instagram hashtags), comments and more markers, plus validation against real outcomes once partner data is available. A first validation of the pipeline and scoring logic was done with a sample group of peers; large-scale validation still needs more data.
The Stakeholder Ecosystem
Creating shared value for users, private insurers, healthcare experts, and society.
Insurance Providers
Gain competitive advantages in risk underwriting, lower long-term claims liabilities, and foster direct customer engagement through wellness incentive loops.
Policyholders / Users
See their own digital habits for free, compared with people like them. Optionally share them with their insurer for premium offsets or vouchers and early, low-threshold support. They can delete everything at any time.
Healthcare Experts
With the user's consent, receive an early risk signal and the behavioural trend behind it, so preventive conversations can start sooner.
The Public & Society
Alleviates systemic pressure on public mental health systems, leading to a healthier, happier, and more productive workforce.
Venture Deployment Roadmap
The three-phase roadmap we pitched at the hackathon for bringing preventive mental wellness insurance to life.
Model Integration & Health Record Pairing
Partner insurers incentivise policyholders to share their data exports. Paired with anonymised claims records, they form the training data for a machine-learning model that replaces the study-based starting model, beginning with depression and later anxiety and stress.
Voucher Incentives & Early Treatments
Using the early-warning models, insurance providers identify high-risk usage patterns in real-time. Insurers offer tailored vouchers and preventative clinical treatments, resolving wellness issues before they become severe, significantly lowering treatment liabilities.
Dynamic Underwriting & Applicant Rewards
We transition to underwriting. New applicants who voluntarily verify their digital footprints receive custom insurance plans and discounts for displaying low-risk habits, creating a substantial competitive advantage for partner insurers.
The Working Prototype
The console is a working prototype, not a trained clinical model. It runs the complete data pipeline end to end: upload your own export, get an anonymous LDiHK-ID, and explore your habits within minutes.
Your history (YouTube, Instagram, TikTok) appears as a timeline, a 24-hour heatmap and a day-by-day breakdown. The population benchmark compares you with people of your age range and sex: about 1,500 synthetic profiles calibrated on the cited studies, plus real users as they join. The health tab shows the study-based risk indicator, for self-reflection, not as a diagnosis.
What the console measures:
- 1Screen Volume Index:Estimated watch time per day across the uploaded platforms. Each additional hour of social media use is associated with higher depression risk in adolescents (Liu et al. 2022).
- 2Circadian Disruption Index:Usage between 23:00 and 05:00, which delays sleep; poor sleep is one of the pathways between heavy social media use and depressive symptoms (Kelly et al. 2018).
- 3Task Fragmentation Index:How many hours of the day are interrupted by usage: frequent checking throughout the day rather than one block.
Questions or Feedback?
LDiHK is a hackathon prototype. If you have questions about the project or the model, found a bug or have an idea for the console, write to us or use the feedback form.