Data Science in Digital Health
89 likes6,022 views
This document discusses using data science and digital technologies to better understand and influence human behavior change. It explores how collecting smartphone and user data can provide insights into behaviors like smoking cessation. Recommendation systems aim to tailor automated support to individuals, but challenges include a "cold start" period with little initial user data and ensuring interventions are grounded in behavioral theory rather than just predictive accuracy alone. Ongoing evaluation is also needed to test whether technologies truly enhance engagement and drive behavior change as intended.
Data Science in Digital Health
- 1. In digital health #wwdh @neal_Lathia DATA SCIENCE
- 4. HOW DOES BEHAVIOUR CHANGE? HOW COULD TECHNLOGY INTERACT WITH PEOPLE? HOW Do PEOPLE INTERACT WITH TECHNLOGY? BEHAVIOUR CHANGE?
- 6. MAKING CHOICES CASE 1: memory & choice (NOT HEALTH)
- 7. “Psychologists have recognized for many years that humans have a limited capacity to store current information in memory.” - “Information Overload” on Wikipedia
- 9. AUTOMATED BY RECOMMENDATION - Neal's slides during his PhD
- 10. AUTOMATED BY RECOMMENDATION - Neal's slides during his PhD Navigating choice ~ Predicting missing data Ranking on predictions
- 11. AUTOMATED BY RECOMMENDATION - Neal's slides during his PhD No “framework” No “item” context No theory/categorisation Simplistic assumption No uniformity 1000 outcomes for 1000 people
- 12. USES BEHAVIOURAL THEORY Online Recommendations EXPLAINS THE BEHAVIOUR ALWAYS GETS IT RIGHT AUTOMATED PROCESS ENHANCES ENGAGEMENT CHANGES BEHAVIOUR NO NO / BADLY NO YES YES YES
- 13. USES BEHAVIOURAL THEORY EXPLAINS THE BEHAVIOUR ALWAYS GETS IT RIGHT AUTOMATED PROCESS ENHANCES ENGAGEMENT CHANGES BEHAVIOUR NO NO NO YES YES YES DOMAIN KNOWLEDGE DATA SCIENCE BOTH Online Recommendations
- 14. “Your decades of specialist knowledge are not only useless, they're actually unhelpful; your sophisticated techniques are worse than generic methods; The algorithms tell you what's important and what's not...” - @jeremyphoward (Interview)
- 15. “...You might ask why those things are important, but I think that's less interesting. You end up with a predictive model that works.” - @jeremyphoward (Interview)
- 17. WHAT SMARTPHONES CAN SENSE THEMSELVES What SMARTPHONES CAN PROMPT YOU TO TELL The Emotion Sense Platform: Location, mobility, sociability, physical activity Mood, symptoms, assessments
- 19. QUITTING SMOKING CASE 2: Automating support
- 20. YOUR SMOKING BEHAVIOUR Smoking Cessation – Ideal + “ReCOMMENDED” SUPPORT = BEHAVIOUR CHANGE
- 21. YOUR SMOKING BEHAVIOUR Smoking Cessation – Ideal + “RECOMMENDED” SUPPORT = BEHAVIOUR CHANGE NO DATA ON THE “USER” WHAT IS THE “ITEM?” NOT POSSIBLE?
- 22. “Cold start is a potential problem in computer-based information systems (...WHERE..) the system cannot draw any inferences for users (or items) about which it has not yet gathered sufficient information.” - “Cold Start” on Wikipedia
- 23. - “Cold Start” on Wikipedia “Cold start is a potential problem in computer-based information systems (...WHERE..) the system cannot draw any inferences for users (or items) about which it has not yet gathered sufficient information.” And beyond: in a given health domain, what information should we (can we) collect?
- 24. HEALTH /SOCIAL SCIENCE DATA SCIENCE HUMAN COMPUTER INTERACTION DIGITAL BEHAVIOUR CHANGE Cold start
- 25. “cue-induced cravings: intense, episodic cravings typically provoked by situational cues associated with drug use (...) smokers exposed to smoking-related cues demonstrate increased craving (...).” - Ferguson, Shiffman. The relevance and treatment of cue-induced cravings in tobacco dependence. In J Subst Abuse Treat. April 2009.
- 26. “cue-induced cravings: intense, episodic cravings typically provoked by situational cues associated with drug use (...) smokers exposed to smoking-related cues demonstrate increased craving (...).” - Ferguson, Shiffman. The relevance and treatment of cue-induced cravings in tobacco dependence. In J Subst Abuse Treat. April 2009. Situation: mood, craving, location, social setting
- 28. Your location + your profile = tailored support EXAMPLE
- 29. USES BEHAVIOURAL THEORY EXPLAINS THE BEHAVIOUR ALWAYS GETS IT RIGHT AUTOMATED PROCESS ENHANCES ENGAGEMENT CHANGES BEHAVIOUR YES NO NO YES YES? YES? Smoking Cessation YES (BUT what DATA!)
- 32. UNDERSTAND IMPLEMENT EVALUATE Design Automate HYPOTHESIS Linear/hypothesis driven research: good for publication, bad for software.
- 33. MONITOR LEARN DELIVER N. Lathia et. al. In IEEE Pervasive Computing. 2013. SOFTWARE IS NEVER FINISHED... ... IT IS UPDATED. HYPOTHESIS
- 35. SCHIZOPHRENIA ANXIETY MOOD ADJUSTMENT ANTI-SOCIAL PERSONALITY ON/oFFLINE MOOD EXPRESSION FREEMIUM Code: http://emotionsense.github.io/
- 36. In digital health #wwdh @neal_Lathia DATA SCIENCE