1. + Tweets about hospital quality: A mixed methods study Felix Greaves Harkness Fellow, Harvard School of Public Health / Imperial College London fgreaves@hsph.harvard.edu @felixgreaves

2. + Co-authors / conflict of interest Anthony Laverty Daniel Cano-Ramirez Stephen Pulman Karo Moilanen Ara Darzi Christopher Millett Dr Pulman and Dr Moilanen are founders of Theysay Ltd, a sentiment analysis spin off company from Oxford University

3. +

4. +

5. + Social media use is normal 81% of UK population use the Internet 61% use a social networking site Source: Oxford Internet Survey 2013

6. + We love describing things on the internet

7. + We compared ratings with traditional measures of quality Vs. Associations between Internet-based patient ratings and conventional surveys of patient experience in the English NHS. Greaves F et al. BMJ Qual Saf. 2012;21(7):600-5.

8. + Cloud of patient experience Free Text Greaves F et al. Harnessing the cloud of patient experience: using social media to detect poor quality healthcare. BMJ Qual Saf.2013 Mar;22(3):251-5.

9. + A new trend: ‘Big Data’ and social media analytics

10. + Sentiment analysis

11. + (

12. + :(

13. + :)

14. + Sentiment analysis of patient experience is possible OverallCleanlinessDignity rude dirty rude excellent floor told hours filthy thank pain bed friendly communication blood attitude Greaves F et al. Use of sentiment analysis for capturing patient experience from free-text comments posted online. J Med Internet Res. 2013 Nov 1;15(11):e239.

15. + NHS Insight dashboard

16. + Aims To describe the frequency with which people are talking about hospitals on Twitter To understand what proportion of tweets to hospitals are related to care quality To examine whether there are associations between Twitter sentiment and other measures of care quality

17. + Methods Collected the Twitter names of all English NHS acute hospitals in April 2012 Collected all tweets ‘mentioning’ hospitals for 1 year Qualitative analysis of 1000 random tweets Automated sentiment analysis of all tweets Compared with hospital quality metrics

18. + Results 76 of 166 hospitals were on Twitter 198,499 tweets Mean tweets per trust: 2647 Median: 796, range: 0 to 88169

19. + Time of tweets

20. + Mean: 508 tweets per day Median: 405 Range: 62 to 3601

21. + Qualitative analysis – Most tweets aren't about quality All tweets: 47% were positive, 47% were neutral and 5.6% were negative. Tweets about quality: 77% were positive, 2% were neutral and 21% were negative. Main theme%Minor theme%Sub theme% Care Quality11.4 Patient centred9.8 Staff5.5 Environment1.1 Access1.4 Effective2.9 Safe0.4 Fundraising38.7 About a patient in hospital10.5 Promotion8.8 Organisational Information7.1 Health Information4.4 Non-specific content20.7

22. + Interaction with staff Home from [@named hospital] after a weeks stay...we feel blessed to have been cared for by such an amazing team. Thank you [named ward] x At the [@named hospital] just had an operation on me foot. Outstanding care as usual, & the nurse has just made me a cracking cup of tea :-) [@named hospital] [named ward] - Disgusted with your treatment of my mother. Will be making huge complaints.

23. + Environment Be nice if this room had been cleaned before we got it. Blood filled cap from an iv on the bedside cabinet, unflushed toilet [@named hospital] Spent a night in [named hospital] with my son. Excellent care - spotlessly clean. Thank you [@named hospital] Don't suppose there is any chance of full english [@namedhospital] Been here since 3 yesterday no hot food or drink #poor

24. + Access/Timeliness [@named hospital] Thanks for squeezing me in with orthoptist [named staff member] today. Great service just so sad that waiting list for [named surgeon] so long :-( [@named hospital] where the waiting time is ridiculous waited 3hr yesterday, 3lots of bloods took, 2hrs so far today for a blood test again! Waiting at [@named hospital]- appointment was over 2 hours ago. can we get takeout delivered??

25. + Effectiveness / Safety [@named hospital]Yes pls. Main concern now is the doctor overprescribing. We worked out the error but vulnerable patient might not [@named hospital] Also looking at a scan from 2010 when u didn't get scanned until 2011 not good, wrong person, terrible, disgusting

26. + Multimedia [@named hospital] Cannot believe I have been served this for my 18 month old. Tastes disgusting and hardly nutritional [permission obtained to use]

27. + People can be rude [@named hospital] [Named chief executive] should come down onto the wards n see what's really going on under her nose. I wish my nan was in [another hospital] [@named hospital] Shit on floor wet sheets, visitors having to change bedding, shit in toilet, ignorant staff- [name ward]!! Stay away !

28. + Twitter sentiment compared to traditional quality measures Twitter MetricConventional Measure Spearman Rho p value Twitter sentimentSurvey measure of “Overall, how would you rate the quality of care you received” 0.150.30 Twitter sentimentRisk adjusted mortality rate0.150.24 Agreement 71%, Kappa 0.39

29. + Limitations Only one way messages, not conversations Ability of sentiment analysis approach Sarcasm / Irony Culturally specific Vulnerable to external effects

30. + Like an angel

31. + Stank of urine

32. + Cup of tea

33. + Tweets about the NHS are affected by external factors

34. +

35. +

36. + Conclusions Twitter is being used by patients to discuss care quality But…it’s only a minority of the tweets Signal at risk of being lost in the noise Twitter may be more useful to drive quality improvement than understand comparative performance Less about Big Data – more about small human stories

37. + Wider lessons from social media A powerful tool for spreading ideas Allows crowdsourcing information Flattens hierarchies Allows new approaches to patient engagement

38. + Caution… Self selecting Fickle New ethical questions Reality vs. hype

39. +

40. + @felixgreaves fgreaves@hsph.harvard.edu

41. + What affects the public’s decision about where to go to hospital