EHR & Social Media Data for Insights in Mental Health

A recent study demonstrated the use of passively sensed Twitter data to better understand the mental health related expressions of women in diverse cultural contexts.

  • What data sources could we draw upon (from EHR to Social Media data) to better understand the mental health of a population?
  • Who owns this data?
  • How can it be accessed?

I’m flagging other work by Munmun De Choudhury (‪Munmun De Choudhury‬ - ‪Google Scholar‬) and will invite her to this forum. She’s one of the authors on the study you mentioned and an expert in this topic.

Thanks @ingmarweber for sharing this insightful resource and also highly appreciate your help in Inviting Munmun De Choudhury.

@farah, @Fatima, @qlong, @kalorenz, @lorenznoe - You might have answers to some of Karan’s question. Please share your thoughts. Thanks.

@Karan There are some studies that have used your phone to track how much you move around, text, go outside, etc and correlated this with changes in depression or other mental health changes.

I still think that the best way to get data is through action projects. I suggested in a new discussion I started a few days ago that we could train retirees to be guidance counselors in the high schools of the USA and they would help students with the arduous and somewhat intimidating process of applying to college or any other post secondary schools. The USA needs a huge number of these guidance counselors. The professional association of guidance counselors says we have half of the number needed. Doubling the current number would yield a ratio of about 1:250. But an independent education consultant I respect thinks that a ratio of one counselor to fifty-five students is about right to do a proper job of counseling. You can pick up a lot of mental health info if yo are doing this kind of counseling! Some of it is simply visual. A smart counselor can sense problems such as abuse of the student by intimates or fellow students etc.
I have to say regrettably no one seemed interested so far in my idea. It was offered to the Future of Work discussion because I have a very different point of view on what are the priorities there than XPRIZE has. The latter sees the future of work initiative as helping young people from working class backgrounds who are lacking in skills and education. I see the future of work initiative as spending its money more wisely by taking a group of smart under utilized older people and using them after suitable training to help our young people prepare for the high tech world of tomorrow. I know I am correct about the need. I just finished a book length monograph already sold to a respectable academic publisher on how to solve the student loan debt crisis in America and Ipresented my idea there. I am sure it makes sense in many other societies that have educated people who are stitll vigorous and interested in keeping busy while making a good living into their eighties and even beyond.

Thanks @sadiew and @boblf029 for sharing your perspective.
@sadiew - Is it possible for you to share links of the studies you mentioned? It would be helpful to go through it. Thanks.

Hello @sarahb, @KarenBett, @stephaniel, @sarahkhenry, @WD_Research and @areff2000 - any suggestions on data sources to draw upon to better understand mental health.

Not as familiar with what types of data collection methods for mental health already exist, but an idea: The ‘sisterhood method’ for maternal mortality data collection, where by interviewing individuals about their sisters maternal experiences you can reduce sample sizes, was a breakthrough in terms of collected maternal mortality data given the various issues with data collection. I can imagine, if it doesn’t exist already, creating a new methodology that leverages peer relationships to collect mental health data (rather than formal entities with some sort of triangulation of data if possible) could possible help and be a new source of indirect mental health data from the population. Factors such as the role of stigma and how accurate peer information is would have to be considered of course.

Social media and mobile data is very interesting, but any solution that involves it must take into account the populations who are less likely to have social media/mobile phones. In particular, girls and women have less access to digital platforms and when they do, more often borrow or share phones and hence data related to movement for example may not be accurate.

I completely agree with WD Research. The suggestion about the sisterhood method could be valuable. But wouldn’t it be better if we could do that interviewing as part of a program of providing counseling on educational opportunities for women beyond high school? And that is exactly what my proposal of training older retirees to be counselors in high school is going to do. Thank you WD Research.

Thanks @WD_Research for sharing sisterhood method of data collection. We would take a note of it as we research further on the prize direction.
Thanks @boblf029 on sharing your perspective.

Hello @DrLiliaGiugni, @Suneetharani, @stellunak, @Tapman and @Andrea - We feel you all might have thoughts on this discussion, would love to hear your perspective on the data source that could be accessed to understand mental health and also on the various comments from experts. Thanks.

When I see the XPRIZE staff ideas on Future of Work and Data Gaps in Mental Health I am reminded of the Green Revolution of maybe fifty years ago. Many developing nations were facing the problem of not having enough food for their population. The population was growing fast thanks to modern medicine conquering many illnesses that kept the population in check. So initially European (including American) agricultural experts recommended big American tractors and fancy harvesters etc. Unfortunately, there was just one problem: the farmers of these nations were still using sticks made of wood to plow their fields. And they were living on maybe a dollar a day. (That certainly bought more then than now but not enough for a middle class lifestyle). When the foreign experts realized that big American tractors and other high tech equipment was not going to be adopted they hit on a brilliant idea. Why not put steel tips on the wooden plows?! It worked. The farmers embraced this innovation. Farm productivity went up. Incomes went up. Then of course better seeds were also invented and that also helped. The staff of XPRIZE need to take a leaf out of the book of the agrarian experts. It is not fancy equipment such as drones and other high tech but simple improvements that are likely to be seen as helpful by people who are very unsophisticated from the standpoint of technology. Even if young women cannot be reached their younger sisters in school can be with counseling just as WD Research and I suggest. And the younger sisters can bring the older less easily reached siblings into contact with mental health experts who can help them. .

Thanks @boblf029 for sharing your thoughts.

Hi @Aria, @panderekha, @AlexandraW, @ddd1, @Kalpana and @lagoble - You might have answers to some of Karan’s question. Please share your thoughts. Thanks.

@ingmarweber thank you for sharing this resource and for inviting your colleague Munmum De Choudhury.

A follow up question here for you regarding use of social media to collect missing data that provides more information about how people experience distress is in the area of gender. We’ve identified that while social media can be helpful broadly in pulling in information on distress, social platforms often do not disclose gender. Twitter is one example of this: gender is not provided through their developer API. Have you seen any examples of social platforms that would allow for researchers to identify the gender of the users that are providing details about the way in which they experience distress?

Many of the studies that we are finding do not place gender at the forefront of this analysis.

@Kathleen_Hamrick - Using data from the advertising platforms partly solves some of these issues, as long as no individual level data is required. For platforms with a real name policy (e.g. Facebook) the self-declared gender is mostly reliable. For platforms that don’t require a real name or gender (e.g. Twitter) this is trickier. However, Twitter still tries to infer this when it is not (optionally) self-declared by the user as it’s an important feature for advertisers. You can see your own “assigned” gender (see How to access and download your Twitter data | Twitter Help) by going to and then account. There it says: “If you haven’t added a gender, this is the one most strongly associated with your account based on your profile and activity. This information won’t be displayed publicly.”

Twitter also makes the gender available, in aggregate and not for individuals, through their advertising API. (Same as Facebook, Snapchat, …) That could not be used for individual users who have posted likely-to-be-in-distress content, but it could be used to get high level insights into what topics men vs. women are interested in on Twitter or potentially into the gender breakdown of the followers of fairly big seed accounts.

Thanks @ingmarweber for sharing insights on gender availability on social platforms.

Hello @shihei, @JennaArnold, @kbeegle, @“ÅsaEkvall”, @erickson - You may have some thoughts on the data that could be analysed to better understand the mental health of a population. Also what do you think about WD_Research and Ingmar’s comment above?

Hi @luisbenveniste, @Vrabec, @dpnichols, @ukarvind and @EVSwanson - what do you think about WD_Research and Ingmar’s comment above?

In reply to @Karan and @Shashi questions about data sources, perhaps looking into Instagram, or phone-based gaming apps as well. An important thing to track (regardless of source) is both the baseline rate of posting/tweeting and their associated ‘sentiment’, to be able to reliably determine any anomalies for an individual. AI technologies for 'sentiment classification" from text/image in conjunction with changes in posting frequency/ posting time etc. may likely be fed into predictive models for possible changes in mental health status.

Thanks @ukarvind for sharing your insightful thought. It would be great to look at an example wherein AI tech and predictive models have been used. Also is it possible to identify gender using the technologies you have mentioned?

@ingmarweber, @adanvers, @Esther_Colwell, @Pavel - You may have thoughts on Arvind’s comment. What do you think?

This is fascinating @ukarvind! Thanks for sharing your insights here.

I’m curious to hear your thoughts on what the potential complications might be with incentivizing the collection of contextual, social determinants of health (specifically, areas where we know that women are especially vulnerable) alongside depression prevalence data? We are finding that having a clear picture of what contributes to the distress, aside from simply evaluating prevalence alone, is an incredibly important part of improving treatment outcomes for people in differing contexts.

Hi @katyaklinova, @rosiecampbell, @Gabriela, @GB2020, @Lemar and @mlow - As you all are from tech background, you’ll may have thoughts on Ingmar, and Arvind’s comments. Please share your thoughts.