Twitter and Mental Health Diagnoses

Something Mindful Data has been thinking about for quite some time: what people are saying on Twitter in association to some of the most major mental illnesses. Unfortunately, the organization doesn’t have access (nor the money) to pay for a Twitter Partnership, so a query of the last 2,000 tweets were pulled. Might create a dashboard though.

What’s the point of this analysis? Namely, text analysis of publicly-facing content by individuals can find insights and help provide the field ways to consolidate language in order to get advocacy and de-stigmatization better across to the larger population. Based on previous analyses of Mental Health Advocates, support to fill that gap is lacking. To visualize language of these tweets, an analysis with WordClouds and basic text analysis was done.

Avg. Polarity
(-1 to 1)

Avg. Subjectivity
(0 to 1)
Avg. Verbosity (0 to 140)107.994.2120.3

More info on the background of the Sentiment analysis via TextBlob can be found on this exceptional community post. So, what’s learned?

  1. On average, the sentiment is surprisingly positive.
  2. On average, the sentiment is surprisingly more objective than subjective.
  3. There’s a difference in verbosity between tweets analyzed related to various MI’s and SMI’s.

What might better help Mental Health advocates on Twitter with this analysis? The most important take-away is the visualization and analysis from the Word Clouds and Verbosity scores. Notice that in Schizophrenia, there is little popularity in most words except the diagnosis term itself compared to Depression and Bipolar. The basic assumption is that Twitter users don’t have a shared language on how that diagnosis is communicated. The Schizophrenia/Psychosis community would lend itself to learn to have a shared language that’s not mixed and verbose.

Scripts found here.

Engaging on Github!

Mindful Data has a Github organization where all the datasets and code created for the public good is housed across all outputs of mining, analysis, and visualizations! It’s taken work for months to get it to this level, but feel free to take a look here (and star/fork)!

Currently, there’s the following:

  • datasets: 20+ datasets for Mental Health collected from the internet
  • charts: D3js charts based on Mental Health datasets
  • suicide-prevention-graphs: Visualizations of US Suicide Rates + National LifeLine Service Data
  • tweetmh: Hourly-updated Twitter analysis on mental health

Opioid Crisis Trends 2006-2017

What is an Opioid? These are substances that act on opioid receptors to produce morphine-like effects. Medically, they are primarily used for pain relief, including anesthesia. Opioids are also frequently used non-medically for their euphoric effects or to prevent withdrawal.

What is the Opioid Crisis? The general trend of overuse of Opioids, primarily, in the United States. It began in the 1990’s and has escalated in recent years with a 2.3x increase in overdoses from 2006-2017 (even higher in 2018). It’s particularly affecting young women and communities within the United States. It’s prompted a significant response from the NIH with the HEAL Initiative starting in 2018.

Though the NIH’s HEAL Initiative started in 2018, this decades old problem has been responded to by doctors in the United States since 2012 (in an effort for the Obama Administration to curb the effects quickly). There’s no doubt there must have been some amazing work, on a national basis, to reduce Opioid prescriptions by a huge 25% between 2012 and 2017.

Meanwhile, Opioid-referenced research, since 2012, has increased 9% (at a time when Mental Health research decreased 75% overall potentially due to the Replication Crisis), but growth flat-lined after 2014. Unfortunately, overdoses…spiked to 85% from 2012 to 2017 (and even higher in 2018) based on data from the CDC. There isn’t any in-house knowledge here to determine what’s causing the significant overdose spike, but if you know the why’s and how’s of this trend, please contact us. This is important and should be analyzed further.

The insight here is that society is responding to a crisis and trying getting ahead of the curve before things get worse than they are. Though, it’s still not enough…

Dataset available here.

Data Advocacy: Opening up CDC Data for US Suicides in 2018

Thanks to data mining and analysis, we found a gap in regards CDC data in their Fatal Injury Report regarding suicide…2018 data just isn’t publicly available. That’s weird; especially being a quarter away from the end of 2019. The original goal was to create a data visualization of US LifeLine efficiency and correlation to outcomes (i.e. Suicide) after finding it in the Mental Health Improvement Act of 2018 Report. That’s been done and available now via a dashboard.

You can find the link to the dashboard here. A couple of data analysis findings are interesting from it:

  1. The suicide rate & answered calls jumped significantly from 2016 to 2017
  2. Funding flat-lined while the problem has gotten worse in the US leading up to 2017
  3. Again, can’t find data on 2018 outcomes

So, Mindful Data has gone ahead and created a petition to request the CDC to release the data(set) for 2018. If not available, then give a reason why. Pretty proud of the effort and getting to this point of our first act of data advocacy in Mental Health.

Trends in Livability with a Cognitive Disability

I didn’t fully understand the difficulties of what it meant to get to work with an illness until having one. On an individual level, it’s a grind to get work. Sometimes you never find it. Sometimes you get lucky and find a savior to help pick you up and give you work. Then, manage the condition (disease + stigma) in order to live a life. A lot is mentioned about this in the previous post.

Though, looking into the issue on a US (aggregate) level, the data shows some promising signs since 2012. Namely, household income (from Cornell’s Disability Statistics) has grown well while disability insurance applications with Social Security (via online applications) has decreased consistently. We’re talking about net changes in the 9-10% variety for some years. That’s the great!

The good? For some reason, employment is getting better. Yes, the number of people disabled and out of the laborforce is still huge (in the 2M+ population range and a 4x difference than those disabled and employed), but employment rates in this group are increasing since 2015, consistently, as well. What might be the cause here?

As this blog is called “Mindful Data” rather than “Disability Data”, I’ll delve into the subset of data points specific around a mental health condition. Namely, what’s been the biggest change since 2012 that might result in more people getting to work and earning more with what’s deemed as a disability? Well, after some data mining on Google Scholar, believe the answer resides in a concerted effort to improve efforts via our Judicial Branch in the United States. Look at this:

2010-2012 not only coincided with the “Replication Crisis”, but also a major uptick in case law on the Federal level for mental health conditions. 2012 alone comprised of a huge 28% y-o-y increase in court cases seen on the Federal level regarding these types of conditions. The following year saw a large ~6.3% y-o-y increase in annual earnings and, further, a consistent y-o-y increase in earning income starting in 2015.

Two points to make on this with a key metric and outcome society, and the field, focuses on:

  • US Suicide Rate saw a significant ~3.6x spike in 2017
  • In 2012, where earnings increased by ~6.3% in the same year, the suicide rate dropped by ~9x (!!!)

What do I make of this based on my experience? Livability is crucial to stay alive. People don’t have a right to work; though, do have a right to live. We’re all living to thrive at some lucky point, but a majority of the population works to survive. The large increase in federal case law looks to have at least helped with increasing annual earnings and, as a result, livability. I do wonder about 2017 though and what happened that year…

Mental Illness and a Lack of Productivity Data

Throughout my journey in finding Tech employment with my own illness, I came across a number of situations. Namely, most jobs available in Tech, for someone with a mental illness, are found via more accepting roles such as Engineering, Customer Support, and Design. No problem, right? Well, that requires a lot of moving levers. Sometimes you’re really not ready for work on a health and productivity level. Other times you’re ready, but employers aren’t ready. That’s a long time to wait to become productive. A couple of factors are important in my view:

  1. Level of openness about a mental illness (MI) to the hiring manager
  2. Ability to empathize and work with the person with the illness
  3. Remote work to limit stress and in workplace and productivity issues
  4. Being a good example so a someone else with a MI can thrive

Why are these important? At the end of the day, as a person with an illness, I went through as many examples and case laws I could find. Found that building a bridge of openness was necessary because of a few reasons:

  1. You’re potentially a workplace/team environment distraction
  2. You’re possibly a management headache (open or not open)
  3. Your health issue may limit your ability to be 100% productive

Now, I won’t go into the appropriate reasoning as to whether to disclose or not. That’s an individual’s choice to make and entirely based on their family, friends, life events, and prioritization on what is important for them.

Though, as a result, there just isn’t a lot of data available on openness on something like a mental illness in tech due to disclosure. Rationally, and under the American Disabilities Act, you shouldn’t need to disclose. Though, that doesn’t help as much in this day and age with hyper-connectivity and sharing. As the common idiom goes: “The Valley is small”. There isn’t a lot of open data about this either. So, to find a way to highlight this issue on a small level, did a data dig in Github for users that have “Mental health” and “Psych” in their profile.

Most accounts with “Psych” related to those who are Researchers and Data Scientists with either Psychology-related backgrounds or interests. Out of the 166,600 jobs in the US for Psychologists, the 1.6M+ Engineers in the United States (2016) and over 24M+ users of Github, these are paltry numbers. Even further, there’s 46.6M+ people with mental illnesses in the United States. The level of openness is astoundingly bad for a productivity platform such as Github compared to a conversation platform such as Twitter. To put it in perspective, in 2016, it’s 1 Github user that denotes “mental health” on their profile for every 40,000 engineers in the USA. What does that mean?

Not enough are looking at this issue and should be from a productivity perspective. That results in a lack of programs and awareness to determine what productivity is possible for sufferers. I’ve seen some/many programs that help in retraining on some level, but it’s not widespread and empowered. We’re washing away a large segment of society from being understood on what productivity looks like with mental illnesses. From one vantage point, you could consider it a human rights tragedy to be largely ignored and suffering without an ability to be open and productive. For example, Thomas Insel wrote that the lifetime economic burden of serious mental illness to individuals and society combined is nearly $250B, or approximately $1.85 million per patient in the United States. That’s awful.

We don’t know what’ll be uncovered by finding what’s happening on a day-to-day basis for life goals and ability to survive and/or thrive for people with mental illnesses. I do know that, at least, data will highlight problems, promote awareness, and improve decision making to further improve the situation and subject area. We need more open data and work about what’s really going on, in an aggregate level, in helping those with MI’s to not only get to healthy, but also help to get us/them to thrive in life.

If you know where I could find productivity and employment data for people with mental illnesses or want to connect, please contact on Twitter.

Dataset available here.

“Replication Crisis” and Tech in Mental Health Research

After continuing to data dig in Google Scholar for research articles referencing “mental health”, there were some interesting observations found. Technology-related publications just increased by 12.7% in the aggregate, but there were increases in Wearables, by ~450%, as well as Apps, by ~650% since 2013. Though, that didn’t paint the entire picture.

Taking a bird’s eye view, there was a significant downtrend in research within Mental Health starting at the beginning of the “Replication Crisis” in 2011-2013.

Trends in Mental Health Publications from 1985 to 2018

Since that year, where the number of research articles reached a staggering 477,000+, research dropped by ~75% within a 5 year block between 2013 to 2018. That’s amazing and paints a dire picture for the field on the surface of it. Though, looking at VC Funding in the space, trend lines look amazing! So, what’s going on here?

Well, tech is actually picking up adoption in total research done within Mental Health with 32.8% of all publications in 2018 even though total research has lowered significantly. A greater than 2x increase since 2013.

From my experience in the mental health system, the “Replication Crisis” has resulted in a few things:

  1. Researchers were/are scrambling for any way to get funding.
  2. Researchers need to focus in the latest technology trends.
  3. Mental Health care is going through many drastic, shifting changes.

This is not a health environment for any field, especially healthcare. You need consistency to provide the right service to patients in order for them to get healthy. I’ve seen so many parts of mental health just focused on surviving day-to-day and reducing the number of really-bad-situations. I’m naive when it comes to healthcare on any level, but understand at least some things from my personal experience and what I see from others. The system just doesn’t work for anyone right now, not just patients…but everyone, including doctors and researchers, who are stuck in a quicksand situation of care.

With that said, the latest trend lines in psychological research towards a leaner and more technology focus shows promise! Talented researchers will get far more data into their hands and derive actionable, repeatable outcomes that have largely been well-developed in mobile and web technology over the years. Also, if VC funding continues on its growth path, then there will no doubt be a lot of monetary opportunity available to researchers going forward to go on a growth curve themselves alongside technology companies. I’m hopeful, but I really hate the current situation.

Again, if you recommend any research or data, feel free to contact on Twitter!

Dataset available here.

Trends in Mental Health Tech Research (1985 to 2018)

One of the clearest problems that stuck out to me going through crisis to recovery was the lack of tools available in Mental Health compared to any other industry I’ve experienced. It was pretty comical some of the conversations my doctor and I have had over the years. I mean, there’s wearables, mobile apps, tablet apps, AI, Big Data, Precision Medicine, etc. So, where’s the technology for Mental Health?

The most important priority in any person with a mental illness and/or serious mental illness is to get healthy and, then, productive. The person needs to know where they’re at in real-time because the disease(s) affect(s) behavior based on internal mechanisms of something that’s, right now, unnoticeable. There just isn’t much there and it leaves a lot of people stuck on both sides (and very political from my perspective because they don’t have enough data to make points and refutations directly to each other). Everyone really means well though.

So, just what is going on here? Google Scholar to the rescue…

Trends in Mental Health Publications from 1985 to 2018

I looked at searches related to wearables and mobile apps as well as the fMRI in research articles found on Google Scholar. So, what is going on? Well, fMRI completely dominated mental health technology research until recently. For those that don’t know, you go into this giant machine and it takes pictures of your brain and it takes about 30 minutes or up to 2 hours depending on what pictures the doctor wants to take. It costs A LOT of money. Now, new technology (e.g. wearables like the Apple Watch) and Mobile Apps (e.g. Calm) are much cheaper and real-time-ish, but don’t give you the full picture (so-to-speak) as a fMRI scanner. If you look into the trends with Mental Health Venture Capital funding, you’ll find that research and funding had a banner year in Mental Health in 2018 too.

Personally, I like where the research trend is going, but I certainly don’t think we completely understand how the brain operates nor how we can improve diagnoses better using fMRI technology and AI for a lot of these illnesses. I worry that there isn’t the right focus in the field, as a person with a lived experience myself, and that tech is eating research and putting it in the wrong direction. I’ll be looking into this more in follow-up posts to get a more defined picture of the field.

Dataset available here.

TweetMH: Hourly Twitter Mental Health Card

So, going through the mental health system, I’ve found that the term “mental health” (much like some aspects of the field) is not 100% agreed upon. From experience and opinion, a lot has to do with a lack of data. Some use “brain”, others” mental” alongside “illness”, “disease”, and “health” in certain cases. So, it makes cohesive efforts for strategy and communications in organizing forces difficult nation-wide and world-wide. In order to get an understanding of the mostly real-time popularity of these generic terms, TweetMH was built as an experiment. It’s on its first couple of iterations and needs feedback, but provides some value add.

Namely, text-based analyses applying the Flesch Reading Ease Scale and Textblob’s Sentiment (i.e. Polarity) were used and averaged across the last hour’s worth of tweets on an auto-updated basis. Some findings:

  • “Mental Health” is far away and the most tweeted term consistently.
  • “Mental Health” twitter comprises of mostly objective tweets.
  • The use of “brain” spikes up polarity of sentiment, but is generally non-normative in its sample size.

The dashboard will go through some more iterations like a daily data Mailchimp newsletter, hopefully, in the future. Though, that’s not certain. Funds are currently at $0 for development, maintenance, and deployment. So, feel free to contact MindfulDataOrg on Twitter for potentially funding this operation and turning it into a 501(c)3!

If you have thoughts on how to improve the dashboard or want to help, follow or contact MindfulDataOrg on Twitter! All of this is on Github too.

Data Mining for Mental Health Advocates on Twitter

How do Mental Health Advocates on Twitter stack up? To look into this question, I ran an API query on the 1st 1000 verified Twitter users with “Mental Health Advocate” in their profile summary against a control sample of verified Twitter users with the letter “a” in their profile summary.

What were the results?

Verified MH AdvocatesVerified Random Users
Profile Summary126.0 (characters)111.2 (characters)

The average sampled Mental Health Advocate had ~2:1 ratio of followers/follows compared to the control which had ~18:1 (!!). At least in this experiment, Mental Health Advocates have far less reach and, based on the generosity with favorites, it suggests they try harder to get less attention as well. Wish there was something Twitter could do to help improve reach and engagement for Mental Health Advocates.