Tracking mental health care

Charts & Graphs, Final Projects, Visualization

Struggling with a mental illness is daunting, but the treatment process isn’t exactly an easy road on its own. After one musters enough courage to face their vitality head-on, treatment tosses the person with mental illness into a pharmaceutical industrial and late-capitalist self-wellness hellscape. Stigmatization reveals itself in dollar signs: According to a recent study conducted by behavioral health experts at The Ohio State University, consumers are spending more out-of-pocket dollars on mental health services than those receiving physical care. How curious for the pharmaceutical providers to keep uninformed customers who unknowingly seek out-of-network care in the dark. 

Doctor-provider incompatibility is one way the process dehumanizes the over 46% of adults aged 18 or older with any mental illness (AMI) that seeks treatment in today’s neoliberal health factory. Other issues of access, information literacy about medication, and cost abound. 

To help shine a light on the cavernous hell marketed as wellness treatment, I focused my final information visualization project on creating an agency-building tool for those seeking mental health care, and for those, in particular, using medications to do so. 

I ethically teetered on choosing mental health tracking as an agenda for an information visualization project because data and technology are commonplace roots of mental strain than anecdotes. Fitness tracking has been called out as fueling obsessive, unhealthy nutritional tendencies amongst even children. Furthermore, third-party apps and gadgets do who-knows-what with the bulk sensor data they gather on our steps or heartbeats. 

But, I decided, this is an opportunity to engage in conversation with my peers in user research and design academia. Can a digital tool really create agency for a recommended user seeking treatment for mental illness? On a personal level, this project means “something” to me. I have a first-hand account of the consequence of blindly strolling into a pharmaceutical treatment path only to be left without a map, and with the door slammed behind me. 

Therapeutic and medicinal care is a strenuous and costly endeavor, so the patient should feel control over how treatment plans impact their daily life or behaviors. Thus, the final dashboard I present is a Mood and Medication tracker with dual views: 90-day and monthly. 

My final project process

Preliminary research


After deciding on a topic, the first step was research. In uncovering the power of tracking, I stumbled upon a group of Dartmouth researchers’ algorithmic approaches in mental health identification. Their analysis of academic, physiological, and daily factors built an equation that predicts student depression with over 80% accuracy.

A more direct inspiration was work by a previous Pratt Information Visualization student. Arushi Jaiswal developed a fitness tracking website prototype with Tableau. Besides revealing technical insights, I took note of the importance her users felt about interactiveness in a prototype.

I also spent quite a lot of time playing with already existing applications for self-tracking on my iPhone, such as the Apple Health and MyFitnessPal, a self-entry log with sensor capabilities. I like the graphical designs of the Apple tracker, but their medication view is lack-luster.

Suggested User and Expert Conversations

Fortunately, this information-gathering period fell over Thanksgiving: an opportunity to corner friends and families for user and expert interrogation.  

I asked, “What types of variables or insights would be beneficial for a universal mental health and medication tracking tool?”

“I encourage all my patients to catalog their moods. A long-term view would be beneficial for treatment, but it would be difficult to standardize a mood scale as everyone feels differently.”

A social worker (my sister)

“Various types of vitals, side effects, and behaviors would make sense to capture. Medications impact patients differently over large swaths of time. I’d focus on the larger picture.” 

A family health physician (my other sister)

“I’m typically apprehensive about health tracking, but I would like a window into my medications impact on mental health, considering the time and money I spend on care.” 

A 24-year-old adult female in treatment for a mental health illness (a friend)

User Goal

After narrowing my topic and hearing from the peanut gallery I needed to develop a user motivation to start gathering data and designing a prototype. I framed the end-goal as: How have my depressive moods, both in intensity, frequency, and duration, correspond with various medications? 

I framed the end-goal as: How have my depressive moods, both in intensity, frequency, and duration, correspond with various medications? 



From the mini, informal interviews, I decided I wanted to develop a dashboard with these factors: 

  1. Mood 3x a day (see Table 1 for scale)
  2. Medication in mg 1x a day
  3. Side effects: irritability, nausea, headaches, suicidality. 
  4. Other behaviors: steps, menstration, social interaction, work. 
Table 1

Originally, I wanted to use three mood counts as people do not have a single mood throughout the entirety of a day. However, I wasn’t sure how to compare moods taken 3x a day with other information, like mensuration occurrence, only is once a day. I decided to weed out that issue after the prototype test and class presentation.

I scoured the internet for sample behavioral data. Unsurprisingly, not many people are offering up their personal mental health data for manipulation by graduate students. I realized the best, and only, the option was to fabricate sample data. 

I read up on key components of major mood disorders. I thought Bipolar II, a type of bipolar that presents through long periods of depression and short bursts of hypomania lasting around 4 days, would make a good sample disorder to map.

 Therefore, I used Excel to create mood ranking data for three times a day over 30 days. In a separate sheet, I created data collected once day on: sleep duration, steps, physical activity, social interaction (if purposely engaged in human contact or not), class attendance, work attendance, mobile screen-time, irritability, and suicidality (see table 2).

Table 2


Tableau Public

To bring this design to life, I used Tableau Public, a free software for information visualizations. 

I uploaded both text files I made in Excel. First, I cleaned up the data. Immediately grouping attributes (behaviors, side-effects, medication, and mood).  


The original sketch and design imagined a 90-day mood and medication tracker using a line graph. I figured mood swings were obviously meant to be represented by large loops next to colored medications graphed by dosage (see Image 1). 

Then, I remembered heat-maps, and the entire design evolved. I parsed the horizontal design by stacking the three months (see Image 2). 

Image 2

Mood was measured with a scale, 3 times of day: Morning (8 am), Afternoon(1:00 pm), and Evening (8:00 pm). I used a color gradient of yellow to blue for the mood scale. 

For the various medication dosages, I used a single grey-warm gradient for mgs (0-900). I lined them horizontally split into three months beneath the main, larger mood-bar. 

The 30-day interactive prototype (which I have accidentally created a design over without capturing an image) included both the line and heat graph. 

User Testing

Design concerns

Again, I was frustrated with how to measure mood three times a day against medication taken daily.  I wanted to use three different time-slots for mood because, as my research indicated, moods vary drastically throughout the day for Bipolar II patients.

User tests

I recruited three users to test my mental health dashboard prototype. The goal of the test was to improve the design.

Takeaways from User 1:

— 30 days is not sufficient to deduct reasonings;  medication is not changed that quickly with out-patient.
— Side effects wouldn’t make a lot of sense over 90 days unless they appeared over a very long period which wouldn’t be the case if monitoring
— It is difficult to get patients to record mood a single time a day, let alone three

Takeaways from User 2:

— Vertical design might be better.
—  Not sure if enough time (data-wise) to answer task question.
— Overall, easy to use. Enjoyed duel view.
— Use generic name for medication

Takeaways from User 3:

— “Not a fan of graphs but enjoy the design and interactivity.”
—   Insightful to see how lows and highs correlate with medication

Class presentation

Due to class design, I got feedback from my peers before finishing the dashboard design. The main takeaway was to negotiate the mood scale problem.

Another insight was to change the color gradient since mood was not defined as happy vs sad but up vs down, since a 2 on the scale can mean users are still uncomfortable.

My peers also echoed the importance of interactivity and also encouraged the addition of an index, particularly on the mood scale.

Another insight was the lack of luster for a line graph. I was on the fence on whether to incorporate this element into the final design or not. 

One of the main takeaways from the user test was the need for a longer timespan. Therefore, I dedicated a large portion of time to adding 60 more days to the excel spreadsheets.

For the mood temporal question, I still captured the level thrice but I decided to average the daily moods taken 3x a day for analysis. I wasn’t sure whether a mid-day calculation would suffice, but since this data is constructed anyway, I wasn’t too concerned with the impact.

90-day View

The first dashboard (see Image 3) has a vertical design, which tested better in the prototype-phase for a 90-day scope.

I used secondary vertical bars with gradients to represent the various medication dosages. I decided to gray-out instead of keep the design transparent for better comparison. It was extremely difficult to periciely line up the bars. 

I removed the behavioral graphs because, as user tests indicated, over 90-days, the message was lost.  

I thought of using one color gradient and using a single legend for dosage in milligrams. However, because of the disparity in doses, the impact was lost. It made sense to assign original colors to match the line graph. 

Ultimately, I included the line graph view. I think it fits in nicely and the dips in the mood graph are impactful. The lifespans of the medication trials are also more apparent than the heat graph. 

30-day view

For the second dashboard for a monthly view, I used a horizontal design.

Instead of a rectangular band that reminds me of Pantone strips, I opted for circles for each day. Although not tested, I think the bubbles allow for better daily comparisons without obscuring weekly or longer-lasting patterns. For call-back, I used the same colors for the drugs. I decided I didn’t have to include the mood-scale legend since you enter the second dashboard by way of the first where it lives currently. 

I think this dashboard allows the patient and practitioner to go beyond paper and pen mood tracking to really understand behavioral patterns on a monthly and tri-monthly level. Treatment is a costly (both emotionally and literally) commitment that can leave a mental-health patient in more distress. This dashboard is an opportunity for autonomy.

The next step in the research process is to user test my hi-fi dashboard. I would like to focus on actual patients instead of practitioners. 

Besides collecting and negotiating user feedback, I think adding additional interactive properties would be beneficial. I am not well-versed in Tableau’s UI, but I am positive I am not making the most use of it. Instead of having separate boards for each month, I would like the user to be able to select a month and filter that way.  The user should be able to add their own side-effects and medication information.

On a design level, I was painfully frustrated with Tableau’s lack of artistic capabilities (Where is the css?! Illustrator integration, anyone?).  I want to investigate learning opportunities to enhance the design of this dashboard. 

From a data perspective, I think the effectiveness of interpretation is skewed depending on the user. I believe expert professionals in the medical community will take the insights as a tool, not a magic diagnosis wand. While I want to empower the user with knowledge about their pharmaceutical intake, I don’t want them to diagnose themselves. This dashboard is a tool that doesn’t work on it’s own, but should be part of a plethora of information used by the patient and their support team to further progress.