Low English Proficiency Application

LEP is a software platform that enables quick translation and communication between nurses and patients/caregivers who don’t speak English.

I joined the Low English Proficiency project just after it was launched with Emory and the Children's Healthcare of Atlanta - CHoA hospitals. The MVP had Spanish-English translations. We now have French, Portuguese, and 6 other languages.

Role: Interaction Designer

Motivation behind LEP

Increase communication from hospital staff towards patients
Improve patient’s experience at the hospital
Improve the experience of healthcare providers, which was previously negative with video interpreters and led to delays in care.
Reduce costs and utilization of video/audio interpreter services for short (1 - 2 min) interactions

The app is not meant to replace in-person interpreters. Rather, it is intended to augment conversation experiences where interpreters are unavailable.

This is what the usage of this tool would look like - I've chosen the Day Surgery Unit of CHoA for this example. Whenever an LEP patient schedules an appointment, they request translator services. On the day of the appointment, once they check-in, they are in the triage area, where the preliminary assessment of patients happens. They check vitals and ask questions like when was the last time you or your child ate, or drank, depending on the age of the child. Most of the time, in-person translators are unavailable during this time and the staff has to rely on video interpreters. This is where our app comes in handy. Every member of the hospital staff has an issued Volte iPhone SE, which they use for their work. The LEP app is already downloaded on the phone. It has preset cards for these questions and statements that nurse practitioners and other staff members can click through and communicate.

When the patients are with the doctor, interpreters are usually available, since this is where the most technical conversations happen.

Once the doctor is done with the appointment, or surgery in the case of this unit, the patient and their caregivers communicate with nurses again, where they check their vitals, and recovery and cater to other requirements. This is another place where our app will be used similar to the triage area.

The duration of this entire interaction can vary from patient to patient and on the unit.

When I joined, the application was limited to static phrases with frequently required conversation points. Through the course of the summer, we have worked on expanding the scope of utilization of the tool.

UI Audit: What comprehensive changes, including visual adjustments, can enhance overall UX?

To get myself acquainted with the application, I explored the usage of the tool in different scenarios, for example, a Nurse is on rounds to check the vitals of a child admitted to the hospital. While clicking through the features, I did a visual audit assessing elements of design components: the hierarchy of information, colors, typography, icons, layout, and overall aesthetics. The primary goal of the audit was to identify areas for improvement in the interface design to enhance user experience.

Suggestions from the audit included changing the way CHOA colors were used in the application to increase contrast and hence accessibility, reducing the amount of scrolling required on the homepage, and increasing consistency in iconography, among others.

I needed to understand the context I had to work with better. This is why I started visiting the hospital to see how staff-patient interactions happen, and to talk to them about their application usage.

How do healthcare providers and low English proficiency patients respond to a translation application in a hospital setting?

Learning from observations and interviews

Broadly, the observation sessions and interviews helped identify design requirements, further research areas, and translation issues. It also helped us work further towards content creation and curation. I'll exemplify some of these.

The research was continued after these initial findings - with patients, interpreters, and staff members. I identified more themes like designing for trust and potential avenues for intervention.

Since the mistrust issue due to incorrect translation was highlighted in the research, I started maintaining a log of Active Translation Issues that were medical, cultural, and grammatical. These came from user research and regular testing that we conducted. We aim to rectify these as soon as possible for better UX.

Increasing Retention Rate

Parallel to conducting research, I started ideating about increasing the retention rate of the application. This stemmed from the observation and conversations at CHoA that staff members knew about the application since when it was first introduced. They explored it once but never used it again. We had to increase awareness about the application, not just for acquisition, but for retention as well.

Popular methods of retention like gamification, referral programs, and having a large acquisition rate won’t work in our case. We have a fixed target audience and were seeing a drop in the retention rate. We saw a peak when we launched the app and went on the ground to talk to the staff about it. The next two weeks saw a significant drop. The number went up slightly two weeks after that - after I went to conduct research and spoke to the staff about the app. However, there has been a steady decline in our Daily Active Users. Thus, instead of focusing on new users, we decided to maintain our focus on retaining the ones we already had.

Elements of gamification and referral programs are extremely popular in the industry today as methods to keep people engaged. They keep the user hooked by playing with progress bars, statistics, and rewards to keep them coming back. However, in our case where we have important healthcare-related conversations happening, we need the app to be streamlined.

I broke down my ideation into design, development, and marketing.

I planned to use onboarding as a potential area to increase user awareness about how the app works, scenarios they could use it in, and new features that we would launch, among others. I conducted a short experiment by launching an onboarding screen for the day surgery department for a week.

We were open to the idea of having personalized interactions, but we have to keep in mind that the staff gets a new Volte phone every time they clock in, so we can not rely on phone memory - every session is new. We can incorporate login functionality, as discussed when talking about research findings. The overall design should be accessible and usable by low-technical-proficiency people. Push notifications were another area that caught my interest. It can be used to remind people subtly about using the app. We later discussed the possibility of linking it with our data tracking. For instance, if a button was clicked multiple times within a few minutes, we could ask the healthcare provider if they had any trouble using the app. Or if they had to scroll a lot, a notification could be sent later to ask if they had trouble finding something they needed.

Tracking and feedback are also included on the development front, along with the live translation feature and introduction of other languages than Spanish - which were in the pipeline.

Seeing the number of daily active users increases slightly after I went for research reinforced the idea of regular, offline feature updates and workshops with staff members. The walls in CHoA are full of colorful posters - which birthed the idea of patient and caregiver awareness through posters and flyers.

Design Tasks

I tackled introducing new languages and ensuring a quick switch between these pages head-on. I explored iterations between popups and drop-downs and introduced flags next to languages as compared to writing the first letter of the script. We chose to stick to popups due to ease of development and replicability across different pages on the app. Using the script was chosen against, since all Latin languages would end up with the letter A - defeating the purpose of quicker recognition. The colors associated with the languages were picked from the flag with changed opacity.

On the cards page, I explored iterations similar to the home page and the option of introducing just the flag in the top right corner. We decided to stick with the former to introduce consistency and familiarity with the color chosen. Clicking the task card, the page in the third image opens up. I added the dialogue box card and reduced the icon size to incorporate bigger sentences in other languages - that are relatively shorter in English and Spanish.

Following the language change functionality was the interesting challenge of designing the active translation page. I spent some time thinking about the constraints I had to work with, which included medically accurate and culturally sensitive content. The overall design had to be usable and accessible by a not-so-technically enabled audience with intuitive navigation. The system would also be technically constrained to pick up phrases to translate in a high interaction volume environment, due to which the design needs to be responsive, with real-time feedback like system status.

On the left of the screen are screenshots of the current feature. Based on the constraints I had defined, I designed the version on the right. It had the same features as the first one, I was playing around with visual styles to communicate the same things.

We decided to explore this further with you all in two brainstorming sessions. I was well within the context, but fresh eyes would introduce new ideas and add value to the design. The first brainstorming session included discussing constraints and ideating on features individually. The following session was focused on the interaction, where we first wrote down elements of interaction, which were followed by the identification of design requirements and research needs. To validate our thoughts further, we separated into pairs and did a role play as a hospital staff member and a patient, both speaking different languages.

Design requirements from the brainstorming session included easy and quick switching between English and the other language used for efficient communication between the 2 stakeholders involved.

We discussed the importance of context. While trying out active translation, we realized that sometimes words are translated. While the translations are correct, they don't often tie together in a sentence correctly. In cases like this, context can help people understand what the other person is saying. Saving chat was a proposal to ensure context was at the forefront.

Another idea generated was to have an “I understand”/ “I do not understand” mechanism to ensure that both parties are on the same page. If not, they could use a different selection of phrases to explain their statements.

On the UI front, it should be indicated when the patient should start talking and when to stop for optimal translation and to avoid the need to repeat their sentences. Having UI elements that indicate the state of the system including when audio is being picked, translating, and error states, among others would increase the responsiveness of the design and trust in technology. This UX heuristic of system status visibility will help users with low technical proficiency be aware of what's going on.

In the coming months, we want the patients and their families to be aware of the app and its usage for ease of use. Hence, the proposal for offline and online onboarding. It was also recommended that we refer to the design of current tools like Google Translate and maintain a level of similarity for quicker adoption and integration into the system.

On the translation front, the need for translation to be grammatically, culturally, and medically accurate was reiterated. Validating the transcript before translating as the first step to correct translations was discussed. This is to increase the patient and their family's trust in the technology. It was suggested that we adopt the method of segmenting speech to allow users to edit the words/phrases instead of re-recording them.

Back translating method - where the translated text is translated back into the first language to ensure its correctness was a suggestion that we could potentially explore.

Content Creation and Curation

Further, catering to the other feature and phrase request segment of our conversations with the staff, I worked on Content Creation and Curation. Here, I added new phrases and reorganized the ones we already had to search for frequent phrases effectively. Since this was manual work, I wrote Python scripts to manage, clean, and reorganize data.

Active Research and Design Areas

In all the mini-projects I was working on for this application, there were research and design requirements that I identified. Some I was actively worked on, others added to future work. I started maintaining pages on notions that catered to the question we had, what research backed up the hypothesis, and the plan of action to work on it. These included questions like "Do patients trust technology?" and "Is the translation accurate?". On the design front, the need for an organized Unit and Role selection for personalization of content and the idea of grouping all food, drink, pain, etc related cards were on the list.

What's next?

While the two lists I just mentioned cater to a lot of the future work for this project, there are a couple of things I started working on that are the immediate next steps, along with locking in the translation page design.

These include removing phrases that haven't been used to reduce visual overload on the home page. This would lead to information reorganizing and locating. We have started tracking scroll and click behaviors on the home page to curate the cards more effectively. Less time spent searching on the app would make the conversation between hospital staff and LEP smoother.
Discussion on the merit of having a search or filtering mechanism.
Explore further what data should be tracked and can be tracked with Pendo, the 3rd party software we have been using. The identified statistics will be used towards the same goal of making conversations systematic.
As pointed out before, handling all the content across different languages is manual work. Writing code to manage some of the data made it efficient. However, the current structure is not scalable. Replacing CSV files with table-based management systems will increase the readability of content.

Stay tuned for more work on this project!