Discover potential diagnoses through user routines

The MySense app works with physical sensors that collect data from individuals' movements and interactions, such as sleeping, using the kettle, or opening the front door. The Trends function aims to provide insights that help stakeholders (clinicians or individuals’ next of kin) discover potential diagnoses earlier by recognise individuals' routine changes.

My Role Contribute research and interviews, define design direction, refine product strategy with PM, drive discovery workshops with related team

Business Goals 1. Create a user experience for the new functionality to address the defined problems. 2. Establish measurements to evaluate the impact and performance of the new design approach

Key Process Stakeholder interviews > discover product strategy and new opportunity with PM > workshops with internal teams > transfer insights and learning into design approach > qualitative testing with stakeholder and performance evaluation

Problem. Impact. Solution

While we had the previous technology to present data to clinicians, user research revealed that they spend most of their time exploring individuals' health data without clear direction. As more individuals onboarded, clinicians struggle to manage their workload while feeling stressed about potentially missing important health issues. We believed this problem is worth to be addressed in order to make the following measurable impacts:

Reduce Workload By providing more efficient use of the product, we can reduce clinicians workload for each individual. This becomes increasingly important as more individuals join the platform.

Reduce Errors Monitoring data from big volume of individuals is stressful for clinicians. A better way of presenting data could help reduce errors that occur when managing large amounts of information.

Improve Data Structure The analysis from our Data Science team shows ways to improve backend data processing to reduce loading times and bugs. This would significantly improve the product performance.

Hypothesis

We believe that improving the data presentation (through data structure and user experiences) will enhance stakeholders' understanding, particularly regarding clinicians' workload efficiency and accuracy

We will validate this hypothesis when we see significant changes in these metrics from clinicians:

⏱️ Decrease in tasks completion time per individual

We applied the raw data processing to the backend and implemented calculations to simplify the data structure. This allows us to present clinicians with summaries based on their criteria, helping them understand the data more easily.

💯 Reduction in clinical errors notes

We believed that errors were related to the complexity of clinician tasks. They mentioned that more data was presented on the page, which required more time to find any health issues and potentially missed out on important health relationships.

⚡️ Faster data loading speeds

Instead of loading raw data directly into various charts, we worked with the team to evaluate the existing data structure and figure out how to transform it into something both lighter and more straightforward.

Exploratory Data vs…

The previous system presented data in an exploratory way, displaying information across different charts for clinicians to explore and identify unusual patterns in individuals' routines. However, as the insights above show, this approach is time-consuming and increases clinician workload, especially since some clinicians need to monitor around 50 individuals per day.

Explanatory Data

Instead of make them search for Wally, what if we bring them Wally?

Based on the defined criteria, we believed an explanatory data approach could work well for the user cases and align with the new data structure. This approach also allow us to simplify the data presentation so we can bring out the core message to the clinician directly.

We began by categorizing the data into groups to connect relevant data with its presentation. This helped us define and prioritise tasks for different teams based on difficulty and potential technical limitations. We then initiated a Proof of Concept (POC) process across Design, Front End, and Data Science teams to evaluate outcomes and measure impact against our defined KPIs.

Sketch it. Make it. Evaluate it

Based on the assigned responsibility for the new product design direction, we first mapped the new data structure into visualisations and communicate it effectively with stakeholders. Trends features 3 types of data visualization, each displaying distinct kinds of data.

🧮 Counting This data visualisation applies to countable data that continuously accumulates from zero.

⚖️ Range/ Rate This data visualisation shows measurements that fall between two specific values (ranges).

⏰ Time This data visualization displays time-based data (specific times or periods during the day).

The System

With the defined data types, we designed a methodology to apply the new data structure as cards. This approach aims to summarise recent significant health pattern changes by comparing the recent collected data to the previous 7 days data. Essentially, a card’s appearance indicates something unusual, which addresses the issue of indirect data delivery.

Level of Complexity

By collaborating with the clinician, we both believed that the explanatory data was achieving our intended purpose. However, we also wanted to preserve the ability of discovery from exploratory data in specific cases. Therefore, we decided to structure our data delivery in two levels, based on the level of complexity that the stakeholders wanted to see.

User Experiences

All Cards in Trends are interactive, enabling clinicians to ‘zoom in’ on a specific topic they wish to deep dive into. This feature allows them to choose the level of complexity they prefer, such as reviewing daily data from the past 7 days or leaving notes. Additionally, it directs them to the relevant sections of the product.

The Results

We launched the first version of Trends following the POC. We implemented some changes based on identified limitations and speed testing of the process to ensure the functionality’s reliability. We monitored Trends’ performance through the backend to assess whether it was meeting the defined metrics from our criteria.

Testing Methodology →

70% Faster

Tasks speed improved

It is socking to see a huge improvement in the completion time of tasks involving over 50 individual tasks. Clinicians have expressed that the new approach enables them to prioritise tasks effectively, as they no longer waste time on individuals who are in a routine of good health. This allows them to allocate more time to those who genuinely require attention. They mentioned it is helpful specially when more individuals on boarded.

35% Less

Errors reduced

With the new approach of the data visualisation, errors are reduced by hiding unrelated data, which is the primary reason. It also helps clinicians focus on actual health issues from individual health changes, allowing them to better examine any unusual changes instead of sifting through all data to find them.

60% Faster

Loading speed improved (landing page)

The simplification and reduction of the data is definitely help a lot with the loading speed. With Data Science and Back End Team amazing work, we managed to achieve 60% faster loading speed. Even complex-level data can now be processed at a faster speed with this new approach.

What’s Next

Although the performance of Trends exceeds our expectations based on our measurements, we’ve identified a few areas that we need to address in the next version, as per clinicians’ feedback.

We’ve analysed all the feedback and come up with the following design directions to enhance the functionality of Trends while also introducing some specific features that we could incorporate into the product.

Data Relation
The new approach simplifies the data structure, making it easier for clinicians to detach individual routine changes and provide a direction for health analysis. However, it becomes challenging to analyze multiple matrices simultaneously.
This insight prompted us to consider the detailed view of each data point and the data relationships that can be established between related data. By connecting these related data points, we can create a comprehensive insight based on a single significant individual routine change.

Data Comparisons
Clinicians found the Trends insights helpful in identifying short-term pattern changes. They can analyse individual health data using this feature. However, they need to compare the recent matrix to the standard matrix for that individual (a baseline representing a healthy day, such as the first day they used the MySense sensor). Current they have a data based of that baseline so they can compare it themselves.
We think this is an interesting feature to be look into but we are not sure if this is beyond the purpose of Trends. We will discuss it with the team to explore how we can develop this feature along with Trends.

Data Timeframe
Clinicians can select a specific date to review any significant routine changes based on the data for that date and the previous 7 days. This short-term pattern changes can help them identify potential diagnoses. However, long-term pattern changes are also necessary for clinicians to monitor for repeated cycles over months to prove the accuracy of the analysis.
We were already working on long-term trends, which cover health data changes for a month. Unfortunately, the performance did not pass QA due to a long loading time and limitations in timeframe selection. We will continue working on this feature and bring it to the next version of Trends.

All user data and quantities in this case study have been adjusted in accordance with individuals’ data protection policies and the company's NDA.

More case studies

🕵️ MySense / Framing Hypothesis →

👩‍✈️ Lufthansa / Cabin System →

👩‍⚕️ NHS / Testing & Feedback →