Prevent pneumonia in nursing homes

Project details





Details disclosable upon request

My Role:

UX Designer

My Tasks:

User Research
Market Research


Alex Heyison
Aubrey Kalashian
Bin Li
Emily Yang
Molly Mercer
Nick Shariat
Stephanie Wang


5 months

A product service system that employs machine learning capabilities for sound to help nurses screen for pneumonia in nursing homes. Combining software and hardware, our system aims to help nurses catch pneumonia early before the infection becomes deadly to the patient and others.


Pneumonia is associated with the highest mortality rate among nursing home residents
Pneumonia is a huge problem in the elder community, particularly within nursing homes, where conditions lead to increased risk for pneumonia and mortality is disproportionately high.

By identifying pneumonia early using breath sounds, interventions for older patients can be initiated earlier, and thus reduce morality and costly hospitalizations.


up to 40%

of mortality rate of nursing home patients with pneumonia



of all hospitalizations come from nursing home residents



most expensive condition for U.S. inpatient hospitalization. Billions of dollars are spent annually.

How can we support nurses to prevent pneumonia in nursing homes?

Solution Key Parts

View patients' health status in a quick-glance calendar overview

How Diagnosis Overview works
After using finger ID logging into a application, the nurse can quickly view the patient's health status, auto-detected symptoms, and recommendations through a calendar.
Why this feature
Nurses see a multitude of patients per day, and they often see different patients each day depending on rounds, this provides a quick summary of the patient's potential symptoms.
Design rationale
We included both color and number signals on the calendar to indicate abnormality levels so it's easy to spot there the level is major.

Review the details of a diagnosis on a specific day

How Diagnosis Detail works
The nurse can check detailed diagnoses, such as symptoms, potential diseases, and recommendations by date.
Why this feature
In the case there is a severe case, the nurse can easily dig deeper and send to doctors if needed.
Design Rationale
We made the call to action for diagnosis detail very clear as it's the most prominent button on the page.

Share multiple historical records with doctors

How the sharing works
The nurse can check and send the diagnostic records on different days to doctors for further guidance.
Why this feature
Nurses in nursing homes do not have frequent access to doctors so we included ways for nurses to send single or multiple date entries to doctors.
Design Rationale
We employed a filtered view pattern for navigation between historical entries. We decided on this navigational view since the content categories are consistent on each dayl

Value Propositions

Identify and flag patients in the early stages of infection to plan interventions and to reduce hospitalizations
“This is helpful because sometimes we might fail to recognize something we’re listening to due to all the sounds around us.” 
- Nurse

“If diagnosis is suggested, I can start doing tests and take precautions early.”
- Nurse

Reduce the stress and free up time for overburdened skilled medical staff
“This is really helpful, especially if you have a lot of patients and if you get new patients, you can easily see their diagnosis, recommendations and assessments.”
- Nurse 

““This avoids errors, mistakes and time wastage.”
- Nurse

Build stronger communications between the patients, providers and caregivers to improve patient outcomes
“This would save me a lot of steps and actually alert doctors immediately to the symptoms we’re seeing.”
- Nurse 

“If I could have this on my medical cart as I’m giving medications, it could save me time from calling or faxing the doctor, I could just share this from one station.”
- Nurse


Scoping the right use case for a sound machine learning algorithm technology
We were first presented with a technology, a sound algorithm and were tasked with scoping an application for it. After doing market research, we rapidly ideated on use cases where certain industries would make sense (e.g. mobility, industrial tech, security, healthcare).
Concept testing with storyboards
We created multiple storyboards for each industry based on our understanding of the competitive landscape and our client's goals. Then, we tested them with users from each area. Healthcare got the most enthusiasm.
Deciding on healthcare based on market trends and  strong positive user feedback
“If sound can pick up heart murmur, it can be very helpful – it’s very hard to hear, normally means faulty value and if can pick up on sooner the better"
- Physician’s Assistant
Glimpsing into the medical professional world to dig deep into how sound is used in diagnosing conditions
Given that our target audience of medical professionals are hard to reach, especially in a pandemic situations, we immersed ourselves in YouTube videos to gain a bare understanding of how stethoscopes work.
Designing mid-fidelity wireframes to test initial assumptions with medical professionals
Assumption 1
It can be difficult for some doctors to identify patient abnormality using their current stethoscope.
Assumption 2
There is clear distinction between the sound of the heart & lung & that it would be beneficial to be able to differentiate both sound. 
Pivoting target audience based on user research insights

Insight 1

Medical professionals find value in reassurance of diagnosis, especially for the untrained ear
“Helpful for people who are not as confident, especially for those that would want to continue to improve”
- Physician's Assistant
“Would be really nice to have a way to double check what you listen to”
- Physician's Assistant

Insight 2

History of the audio files is valuable for all medical providers
“If you listen to multiple patients a day and then a month later you need to know if something has gotten better/worse, history would be really helpful”
- Physician's Assistant
“History allows you to go back and see has this changed from the last time I listened to this persons”
- Physician's Assistant

It's difficult to identify early symptoms of pneumonia in nursing homes

Discovering pain points for variables involved in nursing home care
We conducted secondary research online to understand pain points for providers, patients and the disease when it comes to pneumonia. We also chatted with a medical professional who had experience working in nursing homes.


Nurses have limited time & attention with each patient

Doctors don't visit nursing homes frequently


Older adults are less likely to complain of their symptoms


Pre-existing conditions can mask the onset of infection 

The presentation of symptoms can be much more subtle in the elderly

Updating mid-fi wireframes based on pivot to address nurses' pain points
Simplifying the diagnosis

Since this is now targeted towards the non-specialists, we removed the details that would've been for specialists. This page still show the symptoms recognized by the device, the disease that the patient may have, and recommendations to help the nurse take care of the patient better.

Usability testing on nurses to see if the flow makes sense

Task 1

Find today's diagnosis

Find Richard’s diagnosis for today

Find Julie’s new symptoms

Task 2

Look at history

Find out Julie’s diagnosis from yesterday

Task 3

Share with doctor

Share today’s diagnosis with doctor

Share past three day’s diagnosis with doctor

Final Design

Diagnosis Overview

Based on the smarthoscope algorithm analysis, automatically records the patient's health status and list symptoms and recommendations by date to give nurses a quick overview of the patient's physical condition.
Diagnosis Detail
A detailed diagnosis report by date with detected symptoms, potential disease, recommendations, and original audio files to guide them to take care of the patient.
Previous records by date, can be sorted and filtered by date, abnormality location, level and the nurse who used the smarthoscope.
Nurses can easily share multiple history records with doctors.


If I had more time, I would
  • Visit a nursing home to do contextual inquiry.
  • Design a physical prototype of the hardware component to test with the software.
If I could go back in time, I would
  • Communicate with client and team to scope down use case earlier.
  • Conduct more secondary research on how sound machine learning algorithms work.