Joint Dx

An AI based clinical decision support tool that accurately diagnoses and recommends treatment plans for patients with hip and/or knee pain from arthritis.
Roles
Lead Designer
Researcher
Skills
UXUI Design
UX Research
Collaborators
Product Manager: DJ Vaglia
Project Manager: James Bucki
Software Engineer: August Von Trapp
Medical Director: Dr. Tony DiGioia, MD
Patient Feedback Collection
X-ray Upload with ML Sorting
Care Pathway Recommendation
X-ray Review to account
for ML disagreement

Problem Space

1

An acute need to accurately and efficiently triage patients with hip and knee pain as clinically operative or non-operative before they see a specialist provider.

2

Lack of timely access to all providers. Patients may need to wait months for a proper evaluation and diagnosis.

3

Providers lose the opportunity to intervene. This prevents them from providing care plans early on to slow down the progression of joint degeneration and engage patients in their own care.
"Nearly one third of all patient complaints for PCP office visits are related to musculoskeletal problems, of which the biggest portion is related to the hip and knee."
Solutions for Primary Care Physicians:
  • Find the right provider to refer patients.
  • Educate patients and providers alike to understand their care plan and advocate for  patients.
  • Joint Dx combines feedback directly from patients on their symptoms with standard X-rays to recommend treatment pathways.
“For orthopedic practices, there is a need to efficiently and accurately triage patients that can start non-operative care vs those that are candidates for surgery."
Solutions for Orthopaedic Practices:
  • Designed with busy clinics in mind. Joint Dx is user-friendly to make triaging new patients efficient for clinic staff.
  • Quickly and accurately schedule new patients into surgical or non-surgical providers
Potential systemic impacts in clinic work flows :
  • Avoid or defer 15-30% of hip and knee replacement surgeries
  • Decrease the overall cost of uncoordinated patient care pathways by 10x.
  • Reduce a specialist appointment wait time by 1/2 in high volume clinics.

Leveraging existing features and a new algorithm

Our goal was to leverage 2 key components to mimic an in clinic triage process as closely as possible to create a shared decision making tool:
Patient Reported Outcome Surveys (PROS)
It is common practice for clinics to collect intake surveys that assess the severity of their condition. Using this data will account for the patient’s personal concerns in the final care pathway recommendation.
A Machine Learning Algorithm
Our team trained a Machine Learning algorithm to detect arthritis in knees and hips using over 10,000 clinically evaluated x-rays, with a current accuracy of 96%. Integrating AI into our tool is essential to efficiently determine the level of degradation in a patients joint.

How might we design an MVP that can simulate an orthopedic clinic triage process?

Research

Understanding the clinic flow and stakeholders
We shadowed the triage process of 2 orthopedic clinics to compare the similarities and differences between clinics.
This helped us identify key stakeholders within the clinic flow and each of their actions during an important touchpoint of the triage process.
From this, we compiled a comprehensive stakeholder list and mapped out where each of them make an impact in a customer journey map generalized based off our shadowing experiences at the clinics. Each stakeholder had specific needs they wanted to address throughout the process.
Key Pain Points
Patients
  • Dislikes filling out forms and surveys which are taken as for each clinic
  • Wants to expedite the time between getting a consult and receiving a form of treatment.
  • Majority are indifferent to receiving extensive education.
Clinicians
  • Pressured by hospital to drive revenue.
  • Patients with high priority can’t be filtered quickly to be seen.
Ideal State:
Patients
  • Less waiting time between each appointment.
  • Less filling out paperwork.
  • Less back and forth between different clinics to do testing.
Clinicians
  • Patient reported outcomes are collected as a standardized performance measure and for insurance reimbursements
  • An easy to use platform that centralizes x-ray and survey collection.
  • Configure and adjust care pathway suggestions to each patient’s needs.
Market Research on Product Viability

Learnings

  1. Leveraging AI technology in X-rays along with personalized feedback to triage a patients with bone and joint health is yet to be fully adopted in the medical field
  2. There is still a lot of resistance to change in Orthopedics and Radiology.
  3. Our target product is unique in that it is trained and backed by some of the leaders in the field of Orthopedic Surgery giving the technology credibility and less bias.
  4. Every clinic is different, a successful product will have the ability to white label and customize features based off each clinic’s needs.

What are our goals for the MVP?

MVP Flow Map
Based off existing Electronic Health Record Software and talking with clinical staff, we identified the key features to develop and design to start a pilot with potential collaborators. Important design considerations we had in mind when we started:
  • All patients entered for triaging should be displayed in one screen
  • X-rays should be automatically sorted and reviewed.
  • PROs should be easy to access and fill out for both patients and providers.
  • There should be limited screens and navigation between each should be straightforward.

Initial Iterations

Our initial designs aimed to capture the key features that we outlined in our MVP flow map.

The goal with this round of designs was to give our users some visual reference to give additional feedback on how it could realistically be implemented in their current workflow without disrupting it.
Feedback
Main Dashboard
  • All the essential information required for each patient is correct.
  • Visually overcrowded, this makes navigation unfamiliar.
  • X-rays being sent directly from the Radiologists into the system would be ideal.
  • There is no direct way to edit patient information after creating.
Patient Reported Outcome Scores
  • The most important score that doctors take into consideration is the “Overall Knee Health” the other three scores aren’t looked at in detail.
  • The scores in red immediately draws attention but are not the ones that needs to be considered.
  • Needs better visual hierarchy to distinguish Overall Knee Health from other scores.
After talking to the clinic they said that having patients fill out surveys before arriving to their clinic is ideal, but they don’t have any form collection established with their existing Electronic Health Record system so designing the PRO collection feature should be priority.
Things to consider when we are designing survey collection UI:
  • Legibility for senior aged patients
  • Accessibility across multiple devices
  • Providers should have the option to edit submitted entries if incomplete or it contains misinformation
  • PRO trendine analysis of patient history.

Mid-fidelity Iterations

Interactive Prototypes (Click on the GIF to access the figma file)
Features we focused on:
  • Refining the layout of the patient list dashboard
  • Designing survey UIs for multiple devices as well as improving accessibility for senior users.
  • PRO presentation for providers to quickly assess patient’s care pathway.
  • Data visualizing the PRO scores to track patient’s pain progression.
General Feedback:
  • Showing one PRO score for each degenerative joint per card in the exam room comprehensively captures the patient’s prognosis.
  • The PRO trend line feature isn’t a feature that is integral to the triage process for the MVP so we can remove it from our priority list.
  • Survey sharing, access and usability all well received by users.
  • The primary blue color feels a bit harsh against the otherwise muted color palette.

Final Designs

Features we focused on:
  • Incorporating the X-ray upload and ML tagging feature into the analysis.
  • Refining the visual language across the whole platform.
  • Care Pathway recommendation screen combines both PRO score and X-ray to suggest an accurate treatment.
  • X-rays upload feature interaction and sorting.
General Feedback:
  • Screens feel less crowded with the new blue.
  • There are less screens as we took out the PRO history data visualization. The focus is now on quickly determining a patient’s treatment.
Feature Highlights

1

PRO Survey Collection

2

X-ray Upload

3

Exam Room

Impact

Currently we are piloting at an The Bone and Joint Center Clinic in Pittsburgh. A physician's assistant collects this information via an intake call to determine whether or not a caller is a candidate for surgery.
For Patients
Surveyed 10 patients and asked them to compare their experiences filling out forms on paper and online. 8 out of 10 of the patients preferred having a digital setup on the office tablets, stating better legibility and ease of use compared to the paper version of the form.
For Providers
With the information readily available, the average time it takes to evaluate a patient from beginning to end takes only approximately 5 minutes compared to having a patient having to come in for a 1+ hour appointment not knowing if they need surgery or not.

Clinical staff also praised "the ease of use for someone who isn't technically capable" which makes it quickly adaptable to different clinical settings.