Patient Scheduling Advisor

I led the user-centered design for a proof-of-concept AI scheduling tool that helped cancer ward schedulers create optimal patient schedules while incorporating patient preferences. Working with a large Houston hospital, I designed a complementary application that integrated with Epic medical software to balance over 100 scheduling rules with individual patient needs.

Two computer monitors displaying medical scheduling software, showing patient details, scheduling preferences, and appointment times for a patient named Sam Houston.

My Role & Approach

Key contributions

  • Led user-centered design for initial use case discovery workshop with hospital staff

  • Facilitated proto-persona creation with patient scheduling team and medical staff

  • Conducted user research through interviews and testing with patient schedulers

  • Designed patient preference collection workflow integrated with check-in process

  • Iterated scheduling interface based on changing requirements, API constraints, and user feedback

The Challenge

Design an application for scheduling staff of a large cancer ward to create optimal patient schedules that:

  • Pulled appointment data from Epic medical software

  • Accounted for patient preferences

  • Managed over 100 scheduling rules

  • Fit within existing scheduler workflows

My Approach

I led the user-centered design elements of this proof of concept, which included:

  • Workshop facilitation: Leading use case discovery with hospital staff

  • Cross-functional collaboration: Working alongside the project manager, machine learning experts, and front-end developers

  • User research: Meeting with patient schedulers to conduct testing and validate design assumptions

  • Iterative design: Refining designs based on user feedback and changing technical constraints

Understanding Schedulers and Patients

Initial Two-Day Workshop

I facilitated a two-day workshop with the patient scheduling team and associated medical staff to ground the project in user needs.

Proto-Persona Creation: I led the team through creating proto-personas based on identified patient types. This exercise ensured everyone approached the project from a user-centered perspective rather than starting with technical capabilities.

Goal Evolution: Through the workshop, the project goal evolved from a loosely defined "optimizing of schedules" to a more focused solution: a complementary application that sat alongside Epic medical software. This application would:

  • Incorporate patient preferences

  • Account for the 100+ scheduling rules patient schedulers must consider

  • Integrate with existing workflows rather than replacing them

Key Personas Identified: Based on the workshop, we identified distinct patient types with different scheduling needs:

  • Local patients with flexible availability

  • Out-of-state patients preferring concentrated appointment schedules

  • Working patients with limited weekday availability

  • Patients with mobility challenges requiring specific location considerations

  • Elderly patients with varying levels of technology comfort

Workshop whiteboard capturing proto-persona creation and initial use case exploration with hospital staff

Preference-Informed Scheduling

Collecting Patient Preferences

To incorporate and learn from patient preferences, we first needed to design a mechanism to collect them.

Options Explored:

  • Online portal: Leveraging the hospital's existing patient portal

  • Mail questionnaires: Sending preference forms by mail

Research Insights: Through interviews with schedulers, I uncovered critical concerns:

  • Computer literacy barriers: Many elderly patients struggled with online systems

  • Low response rates: Mail questionnaires historically had poor completion rates

  • Timing issues: Both methods introduced delays before scheduling could begin

Solution: Check-In Integration Based on scheduler insights, I identified an opportunity within the existing check-in process. As patients arrived on the ward, front desk staff could collect preferences as part of standard intake.

Questionnaire Design

I designed a brief questionnaire that front desk staff would complete with patients, capturing:

  1. Permanent address verification - Identifying travel distance considerations

  2. Schedule duration preference - Short (1-2 days) vs. spread out (3-7 days)

  3. Availability - Days/times the patient is unavailable (AM/PM grid by day of week)

  4. Location preferences - Preferred zones for lab work and imaging procedures

This approach:

  • Required no additional technology adoption from patients

  • Fit naturally into existing workflows

  • Achieved high completion rates

  • Provided preferences immediately when needed

Screenshots of a medical appointment scheduling interface with a list of patients, a survey form for patient Sam Houston regarding appointment satisfaction, scheduling preferences, and test timing, and a confirmation message for recording responses.
Medical patient scheduling form on a computer screen. Includes patient name Sam Houston, MRN 123456, and appointment preferences such as address verification, follow-up visit duration, patient availability, and preferred imaging or lab locations.

Scheduler Interface Design

Initial Approach: Multiple Schedule Options I initially explored providing multiple schedule options, allowing schedulers to pick the best fit for each patient. However, user testing revealed critical issues:

  • API overhead: Multiple schedule generation required excessive calls to Epic

  • Analysis paralysis: Schedulers struggled to compare and choose between options

  • Time prohibitive: The comparison process slowed rather than accelerated scheduling

Refined Solution: Single Optimized Schedule Based on this feedback, I pivoted to providing a single recommended schedule. This approach:

  • Reduced technical overhead through fewer API calls

  • Eliminated decision paralysis for schedulers

  • Maintained scheduler control through refresh capability

Handling Real-Time Conflicts: As schedulers worked through appointments, availability in Epic could change. I designed a workflow where schedulers:

  1. Progress through the suggested schedule

  2. When encountering conflicts with Epic availability, lock completed appointments

  3. Refresh remaining appointments to regenerate based on current constraints

This approach:

  • Reduced cognitive load by focusing on one schedule at a time

  • Minimized API calls through selective refresh

  • Preserved scheduler progress through appointment locking

Screen displaying a patient scheduling assistant interface for Sam Houston with appointment details, schedule filter options, and a feedback section.

Reverse Chronological Display

Through scheduler interviews, I learned that all appointments were dependent on the primary care provider appointment, which determined the patient's treatment timeline.

Design Decision: I designed the schedule to display in reverse chronological order, starting with the primary care provider appointment and working backward through prep, travel, imaging, and lab appointments.

Supporting Features:

  • Travel time calculations: Automatically accounted for movement between hospital locations

  • Prep time indicators: Showed required preparation time before procedures

  • Location context: Clearly displayed which zone/building each appointment would occur in

This display order matched schedulers' mental model of how they thought about the scheduling problem, reducing cognitive translation time.

Impact & Outcomes

Proof of Concept Success

Validated Approach:

  • Demonstrated feasibility of preference-informed scheduling

  • Showed integration potential with Epic medical software

  • Proved value of machine learning in healthcare scheduling optimization

Scheduler Feedback:

  • Appreciated the single-schedule approach over multiple options

  • Valued the reverse chronological display matching their workflow

  • Found the lock-and-refresh mechanism intuitive for handling real-time changes

Strategic Value

For Healthcare Schedulers:

  • Reduced manual effort in balancing 100+ scheduling rules

  • Incorporated patient preferences that were previously difficult to track

  • Maintained scheduler expertise and control while providing AI assistance

For Patients:

  • Voice in the scheduling process through preference collection

  • More convenient appointment timing

  • Reduced travel burden through location optimization