Closing the Care Gap
Patients at a free clinic waited more than a year for specialist care. I researched why, then designed the referral workflow that got them seen in about two weeks.
The case study follows the Double Diamond. Each phase opens with the question it had to answer, and every research method shows why I chose it and what it missed. Blocks marked Try it are interactive. Patient details are fictional and figures are illustrative, per HIPAA.
“Can you help with referrals?” There was no workflow, no owner and a drawer full of paper. A process existed on paper, but not in practice.
Discover
What this phase answers: where do referrals actually break, and for whom? I couldn’t study a documented process, so I reconstructed the real one from records, observation and interviews.
Four questions framed the research.
I wrote them down before choosing any method.
Where do referrals stall, and for how long?
RQ1 · Audit, shadowingWho thinks they own each step?
RQ2 · InterviewsWhat does each specialist office need?
RQ3 · Office callsWhat stops patients from doing their part?
RQ4 · Patient interviews
Nine methods, two diamonds.
Go wide to understand the problem, narrow to define it, then wide and narrow again to build the fix. Tap a method to see why I chose it.
The drawer: 214 referrals, coded one by one.
Each dot is one paper referral I found on day one. I logged its age, specialty and the step where it stopped moving. Switch the view to see the same backlog three ways.
Same four questions, eighteen different answers.
I called every partner office in two weeks. Each tile is one office, badged with how it wanted to be contacted.
Define
What this phase answers: what is the real problem, and who are we solving it for? I turned 142 research notes into themes, personas, a journey map and one problem statement.
From scattered notes to five themes.
A sample of 20 notes from the wall. Each note is one observation from one source. Clustering them by what they had in common, rather than by who said them, is what revealed the pattern.
Quotes are composites, paraphrased and de-identified.
Five themes became five design requirements.
Each theme from the wall turned into one thing the design had to do.
Three people, one referral.
Composite personas built from interviews and shadowing. The coordinator is the primary user of what I built. The patient is who it’s for.
The referral journey, before and after.
Same patient, same clinic, same specialists. The bottom row tracks how the patient felt at each stage. Select a stage to see the pain point and the design opportunity it created.
Front-desk coordinators at a free clinic need one visible place to move every referral to its next step, because today referrals have no owner, no status and no end point, and patients wait more than a year as a result.
- HOW MIGHT WEmake every open referral impossible to forget?
- HOW MIGHT WEsend each specialist exactly what they need the first time?
- HOW MIGHT WEtake the paperwork burden off patients who can’t carry it?
Visible by default
Every referral shows its status and age without anyone asking.
One owner, one next action
No row exists without a person and a date attached.
Carry the load for the patient
If a step is hard for a patient, staff do it with them.
Outlive the designer
Anyone on staff can run it from the written SOP.
Develop
What this phase answers: what should we build with no budget and no developers? I generated concepts, ranked them with the team, then prototyped and tested the winner.
Nine concepts, scored for impact and effort.
I scored each concept with the clinic director and two staff members. The flashiest idea, a patient app, scored lowest. Select a point to see the reasoning.
The reasoning, not just the result.
Three decisions carried the project. Each follows the same loop: what the evidence showed, what it meant, the call I made and the tradeoff I accepted.
Three versions of the tracker.
I ran think-aloud tests with four staff members on each version, using the same three tasks and redacted real rows. The numbers are small on purpose. The goal was to find where people hesitated.
Deliver
What this phase answers: what shipped, and how does each part trace back to the research? Five components, one coordinator role.
A standard operating procedure
One written process, with one coordinator accountable from order to appointment.
- Providerwrites the referral
- Coordinatorowns it end to end
- Specialistbooks the visit
- Patient seenreferral closed
From researchEvery role assumed another role owned follow-up.
One referral tracker
Every referral gets a status, a next action and a due date. Clinical detail stays in the EHR.
Rows sort by due date. Overdue ones flag themselves.
From research0 of 214 referrals had a recorded next action.
Reusable building blocks
Standard parts, built once and reused on every referral.
Status taxonomy
Five states replaced 14 free-text spellingsSpecialist contact card
One per office, from the phone survey- Office
- Partner hospital, GI
- Contact
- Fax, then call to confirm
- Aid rep
- Direct line on file
- Requires
- Clinical notes + aid approval
Per-office checklist Try it
Check every item to clear the packetPaperwork, done together
Patients filled out aid forms in person with me beside them. For those who couldn’t come back, I made the calls.
- Bring any ID you have
- Gather documents together
- Fill the form line by line
- Get a call when it’s booked
From research82 referrals stalled on aid paperwork.
A daily follow-up loop
The same four steps every morning. Standing check-ins turned one-off faxes into working relationships.
From researchNo referral had ever been marked done.
What the coordinator sees each morning.
Version 3, sorted by due date. Overdue rows surface themselves. Filter by status to see how each state reads at a glance. All patient data is fictional.
| Next action | Due | Patient | Specialty | Financial aid | Sent | Appointment | Status |
|---|
Measure
What this phase answers: did patients get seen sooner? I defined each metric before rollout so the result couldn’t be cherry-picked afterward.
| Metric | How I measured it | Before | After (wk 12) |
|---|---|---|---|
| Median referral time | Referral date to appointment date, per tracker row | 365+ days | 14 days |
| Open backlog | Rows not Scheduled or Closed, counted each Friday | 214 | 22 |
| Closure rate | Referrals I managed that reached Scheduled or Closed | Unknown | ~90% |
| Referrals with a next action | Rows with a next action and a due date | 0% | 100% |
Illustrative. Shapes are reconstructed from tracker records; exact patient figures are withheld.
The system outlived my role.
The SOP, tracker, checklists and daily loop all stayed in use.
First cycle: a patient with cirrhosis, waiting over a year, saw a hepatologist.
“Since her arrival, she has tremendously improved our patients’ access to specialty care around Athens and has given our partners, like St. Mary’s, a point person for care coordination.”Mercy Health Center · Public staff feature, 2025
Reflect
What this phase answers: what would I keep, and what would I do differently? Good research names its own blind spots.
Visibility beat effort
Making every open case visible changed behavior more than chasing offices harder.
Design for people, not users
The person on the other side of every form has a body and a limit.
Build systems that outlive you
Writing it down so someone else could inherit it was the harder, better choice.
Survivorship bias in patient interviews
I could only interview patients who still answered the clinic’s calls. The patients who had given up entirely, arguably the most important group, are underrepresented. Next time I’d partner with community health workers to reach them.
No baseline instrumentation
The “before” time is reconstructed from referral dates on paper. I’d instrument the old process for two weeks before changing anything, so the comparison is cleaner.
Solo synthesis
I clustered the notes alone, then checked the themes with two staff members. A second researcher in the room would have challenged my framing earlier.
Small usability samples
Four participants per round finds the biggest problems quickly, but it can’t prove an effect size. The timing numbers show direction, not significance.
- Automated remindersA stalled referral surfaces itself.
- Multilingual SMS updatesPlain, shame-free texts when a referral moves.
- Referral-health dashboardClosure rate and aging cases for leadership.
- SOP inside the EHRThe checklist lives where the work happens.