UX Research & Redesign · Cancer Hospital, CAMS
Discharge Follow-up System
Role
Sole designer on a summer engagement — research, IA and interface. Partnered with one attending physician who anchored every decision in the reality of the ward.
Duration
Jun 2024 - Aug 2024
Type
Healthcare UX / Accessibility
Skills
User Research, Information Architecture, Accessibility Design, Prototyping
Tools
Figma
When a cancer patient leaves the hospital, treatment continues, but the system that supports it stops at the door. Follow-up survives on paper forms, phone calls, and whoever remembers to chase whom.
At the Cancer Hospital of the Chinese Academy of Medical Sciences, I spent a summer redesigning what happens after discharge, so the follow-up loop holds without a nurse holding it by hand.
On the design side, I worked solo — partnered with a single attending physician who anchored every decision in the reality of the ward.
王阿姨，您今天
疼得厉害吗？
🎤不想按？按住这里直接说
Daily check-in — one question per screen, faces instead of a 0–10 scale.
记下了，
已经转给您的护士
😣您刚才选的：比较疼
您说的疼痛情况比较重要，李护士会尽快联系您，请留意来电。
您的随访护士
预计 今天下午 4 点前 来电
紧急情况请直接拨打 120 或前往急诊
After a high-risk answer — who will call, and by when.
The deliverable was a mobile mini-program, so follow-up lived in the patient's pocket, not on a clinic desktop.
The challenge
One workflow, two users who can't meet halfway
Follow-up isn't one product; it's a relay between two people with opposite capacities. Any design that helps one side by burdening the other fails both.
Clinicians · no time
Care ran on paper and in series. Histories, current medications, and each day's infusions were entered by hand and re-entered at every step, while patients queued to be seen one at a time. A single attending physician anchored each patient's trust — stretched across inpatient rounds, clinic, and referrals between departments.
Elderly patients · no digital fluency
Patients were mostly older adults managing several conditions at once — chemotherapy alongside things like insulin and blood-pressure drugs. They described how they felt in plain, bodily words — heartburn, no appetite, numb and swollen hands, blisters, “it hurts,” the discomfort that “might just be in my head” — never in the 0–10 clinical scales the records expected.
The clinical constraints
Chemotherapy runs in long, roughly 21-day cycles until discharge, and the fragile stretch falls after the patient goes home, where no one is watching the numbers. Infusions follow a strict safety ritual — the “three checks, seven verifications” protocol — vitals like blood oxygen and blood pressure still need monitoring after discharge, and outcomes vary enough that nothing can be fully standardized.
As-is
Where the loop breaks
Before touching screens, I mapped the existing follow-up journey end to end, from discharge day to the next hospital visit, and marked every point where information dies or a human has to compensate.
Discharge day
Day 0 · leaving the ward
Home
Day 1–2
Side effects hit
Day 3–5
On their own
Day 6+
Next visit
Return to clinic
A paper discharge sheet is handed over at the door.
Patient is on their own; symptoms noticed but not recorded.
Nausea, pain, blisters set in — they call the ward if worried.
No one is watching vitals; family chases whether things are normal.
Everything is reconstructed from memory at the desk.
Paper sheet — the thread is lost at the door.
Symptoms tracked from memory and phone calls.
Vitals (SpO₂, BP) go unmonitored exactly when it matters most.
Day-one baseline is never compared against how they are at home.
The system leaves the patient alone in exactly the window where anxiety peaks — the days at home after discharge. Every redesign decision aims at closing one of these breakpoints.
Mapping one full cycle — consultation and admission → waiting for chemo → nurse-administered infusion → the days at home (Day 1, Day 5, and on) → the next hospital visit — surfaced where the loop breaks. Data is captured by hand and lives on paper. The moment a patient leaves, symptom tracking falls back on memory and phone calls. And the baseline indicators taken on day one are hard to compare against how the patient is actually doing at home.
Research · on-site at the hospital
What 16 interviews taught us
I interviewed both sides of the workflow. Each insight below pairs the finding with what it meant for the design.
01
02
03
04
Body speaks in feelings, not numbers
“疼得没法睡觉。”
It hurts too much to sleep.
— Patient
“一点胃口都没有。”
No appetite at all.
— Patient
“手脚又麻又肿，手上还起了泡。”
Hands and feet numb and swollen, blisters on my hands.
— Patient
→ Symptoms arrive as lived experience, not a score.
The fear is the unknown
“不知道这是不是心理作用。”
I can't tell if it's just in my head.
— Patient
“不想麻烦孩子。”
I don't want to trouble my kids.
— Patient
“要是知道接下来会怎样，就没那么怕了。”
If I knew what came next, I'd be less afraid.
— Caregiver
→ Being informed is itself a form of care.
Alone once I'm home
“出院回家，血氧还得自己盯着。”
Back home, I have to watch my blood oxygen myself.
— Patient
“有点不对就往医院跑。”
If something feels off, I just rush back to the hospital.
— Patient
“总在反复做 CT、复查。”
It's endless CT scans and re-checks.
— Patient
→ The loop breaks the moment they leave.
Grouping the raw quotes surfaced three clusters — which became the insights above. Voices are real and de-identified; no names or records appear.
The pivot
Insights became rules
Each insight was translated into a design principle. These four rules governed every decision that follows.
From insight 01
From insight 02
From insight 03
From insight 04
The redesign · principle 1
Make the next step unmissable
Every day after discharge, the patient gets one push and one question at a time. Pain, appetite, mood — each on its own full screen, answered by tapping a face or holding to speak. Anything above “it hurts a bit” triggers an AI follow-up, and the answer is routed straight to the follow-up nurse's queue.
Before
After
D1–D7 after discharge
faces, not 0–10
NO · ALL MILD
Auto-logged · AI re-checks tomorrow. No human touch needed.
YES · ESCALATE
Patient sees a confirmation naming who will call, and by when.
The whole path is three taps. Complexity only appears when it has to — a mild day closes itself, and only a worrying answer pulls in the AI and a nurse.
The redesign · principle 2
Design for the oldest user first
The oldest, least digitally-fluent patient was the default user, not the edge case: large system-font Chinese, a single column, faces on a green→red tint so severity reads without literacy, an empty radio on every row so it's clearly a choice, a way back to the previous question, and never more than three taps to finish.
王阿姨，您今天
疼得厉害吗？
🎤不想按？按住这里直接说
Faces, not numbers
Wong-Baker faces on a green→red tint carry severity even for users who can't read the labels.
Every row looks tappable
An empty radio on each option signals “this is a choice,” for users who have never touched a smartphone.
A way back, and a voice route
Back to the previous question, plus press-to-talk for anyone who won't tap.
The redesign · principle 3
Never make the clinician repeat themselves
On the clinical side, a patient's history, current medications, and each day's infusions are entered once — infusions by scanning the fluid's QR code under the ward's “three checks, seven verifications” ritual — and everything downstream reads from that record instead of re-collecting it. What the patient logs at home flows into the same place, so the follow-up nurse opens a case, not a blank form.
今日随访队列
7 月 17 日 周五李秀云 · 胃肠外科随访组需优先处理
2
今日待随访
14
已完成
6 / 20
患者自报待读
9
处理队列
按 风险等级 → SLA 剩余时间 排序AI 摘要自述伤口渗液增多、发热 38.6°C；较昨日恶化 ↑；触发红旗规则「持续发热+渗液」，已同步张医生。
SLA 剩余 11 分钟 · 家属已收到就医建议短信
AI 摘要疼痛自评「比较疼，影响睡眠」，连续 2 天上升；进食量减半；无发热。建议当日人工回访评估镇痛方案。
SLA 剩余 2 小时 40 分 · 已向患者承诺 16:00 前回电
AI 摘要情绪低落表述增多（「不想麻烦孩子」出现 2 次），躯体症状平稳。建议回访时增加情绪关注。
今日内完成即可
AI 摘要恢复符合预期，已自动归档。无需人工介入。
还有 10 条今日常规随访 ⌄
The nurse's queue.Not a roster of everyone — a queue sorted by risk, then by how much time is left on the callback promised to the patient. Each case arrives as a three-line AI summary with the rule that flagged it (“sustained fever + discharge”), so the nurse can see why it surfaced instead of trusting a score. Cases that resolved on their own fade out and auto-archive.
张医生，今日院外摘要
32 位随访患者中
2 位需要您关注
需要处理
发热 38.6°C + 伤口渗液增多，护士已回拨确认，建议患者今晚急诊换药。体温近 5 日趋势：
D2 — D6（今天）
疼痛连续 2 日上升、进食减半。护士 16:00 回访后建议调整镇痛方案，等待您确认。
其余患者
The attending physician never opens the queue. Once a day, one push: of 32 patients at home, the two who need a decision — each with the trend that made them exceptional, and the order that answers it.
Exceptions, not a dashboard
The other 30 collapse into a single line. A doctor with eight minutes between patients gets a decision list, not data to interpret.
One tap becomes two tasks
“Bring the visit forward” writes the patient's notification and the nurse's task at once — the order is entered where it's decided, never re-entered downstream.
It lives in WeCom
Chinese hospital workflows already run through WeCom, so this is a mini-program inside it — system capsule and all — rather than one more app to install and forget.
Names, dates and clinical figures in these screens are illustrative. No real patient or staff information is shown.
Validation
Three rounds against real users
I validated the way this kind of product has to be validated — in person, with the people who'd actually use it, changing the prototype between each round (a RITE-style loop: test a few, fix, test again). Three rounds, each asking a different question.
Round 1 · lo-fi wireframes
Elderly patients + a follow-up nurse, moderated one-on-one
Can someone finish a daily check-in with no help?
Changed: dropped the 0–10 pain scale for Wong-Baker faces, and added an empty radio to every option so each row reads as a choice, not just text.
Round 2 · interactive prototype
Returning and new patients
Do they trust what happens after they submit?
Changed: added the confirmation that restates their answer and names the nurse and the callback time, plus a “back to previous question” for mis-taps.
Round 3 · near-production
Follow-up nurses, on the queue side
Does a worrying answer reach the right person, fast?
Changed: ordered the nurse queue by risk then SLA, tightened the AI summary, and auto-archived mild days so the queue only shows who needs a human.
16
Interviews on-site at the hospital
3
Iterative testing rounds
Reflection
Two things stayed with me. First, in a cancer ward accessibility isn't polish — it is the product; a follow-up loop the oldest patient can't close doesn't exist. Second, the sharpest questions were the ones the research left open, not the ones it closed: how often should someone log without it becoming a burden, should this be an app or a mini-program, when does the monitoring wristband go on, and how do you establish a trustworthy day-one baseline to measure the home stretch against. That is where I'd take the next cycle of work.