← All work

Case study · 2-minute summary

Ordering cancer treatment before the start date is known

Oncology teams usually don't know when treatment will start at the moment they order it. The system demanded a date anyway, so everyone made one up. I redesigned how a start date is set, tracked and changed.

Role
Team Lead, UX design and research
Timeline
Aug – Nov 2026
Platform
Enterprise EHR, oncology ordering
Status
Direction validated, in development
This is the short version. The full case study is a 4-minute interactive walkthrough of the working prototype, best on a laptop or desktop. Start the walkthrough →
The new Set Regimen Start Date dialog: an exact date, four scheduling targets, or No date yet (TBD), with a prior authorization blocker recorded below
The new start date step: an exact date, a scheduling target, or “No date yet”, plus what's holding it up. All names and data are fictional.
01

In short

Problem

A required calendar date at ordering, when the real date is unknown 50–80% of the time.

Insight

The empty date field wasn't the real pain. The system hid the whole signed plan from nurses and pharmacy until a date existed.

Design

Separate “signed” from “dated”. Capture intent (a date, a timeframe, or TBD with reasons), let anyone on the team clear blockers, and make date changes visible to everyone.

02

The problem

One required date, and nobody knew it yet

A regimen's start date is Day 1, the first dose. Pharmacy prep, infusion chair bookings, insurance approval and patient safety checks all hang off it. But at the moment a provider orders treatment, that date usually depends on things that haven't happened yet. The legacy flow had one mandatory date picker, so people guessed.

Illustrative recreation
  1. One mandatory date. There was no way to say “not yet”, or why.
  2. So teams typed a placeholder date, and every downstream date inherited the guess.
  3. The real status lived in free text, side spreadsheets, sticky notes and messages, outside the chart.

It's wishful thinking.

Oncology pharmacist, on placeholder start dates
03

Research

What six oncology sites told me

Mixed-methods discovery: a landscape and constraints review, a stakeholder session, a survey of 6 customer sites, and moderated prototype reviews at three: a small US community program (Site A), a US regional hospital (Site B) and a UK public hospital (Site C). I synthesized it into 10 findings, each mapped to a design implication.

5 of 6

Not knowing is the norm

Sites entered a real date at ordering half the time or less. One estimated 80/20.

6 of 6

Many things hold up a start

Prior authorization at every site, infusion chairs at 4, labs at 3, often several at once.

Every site

Everyone guesses, then fixes by hand

Placeholder dates everywhere, with the real status kept outside the chart.

5 of 6

The real pain is visibility

“Dates change and the changes aren't visible to everyone who needs them” ranked top.

No one model

Ownership differs by site

Schedulers, providers, nurses or pharmacists own the date, depending on the site.

Honest note: a small sample (6 survey sites, 3 sessions), and schedulers weren't interviewed yet.

The problem isn't the empty date field. It's that the system hides the plan.

The reframe

Design principles

  • Signing and dating are separate steps.
  • Capture intent, not just absence.
  • Blockers are visible, advisory, and clearable by others.
  • Date status and changes are obvious to everyone.
  • Moves follow how treatment actually happens.
  • Don't make people enter information they don't have yet.
04

The design

Sign the plan now. Date it when you know.

  1. A start date step with real choices. An exact date (with Today as a shortcut), one-click scheduling targets that show the actual dates in each window, or “No date yet”.
  2. Blockers instead of guesses. TBD plus up to three structured reasons from the site's own list. A provisional date is still allowed while a blocker is open.
  3. Closed-loop messages. Saving a blocker or a scheduling target drafts a worklist message to the people who can act on it, editable before sending.
  4. Visible the moment it's signed. The order shows its date status and what it's waiting on, everywhere it appears. Anyone on the care team can resolve a blocker.
  5. Moves that match treatment. Move a day, a cycle, or a cycle and everything after it, with day order kept intact and a “Delayed” tag so changes never happen silently.
A signed regimen on the Orders screen showing 'TBD — 1 blocker pending' as its start date
Signed, not dated. The plan is live and visible, with its open blocker.
The Manage timeline with cycle 3 tagged 'Delayed 7 days'
Changes stay visible. Cycle 3 moved a week, tagged so the whole team sees it.
05

Iteration

How feedback changed the design

  • Site A asked for a 30-day target for monthly infusions, so I added one.
  • Site A found choosing affected medications one by one tedious, which led to an “Entire plan” option.
  • Site C said blockers must never prevent a date, so provisional dates are allowed while blockers are open.
  • Site C preferred a message to schedulers over automatic booking, so “Book first available” sends a task.
  • Site C found a moved Day 1 could end up out of order, so I added a re-sequencing rule.
  • Site B said orders should be visible as soon as they're signed. That became the core principle.
06

Status & reflection

Where it stands

The direction was validated with 3 sites and a 6-site survey, and it's in development. There are no outcome numbers yet. To measure success I'd track how often placeholder dates are entered, time from signing to pharmacy verification, and start-date changes made outside the chart.

Still open: what pharmacy may do with an undated plan (view and verify, or dispense), and who confirms the date at each site. Next research: schedulers and authorization coordinators.

  1. One well-placed question can reframe a whole project.A single question in a Site B review surfaced the 80/20 estimate and the visibility gap.
  2. Workarounds are requirements in disguise.Every placeholder date and side spreadsheet pointed at something the system wasn't doing.
  3. Design for disagreement.When sites conflict, make it configurable by role instead of picking a winner.
Try the interactive walkthrough About 4 minutes · laptop or desktop

Next case study

Regimen Ordering & Management

Contact

If the problem is genuinely hard, I'd like to hear about it.