CASE STUDY — ケーススタディ

Centrova

Centrova

ク リ ニ カ ル リ サ ー チ プ ラ ッ ト フ ォ ー ム

I led design for an AI-powered clinical research platform, from the brand identity to a two-sided enterprise product that connects research sites with CROs.

ROLE

Design Lead, team of 3

TIMELINE

1–2 Years

STATUS

Pilot / Early Customers

COMPLIANCE

HIPAA + 21 CFR Part 11

問題 THE PROBLEM

Clinical trial recruitment keeps failing the people it’s meant to help.

MISS RATE

80%

of clinical trials miss their initial enrollment timelines, costing pharma millions a day in opportunity cost alone

TERMINATION CAUSE

57%

of clinical trial terminations are due to enrollment difficulties. The research just stops, and the funding goes down the drain

Those numbers mean cures arrive late, patients lose access to treatments that could save their lives, and drug prices go up to pay for slow research. The clinical staff doing the work burn out, too.

リサーチ RESEARCH & MANDATE

I was brought in to understand a space the previous design lead couldn’t.

Clinical research has its own language, and its workflows are the way they are for good reasons. Start designing before you understand them and you’ll break something. So I interviewed clinical researchers, PIs, site coordinators, and monitors at all kinds of institutions. I had to account for blinding protocols, randomization structures, and cohort configurations, plus how all of it changes depending on whether the sponsor is a major CRO like IQVIA or a more in-house operation.

The hard part was working out which edge cases were common enough to design for and which were one site’s customization. Get that wrong and you build a tool that works for one workflow and fails everyone else.

I’m fluent in it now, but getting there took a lot of interviews across the US, with researchers at different kinds of institutions, major CROs, and boutique sponsors. I didn’t feel ready to design until I had that wide a picture.

プロダクト THE PLATFORM

One platform for everyone involved in a trial.

Right now, sites patch this together with two or three tools that were never built to work with each other. Centrova puts all of it in one place, which is the main thing that sets it apart.

SMARTMATCH

A bidirectional matching engine. It finds the right patients for a trial, or the right trials for a patient, using a proprietary tagging database built from clinicaltrials.gov and patient records.

AI SCREENING

Evaluates patients against trials in bulk. Every AI recommendation shows the exact line in the patient file it’s based on, and the researcher makes the final call.

TRIAL MANAGER

Browse, configure, and manage every trial, with role-based access, double-blind support, enrollment tracking, and direct criteria editing for when clinicaltrials.gov gets it wrong.

DATA HUB

Shared file storage for visit data, trial documentation, deviation logs, and direct data entry, with one-click export to sponsors.

CLINICAL TEAMS

Trial channels, direct messages, and monitor-to-site lines for sending queries and receiving site reports. Blinded and unblinded chats stay separate to protect double-blind studies.

MONITOR PORTAL

A section where CRO monitors review evaluations, finalize enrollment, and request compliance documents without scheduling a remote visit or sending a single email.

画面 Screens
01Screening: every AI decision links back to the line in the patient file it came from, and the researcher makes the final call.
02Trial overview, with SmartMatch one click away.
03Trial channels keep site staff and CRA monitors in one conversation.

デザイン KEY DESIGN DECISIONS

The hardest problems were about structure.

01

Adding monitors to the platform

This was probably the biggest product call I made, and it came straight out of research. Monitors and sites work on the same trials, but the way their workflows are set up, they’re constantly jumping through hoops to share information: email chains, phone calls, remote visits. For how often they need to talk, it’s wildly inefficient.

The interviews showed me that Centrova sits in an unusual spot. We hold patient data relevant to screening, but we’re not an EHR. That means we can give monitors the access they need without the tight restrictions that slow everything down, and still keep it HIPAA-compliant and properly deidentified.

I took this to the CEO and COO with the interview data to back it up. It was outside the original scope, but they got on board, and in the pilot it turned out to be one of the features monitors liked most.

02

Redesigning the screening flow

My first version was a wizard: pick a patient, pick a trial, run the evaluation. It was clean and simple, and it was wrong. The research showed that real workflows were all over the place: one patient against one trial, many patients against one trial, one patient against many trials, or many patients across many trials at once. The wizard also couldn’t handle adding evaluations mid-session, which is a problem in a clinic where a patient can show up unannounced.

Putting a pre-match score right in patient selection means the screen is never blank. The most likely candidates stand out before the researcher has done anything, and I think it’s one of the strongest features in the platform.

03

AI that shows its work

In a regulated clinical setting, you can’t let an algorithm make enrollment decisions and expect people to trust it. So the AI does the heavy lifting, screening every selected patient against every trial criterion at once, and links each decision to the exact line in the patient’s file it relied on. Researchers then make the final call on each criterion themselves. That’s for compliance as much as trust: clinical responsibility stays with qualified staff, and the double-check is required by design.

04

A visual identity for researchers

Before touching anything, we held a full brand meeting with the design team. We landed on a look that feels like research more than like a hospital, because researchers don’t see themselves as hospital administrators. We also didn’t want to look too tech-forward. In high-stakes settings, being flashy about AI makes people trust you less. So we went with large typography, a top nav, and content front and center, and stayed away from anything that looks like it came out of a Shadcn template.

成果 OUTCOMES

Screening went from about 90 minutes to 5–10.

METRIC

BEFORE CENTROVA

WITH CENTROVA

Time per screening evaluation

Time per screening evaluation

~90 minutes

5–10 minutes

Monitor-to-site communication

Monitor-to-site communication

Email, phone, remote visits

Unified in-platform

Software tools required

Software tools required

2–3 disconnected tools

1 integrated platform

Eligibility documentation

Eligibility documentation

Manual, inconsistent

Auto-generated PDF

In pilot feedback, monitors said having communication, screening history, and files in one place was much simpler than their current multi-tool setup. PIs liked the role-based access and being able to see how the AI reached its screening decisions.

Going from 90 minutes to 5–10 is roughly a tenfold change. For the first time, sites can screen large patient populations in bulk without hiring staff to match.

振り返り REFLECTION

The research did more for this product than any screen I designed.

The interviews are why I pushed to include monitors in the platform. That changed the product’s scope and what it offers, and it’s why Centrova can go after the fragmented communication that slows clinical research down. The original brief hadn’t accounted for that problem, and I only found it in the interviews.

The screening redesign taught me something too. The wizard looked simple on paper and fell apart in practice. The Screening Center looks more complicated, but it matches how researchers really work, so it ends up feeling simpler to use.

Centrova is still in pilot. We’re building one compliant platform, with AI built in, that connects everyone involved in a clinical trial. If we get that right, I think it can shorten the time between a medical discovery and patients getting access to it.