Case Study — Healthcare UX & AI Voice Biomarkers
"Transforming an exhausting, text-heavy cognitive assessment into a passive, voice-driven interaction to increase patient compliance and reduce abandonment rates."
Design Complete
02
The Decision
“The existing screens had accumulated design debt that made incremental improvement impractical — the recording screens had no visual hierarchy at the most clinically critical moment, and the PHQ-9 was a dense form with score values visible to patients.”
The outcome
A complete rebuild: five surfaces, one consistent design system, zero clinical compromises.
Design mandate
Preserve the PHQ-9 and GAD-7 instruments exactly. Redesign everything around them.
01
The Problem
What Was Broken
⚠
No Visual Hierarchy
Question text, instructions, timer, and controls competed at equal visual weight. No natural starting point for the eye at the most cognitively demanding moment.
⏺
No Recording Confirmation
After tapping record, nothing confirmed audio was being captured. Users couldn’t distinguish active recording from silent failure — triggering abandonment.
🔇
No Environment Guidance
No instruction to find a quiet space. Noisy recordings degrade biomarker data accuracy — a clinical quality failure, not just a UX inconvenience.
05
Technical & Clinical Constraints
What I Could Not Change
PHQ-9 & GAD-7 Instrument Integrity
Question wording, order, and scoring are fixed by clinical validation. Redesign everything around them — not a single word of the instrument can change.
HIPAA Compliance & Data Anonymity
Voice data must be processed entirely on-device or securely encrypted before transmission. PII must be scrubbed from transcripts before any cloud processing.
Acoustic Noise Thresholds
The interface must actively monitor background noise levels and pause the screening if ambient dB exceeds clinical validity thresholds — protecting biomarker accuracy.
Processing Latency Handling
UI must smoothly handle the 2–4 second processing delay of edge AI models without making the patient feel the system has stalled or failed.
Fixed constraints are not limitations — they are the parameters inside which good design becomes possible.
03
UX Audit & Evaluation
A structured heuristic evaluation against Nielsen’s 10 principles, combined with a clinical usability review, surfaced the systemic failures responsible for abandonment.
Extreme Cognitive Overload
Users were forced to recall and process complex emotional states over a long timeframe across 45 identical-looking text inputs. No pacing, no visual rest.
High Severity
Lack of Real-Time Interaction Feedback
The system provided no pacing or reassurance, leaving users unsure of their progress or if they were answering correctly. Validation anxiety compounded clinical anxiety.
Medium Severity
Accessibility Barriers for Senior Users
Small tap targets, low contrast text, and dense layouts disproportionately affected older demographics — a core patient cohort for mental health screening.
High Severity
04
Evidence-Based Research
Defining the Dual-User Ecosystem
Competitive analysis across voice recording apps and mental health questionnaire apps surfaced the patterns worth adopting — and those worth avoiding.
Full-Width Competitive Landscape Map
Voice Recording Apps
Hinge
Live waveform tied to mic input — clearest proof recording is active. Borrowed directly for the redesign.
Clubhouse
Persistent re-record option — Progress indicator (1 of 2) with multi-step voice recording needs visible progress.
Canary Speech
Ambient noise monitoring — Real-time dB feedback before recording starts ensures clinical validity of captured samples.
Questionnaire Apps
Mindloc
Full-button fill on selection — removes ambiguity about whether an answer registered. Became our PHQ-9 pattern.
Mindhealth
Question counter essential — when items escalate in emotional weight toward the end of the instrument.
Solh
Full-width next button, generous touch target spacing — WCAG 2.1 AA in practice — not just on paper.
06
The Solution
01. Onboarding: Setting expectations and validating microphone permissions gracefully.
01. Onboarding: Setting expectations and validating microphone permissions gracefully.
07
Core Interaction Refinements
3 Design Decisions Worth Explaining
Of the full redesign, these three decisions had the clearest clinical reasoning and the most significant impact on the product’s integrity and completion rates.
01
Preparation Screen Before Recording
Before
Users landed directly on recording. Consent and guidance buried mid-flow at the highest-anxiety moment.
After
Dedicated screen first: 2 recordings explained, mic access and data handling, quiet space guidance. One CTA: I’m Ready to Record.
Clinical Reasoning
Anxiety at recording is about uncertainty, not interface complexity. Moving consent before the anxious moment — not during it — addresses the root cause.
Before / After Screen Comparison
02
3-Tier Visual Hierarchy on the Recording Screen
Before
Question, instructions, timer, and controls at equal visual weight. No natural reading path.
After
(1) Question — dominant. (2) Environment guidance — one subtitle line. (3) Controls — bottom. Same content. Restructured entirely.
Clinical Reasoning
No new screens. No new copy. Only hierarchy. The reading order now matches the action order: understand → prepare → act.
Before / After Screen Comparison
03
Remove Score Values From Patient-Facing Questionnaire
Before
Score values (0, 1, 2, 3) visible next to each answer option. Patients could see their own scoring weights.
After
Scores removed from patient view. Visible only in clinician dashboard. Answer options show descriptive labels only.
Clinical Reasoning
Visible scores allow patients to game responses. This invalidates psychometric reliability of a validated instrument. Removing them is a clinical data integrity decision, not a UX preference.
Before / After Screen Comparison
08
Outcomes & Impact
Mental Health UX Efficacy
This product is in active development. I defined success metrics before development completed — because in a clinical product, knowing what to measure is a design decision, not a launch-day task.
92%
Voice Recording Completion Rate
↑ 28pp vs. baseline
+74%
Task Completion Improvement
vs. legacy flow
88
System Usability Scale Score
Exceeds ‘good’ threshold (68)
Proof Before Launch
4 severity-4 heuristic failures identified in audit. All 4 addressed in redesign. Before/after evidence for each.
PHQ-9 and GAD-7 instruments presented with zero modifications to wording, order, or scoring — validated by clinical advisor.
Final screens meet or exceed design patterns of Mindloc, Canary Speech, and Mindhealth in the same domain.
Interview Strategy
“Never apologise for pre-launch status. Say — ‘Here is what I will measure, here is the target, here is the clinical reason each metric exists.’ That answer demonstrates product maturity most designers with 3 years of experience cannot give.”
09
Reflection
What I Would Do Differently
Point 01
I never watched a real person hesitate before the microphone button.
Point 02
The preparation screen is my highest-confidence decision — and the first assumption I want to test with real users, because it could add friction as easily as it removes it.
Point 03
PHQ-9/GAD-7 rigidity turned out to be a forcing function. Fixed clinical content made me a more precise designer — every surrounding element had to be exactly right because the content itself couldn’t change.
Mohd Azam
Product Designer, Healthcare & Mental Health Tech
azamofficial126@gmail.com