Intervention Details

Investigational Product

Name: NST-401 — fictional investigational product for demonstration purposes.

Class: Investigational non-opioid therapy for osteoarthritis-associated pain.

Treatment Arm A

NST-401 administered according to protocol for 24 weeks.

Treatment Arm B

Matching placebo administered according to the same schedule for 24 weeks.

Concomitant and Rescue Therapy

Protocol-permitted rescue analgesic medication may be used under defined conditions. Rescue medication use will be recorded and incorporated into efficacy analyses where appropriate. Participants will continue permitted stable background therapies according to protocol requirements.

WOMAC Assessment

Instrument: Western Ontario and McMaster Universities Osteoarthritis Index — WOMAC.

Target Condition: Osteoarthritis of the knee or hip.

Domains: Pain, Stiffness, Physical Function.

Electronic patient-reported outcome assessment delivered through the MOTION WOMAC Assessment Agent. The configured instrument version will determine exact item wording, response scale, recall period, scoring method, language version, and presentation requirements. The agent must administer only the authorized instrument content configured for the study.

Sentiment Measurement

Sentiment analysis is performed at the individual-question level. For each WOMAC item, the system may retain: Question ID, question presented, participant verbal response, structured response value, timestamp, sentiment signals, confidence/uncertainty indicators, dissonance flag, intervention trigger status, intervention performed, and final confirmed response.

Clinical AI Architecture

The trial uses a multi-agent architecture consisting of a Primary Assessment Agent, a Sentiment and Affective Signal Layer (Hume AI), and an Intervention Agent.

1. Primary Assessment Agent

Conducts the assigned clinical outcome assessment: identifies the participant, study, and visit; retrieves the designated index joint and correct assessment; presents questions; captures responses; applies protocol and instrument rules; maintains sequence; records timestamps; passes voice data for sentiment analysis; and generates structured data.

2. Sentiment and Affective Signal Layer

For each spoken response, voice characteristics are analyzed through Hume AI. Signals may include positive/negative affect, confidence, uncertainty, hesitation, stress, frustration, confusion, and emotional intensity. No single sentiment signal automatically changes a participant's response; the sentiment layer functions as an additional signal for detecting situations requiring closer evaluation.

3. Intervention Agent

Operates alongside the Primary Assessment Agent, monitoring assessment rules, responses, sentiment, and dissonance signals. When a predefined trigger occurs, it enters the conversation to repeat questions, clarify response options, confirm intended responses, identify incomplete/invalid responses, and escalate safety issues — then returns control to the Primary Assessment Agent.

Question-Level Dissonance Detection

The system evaluates the relationship between the question asked, the participant's spoken answer, the response value, the voice sentiment signal, and the conversation context. Potential dissonance may be identified when verbal response and vocal expression appear inconsistent. A dissonance signal does not establish that the response is incorrect; it may trigger the Intervention Agent to determine whether clarification is appropriate.

AI Agent Configuration

• Agent Name: MOTION WOMAC Assessment Agent
• Agent Type: Clinical Outcome Assessment Agent
• Protocol: MOTION-OA-301
• Assessment: WOMAC
• Voice Interaction: Enabled
• Hume AI Sentiment Analysis: Enabled
• Question-Level Sentiment Measurement: Enabled
• Dissonance Detection: Enabled
• Intervention Agent: Enabled
• Human Escalation: Enabled according to protocol rules
• Supported Index Joints: Right Hip, Left Hip, Right Knee, Left Knee
• Demo Configuration: Right Hip — Week 12

Agent Restrictions

Neither agent may diagnose the participant, recommend treatment, interpret whether a response is medically good or bad, suggest how the participant should answer, change a participant's response because of sentiment, infer a WOMAC answer from emotion or vocal tone, replace the participant's explicit response with an AI-generated interpretation, modify validated assessment wording outside authorized rules, coach participants toward a particular response, independently change protocol-defined assessment schedules, or replace investigator assessment of adverse events.

Participant self-report remains the authoritative assessment response. Sentiment and dissonance signals are used only to identify situations in which confirmation, clarification, or escalation may be appropriate.