Fixed scopes, starting with a re‑analysis pilot
Another layer of sensitivity and specificity at low cost: readings on your trial’s own scale and rater checks, from audio you hold, site-visit recordings or home sessions. Fixed fee.
Where you start
Recordings from a trial you already ran, a speech vendor already in place, speech models of your own, or no audio yet. For rated interviews, Trial QC can also run on its own. Trial QC ›
Recordings from a past or ongoing trial
Visit or interview audio never read for speech. Readings on your scale, rater flags and planning for the next trial; no new collection. Re-analysis pilot ›
No audio yet
Speech from the site’s own visit recordings and short home sessions, as an exploratory or secondary endpoint. Home sessions run in your eCOA or app; ours only where none exists. Prospective study ›
Already using a speech vendor
A second reading on your existing audio, on the trial’s own scale; your vendor stays. Also a stability check across phone and software changes. Re-analysis pilot › Stability audit ›
Already running your own speech models
Keep your pipeline. When device audio processing changes, your measures come back recalibrated or with a processing-period covariate, fixed before post‑baseline recordings are read. Re-analysis pilot › Stability audit ›
Where it sits in your protocol
- Primary endpoint and SAP hierarchy
- Unchanged
- Rater, scale and scoring
- Unchanged
- Visit schedule
- Unchanged; in a prospective study, site-visit recordings and short home sessions between visits
- eCOA vendor
- Unchanged; where the study uses one, it can record the speech tasks and receives the readings
- Speech readings
- In a prospective study, added as exploratory and pre‑specified in the SAP
- Own speech models
- Your pipeline unchanged; recalibrated values of your measures or a processing-period covariate when device audio processing changes
Re-analysis pilot
For a completed or ongoing study with recorded audio or video, where the participants’ consent covers further analysis. No new collection and no protocol change.
On your scale
A reading per visit on the protocol’s scale. Calibrated on the study’s screening and baseline ratings, blind to arm, and locked before post-baseline recordings are read.
Joined to the visit score
Readings and visit scores in one table, visit by visit, ready for your statisticians.
Trial QC flags included
Where the recordings are rating interviews, Trial QC flags per rater and interview, as in the Trial QC scope.
Sample-size planning
From a completed study: variance components and an effect-size estimate, computed by your statistician on the locked readings. Cephalgo’s analysis team stays blind to arm.
Trial QC
For a trial whose endpoint is a rated interview.
Flags per rater and interview
Each score checked against the recorded interview, the rater’s own other interviews, and other sites at the same visit type.
Before randomization
Screening interviews are read as they arrive, so a flag can reach your blinded team before randomization; baseline interviews are flagged by the next data transfer.
Flag rule fixed first
Set in writing before any audio is read. A between-arm check is pre-specified and run by an independent unblinded statistician; its results never reach the team that receives flags.
Who receives flags
Your blinded study team or central rating vendor, at each data transfer.
- Recorded rating interviews from your sites, eCOA or central rating vendor
- Your authorization to the recording holder; Cephalgo drafts it, sends a one-page file specification and coordinates retrieval
Prospective study
For a protocol in design, including dose-finding: site-visit recordings and short home sessions give each dose arm more readings before the dose decision. Your team sees pooled readings; per‑arm readings stay in an unblinded view.
Capture specification
Recording tasks, cadence and consent wording, ready for the protocol and your eCOA or app vendor; our capture runs home sessions only where the study has none.
Exploratory endpoint
Pre-specified in your SAP; estimand, features and comparison drafted with your statisticians per ICH E9(R1).
Participants’ own devices
Readings recalibrated when a phone or its software changes. Device, OS build and app version where the recording app provides them; otherwise the processing period, read from the audio.
Into your systems
Readings by export, by API or into the study’s eCOA or EDC system.
- The draft protocol: phase, population and endpoint
Stability audit
For a study whose recordings span device or operating‑system changes.
Processing periods
Where, in each participant’s recordings, the device or its OS audio processing changes.
Stability per reading
How stable each reading and its named features are across those periods.
Original and recalibrated
Both sets of readings across the periods, recalibrated blind to arm, for your analysis.
Nothing new to collect
Runs on the recordings you already hold, within the consent they were collected under; no protocol change.
Pre-screening
A reading against your entry criteria, recorded at home, before the screening visit. The same steps in every program; the entry scale is set per program.
Entry range fixed first
Calibrated per study language on paired screening scores and recordings, and fixed in writing before the first candidate records.
Consent
Candidates consent to pre-screening, remotely where approved; trial consent stays at the screening visit. Recordings of those not invited are kept only as the protocol and consent specify.
Reading returned
To the site through the study dashboard, as set in the written scope. Any safety disclosure in a recording goes to the site, under your protocol’s safety procedure. Platform ›
Recording route
A browser on the candidate’s own phone, an automated or site-run phone call, or our app. A failed check gets a retake before the session ends. Recording routes ›
- 01Site
Refers and consents
Pre-screening consent, apart from trial consent
- 02Candidate
Records a few minutes
App, browser or phone call
- 03Cephalgo
Checks and reads
Against the entry range
- 04Cephalgo
Returns the reading
Before the site books the screening visit
- 05Site
Invites and screens
The site decides who is invited. The screening visit is unchanged.
- The program, the entry criteria and the number of sites
What each scope asks of sites and participants
Start-up, training and support wherever a scope adds work at sites.
Start-up
Runs alongside ethics review, on an agreed timeline.
Training
A remote workshop before each site is activated, with a training record kept per site.
Support
A dedicated contact for site coordinators, with an agreed response time.
Nothing new on existing audio
A re-analysis pilot, a stability audit or Trial QC at sites already recording asks nothing new of participants or site staff.
| Scope | Participant | Site staff |
|---|---|---|
| Re-analysis pilot | Nothing new | Nothing new |
| Trial QCSites already recording | Nothing new | Nothing new |
| Trial QCSites not yet recording | Interview recorded under the protocol, with consent from participants and informants | Interview recording added, with one remote training workshop; the scale administered and scored as before |
| Prospective study | Site-visit recordings and short home sessions between visits, at the cadence the protocol sets | One remote training workshop; scale, raters and visit schedule unchanged |
| Stability audit | Nothing new | Nothing new |
| Pre-screeningSeparate scope | A few minutes at home, by app or browser on the candidate’s own device, or by a phone call | Refers and consents; may run the phone call; reviews readings outside the entry range |
What you hold at the end
Readings or Trial QC flags, versions and evidence in the formats your study already uses.
Readings and flags
Readings per session or visit, and Trial QC flags per rater and interview, as CSV, Parquet or API, or into your eCOA or EDC; CDISC ODM, SDTM-aligned or your format.
Version record
Analysis version on every reading; device, OS build and app version where the recording app provides them, otherwise the processing period, read from the audio.
Analysis plan text
Estimand, features and comparison, fixed in writing before post‑baseline recordings are read; the Trial QC flag rule, before any audio is read.
Validation evidence
Each measure evaluated stage by stage following the DiMe V3 framework, one indication at a time, and your study’s validation report follows TRIPOD+AI. Validation approach ›
Also in every scope
- Data protection
- Speech handled as special-category health data under GDPR. DPIA, ROPA and DPA at onboarding; retention period and deletion procedure agreed before transfer. Data protection ›
- Electronic records
- Audit trail, role-based access and versioned analysis, designed to support 21 CFR Part 11 and EU Annex 11. Electronic records ›
- Languages and countries
- Multilingual; tasks and instructions set up in each study language at scoping, with the same protocol definition across sites and countries.
Roles in a study with Cephalgo
Each party keeps the role it has today.
Sponsor
Owns the study data, approves each scope and runs the trial’s statistical analysis.
Cephalgo
Delivers readings, Trial QC flags and their version record, and trains sites wherever a scope adds work at sites.
Sites
Run visits and clinical safety follow-up; scale, raters and visit schedule unchanged.
CRO and eCOA vendor
Site management, monitoring and eCOA, as already contracted; the eCOA can carry the speech tasks.
ContractingCephalgo SAS, Strasbourg
- Registration
- RCS Strasbourg 904 447 323
- Founded
- October 2021
- Public funding
- EIC Accelerator; Horizon Europe
- Vendor qualification
- The qualification pack with the SOC 2 Type II report, then a remote or on-site audit. Security and quality ›
