Quality control of recorded rating interviews in CNS trials
Cephalgo, a speech biomarker company in Strasbourg, France, checks recorded rating interviews for sponsors. A recorded rating interview shows how each scale was given and scored. It can be checked per rater and per site, against each rater’s own usual range, so rater effects surface before they read as participant change.
Explainer for clinical operations, rater quality and central review teams · Updated November 2026
Did the participant change, or only the rating?
Many CNS endpoints are rater‑administered interviews. How a rater asks and scores can move the score, and that movement can look like placebo response or hide a drug effect.
Baseline inflation
A screening or baseline score more severe than the recorded interview supports.
Rater drift
A rater’s scores or conduct moving away from their own early interviews over time, such as follow‑up questions dropped once the scale feels familiar.
Lenient or severe raters
One rater scoring comparable interviews higher or lower than other raters, which can be read as a site effect or a change in the participant.
Off‑protocol administration
Items skipped or given out of order where the manual fixes it, follow‑up questions left out, and questions that steer the participant or informant toward an answer.
- Baseline inflationFlagged
- Rater driftWithin usual range
- Lenient or severe ratersWithin usual range
- Off‑protocol administrationFlagged
Rater‑administered scales across CNS trials
Any assessment where a rater asks, listens and scores leaves a record of how it was done once it is recorded: cognitive tests such as ADAS‑Cog, staging and symptom interviews such as CDR and NPI, and Parkinson’s, mood and psychosis scales such as MDS‑UPDRS Part IA, MADRS, HAM‑D and PANSS.
Alzheimer’s and dementia
CDR and NPI, with a study partner or caregiver as informant, and ADAS‑Cog.
Parkinson’s, Huntington’s and ALS
MDS‑UPDRS Part IA, UHDRS Total Functional Capacity and ALSFRS‑R.
Depression
MADRS and HAM‑D, where most item scores rest on the participant’s own account and some on what the rater observes.
Schizophrenia
PANSS, where scores rest on both the interview and what the rater observes.
Cephalgo Trial QC covers CDR, NPI, ADAS‑Cog, MDS‑UPDRS Part IA, ALSFRS‑R, UHDRS TFC, MADRS, HAM‑D and PANSS, in every language your sites interview in.

Wording, timing and tone in each interview
Each check follows the study’s scale version and interview guide, and names the measure behind it and the time in the recording.
Wording
Which items and follow‑up questions were asked, how each question was put, and whether the rater named a symptom before the participant or informant did.
Tone and timing
How loudly and how fast the rater speaks, how their tone changes after a symptom report, and how long they wait when an answer comes slowly.
Agreement, item by item
What was said about each item, read against the scale’s anchors and set beside the recorded score, with the passages behind it.
Whose account carries each item
The participant’s account, the informant’s account, what the rater observed, or a mix, so each score is checked against the evidence it rests on.
Interview length
Read against the rater’s usual range or your rater guidelines, and each site against other sites at the same visit type.
Informant interviews
Interviews with a CDR study partner or an NPI caregiver are split by speaker, and the rater’s conduct is read the same way.

Per rater and per site, against each rater’s usual range
A rater is compared first with their own interviews, so a personal style is not mistaken for a problem, and a real change in conduct still shows.
Ranges built within your study
Each rater’s usual range comes from their own interviews at the same visit type in this study. How a rater’s first interviews are read is fixed with your statistician.
Raters and sites side by side
Each rater’s item scores are set against other raters reading comparable interviews, item by item and site by site.
Treatment arm not used
Arm is not used to set any rater’s range or any flag. The flag rule is fixed in writing with your statistician and rating vendor before any audio is read.
Between‑arm check, behind a firewall
Rater conduct is compared between arms only as pre‑specified in the analysis plan, run by an independent unblinded statistician. Its results never reach the team that receives flags.
A listing at each data transfer
Flags come per rater, site and interview, as CSV or Parquet, by API or into your eCOA or EDC.
- Rater, site, visit and interview, with pseudonymized participant and rater IDs.
- The flag, and the item score it concerns where one applies.
- The measure behind the flag, with its value against the rater’s usual range.
- The time in the recording, so a reviewer can listen at the flagged moment.
- Before any flag, a design map of which comparisons your recordings support, and a report on which interviews can be read.
- At review points set in the analysis plan, your reviewers keep, tune or drop each class of flag, and each change is versioned.
Screening interviews are flagged on arrival, so a flag can reach your blinded team before randomization. Baseline interviews are flagged by the next data transfer.

How it fits central rating review
Raters give and score the scale exactly as before. Flags go to your blinded study team or central rating vendor, who listen at the flagged time and decide any follow‑up. Cephalgo never changes a score or contacts a rater.
It works on interview audio your study records or already holds, from your sites, eCOA or central rating vendor, or a finished study’s archive. Where sites do not yet record, the site or your eCOA vendor adds recording under the protocol, with consent from participants and informants.
Trial QC- Interviews reviewed
- As your vendor’s plan sets
- Compared against
- An expert review of the same interview
- Flag rule
- Set by your vendor’s review plan
- Interviews reviewed
- Every recorded interview
- Compared against
- What was said in the interview, the rater’s own other interviews, and other sites at the same visit type
- Flag rule
- Fixed in writing with your statistician and rating vendor before any audio is read
Rating interview quality control

What can be checked on a recorded rating interview?
Which items and follow‑up questions were asked and how, the rater’s tone and timing, whether each item score matches what was said, whose account carries each item, and the length of the interview. Each is read per rater and per site, against the rater’s own usual range.
Which rating scales can Cephalgo Trial QC check?
Cephalgo Trial QC covers CDR, NPI, ADAS‑Cog, MDS‑UPDRS Part IA, ALSFRS‑R, UHDRS TFC, MADRS, HAM‑D and PANSS, in every language the sites interview in.
Does rating interview quality control replace central rating review?
No. With Cephalgo Trial QC, flags go to the blinded study team or the central rating vendor, who listen at the flagged time and decide any follow‑up. Cephalgo never changes a score or contacts a rater.
What does the sponsor receive from rating interview quality control?
A listing at each data transfer with the rater, site, visit and interview, the flag and the item score it concerns, the measure behind the flag with its value against the rater’s usual range, and the time in the recording, as CSV, Parquet, API or into the eCOA or EDC.
Can rater flags arrive before randomization?
Yes. Cephalgo reads screening interviews once audio and scores arrive, so a flag can reach the blinded team before randomization. Baseline interviews are flagged by the next data transfer.
Can interviews from a finished trial be checked?
Yes. Cephalgo reads recordings from sites, the eCOA or central rating vendor, or a finished study’s archive, with the scores, in any language.
