How to add a speech measure to a trial as an exploratory endpoint

Name the speech measure as an exploratory endpoint in the protocol, define it in the statistical analysis plan before the blind is broken, and keep evidence that the recording and the analysis are fit for purpose in the trial’s context of use. FDA applies the same principles to endpoints from digital tools as to any other endpoint and states that qualification is not required for a clinical outcome assessment to be used in trials. EMA does not require qualification for exploratory use in early trials.

Guide for trial sponsors, statisticians and digital endpoint teams, by Cephalgo · Updated November 2026

Endpoint hierarchy

Primary, secondary and exploratory endpoints

Trial endpoints sit in three families. Where a speech measure sits decides how much evidence it needs and how its results are read.

“The third category in the hierarchy includes all other endpoints, which are referred to as exploratory. Exploratory endpoints can include endpoints for research purposes or for new hypotheses generation.”
FDA, Multiple Endpoints in Clinical Trials, final guidance, October 2022

Primary

The endpoint that establishes the effect of the drug and decides whether the study meets its objective. ICH E9 asks for one primary variable where possible, specified in the protocol with its rationale, and recommends a reliable, validated variable with experience from earlier studies.

Secondary

Endpoints that support the primary result or show additional clinically important effects. FDA asks that secondary endpoints able to show additional effects be included in the trial’s Type I error control plan.

Exploratory

All other endpoints. FDA’s guidance states that exploratory endpoints do not need multiplicity adjustment because they are generally not used to support conclusions. This is where a new speech measure usually starts.

The ICH E9(R1) addendum adds the estimand: a precise statement of the treatment effect, built from five attributes, namely the treatment, the population, the variable (endpoint), how intercurrent events are handled, and a population‑level summary. FDA’s digital health technology guidance asks late‑phase trials to describe endpoints from digital tools with this framework. Writing an estimand for an exploratory speech endpoint early makes a later move to the secondary family simpler.

Protocol and SAP

What the protocol and SAP should pre-specify

ICH E9 asks that the statistical analysis plan be finalised before the blind is broken. FDA asks that the definition of a digital endpoint and the source data behind it be pre‑specified in the plan. For a speech measure, that covers eight points.

  1. Context of use

    The indication, population, language and purpose the reading serves, such as tracking change in speech or cognition in early Alzheimer’s disease. Both FDA and EMA assess a measure against its stated context of use.

  2. Speech task and recording

    Which tasks are used, such as picture description, reading a passage or sustained vowels, the instructions given, and how recordings are captured and checked for quality. The same collection method is used in all study arms.

  3. Schedule

    Where and when readings are taken: from recordings the trial already holds, at site visits, or in short home sessions between visits. More readings per participant can lower the noise in each participant’s estimate of change. Speech readings and sample size ›

  4. The clinical scale it reads against

    Which scale the reading lands on or sits beside, such as MMSE, MDS‑UPDRS item 2.1 or the ALSFRS‑R speech item, and at which visits the two are joined. A reading in units reviewers know is easier to place in the analysis.

  5. Analysis model and version

    The model for change over time, such as a mixed model for repeated measures, how practice effects are handled, and the version of the speech analysis, fixed so that the first and last readings come from the same analysis.

  6. Missing data and intercurrent events

    How missed or unusable sessions are handled, and post‑baseline events that affect the reading or its interpretation, such as starting speech therapy, an intercurrent event under ICH E9(R1). FDA asks for a plan to reduce missing data and for the SAP to say how such events enter the primary and sensitivity analyses.

  7. Phone and software updates

    How the timing and nature of each update is recorded, how measurements are checked after an update, and how any difference before and after it is handled in the analysis, specified before unblinding. Device changes ›

  8. Blinding

    Who may see readings by arm during the study. Site staff, raters and the analysts producing speech readings work blind to arm, and any interim view by arm follows the trial’s charter for unblinded review.

FDA

What FDA expects

Three sets of FDA documents apply to a speech measure in a drug trial: the digital health technology guidance, the patient‑focused drug development series on clinical outcome assessments, and the biomarker definitions and qualification program.

Digital health technologies, December 2023

The final guidance covers primary, secondary and exploratory endpoints from digital tools. It asks for verification and validation showing the tool is fit for purpose, a precise endpoint definition, analysis in the SAP, and a record of updates with a before and after comparison where measurements change. Sponsors should discuss endpoint plans with the relevant review division.

Fit-for-purpose clinical outcome assessments

The third guidance in the patient‑focused drug development series, final in October 2025, covers how to select, develop or modify a clinical outcome assessment that is fit for purpose. Depending on its use, a speech task scored from a recording may be treated as a performance outcome.

From assessment to endpoint

The fourth guidance in the series, a draft from April 2023, covers how an assessment becomes an endpoint, how its data are collected and analysed, and how meaningful within‑patient change is defined and interpreted.

Biomarkers and qualification

The BEST glossary defines a biomarker and notes that it is not an assessment of how a person feels, functions or survives. FDA’s biomarker qualification is a three‑stage process that sets a biomarker’s interpretation within a stated context of use; FDA notes that the biomarker is qualified, not the measurement method. For clinical outcome assessments, FDA states that qualification is not required for use in clinical trials; qualification is a voluntary route to use a tool across drug development programs.

A speech measure may be treated as a clinical outcome assessment, a biomarker, or both. The context of use decides which documents lead. What is a speech biomarker ›

EMA

What EMA expects

EMA reviews digital measures through its qualification of novel methodologies procedure, and has set out how it applies to digital technologies.

“The exploratory use of digital technologies in early trials (e.g. proof of concept, before pivotal development) does not require the qualification of a technology: the need arises once the technology is used to support or collect the main body of data that will be considered pivotal to the assessment of the benefit-risk balance of a new medicine.”
EMA, Questions and answers: Qualification of digital technology‑based methodologies to support approval of medicinal products, EMA/219860/2020

Qualification advice

Feedback on preliminary data and on the qualification plan needed for a future opinion. It can be requested at any time during development (EMA qualification of novel methodologies).

Qualification opinion

CHMP’s assessment of whether a methodology is acceptable for a specific context of use, based on the supporting data, published on EMA’s website.

Context of use and changes

EMA asks for a full description of the use, including the nature of the endpoint (primary, secondary, exploratory). It describes an iterative qualification process as often desirable, and asks for a risk assessment of changes and software updates during development.

Exploratory to secondary

How sponsors move a measure from exploratory to secondary

A digital measure earns a higher place in the hierarchy by accruing evidence for one context of use, study by study, and by agreeing each step with regulators.

  1. Pre-specify it as exploratory

    Run the measure in early or ongoing trials with a fixed task, schedule and analysis in the SAP, so its results can be read as evidence later.

  2. Build V3 evidence

    Document verification, analytical validation and clinical validation following the DiMe V3 framework, and how readings hold across phone and software changes. Validation approach ›

  3. Replicate and define meaningful change

    Show the same relation to the clinical scale in further studies and languages, and define what change in the reading is meaningful to patients.

  4. Agree the next step

    Discuss the endpoint with the FDA review division, request EMA qualification advice, then name the measure as secondary, inside the Type I error plan.

Stride velocity 95th centile, a wearable measure of walking in Duchenne muscular dystrophy, followed the EMA route. CHMP qualified it as a secondary endpoint in pivotal or exploratory studies in patients aged 5 and above in April 2019, and as a primary endpoint in ambulant patients aged 4 and above in July 2023, provided it is supported by consistent findings in established secondary endpoints.

Sources

Guidance, opinions and frameworks

The requirements on this page come from these documents.

Questions

Speech as an exploratory endpoint: common questions

Can a speech measure be an exploratory endpoint in a clinical trial?

Yes. FDA's December 2023 guidance on digital health technologies covers primary, secondary and exploratory endpoints measured with data from digital tools, and applies the same principles as for endpoints captured by other means. The protocol names the speech measure as an exploratory endpoint and the statistical analysis plan defines it before the blind is broken.

Does an exploratory speech endpoint need multiplicity adjustment?

FDA's 2022 guidance on multiple endpoints says exploratory endpoints do not need multiplicity adjustment because they are generally not used to support conclusions. A speech measure moved to the secondary family would enter the trial's Type I error control plan.

What should the SAP pre-specify for a speech measure?

The definition of the endpoint and the source data it comes from, the speech task and schedule, the clinical scale it is read against, the analysis model and analysis version, how missing sessions and intercurrent events are handled, how phone and software updates are recorded and analysed, and who may see readings by arm.

Does a speech measure need FDA or EMA qualification before it is used in a trial?

Not for exploratory use in early trials. EMA's 2020 questions and answers on digital technologies state that exploratory use in early trials does not require qualification, and that the need arises once the technology supports pivotal data. FDA states that qualification is not required for a clinical outcome assessment to be used in clinical trials, and recommends that sponsors discuss endpoint plans with the relevant review division. Qualification is a voluntary route that lets a measure be used across drug development programs for a stated context of use.

How do phone and software updates affect a speech endpoint?

Updates can change the audio before analysis. FDA's December 2023 guidance asks sponsors to record the timing and nature of updates, to ensure the measurements are unaffected, and, where measurements change, to compare data before and after the update and specify how differences are handled before unblinding.

How does a digital measure move from exploratory to secondary or primary?

By accruing evidence across studies for one context of use, and by agreeing the next step with regulators. Stride velocity 95th centile is the known EMA precedent: CHMP qualified it as a secondary endpoint in Duchenne muscular dystrophy patients aged 5 and above in April 2019, and as a primary endpoint in ambulant patients aged 4 and above in July 2023.

How Cephalgo supports this

Speech readings written into your SAP

Cephalgo provides speech readings pre‑specified in your statistical analysis plan, read on, or beside, the scale the study already uses, calibrated on the study’s screening and baseline ratings, blind to arm, from recordings a trial already holds, site‑visit recordings and short home sessions. The readings stay stable across phone and software changes, and every reading carries its analysis version. Cephalgo evaluates each reading stage by stage following the DiMe V3 framework, and your study’s validation report follows TRIPOD+AI. Validation approach ›Device changes ›Readings and sample size ›

Planning speech in your next trial?

Share your indication, your scale and any audio you hold, and get a written scope.

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.