Food Noise: What the Term Means in GLP-1 Research

Food noise describes an experience that researchers are learning to measure. A reliable questionnaire and a persuasive treatment study answer different questions.

Imagine someone saying that thoughts about the next meal keep interrupting work, even when they are trying to concentrate on something else. Another person may enjoy planning dinner without finding those thoughts intrusive. “Food noise” is an attempt to describe the first kind of experience: persistent thoughts about food that can interfere with daily life. That is how Arnaut and colleagues frame the term in their 2026 INFORM survey of adults receiving semaglutide for weight management.^[2]

Turning that description into research requires more than giving it a name. Researchers need a way to measure the experience consistently, establish what a change means, and test whether treatment caused it. Two recent studies address different parts of that sequence.

An experience becomes a research construct

A construct is a concept researchers define so it can be investigated. Food noise is being studied as a reported experience, rather than as a new chemical concentration measured in blood. The Food Noise Questionnaire, or FNQ, asks participants to report that experience through five items. Its development study examined whether those responses behaved like a coherent measure.^[1]

This distinction matters because subjective experiences can be important outcomes without being laboratory measurements. A person's report describes something an assay cannot directly capture. The scientific question is whether the instrument captures the intended experience consistently and usefully. Calling it subjective does not make it unimportant; calling it measurable does not establish a diagnosis.

Neither study reviewed here establishes a universal diagnostic threshold for food noise. The original FNQ authors explicitly state that further research is needed to evaluate its clinical utility.^[1] That leaves room for a useful research instrument while keeping claims about diagnosis and clinical decisions appropriately limited.

What the questionnaire study established

In the 2025 development study, Diktas and colleagues initially had 400 participants; data from 396 were analyzed because four had not completed all FNQ items. The study used separate participant subsets to examine the questionnaire's structure and its relationships with other measures. A subset repeated the FNQ approximately a week later.^[1]

The five items loaded onto a single factor, meaning their response patterns were consistent with a shared underlying concept. Internal consistency was high, with Cronbach's alpha of 0.93. Internal consistency concerns how closely the items relate to one another; it is not a statement that the questionnaire diagnoses people with 93% accuracy. Test-retest reliability was also high, with a correlation of 0.79.^[1]

That second result indicates that scores from the two administrations were strongly related. It does not mean every person's score stayed identical, or that a particular change must represent a treatment effect. Repeatability and sensitivity to meaningful change are separate measurement questions.

The FNQ also correlated strongly with other food-preoccupation questionnaires and less strongly with mood, anxiety and stress measures. The authors interpreted these findings as support for convergent and discriminant validity: connection to related concepts, with some distinction from other experiences.^[1] Their favorable conclusion about measurement properties should be retained alongside their caution about clinical utility.

What people reported in INFORM

Arnaut and colleagues' 2026 INFORM study surveyed 550 US adults receiving injectable semaglutide for weight management. Participants came from an opinion panel and completed questions that included the FNQ. Respondents were predominantly women, at 86%, and white, at 79%; their mean age was 53 years. Most, 81%, had used semaglutide for at least four months.^[2]

Participants recalled their food noise before treatment and reported their experience after starting it. The median recalled pretreatment FNQ score was 13, and the median after-initiation score was 6. The reported median individual change was −5. These are favorable reported changes in the surveyed group, and they should not be dismissed simply because the outcome concerns thoughts rather than a blood test.^[2]

The authors also reported that 83% were satisfied with treatment, and that reductions in food-noise responses were present across treatment-duration and body-mass-index subgroups.^[2] These findings describe respondents' perceptions. They do not establish that all users experience the same change, that the change persists indefinitely, or that the survey isolated semaglutide's causal effect.

Why 13 to 6 does not make the median change −7

A median is the middle value after observations have been put in order. The middle pretreatment score and the middle post-treatment score need not belong to the same person. Subtracting those two group summaries therefore does not generally produce the median of each person's change.

For a paired change, researchers first subtract each participant's earlier score from that participant's later score. They then summarize those individual differences. INFORM reports a median paired change of −5, with an interquartile range of −10 to −2, even though its separate group medians were 13 and 6.^[2] The interquartile range describes the middle half of the reported changes, not the range for every participant.

This is a useful example of why an article should preserve the statistic the authors actually calculated. Replacing −5 with −7 would change the reported result. Neither value, by itself, establishes a clinical threshold or the size of a causal medication effect.

Research question	Evidence here	Remaining limit
Is measurement consistent?	FNQ reliability study	Clinical utility
Did respondents feel change?	INFORM recalled scores	Recall and selection
Did treatment cause change?	No randomized comparison	Causal attribution

What a retrospective survey cannot resolve

INFORM was retrospective: its “before” measurements depended on recollection after treatment began. They were not FNQ scores collected prospectively before participants started treatment. Recall can be influenced by current experience or expectations. That possibility does not prove respondents remembered incorrectly; it explains why the design cannot resolve the issue.^[2]

Selection also matters. An opinion-panel survey of people receiving treatment may not represent those who stopped early or chose not to participate. Continued users can differ from early discontinuers, creating a potential survivorship effect. The predominance of women and white respondents further limits direct transfer to populations represented less often. These are interpretation limits, not measured estimates of how much bias occurred.^[2]

A proposed next study would collect pretreatment scores in real time, follow participants repeatedly, include those who discontinue, and use an appropriate comparison group. Such features would help distinguish change over time from treatment-related change. The INFORM authors themselves call for prospective, longitudinal research and investigation of relationships with clinical outcomes.^[2]

Patient reports deserve careful evidence standards

FDA's December 2009 guidance concerns patient-reported outcome instruments used to support medical-product labeling claims. Its public description treats an instrument as the questionnaire plus the documentation supporting its use. It describes evaluation in medical-product development, rather than a general endorsement of every patient questionnaire.^[3]

That guidance is not FDA approval of the FNQ. The two food-noise studies support a narrower account: a promising measure with reported reliability and validity, followed by a survey in which respondents described substantial improvement. Reliable measurement and causal treatment benefit remain distinct questions. Neither paper supplies instructions for changing a prescription on the basis of a questionnaire score.

The two studies were reviewed through their original published abstracts. The FDA source reviewed was the guidance's public overview page, not the full guidance document. Detailed questionnaire content, complete study methods and a clinical interpretation threshold were not independently assessed here.