Commentary

Notes from the dietary-assessment poster session at the 2024 ACSM annual meeting

Field observations from Boston, 28 May – 1 June 2024

Helena Weiss, PhD, MPH, RD
Published July 8, 2024

The dietary-assessment poster rows at the American College of Sports Medicine annual meeting rarely draw the crowds of the sport-science sessions, but the 2024 meeting in Boston was, by my informal count, the densest assembly of image-based assessment work the Initiative has observed at a mainstream exercise-science venue. I walked the session twice on 30 May and took notes; what follows is a field-note summary, not a systematic review.

What was on the boards

Of the 37 posters I identified as relevant to image-based dietary assessment or mobile app validation, 19 reported some form of primary validation data against a reference (weighed, duplicate plate, or 24-hour recall). The remaining 18 were either descriptive (adoption, acceptability, feasibility) or methodological without new empirical data. Several groups — including two from the Nordic countries and one from Singapore — presented staged validation work in which the first poster of a multi-year program simply established the reference-meal protocol.¹ This is, in our view, a healthy sign; multi-stage validation was uncommon at ACSM as recently as 2021.

Reference methods varied considerably. Among the 19 validation posters, the reference was weighed food in 6, a 24-hour dietary recall administered by a dietitian in 7, a 3-day food record in 4, and — in two cases — a prior self-reported estimate from the same participant, which is not a reference method by any conventional definition. The field continues to tolerate a level of terminological slippage that would not pass review in, for instance, laboratory-medicine validation work.²

Where sample sizes landed

Sample sizes for the validation posters ranged from n = 11 (a pilot from a European group) to n = 312 (a multi-site study). The median was n = 42 participants and approximately 180 meals. This is roughly consistent with what Henriksen and colleagues reported in the 2023 systematic-review update for the field.³ Confidence intervals on point estimates of mean absolute percentage error (MAPE) or limits of agreement were, in many cases, not reported; where they were, they tended to be wider than the text of the poster abstract suggested.

Vendors in the room

Four posters were first-authored by industry affiliates, and a further six had at least one industry co-author. The industry-affiliated posters were not systematically less rigorous than the academic posters — one of the more careful Bland-Altman analyses of the session was an industry poster — but they were systematically more likely to foreground a single favourable point estimate in the headline. Readers who have followed the Initiative’s prior commentary on vendor-reported numbers will not find this surprising.⁴

What was missing

Three things struck us as absent. First, pre-registration: of the 19 validation posters, two referenced a prospectively registered protocol. Second, independent replication: no poster I saw presented an independent replication of a previously published validation. Third, a standardized outcome set: the primary outcomes varied across energy (kcal), macronutrients, portion weight, and — in one poster — a composite “nutrition accuracy score” that combined six outputs with unequal weights.

We intend to return to this last point in a separate brief. A field cannot compare apps across studies if each study defines “accuracy” differently, and conference programmes are a reasonable place to begin arguing for convergence.

Notes