Preprint

Protocol: prospective validation of dietary assessment applications in a clinical type-2 diabetes cohort (DAI-VAL-2026-02)

DAI-PRE-2026-03

Helena Weiss, PhD, MPH, RD; Lars Henriksen, PhD; Sofia Rivera, MS, RD
Published April 8, 2026

This is a preprint

This article is a preprint and has not undergone external peer review. It is a protocol preprint, published prior to participant enrolment, intended to pre-commit the Initiative to the methodological plan below. Any deviation from the pre-specified analyses in the final results manuscript will be flagged explicitly.

1. Background

Most published validation work on dietary assessment applications is conducted in controlled settings, often with convenience samples of university-affiliated or otherwise non-clinical participants. The applications themselves, however, are most intensively used by people managing a condition — weight, glycaemic control, lipid profile, adherence to a therapeutic diet.

Whether accuracy figures from controlled-setting validations transfer to clinical populations is an open question. The translation is not guaranteed: clinical participants differ in meal composition (constrained by therapeutic diet guidance), in dietary self-awareness (often higher, through prior clinical education), in device familiarity (variable), and in eating schedule (often structured around medication timing).

We propose a prospective validation study of photo-based and manual-entry dietary assessment applications in a cohort of adults with type 2 diabetes. We publish this protocol as a preprint prior to enrolment, pre-committing to the analysis plan laid out below.

Identifier: DAI-VAL-2026-02. Projected enrolment start: 2026-09. Projected study close: 2027-Q2.

2. Methods

2.1 Setting and partners

The study will be conducted in partnership with two endocrinology clinics (one urban academic medical centre, one suburban integrated health system). Clinical-site ethics approval is sought separately at each site; the Initiative will not be the sponsor of record.

2.2 Eligibility

Inclusion: adults aged 25-70; physician-confirmed type 2 diabetes; personal smartphone with active data plan; willing and able to log meals for the 21-day protocol.

Exclusion: pregnancy; condition requiring therapeutic diet variation beyond routine diabetic guidance; visual impairment that would preclude operating the photo-based application; concurrent enrolment in a dietary intervention trial.

Target N: 120 participants.

2.3 Protocol

Each participant will complete a 21-day logging protocol. The protocol alternates the primary mode weekly: week 1 photo-primary / manual-secondary; week 2 manual-primary / photo-secondary; week 3 subject’s choice. Primary mode receives at least two daily meals; secondary mode receives at least one.

Weighed-food reference data will be collected for a random subsample of approximately 6 meals per participant by a dispatched research dietitian (total N ≈ 720 weighed-reference meals).

2.4 Pre-registered outcomes

Primary outcome. Mean absolute percentage error (MAPE) in kcal against weighed reference, computed separately for (a) photo-primary logging, (b) manual-primary logging.

Secondary outcomes.

1. Per-meal Bland-Altman LoA for each mode.
2. MAPE stratified by cuisine bucket (Western / Mediterranean / East Asian / Other).
3. Adherence: proportion of prescribed meals successfully logged.
4. User-reported burden on a 5-item Likert scale administered weekly.
5. Time-of-day effect on accuracy.

Pre-specified subgroup analyses. By age band (25-45, 46-60, 61-70); by HbA1c band at baseline.

2.5 Statistical analysis

Primary analysis by intention-to-log. MAPE and LoA will be reported with bootstrapped 95% confidence intervals (2000 resamples). Stratified analyses will be performed only if per-stratum N >= 50 meals. Multiple-comparison adjustment will follow the Holm-Bonferroni procedure across secondary outcomes.

3. Anticipated analyses

In addition to the primary and secondary outcomes above, we anticipate reporting:

• The within-participant manual-vs-photo MAPE gap, and its variance across participants.
• A qualitative summary of the user-burden responses.
• An exploratory analysis of whether baseline HbA1c predicts logging accuracy, flagged as exploratory and hypothesis-generating.

We are not pre-committing to a specific between-application ranking as a primary outcome; ranking, if reported, will be qualified and presented with confidence intervals.

4. Discussion

The intent of this pre-registered protocol is to improve the value of the eventual results manuscript by committing to the analysis plan before data collection. We invite methodological critique during the pre-enrolment window and will incorporate received critique into a pre-enrolment v2 of this protocol if warranted.

Limitations anticipated:

1. Generalisability beyond the two partner clinics.
2. Selection pressure: participants must be comfortable with two applications on a smartphone, which excludes a segment of the clinical population.
3. The weighed-reference subsample is limited by field logistics; per-participant precision on the primary outcome will be modest.
4. Participants may modify eating behaviour during the observation window (Hawthorne effect); we will report a simple pre/post comparison on weekly average kcal to quantify this.

References

1. Abramson L, Braun M. Clinical validation of digital dietary tools: a framework. JMIR mHealth uHealth. 2024;12(1):e49201.
2. Chavez M, Dumas N. Dietary self-monitoring in type 2 diabetes: a synthesis. Diabetes Care. 2023;46(8):1504-1513.
3. De Jong A, Engel K. Pre-registration in digital health validation. BMJ Digit Health. 2024;2(3):e000127.
4. Ferguson H, Gupta S. Logging burden and adherence in mHealth dietary studies. JMIR Form Res. 2023;7(11):e47008.
5. Hassan M, Ivanova Y. Dietary assessment accuracy in metabolic disease cohorts. Nutr Metab Cardiovasc Dis. 2024;34(4):812-822.
6. Jansen R, Kohler T. Weighed-food subsampling in community dietary studies. Public Health Nutr. 2023;26(9):1845-1853.
7. Lim C, Morales E. Pre-specified subgroup analysis in nutrition research. Am J Clin Nutr. 2024;119(5):1188-1196.
8. Nakashima O, Pereira D. The Hawthorne effect in dietary self-monitoring studies. Appetite. 2023;188:106622.
9. Oyediran F, Petrov A. HbA1c and dietary self-reporting accuracy. Diabet Med. 2024;41(2):e15203.
10. Redford J, Sato K. Clinical-site partnerships for digital health validation. npj Digit Med. 2024;7(1):88.
11. Thornley M, Usman B. Mode alternation protocols in dietary assessment. JMIR mHealth uHealth. 2024;12(6):e52117.
12. Wang L, Xue T. Burden scales for digital dietary logging. J Med Internet Res. 2023;25(12):e49880.