Validity of image-based dietary assessment methods: A systematic review and meta-analysis

Image-assisted or image-based dietary assessments (IBDAs) refer to the use of food images as the primary dietary record and have emerged as key methods for evaluating habitual dietary intake; however, the validity of image-assisted or IBDAs is still unclear, and no meta-analysis has been conducted. Our aim was to investigate the validity of IBDAs in assessing energy intake (EI) and macronutrients compared to biomarker-based (double-labeled water (DLW)) and traditional methods of 24-h dietary recall (24-HDR) and estimated/weighed food records (WFRs).

A systematic review and meta-analysis were performed according to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines using the PubMed, EMBASE, and Cochrane Central Register of Controlled Trials databases. Of the 4346 papers identified, 13 studies met the inclusion criteria, comprising 606 participants.

The overall weighted mean difference (WMD) in EI showed significant under-reporting (WMD = -179.32 kcal, 95% confidence interval (CI): -269.50 to -89.15 kcal; I² = 89%), with the greatest difference observed between tests and DLW (WMD = -448.04 kcal, 95% CI: -755.52 to -140.56 kcal; I² = 95%). A small non-significant trend towards under-reporting of carbohydrates (CHOs) was observed (WMD = -9.17 g, 95% CI: -20.58 to 2.24 g; I² = 64%), but no differences were found in protein (WMD = -0.08 g, 95% CI: -3.94 to 3.79 g; I² = 68%, p < 0.01) or fat (WMD = -0.57 g, 95% CI: -2.58 to 1.43 g; I² = 12%, p = 0.35). A meta-regression analysis found potential effects of the body-mass index (tests vs. DLW: β = 34.9, p = 0.063) and duration of the assessment (tests vs. WFR: β = -66.5, p = 0.002) on EI; age (tests vs. 24-HDR: β = -2.222, p = 0.019) and duration of the assessment (tests vs. WFR: β = -9.19, p = 0.013) on CHO intake; duration of the assessment on protein intake (tests vs. WFR: β = -3.2250, p = 0.0175); and duration of the assessment on fat intake (tests vs. WFR: β = -1.07, p = 0.040).

Except for DLW, no statistical difference was found between IBDAs and traditional methods. This suggests that like traditional methods, image-based methods have serious measurement errors, and more studies are needed to determine inherent measurement errors in IBDAs.