Validity and reliability of tools to measure ultraprocessed food intake under the NOVA system: A systematic review

Abstract

Background: Ultraprocessed foods (UPFs) contribute substantially to global energy intake and are consistently linked to adverse health outcomes. However, accurately assessing UPF intake remains challenging due to reliance on self-reported data and retrospective processing classification, with performance highly dependent on reference method quality. Objectives: To evaluate the validity, reliability, agreement, and risk of bias of instruments used to quantify UPF intake according to the NOVA classification and to examine the agreement between food processing classification systems. Methods: We conducted a 2020 PRISMA–compliant systematic review of PubMed/MEDLINE, Scopus, and Web of Science from inception to 12 January, 2026, without language restrictions. Two reviewers (IC-R and AS-L) independently screened studies, extracted data, and assessed methodological quality via the COnsensus-based Standards for the selection of health Measurement INstruments (COSMIN) risk of bias checklist. The evidence was synthesized narratively and through structured visual summaries of test–retest reliability (intraclass correlation coefficients), relative (criterion/convergent) validity against dietary reference methods, classification agreement (Cohen’s κ and prevalence-adjusted bias-adjusted κ), and internal consistency (Cronbach’s α or McDonald’s ω). Results: Thirty studies were included: 19 evaluated basic dietary intake instruments (NOVA-oriented food frequency questionnaires, short screening tools, or UPF scores), 4 examined biomarker-based or indirect tools, and 7 assessed contextual or auxiliary instruments. The test–retest reliability of the dietary intake instruments was generally moderate-to-high (intraclass correlation coefficients ≈ 0.46–0.94). Relative validity against dietary reference methods was modest to moderate (correlations typically r ≈ 0.47–0.72), and agreement between processing classification systems ranged from fair-to-substantial (Cohen’s κ and prevalence-adjusted bias-adjusted κ ≈ 0.36–0.84). The contextual and auxiliary instruments showed high internal consistency (Cronbach’s α/McDonald’s ω ≥0.74). Fifteen studies were rated as having a low risk of bias, and 15 were rated as unclear. Conclusions: Instruments assessing NOVA-defined UPF intake demonstrate acceptable reproducibility but variable, reference-dependent validity and system-specific classification uncertainty. Future research should prioritize higher-quality reference methods, transparent coding rules, calibration strategies, and routine evaluation of measurement error and responsiveness.