**Background:** Accurate dietary assessment is crucial for studying diet-disease relationships. Traditional food-based dietary records (DR) require detailed reporting of all ingredients, which is burdensome and prone to error, especially with mixed dishes common in Asian cuisines. Dish-based methods using a dish composition database (DCD) may reduce this burden by focusing on whole dishes rather than individual ingredients. However, the validity of DCD-estimated intakes against objective biomarkers had not been evaluated. This study aimed to validate a recently developed Japanese DCD against 24-hour urinary excretion of protein, sodium, and potassium, and compare its performance with a standard food composition database (FCD).
**Methods:** Data came from a nationwide cross-sectional survey in Japan (2013) involving 199 dietitians who recruited co-workers and family members. Participants completed a 4-day weighed DR (three working days, one non-working day) and two 24-hour urine collections. After excluding those with incomplete urine collections (creatinine ratio <60% or >140%) or insufficient measurement intervals, 161 men and 163 women (aged 20–69 years) were analyzed. Dietary intakes were estimated using: (1) the Japanese standard FCD (1878 food codes), (2) the DCD (128 dish codes with standard portion sizes), and (3) the DCD with reported portion sizes (adjusting standard portion sizes by the ratio of reported to standard dish weight). Urinary biomarkers were used to estimate dietary protein (from urea-N, using conversion factors: urinary urea-N/0.85/0.81 × 6.25), sodium (urinary Na/0.86), and potassium (urinary K/0.77). Statistical analyses included Wilcoxon signed-rank tests for median comparisons, Spearman correlation coefficients (crude and deattenuated to correct for within-person variation), and Bland-Altman plots to assess agreement.
**Key Results:** Median intakes estimated by the DCD and FCD differed significantly from urinary biomarker estimates for most nutrients. For men, protein (DCD: 74.7 g/d vs. urine: 89.8 g/d, p<0.0001; FCD: 82.9 g/d, p<0.0001) and sodium (DCD: 5017 mg/d vs. urine: 5240 mg/d, p=0.03; FCD: 4442 mg/d, p<0.0001) were underestimated. For women, sodium (DCD: 4765 mg/d vs. urine: 4442 mg/d, p=0.01; FCD: 3586 mg/d, p<0.0001) and potassium (DCD: 2595 mg/d vs. urine: 2305 mg/d, p=0.0002; FCD: 2560 mg/d, p=0.0002) were overestimated. The DCD with reported portion sizes improved median estimates for protein in men (88.2 g/d, p=0.07) but not for other nutrients. Deattenuated Spearman correlations with urinary biomarkers were lower for the DCD (protein: men 0.26, women 0.22; sodium: men 0.15, women 0.27; potassium: men 0.44, women 0.22) than for the FCD (protein: men 0.43, women 0.33; sodium: men 0.40, women 0.45; potassium: men 0.59, women 0.42). Using reported portion sizes improved DCD correlations (protein: men 0.32, women 0.31; sodium: men 0.31, women 0.41; potassium: men 0.56, women 0.39), approaching FCD levels. Bland-Altman plots showed wide limits of agreement for all methods, indicating poor individual-level accuracy. Proportional bias was observed for sodium in men (FCD slope = -0.659; DCD with reported PS slope = 0.410) and for protein and sodium in women (FCD slopes = -0.269 and -0.547, respectively).
**Clinical Implications:** The DCD, even with reported portion sizes, cannot accurately estimate absolute intakes of protein, sodium, and potassium at the individual or population level, similar to the FCD. However, the DCD with reported portion sizes demonstrates fair to moderate validity for ranking individuals, which is often sufficient for epidemiological studies examining diet-disease associations. This suggests that dish-based dietary assessment tools using the DCD could be a less burdensome alternative to traditional food-based methods, particularly in populations where mixed dishes are common. The findings highlight the need for further validation in real-world settings where participants select dish codes themselves, and caution against using the DCD for absolute intake estimation without additional calibration. The study underscores the inherent limitations of self-reported dietary data, regardless of the database used.