Table 2.
Reference | Name of Tool | What Is Measured | Reliability | Validity | ||
---|---|---|---|---|---|---|
Statistical Method Used | Result | Statistical Method Used | Result | |||
[28] | Automated Wrist Motion Tracking | EI | Sensitivity (true detection rate) = (total true detection)/(total true detection + total undetected bites); Positive Predicted Value (PPV) = (total true detection)/(total true detection + total false detection); compared recorded bites with direct observation. | Control setting: Sensitivity = 94% PPV = 80% Semi-controlled setting: Sensitivity = 86% PPV = 81% |
Pearson correlation of EI estimated by device vs. direct observation (r) | R = 0.6 |
[30] | The bite-based model of kilocalorie intake | EI | Pearson’s correlation of device compared with direct observation; shrinkage value | R = 0.374 Shrinkage value (difference in R2) = 0.014 | Independent t test Paired sample t test | Mean estimation error kilocalorie information group: −185 ± 501 kcal; Mean estimation error no kilocalorie information group: −349 ± 748 kcal (p < 0.05); Best human-based estimation (kilocalorie information group) mean estimation error: −257 ± 790 kcal; Bite-based method (predicted formula) mean estimation error: 71 ± 562 kcal; (p < 0.001). |
[31] | Automatic Ingestion Monitor (AIM) | EI | N/A | N/A | Accuracy = average between precision (P) and recall (R). | Accuracy of food ingestion = 89.9%, range from 75.82–97.7%. |
[32] | Counts of Chews and Swallows Model | EI | A 3-fold cross validation technique, one sided Wilcoxon-Mann-Witney, Bland-Altman analysis and t-Test analysis. | Reporting error for the CCS model was lower than that of the diet diary (p < 0.01). The model underestimated EI. Energy intake estimation had the lowest bias. | A 3-fold cross validation technique, one-sided Wilcoxon-Mann-Witney, Bland-Altman analysis and t-Test analysis. | No statistical differences were found between the CCS model and either diet diary or photographic records. |
[33] | Intelligent food-intake monitor | Food intake | Correlation: Proportion of food consumed from sound (auditory based) and image sequence (vision based) compared to the ground truth: proportion of food consumed. | Data not shown | N/A | N/A |
[34] | DP + R | EI | Inter-rater reliability coefficients | Error rate ≤5%, Recall assessments ≥0.95 | Dependent t-test comparing device to DLW method; Bland-Altman plots; Limits of agreement | Differences between methods in the total sample was not significantly different (DP + R = 2912 ± 661 kcal/day; TDEEDLW = 2849 ± 748 kcal/day, p = 0.42); DP + R was found to overestimate EI compared to TDEEDLW by 63 ± 750 kcal/day (6.8 ± 28%; limits of agreement: −1437, 1564 kcal/day). The Bland-Altman plot indicated no proportional bias variation as a function of the level of EI in the total sample (R = −0.13, p = 0.21). |
[35] | RFPM | EI | Bland & Altman analysis | Significant difference: p < 0.0001 between the RFPM and DLW in the standard prompt group. No significant difference in the customized group: p = 0.22. The level of bias in both groups was not influenced by the amount of EI (Adj. R2= −0.03, p = 0.55; Adj. R2 = −0.08, p = 0.78) | Independent sample t-test for error between methods = EI estimated with the RFPM-EI measured with DLW | Significant smaller underestimation in the customized group (270 ± 748 kcal/day or 8.8 ± 29.8%) when compared to the standard prompt group (895 ± 770 kcal/day or 34.3 ± 28.2%), t (33) = −2.35, p < 0.05 with RFPM. |
[36] | Real-time Food Recognition System | EI | Test-retest reliability | 79.2% classification rate | N/A | N/A |
[37] | Snap-n-Eat | Energy/dietary intake | Test-retest reliability | Classification accuracy (% of correctly classified images categories) = 85% | N/A | N/A |
[38] | GoCARB | Carb EI | Comparison to actual foods/database | Automatic segmentation (portion size) = 75.4% (86/114); Food item recognition = 85.1% (291/342) |
Mean absolute error; Relative error |
Mean absolute error = 27.89 (SD 38.20) and 12.28 (SD 9.56) grams of carbohydrates; Mean relative error = 54.8% (SD 72.3%) and 26.2% (SD 18.7%). A significant error between estimations was found (p = 0.001). In general, 60.5% (69/114) of the participants underestimated carbohydrate content. |
[39] | Mathematical method | Change in EI | Test-retest reliability; Mean difference |
40 kcal/day of mean difference between the gold standard and the mathematical model; No significant difference between the methods for any of the time segments was found (weeks 0–26: p = 0.14; weeks 26–52: p = 0.34; weeks 52–78: p = 0.32; weeks 78–104: p = 0.11). | Paired, 2-sided t test; Pearson correlation (r) Spearman’s corrected (rs) | Change in EI values calculated by the mathematical method or the gold standard DLW/DXA weren’t significantly different; The mathematical model had an accuracy within 132kcal/day for predicting changes in EI; The magnitude of correlation of the change in EI values between models were correlated (weeks 0–26: r = 0.57 (95% confidence interval 0.45, 0.68); p =≤ 0.0001; weeks 78–104: r = 0.19 (0, 0.36); p = 0.05). |
1 N/A = Not applicable.