| Characteristic | Acquisition 1, N = 1031 | Acquisition 2, N = 1031 | Acquisition 3, N = 1031 | Acquisition 4, N = 1031 | Acquisition 5, N = 1031 |
|---|---|---|---|---|---|
| 5C1 - ALGEBRA | |||||
| ABSC (dB) | -27.00 (6.74) | -26.92 (6.81) | -26.90 (7.06) | -27.10 (6.87) | -27.13 (6.79) |
| DAX - ALGEBRA | |||||
| ABSC (dB) | -35.72 (7.75) | -35.75 (7.75) | -35.68 (7.42) | -35.43 (7.40) | -35.58 (7.48) |
| 5C1 - SDM | |||||
| ABSC (dB) | -27.89 (7.16) | -27.92 (8.52) | -27.40 (8.53) | -27.64 (7.26) | -27.87 (7.32) |
| DAX - SDM | |||||
| ABSC (dB) | -18.93 (11.23) | -20.03 (10.79) | -20.63 (10.31) | -19.17 (10.60) | -20.47 (9.35) |
| 1 Mean (SD) | |||||
Comparison of Ultrasound Scattering Estimates in Subjects Undergoing Weight Loss Surgery
Statistical analysis
Continuous variables are summarized as mean (SD) and categorical variables as frequency (%). Linear mixed effects models (LME) are used to analyze the longitudinal data of average backscatter coefficient (ABSC) estimated from ultrasound data. P values < 0.05 are considered statistically significant. All analyses are performed in R version 4.3.1 (R Core Team, 2023), using the lme4 package for LME models.
Baseline characteristics
There are \(n=\) 103 unique subjects across 2 sites. Table 1 below summarizes the baseline (day 0) ABSC (dB) values by transducer and algorithm.
Longitudinal analysis of ABSC
ABSC (dB) trajectories over time
The longitudinal trajectories of average ABSC (dB) by transducer, algorithm, and site are shown in Figure 1.
Without applying image critera
A preliminary analysis shows that
- The only significant interaction is between transducer and time;
- No significant difference between acquisitions.
A linear mixed effects model (LME) was fitted with subject random effects and relevant fixed effects. The results are summarized in Table 2 below.
| Characteristic | Beta | 95% CI1 | p-value |
|---|---|---|---|
| Time (months) | -0.03 | -0.05, -0.01 | 0.009 |
| transducer | <0.001 | ||
| 5C1 | — | — | |
| DAX | -0.65 | -0.81, -0.50 | |
| Site | <0.001 | ||
| Site 1 | — | — | |
| Site 2 | 2.9 | 2.1, 3.8 | |
| algorithm | <0.001 | ||
| ALGEBRA | — | — | |
| SDM | -1.1 | -1.3, -1.0 | |
| Skin to Liver Distance (cm) | -0.71 | -0.84, -0.57 | <0.001 |
| PDFF (%) | 0.14 | 0.13, 0.16 | <0.001 |
| TAL (dB) | 0.20 | 0.20, 0.20 | <0.001 |
| Time (months) * transducer | <0.001 | ||
| Time (months) * DAX | 0.11 | 0.08, 0.13 | |
| 1 CI = Confidence Interval | |||
Applying image quality criteria 1
Table 3 summarizes the results of the LME model fitted to the subset of data with image quality criteria 1.
| Characteristic | Beta | 95% CI1 | p-value |
|---|---|---|---|
| Time (months) | -0.05 | -0.08, -0.03 | <0.001 |
| transducer | <0.001 | ||
| 5C1 | — | — | |
| DAX | -0.73 | -0.89, -0.56 | |
| Site | <0.001 | ||
| Site 1 | — | — | |
| Site 2 | 2.7 | 1.8, 3.5 | |
| algorithm | <0.001 | ||
| ALGEBRA | — | — | |
| SDM | -1.1 | -1.2, -0.97 | |
| Skin to Liver Distance (cm) | -1.2 | -1.3, -1.0 | <0.001 |
| PDFF (%) | 0.16 | 0.14, 0.18 | <0.001 |
| TAL (dB) | 0.20 | 0.20, 0.20 | <0.001 |
| Time (months) * transducer | <0.001 | ||
| Time (months) * DAX | 0.11 | 0.09, 0.14 | |
| 1 CI = Confidence Interval | |||
Applying image quality criteria 2
Table 4 summarizes the results of the LME model fitted to the subset of data with image quality criteria 2.
| Characteristic | Beta | 95% CI1 | p-value |
|---|---|---|---|
| Time (months) | -0.05 | -0.07, -0.02 | <0.001 |
| transducer | <0.001 | ||
| 5C1 | — | — | |
| DAX | -0.73 | -0.90, -0.57 | |
| Site | <0.001 | ||
| Site 1 | — | — | |
| Site 2 | 2.7 | 1.8, 3.6 | |
| algorithm | <0.001 | ||
| ALGEBRA | — | — | |
| SDM | -1.1 | -1.2, -0.96 | |
| Skin to Liver Distance (cm) | -1.1 | -1.3, -0.96 | <0.001 |
| PDFF (%) | 0.15 | 0.13, 0.17 | <0.001 |
| TAL (dB) | 0.20 | 0.20, 0.20 | <0.001 |
| Time (months) * transducer | <0.001 | ||
| Time (months) * DAX | 0.11 | 0.09, 0.14 | |
| 1 CI = Confidence Interval | |||