Descriptive analysis of the clinical data
All the 742 cases were female, ranged from 18 to 59 years, and the average age was 36.34 ± 8.73 years old. Normality tests and descriptive statistics of age, height, weight, BMI, WC, WHtR and fat volume are listed in Table 1. After the normality tests, we found that such variables as age, height, weight, WC, BMI, WHtR and intraoperative liposuction volume were skewed distribution. The fat volume removed from the waist and abdominal areas and other variables are shown by a distribution histogram in Figure 3. Fat volume was not a normal distribution (KS test, P < 0.001) but a positively skewed distribution with the long right-side tail, and the kurtosis and skewness were 2.759 and 1.431 respectively.
Correlations between fat volume, age and other anthropometric measurements
All the data were skewed distribution, thus, the Spearman correlation coefficients were used (Figure 4). There was a weak correlation between liposuction volume and age (r = 0.150, p < 0.05), and there was no correlation between fat absorption and height (r = 0.045, p > 0.05). Body weight, BMI and WHtR correlated significantly with liposuction volume (r = 0.608, 0.646 and 0.699, respectively, p < 0.05), while WC displayed the strongest correlation (r = 0.705, p < 0.05), thus suggesting that the size of preoperative WC may be the most important factor in determining liposuction volume, not BMI and WHtR. Based on that, in order to ensure the robustness of the data, outliers were removed for analysis, and the results were consistent with the above results (Table S2).
Besides, there exhibited high correlations between BMI and weight (r = 0.897, P < 0.01), between BMI and WC (r = 0.8, P<0.01), between BMI and WHtR (r = 0.820, P < 0.01), and between WC and WHtR (r = 0.928, P < 0.01). The above cross correlation suggested that there may be a collinear relationship between these preoperative measurements.
Partial correlations between preoperative measurements and fat volume
Adjusted for WC, weight, BMI and WHtR, partial correlation analysis was performed to evaluate the association between preoperative measurements and liposuction volume. As shown in Table 2. the correlation coefficients between all included variables and fat extraction declined to some extent, no matter which variables were adjusted. In addition, when WC was adjusted, the correlation coefficients between such variables as body weight, BMI, WHtR and fat volume were most significantly reduced, from 0.608 to 0.079, from 0.646 to 0.193, from 0.699 to 0.144, respectively, indicating that WC was the most important factor in determining fat volume. Also, we got rid of the outliers and the same results were found.
Multiple linear regression analysis
Multiple linear regression models were used to evaluate the association between WC and liposuction volume. At the same time, the non-adjusted and adjusted models are shown in Table 3. In the crude model, WC displayed significant correlation with liposuction volume (β = 108.36, 95% CI: 101.35 to 115.38, P < 0.0001). In the minimally adjusted model (adjusted age), the effect size did not show obvious changes (β = 110.32, 95% CI: 103.18 to 117.46, P < 0.0001). Nevertheless, after adjusting for other covariates, we did not detect any connection in a fully adjusted model (β = 250.79, 95%CI: -126.12 to 627.70, P = 0.1926). To ensure the robustness of data analysis, we also handled WC as a categorical variable (quintile) and found the same trend (p for the trend was 0.582).
Relationship between WC and fat volume in subgroup analyses
Interactions with all covariates are presented in Figure 5. Subgroup analyses showed that the association of WC with fat volume remained stable in different subgroups when grouped by age, height, weight, BMI, WHtR. There was evidence for an interaction between WC and age (P for interaction = 0.0065). Besides, the tests for interactions were statistically significant for height, weight, BMI and WHtR with WC (P for interaction < 0.05). The effect sizes of WC on fat volume showed significant differences in diverse degrees of obesity. Compared with normal BMI (<18.5) and lower BMI (18.5-24.0), the strongest effect size between WC and fat volume was observed on higher BMI (> 24.0) (β = 111.28, 95% CI: 97.88 to 124.69). Besides, WC was the most significantly associated with fat volume in individuals who had a higher WHtR (β = 114.21, 95% CI: 90.71 to 137.71).
Predicting total liposuction volume
Because there was severe collinearity between the included variables, it is impossible to provide a viable equation for fat volume estimation. According to the amount of lipoaspirate, patients were divided into three surgical grades: level 1 < 3000ml, level 2: 3000ml to 5000ml level 3 > 5000ml. In these 742 cases, 68.60% of the patients' fat absorption was less than 3000ml, 26.95% was between 3000ml and 5000ml, and only 4.45% was greater than 5000ml, among which the maximum volume was 8300ml. The WC was divided into nine consecutive groups with 5cm equidistance, and fat volume sucked out was classified into three grades to form a contingency table (Table 4). The linear correlation trend test was carried out based on the two-way grade data (linear-by-linear association chi-square statistic, X2 = 535.282, df = 1,p < 0.001). The results showed that there was a significant linear correlation between WC group and fat aspiration grade group.
The bubble diagram of waist circumference and fat volume within the group can be seen in the Figure 6. The liposuction volume of small WC group (< 90cm) was less than 3000ml, and the liposuction volume exceeding 3000ml emerged with the increase of WC. Besides, the proportion of large WC group (> 105cm) was mainly accompanied by liposuction volume of level 2 and level 3. The main trend indicated in this figure, with some variation, was that with greater WC, liposuction volume became larger. The Goodman-Kruskal gamma test showed statistically significant positive correlations between the WC rankings and the degree of liposuction volume (gamma = 0.802, P < 0.001), indicating that there was a high correlation and consistency between these two variables. The purpose of this study was to provide an evidence for WC to predict and estimate the grading distribution of liposuction volume.
Figure 7 displayed that the average liposuction volume in each group had a linear trend with the WC grouping, the Pearson correlation coefficient was 0.9817, 95% CI: 0.9125 to 0.9963 (P < 0.001), and the linear regression formula was as follows: liposuction volume (mean) = 106.3WC (mean)-7497; P<0.001, adjusted R2 = 0.9638. However, Figure 8 showed that the variation coefficient of WC was relatively small, 1.07%-3.60%, while the liposuction volume has a larger coefficient of variation, distributed in the range of 21.20%-37.30%.
Although there was a linear correlation between WC and liposuction volume in each group, the variation of liposuction volume within the group was high, demonstrating that there were great differences in liposuction volume among individuals with the same WC. Therefore, it is impossible to accurately predict the liposuction volume in terms of the individuals’ WC. However, it is feasible to predict the probability of liposuction volume grade distribution by the grouped WC.