A J-shaped relationship between body mass index and the risk of elevated liver stiffness: a cross-sectional study

Repeated liver injury secondary to any cause would lead to progressive fibrosis, and eventually resulting in liver cirrhosis [1]. Even though advanced liver cirrhosis could manifest in many systems, there is no specific clinical symptoms or signs indicating early stage liver fibrosis [2]. It has been reported that using the degree of liver fibrosis can predict adverse patient outcomes, and thus monitoring liver stiffness (LS) in the common community residents without advanced liver diseases may be a cost-effective way in the secondary and tertiary prevention for liver diseases [3].

Liver biopsy has always served as the golden standard for the diagnosis of liver fibrosis. However, this procedure is not only invasive and painful, but may also cause potentially life-threatening complications. Many studies reported that TE-based liver stiffness measurement could be used as a rapid, accurate and most importantly, non-invasive way to measure liver fibrosis and liver cirrhosis in people with or without diseases [4, 5]. Notably, the strong association between LSM values and the histological stages of liver fibrosis makes it an ideal alternative for liver biopsy [6, 7], especially in people with no or few symptoms.

Obesity, classified as BMI ≥ 30 kg/m² in Caucasian populations, or BMI ≥ 25 kg/m² in Asian–Pacific populations [8], is related to higher risk of advanced liver disease [9‐11]. Several studies have explored the relationship between BMI and the risk of liver fibrosis [12, 13]; however, the conclusions remain controversial. Here, we took advantage of TE technique to measure LS in common community-based residents, and to further analyze the association between BMI and the risk of elevated LS in this population.

The average age of the 5,791 participants enrolled in the final analyses was 47.97 (± 12.59 SD) years, with 3662 (63.24%) being male. Among these 5791 participants, elevated LS was detected in 230 (3.97%). We first classified these participants according to their BMI into underweight (BMI < 18.5 kg/m²), normal weight (BMI 18.5 ~ 23.9 kg/m²), overweight (BMI 24.0 ~ 27.9 kg/m²) and obese (BMI > 28.0 kg/m²) groups. Then, we compared their demographic features, as well as clinical and biochemical test results in each group (Table 1). Significant differences were observed among these groups. We found that participants in the obese group are more prone to have comorbidities, such as hypertension, coronary heart disease, diabetes and dyslipidemia. These people also present with elevated MAP, FBG, as well as AST and AST levels, which indicate worse liver functions. To further clarify the correlations between BMI and liver damage, we then compared LS measured by TE in each group. We found that participants in the normal weight group had a lowest prevalence of 1.72% for elevated LS, while abnormal BMI, whether increased or decreased, was associated with higher risk of elevated LS (2.59% for underweight, 4.54% and 14.37% for overweight and obese, respectively).

To eliminate interference of different baseline variables and to distinguish the relationship between each of them and elevated LS, we first used univariate logistic regression models. The univariate analysis indicated that age, history of coronary heart disease, diabetes, hypertension, MAP, ALT, AST, FBG, creatinine and TG were positively correlated with the prevalence of elevated LS; however, female gender and HDL were negatively correlated with it (Table 2).

The relationship between BMI and LS was further analyzed using logistic regression models. As shown in Table 3, these models were adjusted based on the participants’ demographic variables, underlying comorbidities and their blood biochemical test results. We noticed that with every unit of BMI increase, the prevalence of elevated LS increased by 29% (OR = 1.29; 95% CI 1.24–1.34; P < 0.0001) in the original, unadjusted model. Besides, compared to the underweight group, OR for the normal weight group (OR = 0.66; 95% CI 0.26–1.67; P = 0.3775) was lower, and ORs for the overweight (OR = 1.79; 95% CI 0.72–4.45; P = 0.2105) and obese groups (OR = 6.31; 95% CI 2.51–15.85; P < 0.0001) were higher. We then made partial adjustment to exclude the influence of demographic variables and underlying comorbidities of the participants in Model 1, and full adjustment by adding the baseline levels of ALT, FBG, Creatinine, TG and HDL in Model 2. Our analysis results showed that the correlation remains in this situation (OR = 1.29; 95% CI 1.23–1.36; P < 0.0001 for Model 1, OR = 1.23; 95% CI 1.16–1.30; P < 0.0001 for Model 2), which confirmed the relationship between BMI as a continuous variable and increased risk of elevated LS. Moreover, in the fully adjusted Model 2, the normal weight participants exhibited the lowest risk of elevated LS in comparison with the underweights and the obese participants, although the P values were not statistically significant.

Since BMI is a continuous variable, the non-linear relationship should also be analyzed. Using spline smoothing fitting with adjustment of the previously mentioned demographic variables, underlying comorbidities and the participants’ blood biochemical test results, a J-shaped relationship between BMI and the prevalence of elevated LS clearly emerged (Fig. 2). Further analyses using a two-piecewise linear regression model revealed the inflection point was 23.05 kg/m² (Table 4). Based on this J-shaped curve, to the left of the inflection point, per unit decrease of BMI is associated with 16% increase of the prevalence of elevated LS (adjusted OR = 0.84, 95% CI 0.71–0.98, P = 0.027); on the other side, to the right of the inflection point, per unit increase of BMI is associated with 32% increase of the prevalence of elevated LS (adjusted OR = 1.32, 95% CI 1.24–1.41, P < 0.0001).

In this study, we examined the relationship between BMI and the risk of elevated LS in community-based residents, using TE as our measurement tool, and observed a J-shaped curve between them. An inflection point was detected at 23.05 kg/m², which is in the ‘normal BMI’ range. BMI seemed to have opposite influences on the risk of elevated LS on both sides of the inflection point. On the right side, BMI was positively associated with the risk of increased LS; however, a significantly negative association was found on the left side. That is to say, people with very low or very high BMI are more prone to present with elevated LS, while a normal BMI could be treated as a protective factor for liver function.

Previous researches have been conducted to detect the prevalence of liver fibrosis measured with TE in general populations. Two researches based on the Rotterdam Study, a European cohort, used LSM ≥ 8.0 kPa as their cutoff to suggest clinically relevant fibrosis [24, 25]. The use of a lower cutoff value would lead to a higher estimated prevalence of fibrosis in their studies (5.6% and 9.6%) in comparison with ours. When the cutoff value was raised to ≥ 9.0 kPa, Caballeria L et al. reported that the estimated prevalence of elevated liver stiffness was 3.6% [26], which is similar to our result of 3.97%. They have also suggested that the best cutoff of liver stiffness for significant liver fibrosis (F2–F4) should be 9.2 kPa, but that study was conducted in southwest Europe and did not include Asian participants. Recently, Serra-Burriel M et al. published their study result with much larger cohorts from Europe and Asia included [22]. They suggested using a TE cutoff of 9.1 kPa was more accurate to diagnose significant fibrosis (≥ F2) in the European and Asian populations. Therefore, we endorsed this value in our study.

Many studies have been devoted to address the association between BMI and the risk of elevated LS evaluated by TE in general population. Conti F, et al. explored factors influencing LSM in healthy population [27], which was based on the Bagnacavallo study. The study was a cross-sectional, community-based investigation among the dwellings of Bagnacavallo, Italy. Due to limited population in their study, participants with low and high BMI were both underrepresented, and unfortunately no significant correlation was observed. The Rotterdam Study conducted on another European cohort proved that there was no association between BMI and increase of LSM [24]. However, a study conducted in India argued that healthy subjects, either lean or obese, present with higher LSM values than those with normal BMI [28], which was in accordance with our finding of a J-shaped relationship between BMI and risk of elevated LS.

Even though debates about the influence of obesity on LSM have never stopped, studies indicated that obesity could act as a single or additive risk factor of elevated TE reading [29]. Especially in people with hepatitis B and C, life style modification with a purpose to reduce BMI should be encouraged to promote fibrosis improvement [4, 30, 31]. Unlike patients with hepatitis, in which LSM is widely used to indicate early stage of liver cirrhosis, the LSM of relatively healthy population has been overlooked. Moreover, the effect of underweight on LSM values in a healthy liver is yet unsolved. Large numbers of researches focusing on the effect factors of LSM were conducted in patients with liver diseases. The degree of hepatic fibrosis, central venous pressure, extrahepatic cholestasis and inflammation have been reported positively correlated with LSM values [32‐35]. Individuals with the above conditions tended to have a lower BMI or be leanness as a result of chronic liver diseases. However, whether underweight would affect the LSM via the above mechanisms in the healthy population is remained to be explored. Researchers speculated that this may be due to different viscoelastic properties in the normal liver, where cellular components, cytoskeleton as well as Glisson’s capsule contributed more than collagen tissue to the exact LS value [28]. Further exploration into the biophysical properties of the liver would help to unravel the determinants of tissue elasticity at physiologic state in the future.

Here, we reported three major strengths in this study. First of all, although this was an observational study, which could be susceptible to various confounders, we managed to adopt hierarchical adjustments in the process of statistics to minimize residual confoundings. To do this, we not only selected the variables on the basis of their associations with the BMI and/or elevated LS, but also the ones changed in effect estimates of more than 10%. We found that compared with the crude regression analyses, the relationship remained unchanged after demographic and clinical variables were further adjusted. Second, we used both generalized linear model and generalized additive model to evaluate the linear and nonlinear relationship between BMI and the risk of elevated LS, respectively. GAM has been regarded as a persuasive way in the assessment of non-linear relations, As GAM can deal with non-parametric smoothing and fit regression splines to the data, the use of GAM could help us to discover the true relationships between exposures and outcomes. Finally, we evaluated the robustness of the findings by treating BMI both as a categorical and a continuous variable.

Some limitations should not be overpassed. First, liver biopsy is still widely accepted as the golden standard to diagnose liver fibrosis and liver cirrhosis, which could not be fully replaced by TE. However, studies have shown that elevated LS measured by TE could probably serve as an early predictor even before fibrosis was detected by liver biopsy [4‐7]. With its rapid and non-invasive procedure, liver stiffness measurement has now been regarded as more cost-effective, applicable and potentially cost saving method to identify high-risk groups in the common residents. Second, an analytical cross-sectional study has its own limitations. Because it only provided weak evidence in analyzing the relationship between exposure and outcome and thus made it hard to differentiate between the cause and effect. Third, limited by raw data, we cannot elucidate the influence of waist-to-hip ratio and body weight changes on liver stiffness. There could still be some residual yet relevant confoundings needs to be considered. Finally, as all participants in this study were Chinese, it is not appropriate to apply the results directly to other ethnicities.

BMI was independently related to the risk of elevated LS. The current study reported a novel J-shaped relationship between BMI and the incident of elevated LS measured by TE. A higher or lower BMI was associated with an increased risk of elevated LS, which is further associated with impaired liver function.