Plasma Adiponectin Levels in Relation to Chronic Hepatitis B Infection Progression to Liver Cancer Milestones: A Prospective Study Open Access

Hepatitis B virus (HBV) infection poses a serious public health threat in the world, with an estimated 316 million people living with chronic hepatitis B infection [1]. They are at much higher risk of progressing to cirrhosis, hepatic decompensation, hepatocellular carcinoma, and even liver-related deaths. In 2019, hepatitis B was estimated to cause 555,000 deaths, mostly from cirrhosis and hepatocellular carcinoma [1], and without intervention, deaths are expected to peak at 1.14 million by 2035 [2]. Current treatment regimen can effectively control HBV viral replication and suppress the viral load to obscure the progression of chronic HBV infection to liver cirrhosis and HCC, but cannot eradicate the virus totally [3]. This leads to slow decline in the prevalence of chronic HBV infection in Taiwan and other HBV endemic areas [2, 4]. Thus, providing treatment and surveillance to chronic HBV carriers remains the prominent option to prevent HBV-related mortality. In addition to the use of virus-related (HBV viral load, HBsAg titer, HBV core-related antigen, etc.) or liver function-related (ALT, AFP, etc.) factors, finding biomarkers that can help in risk stratification of HBV patients is an important issue in the goal of HBV elimination set by WHO [1].

Adiponectin – an adipose tissue-specific protein, which is largely determined by body fatness – may partially explain the observed associations between cancer and obesity and/or insulin resistance-related factors. Our previous nested-case-control study showed that elevated adiponectin levels at baseline were associated with an increased risk of HCC and liver cirrhosis over time with a significant dose-response trend [5]. Similar results were found in patients with chronic hepatitis C [6], HBV and HCV [7], as well as among those without hepatitis [8]. Although the temporal sequences of the adiponectin levels and the development of HCC were assured by the nested-case control study design, the interpretation of the outcomes of circulating adiponectin levels and HBV infection progression was hindered by the matching of controls. We conducted current longitudinal analysis to further investigate whether the increased cirrhosis and HCC risk associated with plasma adiponectin levels are through progression milestones of chronic HBV infection including HBeAg, HBsAg, and HBV DNA seroclearance on the REVEAL-HBV cohort.

A total of 89,293 residents aged 30–65 years living in seven townships in Taiwan were invited to participate in a follow-up study. During February 1991 to December 1992, 23,820 (11,973 males and 11,847 females) agreed to participate and provided their written informed consent. Detailed description of the recruitment procedures of the cohort has been published previously [9, 10]. The data collection procedures were reviewed and approved by the IRB of College of Public Health and College of Medicine, National Taiwan University (Approval No. 201212085RINB). Among 3,931 HBsAg(+)/anti-HCV(−) participants, 3,692 had sufficient biosamples left for adiponectin assay.

Personal interviews using a structured questionnaire were administered by well-trained public health nurses at recruitment. Information collected included sociodemographic characteristics, dietary habits, habits of cigarette smoking, alcohol drinking, medical and surgical history, and family history of HCC and liver cirrhosis. Using standard sterile techniques, a 10-mL blood sample was collected at cohort entry and follow-up examinations, fractionated on the day of collection and stored in a deep freezer (−70°C) until examination.

The metabolic factors available included obesity, history of diabetes and hypertension, and serum levels of total cholesterol and triglycerides. The presence of obesity was grouped by the measurement of body mass index (BMI, kg/m²) and waist circumference (cm). Central obesity was defined as a waist circumference >90 cm for men and >80 cm for women. BMI was categorized into normal (<23 kg/m²), overweight (23–<25 kg/m²), obese (25–<30 kg/m²), and extreme obese (≥30 kg/m²). Personal histories of diabetes mellitus and hypertension were obtained from the interview. Serum levels of triglycerides ≥150 mg/dL and cholesterol ≥240 mg/dL were considered hyperlipidemia. Plasma adiponectin levels were measured using the samples collected and stored at baseline by commercially available enzyme-linked immunosorbent assay (ELISA) kits (B-Bridge International, Inc., USA).

HBeAg and HBV DNA viral load and genotypes were tested using commercial kits (HBeAg: radioimmunoassay with the cutoff value of the net counts per minute of the negative control times the factor 2.1, Abbott Laboratories, North Chicago, IL, USA; HBV DNA: Cobas Amplicor HBV Monitor test kit by Roche Diagnostics, Indianapolis, IN, USA). HBV genotypes were assayed using a melting curve analysis [11]. Qualitative HBsAg was tested using a commercial kit with a lower limit of detection of 230 pg/mL (AUSRIA-II; Abbott Laboratories), while HBsAg titers were tested using automated chemiluminescent microparticle immunoassay (Architect HBsAg, Abbott Laboratories). Using HBeAg, ALT, and HBV DNA, we divided them into 4 groups, indicating different phases of HBV infection at baseline, including (1) HBeAg (−) and normal ALT and HBV DNA <2,000 IU/mL, (2) HBeAg (−) and ALT>1x ULN and HBV DNA >2,000 IU/mL, (3) HBeAg (+) and normal ALT and HBV DNA >200,000 IU/mL and (4) HBeAg (+) and ALT>1x ULN and HBV DNA >200,000 IU/mL. When the binary HBV DNA level was indicated in the analysis, the HBV DNA >200,000 IU/mL was chosen to indicate high viral load, which was converted from >10⁶ copies/mL. HBeAg and HBsAg seroclearance were defined as attaining seronegative in any of the follow-up tests. Newly diagnosed liver cirrhosis cases via high-resolution real-time ultrasound adapting a quantitative scoring system derived from the appearance of the liver surface, liver parenchymal texture, intrahepatic blood vessel size, and splenic size during the follow-up were also identified [12]. All cirrhosis cases were confirmed by reviewing of the sonographic records by a specialized gastroenterologist, and by computerized linkages to the National Health Insurance profiles of Taiwan until June 30, 2004 [12]. The earlier HCC cases were detected by both ultrasound and AFP testing through follow-up examinations and were confirmed by gastroenterologists based on histopathological confirmation, positive lesions detected by at least two different imaging techniques (abdominal ultrasonography, angiogram or CT), or positive lesions detected by one imaging technique combined with a serum α-fetoprotein level greater than 400 ng/mL [10]. The unique national identification numbers were used to access computerized national cancer registry profiles and death certificate profiles in Taiwan to identify newly diagnosed HCC cases and liver-related deaths among cohort participants from their dates of enrollment until December 31, 2014 for cancer and December 31, 2016 for death profiles. The liver-related causes of death included liver cancer (ICD 10: C22-C229) and chronic liver diseases and cirrhosis (ICD 10:K70-K709, K730-K746).

Person-years for each participant were calculated from the dates of enrollment to the dates of the occurrence of the designated end points including HBeAg seroclearance, HBsAg seroclearance, HBV DNA clearance, diagnosis of liver cirrhosis and HCC, death, or last censored dates, whichever came first. We defined the last dates of data linkage (HCC and death) and examination (cirrhosis) and blood tests (HBVDNA, HBsAg, and HBeAg) as the censored dates for respective survival analysis. For incident liver cirrhosis and HCC, liver-related causes of death, HBV DNA clearance, HBsAg seroclearance, and HBeAg seroclearance, they were June 30, 2004, December 31, 2014, December 31, 2016, June 18, 2015, May 11, 2015, and June 30, 2014, respectively. Established and suggested risk factors for HBV progression were considered potential confounders, including age at reference, gender, education years and residence, cigarette smoking, habitual alcohol consumption and betel nut chewing at enrollment, liver function measurement, and history of liver cirrhosis. The hazard ratios (HRs) and 95% confidence interval (CI) were estimated by Cox proportional hazards regression models. Since HBsAg and HBV DNA seroclearance were important predictors of liver cirrhosis and HCC development, as well as the liver-related death, the duration to HBsAg and HBV DNA seroclearance should be considered as confounding factors. We added HBsAg and HBV DNA seroclearance as time-varying covariates into the original multivariable adjusted Cox models. In addition, a propensity score-matched analysis on HCC development was conducted to clarify whether multivariable adjustment models were sufficient to control the potential confounding effects. The propensity score was derived from a multiple probit regression model with an adiponectin level greater than median as exposure and incorporated all adjustment variables. A nearest-neighbor matching algorithm with replacement and the caliper of 0.001 was applied, which resulted in 980 with high adiponectin and 980 with lower adiponectin levels. The Cox proportional hazards regression model stratified by the matching pairs was conducted. Adiponectin levels were categorized into quintiles and the linear trends across quintiles were assessed by testing the statistical significance of a single trend variable coded as the category of exposure (1, 2, etc.). The binary variable with the highest quintile of adiponectin as a cutoff point was used to evaluate the presence of effect modifications, i.e., a different relationship of adiponectin and progression milestones between preselected viral factors, and the statistical significance was tested using an interaction term with a p value derived from the Wald test. The Nelson-Aalen cumulative hazard method was used to estimate the cumulative incidence of respective milestones and log-rank tests were conducted to compare incidence between the combination of the highest quintile versus the remaining quintiles of plasma adiponectin levels and HBV viral load ≥200,000 versus <200,000 IU/mL. Because the highest quintile of adiponectin level and HBV viral load ≥200,000 IU/mL consistently showed strong association with all the progression milestones, we used these two binary variables for subgroup and combined analysis to assure valid estimations when a small number of events occurred in some subgroups. All analyses were conducted using Stata statistical software (Version 13.1, Stata Corp., College Station, TX, USA).

There were 333 newly diagnosed HCC and 328 died from liver-related causes, with respective 75,887 and 82,623 person-years of follow-up, yielding an incidence of 4.39 (95% CI:3.94–4.89) HCC cases and 3.97 (95% CI: 3.56–4.42) liver deaths per 1,000 person-years. Older age and females had a higher level of adiponectin, while those with higher education levels, habits of cigarettes smoking, alcohol consumption and betel nut chewing, obesity, and ultrasonographic fatty liver tended to have lower adiponectin levels (Table 1). Cross-sectionally, elevated adiponectin was associated with higher HBeAg-seropositive rate, high HBV viral load (≥2 × 10⁵ IU/mL) and high HBsAg titers (≥1,000 IU/mL) in a dose-response manner, with a twofold increased risk (OR = 2.25, 95% CI: 1.55–3.28; OR = 2.11, 95% CI: 1.47–3.04 and OR = 1.92, 95% CI: 1.47–2.52) for Q5 versus Q1, respectively (Table 2). Higher adiponectin levels were also associated with higher serum ALT levels and more advanced HBV infection status indicated by either HBeAg seropositive or abnormal ALT or higher HBV viral load.

Table 3 shows the association of adiponectin levels and HBV progression milestones longitudinally. Those with the highest quintile of adiponectin levels had a lower chance of achieving HBsAg (HR = 0.69, 95% CI: 0.49–0.97), HBeAg seroclearance (HR = 0.48, 95% CI: 0.27–0.85), and undetectable status of HBV DNA (HR = 0.63, 95% CI: 0.43–0.90) during the follow-up. Higher adiponectin levels were associated with a higher chance of eventually developing liver cirrhosis (HR = 2.88, 95% CI: 1.98–4.20), HCC (HR = 2.38, 95% CI: 1.52–3.73), and even death from liver-related causes (HR = 2.32, 95% CI: 1.51–3.54). The association of adiponectin level and all progression milestones showed significant linear trends when treating adiponectin as a continuous variable. After adding HBsAg and HBV DNA seroclearance as time-varying covariates into multiple Cox regression models with all adjustment variables, the associations of high adiponectin and cirrhosis, HCC, and liver-related death remain similar (Table 3). The HRs did not change much after the propensity score matching on HCC development, with the HR of 2.69 (95% CI: 1.11–6.49) for the highest quintile compared to the lowest, and the dose-response trend remained statistically significant (p = 0.052) (Table 3).

The association between the highest quintile versus remaining adiponectin levels and these HBV progression milestones stratified by selected HBV-related factors is shown in Figure 1. For liver cirrhosis, the association was stronger among carriers with genotype C HBV infection (HR = 2.53, 95% CI = 1.74–3.58) and ultrasonographic fatty liver (HR = 2.73, 95% CI = 1.63–4.57). The adiponectin-HCC association was higher among those who were HBeAg seropositive (HR = 2.52, 95% CI = 1.53–4.15), those with fatty liver (HR = 2.41, 95% CI = 1.46–4.03), those with elevated ALT (HR = 2.02, 95% CI = 1.07–3.81), and those infected with genotype C HBV (HR = 2.25, 95% CI = 1.54–3.29). HBV genotype significantly modified the association between adiponectin and HCC (p for interaction = 0.005) as well as liver-related deaths (p for interaction = 0.0157). Carriers with high adiponectin and infected with C genotype HBV were at more than twofold higher risk of liver-related death (HR = 2.54, 95% CI = 1.67–3.87), while those with genotype B had 60% increased risk (HR = 1.60, 95% CI = 1.03–2.47). Although the interaction is not significant (p = 0.063), those who were HBeAg positive and with high adiponectin were at threefold increased risk of dying from liver-related causes (HR = 2.99, 95% CI = 1.77–4.05) and an about 54% elevation from HBeAg negative (95% CI = 1.01–2.30). Among HBV carriers with elevated ALT, those with higher adiponectin levels had a much lower chance achieving HBV viral and HBsAg clearance. Among HBeAg positive carriers with lower HBsAg titer and viral load, higher adiponectin levels hinder their chance of clearing the HBe antigen (HR = 0.20, 95% CI: 0.06–0.66, HR = 0.06, 95% CI: 0.004–0.98, respectively). The association of all liver progression milestones remains similar when stratified by age (<50, ≥50), gender and BMI (<25, ≥25) (online suppl. Table 1; for all online suppl. material, see https://doi.org/10.1159/000539909).

We further combined adiponectin and HBV DNA viral loads to exam whether there was an additive effect of adiponectin with the most important viral risk factor known to affect HBV progression (Fig. 2). Compared to HBV carriers with low viral load and plasma adiponectin, HBV carriers with high viral load (>200,000 IU/mL) and high adiponectin levels were at much higher risk of liver cirrhosis (HR = 5.87, 95% CI: 4.25–8.10), HCC (HR = 8.61, 95% CI: 6.02–12.3), and death from liver-related causes (HR = 9.82, 95% CI: 6.82–14.1). Among HBeAg-positive carriers, those with high viral load and high adiponectin levels had a much lower chance of reaching HBeAg seroclearance (HR = 0.15, 95% CI: 0.08–0.26) when compared to those with low levels of both.

In an extension of the previous nested-case control study, we prospectively evaluated the relationship between plasma adiponectin levels and the key HBV infection progression milestones in the whole REVEAL-HBV cohort. In this comprehensive longitudinal analysis, those with high adiponectin at baseline had lower chances of achieving HBeAg, HBV DNA, and HBsAg seroclearance and thus higher chances of developing liver cirrhosis, HCC, and died from liver-related causes. In addition, the HBV genotype was found to significantly modify the association between adiponectin and HCC (p for interaction = 0.005) as well as liver-related deaths (p for interaction = 0.0157). Carriers with high adiponectin and infected with C genotype HBV were at more than twofold higher risk of HCC and liver-related death, in contrast to a 60% increase in liver-related death and no association of HCC in genotype B. Fatty liver appeared to affect those with high adiponectin by reducing the HBV DNA clearance chance and increasing cirrhosis and HCC risk, though the interaction is not statistically significant.

The association between adiponectin and liver diseases appeared conflicting between metabolic and virus originated. Adiponectin levels decrease as obesity levels increase, and was thus found to be lower in metabolic liver diseases including NAFLD and NASH [13]. But it was elevated in virus and autoimmune-related liver diseases [14‒16] and increased along with the disease severity [17]. However, adiponectin concentration was similar between patients with NASH-induced cirrhosis and cirrhosis from other causes [14, 16]. It was also found that adiponectin was positively correlated with surrogate markers of hepatic fibrosis including transient elastography, fasting serum bile acids, and hyaluronate [16]. Another cross-sectional study observed that patients with HBV-related HCC had higher adiponectin levels than healthy controls, patients with CHB, and patients with cirrhosis [15]. Consistent with our results, one hospital-based prospective study conducted in Japan reported that among patients with chronic hepatitis C, higher serum adiponectin levels were associated with a higher risk of developing HCC [6]. Interestingly, a similar association was found among a European cohort with most members without HBV/HCV infections [8]. They reported a twofold liver cancer risk among those tested with higher non-HMW adiponectin levels (IRR per doubling of concentrations = 2.09, 95% CI = 1.19–3.67) [8].

In this report, we further showed that patients with higher adiponectin at baseline had a lower chance of achieving HBeAg, HBVDNA, and HBsAg seroclearance. As a result, they had higher chances of having cirrhosis and HCC during the follow-up. This implies that elevated adiponectin may directly or indirectly abrogate the immunoclearance of HBV. We propose three potential mechanisms to explain this link, including direct virological actions, indirect effects via hepatosteatosis, or immunoregulation. First, adiponectin may directly upregulate HBV DNA replication. In HBV-infected cellular models, modulating adiponectin gene expression caused parallel changes in HBV DNA synthesis [18]. It was demonstrated that downregulation of adiponectin gene expression by RNA interference in HepG2-HBV-stable cells resulted in an 80% reduction of HBV DNA synthesis without changing pgRNA encapsidation. Furthermore, adding adiponectin to cytoplasmic core particles isolated from HBV-infected hepatocyte cell lines increased HBV DNA synthesis probably by enhancing endogenous polymerase activity. Second, adiponectin may indirectly regulate HBV replication via affecting hepatic steatosis. Recent reports that concurrent hepatosteatosis in chronic HBV-infected patients had higher rates of HBsAg seroclearance and seroconversion, and therefore a lower risk of HCC is relevant to our reports [19, 20]. It is well established that lower adiponectin is related to increased risk of hepatosteatosis [21]. Adiponectin acts on its receptor in liver through activating AMP-activated protein kinase, PPAR-α, and PGC1-α and altogether increases fatty acid β-oxidation [22‒25]. Therefore, adiponectin has direct effects on reducing liver fat accumulation. It was even demonstrated that synthetic adiponectin receptor agonist reduces obesity, insulin resistance, and ameliorated hepatic steatosis and inflammation in a high-fat-diet obesity animal model [26]. As a result, higher adiponectin leads to lower chance of hepatic steatosis and thus lower chance of HBV seroclearance and higher risk for cirrhosis and HCC.

How hepatic steatosis increases HBV seroclearance remains unclear. HBV infection has been reported to alter lipid metabolism in hepatocytes [27]. Human lipidomic studies have shown that upregulation of phosphatidylcholine and choline plasmalogen and downregulation of free fatty acids and lysophosphatidylcholine are the dominant findings in NAFLD-HBV subjects [27]. Interestingly, the levels of serum hepatotoxic lipids were significantly lowered in the NAFLD-HBV group, when compared to those of NAFLD group [27]. Hepatosteatosis was further shown to reduce viral particle secretion [28]. These data suggest that hepatosteatosis may be beneficial for HBV-infected hepatocytes. Nevertheless, we observed that the effect of high adiponectin in reducing the chance of viral clearance and increasing risk of cirrhosis and HCC appeared to be more profound in HBV patients with fatty liver than those without. It is plausible that adiponectin may directly affect at least a part of the viral lifecycle and/or liver cellular functions, suggesting that the direct and indirect effects of adiponectin are not mutually exclusive.

On the other hand, altered lipid metabolism was associated with immunosuppression in HCC tumor samples [29]. In an NAFLD-HBV transgenic model, fatty acid-induced immune responses via Toll-like receptor 4 may be responsible for reduction in serum levels of HBsAg, HBeAg, and HBV DNA [30]. Furthermore, direct inhibition of the lipid biosynthetic pathway using inhibitors of acetyl-CoA carboxylase and fatty acid synthase reduced extracellular HBsAg in a cellular model. However, a liver-targeted acetyl-CoA carboxylase inhibitor failed to produce antiviral activity in HBV-infected liver chimeric mice [31]. These data imply that the existence of extrahepatic mechanisms, probably in the immune system, may help explain the link between adiponectin and HBV. Adiponectin has been shown to regulate various immune cells in innate and adaptive immunity [32‒34]. Therefore, the third potential mechanism is that adiponectin may modulate HBV viral clearance by way of its immunoregulatory roles. The exact immunoregulatory mechanisms that underlie how higher adiponectin can be linked to lower HBV viral clearance also await further investigation. Whether it is a direct effect on viral replication or a secondary effect mediated by hepatosteatosis or immunity is an intriguing question. Exploring these issues will absolutely provide in-depth insights about the connection between hepatocyte energy metabolism and HBV lifecycle. It should be reminded again that these competing hypotheses, however, are not mutually exclusive.

Previous cross-sectional study showed that patients with HBV genotype C had a higher serum adiponectin level than genotype B [15]. The association of high adiponectin level and HCC was more profound in genotype C in our previous nested-case-control study [5]. In current analysis, although those with genotype C had higher adiponectin levels, there was no association of adiponectin and HBV genotype after the adjustment of other covariates. Nevertheless, we observed that high adiponectin levels are associated with all important progression milestones more profoundly in genotype C. The mechanisms behind these are not clear at present. As addressed above, adiponectin may directly enhance HBV DNA synthesis, probably by increasing endogenous polymerase activity [18]. We speculated that increasing the rate of DNA synthesis would presumably accelerate the evolution of HBV into genotype C, since genotype C was known to be with more genetic variability. Furthermore, adiponectin may increase the DNA synthesis of preexisting genotype C HBV, thus exacerbating clinical outcomes.

The main strength of our current analysis is that we were able to capture all the important HBV progression milestones to the end-stage liver disease and establish the correct temporal relationship. With this, we can further clarify whether adiponectin plays its role only on end-stage liver disease or starts early in the viral progression leading to cirrhosis and HCC. We were also able to evaluate the adiponectin-associated HBV progression risks stratified by potential effect modifiers and indeed found that HBV genotype plays some role in modifying this risk. Therapeutic intervention with nucleotide analogues on HBV can interrupt the progression of HBV infection to end-stage liver diseases including HCC. Most of the REVEAL-HBV participants were confirmed to be treatment-free, as the Taiwan National Health Insurance (NHI) system only began offering antiviral treatment coverage for chronic hepatitis B patients meeting treatment guidelines in October 2003. Furthermore, according to internal surveys from REVEAL-HBV, even after the NHI started coverage, only a very small number of patients met the treatment guidelines and underwent antiviral therapy. Nevertheless, since those with high adiponectin level also had high viral load, thus more likely to receive antiviral treatment, their HCC risk should be reduced. With more under anti-HBV treatment, the adiponectin-HCC association would be more likely to be underestimated, while we still see association of high adiponectin and HCC. With the trade-off of sample size, the adiponectin-progression milestone associations remained after the adjustment of HBsAg and HBV DNA seroclearance, as well as after the propensity score matching, indicating that the multivariable adjustment models were sufficient to control the confounding in this analysis. Further sensitive analyses including using inverse-propensity score weighting method instead of matching, using median and quintiles adiponectin level as cutoffs for subgroup analysis, all showed similar results. Since the current analysis used adiponectin levels measured for the nested-case control study and the remaining cohort participants at different times, we added the instrumental variable of testing batch to account for the potential confounding in all regression models. In addition, all regression models were conducted with the exclusion of the nested-case control study data, and all the results remain the same. Presenting the adiponectin with quintile distribution, rather than absolute levels, can help in alleviating potential errors in risk estimates. One limitation is that we only have baseline measurements of adiponectin levels, and the fluctuation of adiponectin levels is a possibility and may obscure the results. Nevertheless, since the misclassification was random, it would skew the results toward the null. Another limitation is that we only have total cholesterol levels measured, without assays on HDL and LDL.

In conclusion, we demonstrated that elevated adiponectin levels could predict the severity of HBV-related liver disease progression, thus leading to end-stage liver diseases. The exact mechanism of how adiponectin mediates HBV infection toward carcinogenesis remains unclear and was mostly based on speculations from experimental and observational studies. Nevertheless, this is still a comprehensive observational study with valid methodology and further confirmation from definite laboratory data to validate our observation is warranted. Disentangling this may help us in finding new HBV treatment targets, biomarkers in HBV surveillance to identify high risk patients, or even cancer prevention.

This study protocol was reviewed and approved by the National Taiwan University Hospital Institutional Review Board (Approval No. 201212085RINB). The National Taiwan University Hospital Institutional Review Board waived the requirement for informed consent (Approval No. 201212085RINB).

All authors of this manuscript had no conflicts of interest to declare.

This study was supported by Grants NSC-94-2314-B002-268 from the National Science Council, Taipei, Taiwan.

C.-L.C., W.-S.Y., and C.-J.C. contributed to the design and execution of this work. C.-L.C., W.-S.Y., H.-I.Y., J.-H.K., and P.-J.C. contributed to the interpretation of the results and the preparation of this manuscript. H.-I.Y., C.-F.C., L.-Y.W., S.-N.L., and C.-J.C. contributed to the recruitment of study participants and data collections. The final draft was approved by all authors.

The data that support the findings of this study are not publicly available due to their containing information that could compromise the privacy of research participants, but are available from the corresponding author (C.-L.C).