Study population
GLCC is an ongoing prospective cohort study conducted at Sun Yat-sen University Cancer Center (SYSUCC), the detailed protocols have been described previously23, 24. The present analyses included adult HCC patients enrolled between September 2013 and February 2017 if they met all the following criteria: 1) diagnosed within 1 month, 2) had no history of other carcinoma or any anticancer therapy, 3) had donated available fasting blood samples. HCC was diagnosed by either imaging or histopathology based on National Comprehensive Cancer Network Clinical Practice Guidelines in Oncology: Hepatobiliary Cancers25.
Serum choline and betaine measurement
Isolated fasting serum samples were assessed using high-performance liquid chromatography with online electro-spray ionization tandem mass spectrometry to measure the concentrations of choline, and betaine. Detailed protocol of choline measurement has been described previously19. Detection of serum betaine was similar to choline, except for the internal standard was d9-betaine (obtained from Sigma-Aldrich, St. Louis, USA). The coefficients of variation for the between-run assays were 4.91% and 6.21% for choline and betaine, respectively.
Survival data collection
The date of recruitment was used as the start date of follow-up for GLCC patients, and the last outcome follow-up of the present study was conducted on September 26, 2017. We assessed liver cancer-specific survival (LCSS) and overall survival (OS), and deaths from liver cancer or all cause were assigned as outcome event of LCSS or OS, respectively. The ascertainment process of deaths has been described previously23, 24.
Other covariates
A structure questionnaire was used to collect sociodemographic and lifestyle data face-to-face by well-trained research staff. Current smokers or current alcohol drinkers were patients who smoked more than one cigarette per day or drank alcohol more than once per week continuously during past six months, and patients who had abstinence of the habits were former smokers or former drinkers. Body mass index (BMI) was calculated based on height (m) and weight (kg) at diagnosis. In addition, education level, age at diagnosis, sex, residence and family history of primary liver cancer were also collected.
We retrospectively extracted examination, diagnostic and treatment data from the SYSUCC electronic medical records. Laboratory data at diagnosis included albumin (ALB), total bilirubin (TBIL), alkaline phosphatase (ALP), γ-glutaryl-transferase (GGT), aspartate aminotransferase (AST), alanine aminotransferase (ALT), prothrombin time (PT), alpha fetoprotein (AFP), C-reactive protein (CRP), hepatitis B surface antigen (HBsAg) and hepatitis C virus antibody (Anti-HCV). A liver damage level was computed based on six liver function tests (ALB, TBIL, ALP, GGT, AST and ALT), cutoffs were described previously23, the level was presented as 1 (no liver injury), 2 (possible minor injury), and ≥3 (possible injury) in this study. Characteristics of HCC tumor and the non-tumoral liver were reviewed through imaging or pathology data at diagnosis, then Child-Pugh class26 and Barcelona Clinic Liver Cancer (BCLC) stage26 were determined. Treatments mainly included liver resection, intervention, radiofrequency ablation and others. In addition, presence of chronic diseases which can affect the prognosis of HCC were also collected, including hypertension, diabetes mellitus, fatty liver and cirrhosis.
Statistical analysis
The statistical analyses were computed in SPSS version 20.0 for WINDOWS (SPSS Inc, Chicago, IL, USA). All P-values were based on two-sided tests, P <0.05 was considered statistically significant.
The serum levels of choline and betaine at diagnosis were mostly rescaled to sex-specific tertiles (T) for analysis, except for the stratified analysis. Differences in basic characteristics across sex-specific tertiles of serum choline and betaine were compared by one-way analysis of variance (ANOVA), Kruskal-Wallis tests and Pearson’s Chi-Squared tests, where appropriate. Multivariate Cox proportional hazards models were used to examine the relationship between serum levels of choline and betaine and HCC survival outcomes. Hazard ratios (HRs) and 95% confidence interval (CI) were calculated with the first tertiles as the reference. Covariates were included into the models if the P-values of the nonparametric log-rank tests were <0.05. Finally, in model 1, nonclinical covariates such as age at diagnosis, sex, education level, residence, BMI and smoking status were selected to adjust. In model 2, significant clinical covariates were further adjusted, including AFP level (≥400 or <400 ng/L), CRP level (≥3.0 or <3 mg/L), baseline liver damage level, BCLC stage and treatments. Although the P-values of the nonparametric log-rank tests for the presence of fatty liver, presence of cirrhosis and Child-Pugh class were <0.05, they were not included in the multivariate analyses due to the multicollinearity with the baseline liver damage level or BCLC stage. Linear trends (P-trend) were evaluated by entering sex-specific tertiles of serum choline and betaine as continuous variables in the corresponding models.
Stratified analyses were conducted to evaluate whether the relationships between serum choline/betaine and HCC survival would be modified by potential prognosis factors for HCC, including sex, age at diagnosis, BMI, smoking status, alcohol drinking status, serum folate level, AFP level, CRP level, baseline liver damage level, presence of fatty liver, presence of cirrhosis, BCLC stage and treatments. Interactions were estimated by including the multiplicative interaction terms in the multivariate models.