A prospective follow-up study of the relationship between high-sensitivity C-reactive protein and primary liver cancer

The data was obtained from the Kailuan Study, which is a prospective cohort study based on the population of the Kailuan community in Tangshan (Hebei Province), 150 km southeast of Beijing. Kailuan Group, a coal industry company, has branched out into a range of other fields including healthcare, education, manufacture, etc. This study was designed to investigate the risk factors associated with chronic diseases such as cancers, arteriosclerosis, hypertension, and diabetes mellitus. From July 2006 to October 2007, all 155,418 employees (including retirees) aged 18 to 98 years from Kailuan Corporation were invited to participate the physical examinations (the baseline examination) at Kailuan General Hospital and its 10 affiliated hospitals. A total of 101,510 participants (65.3%) aged 18–98 years agreed and were enrolled after written informed consent was obtained. All participants were then followed up biennially to collect information of potential risk factors and newly diagnosed PLC cases [19].

In this study, 543 participants with a diagnosis of PLC, 1830 participants without hs-CRP measurement, and 3378 participants without measurements for traditional risk factors for PLC at the baseline examination were excluded. Traditional risk factors include age (N = 735), gender (N = 443), waist circumference (WC, in cm, N = 258), body mass index (BMI, in Kg/m², N = 312), total cholesterol (TC, in mmol/L, N = 133), triglyceride (TG, in mmol/L, N = 290), fasting blood glucose (FBG, in mmol/L, N = 90), alanine aminotransferase (ALT, in u/L, N = 107), high-density lipids (HDL, in mmol/L, N = 88), low-density lipids (LDL, in mmol/L, N = 52), HBV infection (N = 242), systolic blood pressure or diastolic blood pressure (SBP or DBP, N = 44), cirrhosis (N = 193), nonalcoholic fatty liver disease (NAFLD, N = 83), family history of cancer (N = 59), smoking (N = 113) and drinking status (N = 87) and physical activity (N = 49). The remainder of 95,759 participants (76,540 males and 19,219 females) participated in this study. Those who were excluded in this study were older (54.13 ± 11.30 years versus 53.12 ± 9.12 years, P < 0.001), and had higher levels of SBP (133.75 ± 19.15 mmHg versus 132.12 ± 20.12 mmHg, P < 0.001), BMI (26.19 ± 4.07 Kg/m² versus 25.63 ± 3.23 Kg/m², P < 0.001), WC (87.92 ± 11.38 cm versus 86.33 ± 11.40 cm, P < 0.001), and exhibited higher prevalence of cirrhosis (8(0.14) versus 97(0.10), P < 0.001) and HBV infection + (167(2.90) versus 2616(2.73), P < 0.001). The details of the participants’ screening were shown in Fig. 1. Based on guidelines from Disease Control and Prevention and the American Heart Association, participants were divided into three groups according to their hs-CRP concentration: < 1 mg/L, 1-3 mg/L, and > 3 mg/L [20]. This study complied with the Declaration of Helsinki and was approved by the Ethics Committee from both Kailuan General Hospital, Aerospace Center Hospital, and Beijing’s Children Hospital. Informed consent was obtained from all individual participants included in the study.

Participants had in person interviews with qualified physicians and nurses to complete a standardized questionnaire to obtain their baseline information. Information obtained from the questionnaire at baseline included age, gender, family history of cancer, smoking status, drinking status, and physical activity level. Drinking was classified as having consumed at least 100 ml/day of alcohol for longer than 6 months. Smoking was classified as having at least 1 cigarette/day for longer than 6 months. Evaluation of physical activity was carried out using responses regarding the frequency of physical activity (≥3 times/week, ≥30 min/time).

During baseline and all subsequent biennial interviews, height, weight, WC and blood pressure (BP) measurements for the participants were taken. A tape rule was used to measure height and was rounded to the nearest 0.1 cm. Weight was rounded to the nearest 0.1 kg. A calculation of weight/square of height was carried out to obtain body mass index (BMI). WC was measured at the narrowest point between the lowest rib and pelvis during expiration and was rounded to the nearest 0.1 cm. Blood pressure was taken when participants were seated and with a 5 min interval in between. The average of the two readings, both taken from the left arm, was used for analysis. Hypertension was classified as having a history of hypertension, systolic blood pressure ≥ 140 mmHg, diastolic blood pressure ≥ 90 mmHg, or the use of antihypertensive medications.

Vacuum tubes with EDTA were used to obtain overnight (≥8 h) fasting venous blood samples. Separation of the plasma was carried out, and plasma was kept at − 80 °C for further analysis. Hs-CRP concentrations were measured using a commercial, high-sensitivity particle-enhanced immunonephelometry assay (Cias Latex CRP-H, Kanto Chemical Co. Inc., Tokyo, Japan). The lower limit of detection was 0.1 mg/L. TC and TG concentration was measured using the colorimetric enzymatic method (Mind Bioengineering Co Ltd., Shanghai, China). The upper limit of detection was 20.68 and 11.30 mmol/L, respectively. LDL-C and HDL-C concentration were measured using the direct test method (Mind Bioengineering Co. Ltd., Shanghai, China), and there was a respective detectable upper limit of 12.9 and 3.88 mmol/L. The hexokinase/glucose-6-phosphate dehydrogenase method and the enzymatic method was used to measure FBG and ALT respectively. The inter-assay coefficient of variation for each measurement was lower than 10%. All plasma samples were analyzed using an auto-analyzer (Hitachi 747; Hitachi, Tokyo. Japan) at the central laboratory in Kailuan General Hospital. Diabetes mellitus was defined as a fasting blood glucose level ≥ 7.0 mmol/L, taking oral hypoglycaemic agents or insulin, or a validated physician diagnosis.

All PLC cases were identified via the subsequent biennial follow-up, either through clinical examinations or in person interviews, and were classified according to ICD-10 numbers. Additional information from Kailuan General Hospital and the 10 affliated hospitals’ medical records were taken into consideration to prevent missed diagnosis. In cases where outcome information of participants were unavailable, death certificates from the Provincial Vital Statistics Offices (PVSO) were also taken into consideration. Review of medical records and consultation of post-surgical pathology reports confirmed PLC diagnosis.

A commercially available software program (SAS software, version 9.4) was used to complete the statistical analysis in this study. Calculation of person-year was based on the date of their first examination until the first occurrence of the following circumstances: date of PLC diagnosis, death, or end of follow up (31 December 2018). Mean ± standard deviation was used to describe normally distributed variables and one-way analysis of variance (ANOVA) was used to make comparisons among subgroups. Median (interquartile range) was used to describe the skewed distribution (TG and ALT) and compared using nonparametric tests. Categorical variables were represented by percentage and the χ² test was used for comparison among groups. Kaplan-miere was used to calculate the cumulative incidence of PLC by hs-CRP, and the differences of cumulative incidence were tested by log-rank test [19]. The associated risk of hs-CRP for new-onset PLC was estimated using Cox proportional hazard models, and hazard ratios (HRs) and 95% confidence intervals (95% CI) were calculated. Adjustments for confounding variables were made when fitting three models. Model 1 was a univariate analysis, model 2 was adjusted for age and sex, and model 3 was further adjusted for BMI, ALT, cirrhosis (yes/no), HBV infection (positive/negative), NASH/NAFLD (yes/no), alcoholic liver disease (yes/no), diabetes mellitus (yes/no), family history of cancer (yes/no), smoking status (yes/no), drinking status (yes/no), and physical activity (yes/no). We performed tests for linear trends by entering the median value of each hs-CRP category in the models. Acknowledging that epidemiologic data is limited because competing events (death) can preclude PLC from occurring, in the presence of competing risks, the traditional multivariate COX regression model has the potential to overestimate the absolute risk. For extrapolation purposes, prognostic models that estimate individual risk need to be as precise as possible. Because all-cause death (except PLC related death) may preclude the occurrence of PLC, the existence of competing events might lead to inaccurate estimation of the risk in traditional multivariate COX regression. A more realistic method of estimating the relationship between hs-CRP and new-onset PLC is the application of competing risk models. Application of the cause-specific hazard model (CS model) is suitable when studying the etiology of disease, whereas the sub-distribution hazard function model (SD model) are more suitable when predicting the outcome risk of an individual, indicating the fundamental difference between these two models [21, 22]. Thus, competing risk regression models (CS and SD model) were used to calculate the absolute risk of PLC. To reconfirm the association of hs-CRP with the risk of PLC, subgroup analyses were carried out by stratifying participants according to sex, age (in years) (youth: ≤45, middle age: 45–59 and old age: ≥60) or several important risk factors, including HBV infection (positive/negative) and cirrhosis (yes/no). The dose-response association of hs-CRP with PLC was calculated by restricted cubic spline regression (RCS). Statistical tests were 2-sided, and P less than 0.05 was deemed as statistically significant.

The mean age at baseline was 51.69 ± 12.48 years for the study population, of which 76,540 (79.9%) were males. Participants in the hs-CRP < 1 mg/L group had a greater prevalence of HBV (+) and higher alcohol and tobacco consumption. Participants in the hs-CRP of 1-3 mg/L group had the highest DBP, BMI, TC, TG, ALT, LDL-C levels, the highest prevalence of a family history of cancer and physical activity. Participants in hs-CRP > 3 mg/L group were older, with the highest SBP, WC, FBG, and HDL-C and with a higher prevalence of hypertension, diabetes, cirrhosis, and NAFLD. Baseline characteristics of participants stratified by hs-CRP levels (< 1 mg/L, 1-3 mg/L, and > 3 mg/L) are showed in Table 1.

The median follow-up time was 11.07 years per participant. At the end of the study, 357 new-onset PLC cases were diagnosed among 95,759 participants. The crude incidence per 10,000 person-years of PLC was 3.45 in all participants (4.03 in male participants, 1.11 female participants). Age- and sex- standardized incidence of PLC increased from 2.73 to 3.14 and 4.07 per 10,000 person-years in the hs-CRP < 1 mg/L, hs-CRP of 1-3 mg/L and CRP > 3 mg/L group, respectively. Figure 2 is a smooth model-based graph and describes the details of the cumulative incidence of PLC stratified by hs-CRP. The log-rank test indicated that the cumulative incidence of PLC among different hs-CRP subgroups had a significant statistical difference where the incidence increased as hs-CRP levels increased.

Commonly used categories for hs-CRP were analyzed in univariate and multivariate proportional hazards analyses to test for associations of risk for new PLC events (Table 2). Compared with CRP < 1 mg/L group, no significant associations were observed for hs-CRP of 1–3 mg/L on PLC risk in the univariate analysis (HR, 1.08; 95%CI, 0.84 to 1.40) and multivariate analysis (HR, 1.07; 95%CI, 0.82 to 1.38). Statistically significant increased crude and multivariable HRs were observed in people with CRP > 3.0 mg/L for PLC with corresponding HRs (95% CI) of 1.38(1.06 ~ 1.81) and 1.51(1.15 ~ 1.98), respectively. A positive linear association between hs-CRP levels and the risk of PLC was observed in the RCS model showed (P-overall = 0.0037, P-nonlinear = 0.0550; Fig. 3).

The results of the competing risk regression models were displayed in Table 2. There was an identification of 9232 death cases before the occurrent of PLC at the end of the follow-up period. Compared with the hs-CRP<1 mg/L group, the multivariable HRs (95%CI) for the association of hs-CRP of 1–3 mg/L group and hs-CRP>3 mg/L with PLC were 1.06(0.81 ~ 1.40), 1.50(1.14 ~ 1.99) in the multivariate-adjusted analysis in the CS model. Similar results were also observed in the SD model with minor differences.

The effects of hs-CRP on new-onset PLC cases after stratifying the participants by sex, age, HBV infection, and cirrhosis were shown in Table 3. The effect of hs-CRP was not modified by sex both in the univariate and multivariate analyses. Similar results were also observed when participants were stratified by HBV infection or cirrhosis. However, when stratified by age, HR values increased only in the middle-aged and elderly groups.