Prospective cohort studies of association between family history of liver cancer and risk of liver cancer

Abstract

Uncertainty remains on the relationship between a family history of liver cancer and liver cancer risk in prospective cohort studies in a general population. Thus, we examined this association in 133,014 participants in the Shanghai Women’s and Men’s Health Studies. Family history of liver cancer was categorized through dichotomous and proportional score approaches. Hazard ratios (HRs) and 95% confidence intervals (CIs) were derived using the Cox proportional hazards models with adjustment for potential confounders. A meta-analysis of observational studies through December 2013 on liver cancer risk in relation to family history of liver cancer was also performed. Study-specific risk estimates were combined using fixed or random effects models depending on whether significant heterogeneity was detected. For the Shanghai Women’s and Men’s Health Studies, 299 liver cancer cases were identified during follow-up through 2010. Family history of liver cancer was associated with liver cancer risk using both binary indicator (HR=2.60, 95%CI: 1.77–3.80) and proportional score (high-risk vs. minimal-risk category: HR=3.03, 95%CI: 1.73–5.31), with increasing HRs for increasing score categories. The Meta-analysis also showed an increased risk for those with a family history of liver cancer (RR=2.55, 95% CI: 2.05–3.16). Family history of liver cancer was related to increased risk of liver cancer in Chinese population. This risk is particularly high for those with an affected mother. The “dose-response” of risk with an increasing family history score of liver cancer might further facilitate future cancer prevention programs on identifying individuals with the highest potential liver cancer risk.

Background

Liver cancer is the sixth most common cancer and the third most common cause of cancer-related deaths worldwide¹. An estimated 748,300 new liver cancer cases and 695,900 liver cancer deaths occurred worldwide in 2008, about half of which were estimated to occur in China¹. Although the age-standardized incidence rates in urban Shanghai showed decreased trends among both men and women from 1976 to 2005, the incidence rates were still relatively high compared to most Western countries².

Well-established liver cancer risk factors include hepatitis B virus (HBV) and hepatitis C virus (HCV), cirrhosis, aflatoxin exposure, heavy alcohol drinking, tobacco smoking and some rare monogenic syndromes such as hereditary hemochromatosis and α-1 antitrypsin deficiency³. Among these factors, HBV and HCV infections play the most important roles in liver carcinogenesis³.

Liver cancer has been reported to cluster within families, which may be caused by genetic and/or shared environmental risk factors, and especially related to a high prevalence of chronic HBV infection within families^{4, 5}. A number of studies suggested that heritable factors likely contribute to the risk of liver cancer, possibly with effect modification by environmental factors^6–9. A recent hospital-based case-control study of mostly Caucasian individuals reported a link between family history of liver cancer and risk of hepatocellular carcinoma (HCC) in those without HBV or HCV infection, possibly suggesting an independent genetic effect¹⁰.

A previous meta-analysis including nine case-control and four cohort studies showed that individuals with a family history of liver cancer doubled the liver cancer risk in comparison to those without a family history¹¹. For previous case-control studies, an increased risk of liver cancer associated with a family history of liver cancer was extensively reported in both the HBV-related and the general population^{5, 10–14}. However, for previous prospective cohort studies, consistent results could be suggested only in the HBV-related population^{5, 12}, whereas the association in the general population was not clear^15–17. As a case-control study may be more likely to introduce selection and recall bias compared to a cohort design, thus, we used data from two prospective cohort studies in Shanghai to evaluate the association between a family history of liver cancer and liver cancer risk in general Chinese population. To classify the risk categories of the family history of liver cancer, we used both a standard approach in which the exposure was dichotomous (positive when at least one relative was affected) and a proportional family history score (FH score)¹⁸ taking into consideration certain family characteristics. In addition, we conducted an updated meta-analysis to quantitatively combine the results from all published case-control and cohort studies, including results from our study.

Methods

1. The Shanghai Women’s and Men’s Health Studies

1.1 Study population

Participants included in this study were enrolled in either the Shanghai Women’s Health Study (SWHS) or the Shanghai Men’s Health Study (SMHS). Details on these two ongoing population-based cohorts in urban Shanghai have been described elsewhere^{19, 20}. Briefly, a total of 74,941 women aged 40 to 70 years (response rate = 92.6%) at recruitment were enrolled in the SWHS between 1997 and 2000. A total of 61,482 eligible men (response rate = 74.0%) aged 40 to 74 years were enrolled in the SMHS from 2002 to 2006.

We excluded participants from this analysis because of (a) a diagnosis of cancer before enrollment (1,598 women and 0 men), (b) loss to follow-up shortly after enrollment (5 women and 14 men), (c) diagnosis of cancer in situ during follow-up (69 women and 5 men), (d) cancer diagnosis that could not be confirmed (15 women and 5 men), (e) death from cancer with no cancer type or diagnosis date (114 women and 126 men), (f) extreme values for total energy intake (outside the range of 500 – 4000 kcal/d) (50 women and 91 men), (g) missing data for any of the covariates of interest (78 women and 218 men). We further excluded 357 women and 664 men with less than 2 years of total observation from the study. After exclusion, a total of 133,014 participants (72,655 women and 60,359 men) were included in the current analysis.

1.2 Follow-up and data collection

Cohort members were followed for cancer occurrence through in-person follow-up surveys every 2–3 years and annual record linkage with databases of the population-based Shanghai Cancer Registry, Shanghai Vital Statistics Registry, and Shanghai Resident Registry. For the SWHS, the response rates for the first (2000–2002), second (2002–2004), third (2004–2007), and fourth (2007–2011) in-person follow-up surveys were 99.7%, 98.7%, 94.9%, and 92.3%, respectively. For the SMHS, the response rates for the first (2004–2008) and second (2008–2011) follow-up surveys were 97.6% and 93.7%, respectively.

All possible cancer diagnoses were verified through home visits and review of medical charts by a panel of clinical and pathological experts. Cancers were coded according to the International Classification of Disease, Ninth Revision (ICD-9). Liver cancer was defined as primary tumor with an ICD-9 code of 155.

Information on family cancer history in first-degree relatives was collected at baseline survey using structured questionnaire. Each subject was asked whether their first-degree family members, including parents, siblings, and children, had ever been diagnosed with cancer. For those who reported a positive family cancer history, further information on the affected family members, types of cancer, and ages at diagnosis was collected. In addition, we also obtained information about demographic characteristics, anthropometric measurements, lifestyle, dietary habits, medical history, occupational history and physical activity habits in baseline survey.

1.3 Statistical analysis

Person-years of follow-up time were calculated for each participant from 2 years after the date of enrollment to the date of cancer diagnosis, death, date of loss to follow-up (if applicable), or December 31, 2010, whichever came first. For the baseline characteristics between the liver cancer cases and non-cases in women and men, continuous variables were compared with use of t test and categorical variables were compared with use of chi-square test. The Cox proportional hazards models were used to estimate hazard ratios (HRs) and its corresponding 95% confidence intervals (CIs), with age as the time scale.

Covariates were selected based on their potential to confound or modify the association between a family history of cancer and liver cancer. All the selected covariates were modeled using data from the baseline survey. The covariates included in the final multivariate-adjusted models were: age (years, continuous variable); sex (male or female); education level (four categories: elementary school or less, middle school, high school, and college or above); family income level [four categories: low [<10,000 Yuan per family per year in the SWHS and <500 Yuan per person per month in the SMHS], low to middle [10,000 to 19,999 Yuan per family per year in the SWHS and 500 to 999 Yuan per person per month in the SMHS], middle to high [20,000 to 29,999 Yuan per family per year in the SWHS and 1,000 to 1,999 Yuan per person per month in the SMHS]), and high [≥30,000 Yuan per family per year in the SWHS and ≥2,000 Yuan per person per month in the SMHS]; body mass index (BMI, kg/m², continuous variable); vegetable and fruit intake (g/day, continuous variable); self-reported history of chronic hepatitis (yes or no); cirrhosis or other chronic liver disease (yes or no); diabetes (yes or no); cholelithiasis or cholecystectomy (yes or no). Smoking and alcohol consumption were not associated with liver cancer risk in our study populations. Further sensitivity analysis with adjustment for smoking and alcohol consumption did not indicate a material alteration on our findings; therefore, we did not adjust for them in the final model.

Firstly, we used a standard dichotomous method to assess the liver cancer risk associated with a family history of liver cancer. Subjects were classified into two categories: (1) no first-degree relative with liver cancer; (2) one or more first-degree relatives with liver cancer. Participants without family history of liver cancer were treated as reference group. When further considering the effects of family size and number of affected first-degree relatives, family risk of liver cancer for each study participant was further classified according to the proportional FH score approach (FH score=No. of affected first-degree relatives / total No. of first-degree relatives excluding the index one), as proposed by A van Esch et al.¹⁸. For example, the proportion (FH score) would be 1/7 if one had five unaffected siblings and two parents, one of whom was affected. In present study, the ranges of proportional FH score were from 0 to 2/3 and 0 to 3/4 for men and women, respectively. The minimal-risk population represented cohort members without a family history of liver cancer. The median FH score (FH score _median=1/7) in those with a family history of liver cancer was treated as the cutoff point to distinguish intermediate-risk population and high-risk population. Thus, subjects were then divided into three categories by using the proportion approach: (1) the minimal-risk group (FH score=0); (2) the intermediate-risk group (0< FH score ≤1/7); (3) the high-risk group (FH score >1/7). The minimal-risk group was taken as reference group in the analyses. Testing for linear trend was done by assigning an ordinal value (1, 2 and 3) to each range of FH score and treating it as a continuous variable in the regression models.

Besides, we also evaluated liver cancer risk with a family history of some common cancers using a standard dichotomous method including a family history of any cancer and cancers of esophagus, stomach, colorectum, pancreas and lung. A family history of other cancer sites, including cancers of gallbladder, thyroid, larynx, kidney, urinary bladder, and leukemia, as well as prostate cancer among men and cancers of the breast, cervix uteri, corpus uteri, and ovary among women, was not evaluated due to small numbers (n≤2) of liver cancer cases with positive family history.

We also tested for potential interactions between family history of liver cancer and the covariates adjusted in the model, but no interaction was found based on the first-degree multiplicative model. Analyses of our cohort data were conducted using SAS 9.2 (SAS Institute, Inc., Cary, NC). All tests were two sided and P value <0.05 was considered statistically significant.

2. Meta-analysis

2.1 Study selection and study quality assessment

A comprehensive, computerized literature search was performed in PubMed and Web of Knowledge using the following key words: (liver OR hepatocellular OR hepatic) AND (cancer OR neoplasm OR tumor OR carcinoma) AND (family history) without language limitation. The search of the publications was through December 2013. The identified publications were reviewed independently for their relevance to the research topic by two authors (Y.Y., W.Q.J.). A set of pre-specified criteria was applied during the review and discrepancies were resolved by discussion. We included studies that fulfilled the following criteria: (a) had information on a family history of liver cancer among first-degree relatives as the exposure of interest, (b) had liver cancer as the outcome of interest, and (c) provided odds ratio (OR), relative risk (RR) or HR estimates with CIs or standard errors or data necessary to calculate these.

A prior consideration was given to the multivariate risk estimates, adjusted for the largest number of potential confounding factors. If more than one study was conducted in the same population, we selected the most informative or most recent report. A flow diagram for the search is presented in Figure 1. In total, 1421 articles were obtained from PubMed (n=571) and Web of Knowledge (n=850) databases searches. We further excluded 581 records with duplicate titles/abstracts. After screening the titles and abstracts of all the obtained records, we selected 31 articles for further assessment and 809 records with minimal possibility to report the association between family history of liver cancer and liver cancer risk were excluded. Based on the pre-specified inclusion criteria, 22 reports were identified by screening the full texts of the 31 selected records^{5, 10–17, 21–42}, i.e. nine articles were excluded because their data was later updated^34–40 or reporting data from case-control studies nested in previously identified cohorts^{41, 42}. Moreover, 4 additional articles were identified through checking the reference lists of the retrieved articles^43–46. Thus, besides our current cohort study, 26 articles (21 case-control studies and 5 cohort studies) met our inclusion criteria and were finally included in present meta-analysis^{5, 10–17, 21–33, 43–46}.

Figure 1.

References searched and selection of studies in the meta-analysis.

To assess the study quality, an evaluation system based on the Newcastle-Ottawa Scale⁴⁷ was adopted and assessed independently by two reviewers (Y.Y., W.Q.J.). In this evaluation system, a study was judged on 3 broad perspectives as follows: the selection of study groups, comparability of groups, and ascertainment of either the exposure or outcome of interest for case-control or cohort studies respectively⁴⁷. The scores of the included studies ranged from 3 to 9 stars and the median value was 6 stars. Thus, the high-quality studies were defined as studies with ≥6 stars^{5, 10–12, 14, 15, 17, 23, 25, 28, 45, 46} and the low quality study were those with <6 stars^{13, 16, 21, 22, 24, 26, 27, 29–33, 43, 44}.

2.2 Statistical analysis

The heterogeneity among studies was assessed with the Q and the I² statistic (results were defined as heterogeneous for a P value <0.10 or an I² >50 %.)⁴⁸. We pooled the study specific estimates using the fixed effect model or the random effect model proposed by DerSimonian and Laird depending on whether a significant heterogeneity was detected⁴⁹. Publication bias was evaluated with Begg’s and Egger’s tests^{50, 51}. The Meta-analysis was performed with STATA version 11.0 (StataCorp, College Station, TX). All tests were two sided and P value <0.05 was considered statistically significant if not specified.

Results

1. The Shanghai Women’s and Men’s Health Studies

After a median follow-up time of 12.2 years for the SWHS and 6.5 years for the SMHS, 128 and 171 liver cancer cases were identified in the SWHS and SMHS, respectively. Compared with cohort members without liver cancer, patients with liver cancer were generally older and had lower socioeconomic status as indicated by lower income and education levels at baseline (Table 1). Liver cancer patients were more likely to have reported a history of other chronic diseases, including hepatitis, cirrhosis or other liver disease, diabetes, and cholelithiasis or cholecystectomy. Among women, liver cancer patients tended to have higher BMI status and less vegetable and fruit intake than other cohort participants, while the distributions of BMI status and vegetable and fruit intake were comparable in men with and without liver cancer.

Table 1.

Baseline characteristics of the Shanghai Women’s (1997–2000) and Men’s (2002–2006) Health Studies

	Women			Men
Characteristic a	Non-cases (n=72527)	Liver cancer cases (n=128)	P value b	Non-cases (n=60188)	Liver cancer cases (n=171)	P value b
Age, y	52.45 ± 9.03	59.08 ± 8.51	<0.01	55.26 ± 9.69	59.76 ± 9.81	<0.01
body mass index, kg/m2	24.01 ± 0.01	24.70 ± 0.29	0.02	23.73 ± 0.01	23.82 ± 0.23	0.70
Family income (%)
Low	16.01	27.39		12.52	11.84
Low to Middle	38.20	35.19		42.46	48.91
Middle to High	28.17	29.62		35.22	30.28
High	17.62	7.80	0.10	9.80	8.96	0.24
Education level (%)
Elementary school or less	21.14	27.41		6.54	6.60
Middle school	37.29	33.23		33.52	31.17
High school	28.01	31.48		36.16	43.10
College or above	13.56	7.88	<0.01	23.78	19.13	0.08
Ever had chronic hepatitis (%)	2.59	23.38	<0.01	6.83	40.96	<0.01
Ever had cirrhosis or other chronic liver disease (%)	0.81	4.35	<0.01	3.27	15.03	<0.01
Ever had diabetes (%)	4.21	7.19	0.13	6.13	8.81	0.01
Ever had cholelithiasis or cholecystectomy (%)	11.31	15.14	<0.01	7.73	14.52	<0.01
Vegetable and fruit intake, g/day	561.56 ± 1.07	517.04 ± 25.47	0.08	495.96 ± 1.06	489.73 ± 19.99	0.76
Ever smoker (%)	2.76	1.55	0.29	69.61	68.55	0.76
Ever alcohol drinker (%)	2.25	1.06	0.90	33.67	30.40	0.94

^aAll variables were standardized to age distribution at baseline. Continuous variables are presented as the mean ± standard deviation.

^bContinuous variables were compared with use of t test; categorical variables were compared with use of chi-square test.

Overall, a family history of liver cancer in first-degree relatives was associated with a significantly elevated risk of liver cancer (Table 2). When we adjusted for all potential confounding factors, the HR was 2.60 (95% CI: 1.77–3.80) for both sexes combined. The results were similar for men (HR=2.78, 95% CI: 1.72–4.50) and women (HR=2.39, 95% CI: 1.28–4.48). The excess risk was confined to individuals reporting liver cancer in their mothers (HR=5.35, 95% CI: 3.10–9.21) and their siblings (HR=2.70, 95% CI: 1.54–4.72), but not in their fathers. In both men and women, the risk of developing liver cancer was particularly high for those with an affected mother. Moreover, increasing categories of the number of affected relatives was associated with an increased risk of liver cancer in both men and women (P_trend <0.001). Compared to individuals without positive family history of liver cancer, the relative risk of liver cancer for those who had ≥2 affected relatives in the family was 5.11 (95% CI: 2.10–12.45) for both sex combined. When we further considered family size using the FH score, risk of liver cancer significantly increased with increasing levels of the score (P_trend <0.001). Compared to the minimal-risk category, the HRs were 2.34 (95% CI: 1.43–3.83) for the intermediate-risk category and 3.03 (95% CI: 1.73–5.31) for the high-risk category (Table 2).

Table 2.

Adjusted hazard ratios for liver cancer by family history of liver cancer in the Shanghai Women’s (1997–2000) and Men’s (2002–2006) Health Studies

Family history of liver cancer	Both sexes combined			Women			Men
	Cases	HR a	95% CI	Cases	HR a	95% CI	Cases	HR a	95% CI
Dichotomous method
No family history	269	1	Reference	117	1	Reference	152	1	Reference
Any first-degree relatives b	30	2.60	(1.77, 3.80)	11	2.39	(1.28, 4.48)	19	2.78	(1.72, 4.50)
Mother	14	5.35	(3.10, 9.21)	7	7.42	(3.42, 16.11)	7	4.68	(2.18, 10.06)
Father	5	0.99	(0.41, 2.40)	1	0.55	(0.08, 3.97)	4	1.23	(0.46, 3.33)
Siblings	13	2.70	(1.54, 4.72)	4	2.00	(0.73, 5.45)	9	3.20	(1.63, 6.28)
No. of affected relatives
0	269	1	Reference	117	1	Reference	152	1	Reference
1	25	2.37	(1.57, 3.58)	8	1.88	(0.91, 3.86)	17	2.76	(1.67, 4.57)
≥2	5	5.11	(2.10,12.45)	3	9.35	(2.93, 29.86)	2	2.97	(0.73, 12.07)
Ptrendc	<0.001			<0.001			<0.001
FH score methodd
Minimal-risk	269	1	Reference	117	1	Reference	152	1	Reference
Intermediate-risk	17	2.34	(1.43, 3.83)	6	1.93	(0.85, 4.41)	11	2.77	(1.50, 5.13)
High-risk	13	3.03	(1.73, 5.31)	5	3.37	(1.36, 8.36)	8	2.80	(1.37, 5.73)
Ptrende	< 0.001			0.003			< 0.001

^aAdjusted for age, sex (for both sexes combined only), education, income, BMI, vegetable and fruit intake, history of chronic hepatitis, history of cirrhosis or other chronic liver disease, history of diabetes, history of cholelithiasis or cholecystectomy.

^bFirst-degree relatives included mother, father, sisters, brothers, daughters and sons. As few cancer cases were in offspring group, HRs and 95% CIs were not available.

^cTwo-sided P_trend values were calculated by assigning an ordinal value (1, 2, 3) to each subcategory of the No. of affected relatives and treating it as a continuous variable in the regression models.

^dFH score, family history score (FH score=No. of affected first-degree relatives /total No. of first-degree relatives excluding the index one ); minimal-risk (FH score = 0); intermediate-risk (0 < FH score ≤1/7 ); high-risk (FH score > 1/7).

^eTwo-sided P_trend values were calculated by assigning an ordinal value (1, 2, 3) to each range of the FH score and treating it as a continuous variable in the regression models.

Table S1 shows the liver cancer risk associated with family history of some common cancer sites identified in our cohorts, including any cancer, as well as cancers of esophagus, stomach, pancreas, colorectum and lung. Excess liver cancer risks were observed in men with first-degree relatives affected by stomach cancer (HR=1.66, 95% CI: 1.03–2.68), particularly for men with affected siblings (HR=2.62, 95% CI: 1.33–5.16). Additionally, liver cancer risk was also observed to be associated with a family history of pancreatic cancer (HR=2.52, 95% CI: 1.25–5.10) in women and men combined, particularly for those with affected parents (HR=2.65, 95%CI: 1.18–5.97).

2. Meta-analysis

The main characteristics of the studies included in the meta-analysis are shown in Table S2. Figure 2 shows the forest plot for the association between family history of liver cancer and liver cancer risk. The pooled RR was 2.55 (95% CI: 2.05–3.16) for all studies combined with a significant heterogeneity (I²=78.4, P_{heterogeneity}<0.001). The pooled RRs were 2.56 (95% CI: 1.91–3.43, I²=81.6%, P_{heterogeneity}<0.001; n=21) for case-control studies^{10, 11, 13, 14, 21–33, 43–46} and 2.64 (95% CI: 1.96–3.56, I²=60.1%, P_{heterogeneity}=0.020; n=6) for cohort studies (including the present study)^{5, 12, 15–17}. When we excluded 14 low-quality studies^{13, 16, 21, 22, 24, 26, 27, 29–33, 43, 44} from the analysis, the summary estimate for the 13 high-quality studies (including the present study) (RR=2.48, 95%CI: 2.09–2.93, I²=15.1%, P_{heterogeneity}=0.288; n=13)^{5, 10–12, 14, 15, 17, 23, 25, 28, 45, 46} was slightly lower than the overall result and the statistical significant heterogeneity was not detected. When we considered studies reporting risk estimates for men and women separately, the pooled RRs were 2.59 (95% CI: 1.74–3.86; I²=79.5%, P_{heterogeneity}<0.001; n=10) for men (including the present study)^{5, 10–12, 15, 32, 44–46} and 1.76 (95% CI: 1.25–2.48; I²=0%, P_{heterogeneity} =0.703; n=7) for women (including the present study)^{10, 11, 15, 32, 44, 45}. When stratified by study location, studies conducted in Mainland and Taiwan China (RR=2.66, 95%CI: 2.07–3.43, I²=81.9%, P_{heterogeneity}<0.001; n=20) (including the present study)^{5, 12–17, 21, 22, 24, 26–29, 31–33, 43, 46} and Italy (RR=2.55, 95%CI: 1.70–3.81, I²=0%, P_{heterogeneity}=0.870; n=3)^{11, 23, 25} both showed significant positive associations between family history of liver cancer and liver cancer. The pooled RR was 1.83 (95%: 0.50–6.76; I²=80.4%, P_{heterogeneity} =0.024; n=2) and 2.38 (95%CI: 0.90–6.32, I²=55.3%, P_{heterogeneity}=0.135; n=2) for studies conducted in the United Stated^{10, 30} and Japan^{44, 45}, respectively. Moreover, the pooled RR was 2.84 (95% CI: 2.32–3.49; I²=64.9%, P_{heterogeneity}<0.001; n=21) after excluding 6 studies conducted in subjects with positive HBsAg or other liver diseases^{5, 12, 13, 28, 30, 32}. We further explored the publication bias by both Egger’s test (P=0.248) and Begg’s test (P=0.678), and no publication bias was detected in this analysis.

Figure 2.

Results from a meta-analysis of association between family history of liver cancer and liver cancer risk. Squares represent study-specific estimates (size of the square reflects the study-specific statistical weight); horizontal lines represent 95% CIs; diamonds represent summary estimates with corresponding 95% CIs.

Discussion

We investigated the association between a family history of liver cancer and liver cancer risk in two prospective, population-based cohort studies. Family history of liver cancer was associated with an increased risk of liver cancer using both binary indicator and proportional FH score, with increasing HRs for increasing score categories. This risk is particularly high for those with an affected mother or those with ≥2 affected relatives in the family. The meta-analysis also showed an increased risk for those with a family history of liver cancer.

Our observation of an increased risk of liver cancer associated with a family history of liver cancer was consistent with the overall evidence from the meta-analysis. It was just slightly higher in comparison with the pooled estimate of the high-quality studies^{5, 10–12, 14, 15, 17, 23, 25, 28, 45, 46}. Moreover, our results were also in agreement with the meta-analysis even after excluding 6 studies among subjects with positive HBsAg or other liver diseases^{5, 12, 13, 28, 30, 32}. Besides, our results were also consistent with the pooled estimate of the five case-control studies from the Western population (RR=2.12, 95% CI: 1.29–3.51)^{10, 11, 23, 25, 30}.

Familial aggregation of liver cancer in our study population may result from the complex effects of shared common environmental and/or genetic factors in families. Firstly, given the high prevalence of chronic HBV infection and vertical transmission of HBV being the major route of viral transmission in the population, the reported association of an elevated liver cancer risk among those with a family history of liver cancer could be partly explained by clustering of HBV infection among members of the same family⁴. This assumption is consistent with our findings that those with an affected mother had a higher risk of developing liver cancer than those with an affected father or siblings. Secondly, genetic factors may also independently modulate liver cancer risk. A Swedish study using a nationwide registry database reported that “true” genetic effects likely confer liver cancer risk, possibly modified by environmental factors⁶. The prevalence of environmental risk factors for liver cancer is low in Sweden. The study assessed the environmental effects by comparing familial risks and spouse correlations for liver cancer. No spouse correlation was observed, suggesting that shared environmental effects are small between spouses. As such, “true” genetic effects may contribute to the familial aggregation of liver cancer⁶. A previous case-control study conducted in the United States also suggested that a family history of liver cancer in first-degree relatives was significantly associated with HCC development independent of HBV and HCV infections¹⁰. Thirdly, genetic factors may interact with HBV infection to enhance familial aggregation of liver cancer occurrence. For instance, synergistic effects of family history of HCC and HBV infection on risk of incident HCC has been reported in a population-based cohort study of 22,472 participants¹⁷. Other studies suggested that single nucleotide polymorphisms in the PAPSS1 gene, which is a co-substrate for all sulfation reactions and has been implicated in cancer, may modify the response to viral infection in Chinese families^7–9. Recently, two genome-wide association studies suggested that chronic HBV carriers with several susceptibility locus in the DEPDC5, STAT4 and HLA-DQ genes are more likely to develop HCC^{52, 53}. Xie et al. have detected an interaction effect of signal transducer and activator of transcription 3 polymorphisms with HBV mutations on HCC risk⁵⁴. Finally, other shared genetic and environmental factors in families, such as shared dietary habits, exposure to carcinogens such as aflatoxin, heavy alcohol drinking and tobacco smoking, may also act individually or jointly in the development of liver cancer in family members⁵⁵.

Other intriguing findings in our cohorts include an increased risk of liver cancer associated with a family history of stomach or pancreatic cancer. In some Western countries such as Italy and the United States, the relationships between liver cancer and family history of stomach or pancreatic cancer have also been examined in three case-control studies, while no significant associations were found previously^{10, 11, 23}. The co-aggregation of liver and stomach cancer, and that of liver and pancreatic cancer in our cohorts may be attributable to a number of shared environmental risk factors, including cigarette smoking, alcohol drinking, and carriage of Helicobacter pylori, a carcinogenic stomach pathogen^{3, 56, 57}. Of note, the increased risk of liver cancer in association with family history of pancreatic cancer could also be possibly attributable to the association between HBV infection and pancreatic cancer⁵⁸. The shared environmental exposures among family members could have contributed to the associations observed in our study, although we cannot rule out any genetic causes underlying the observed associations.

Our study has several strengths compared to previously published studies on the association between family history of liver cancer and liver cancer risk, such as the recent two case-control studies conducted in Western population by Turati et al.¹¹ and Hassan et al.¹⁰. Firstly, the prospective study design and high participation rates minimized recall and selection biases which may distort the associations in previous case-control studies^{10, 11}. Secondly, we applied a proportional FH score approach to classify the risk categories of family history of liver cancer, taking into consideration family size and number of affected first-degree relatives in our study population. By a statistical comparison of different family history scores (Proportion, SR, Reed, Williams, Schwartz and Silberberg scores), Murad et al. have found surprisingly little difference in the performance of various scores, and strongly recommended the usage of the simpler proportion FH score⁵⁹. The internal consistency of findings in both men and women and the “dose-response” of risk with increasing FH score lend credence to our findings. Moreover, the FH score might be helpful for us to find out individuals with the highest potential liver cancer risk in future cancer prevention and screening programs. The external consistency with the results from the meta-analysis could further support our findings.

Information on family history of cancer was collected from subjects’ recall and no further efforts were made to verify the diagnosis. Since the baseline information was collected before the disease onset using a structured questionnaire, any bias resulting from subjects’ inaccurate recall would likely to be non-differential and therefore may result in an underestimation of the observed association. In addition, there is a possibility of under-reporting of family history of cancer due to under-diagnosis and limited communication about cancer in families. As such, our findings are likely to be conservative. We restricted our analysis to first-degree relatives; previous studies have shown that family history for first-degree relatives is accurate and valid for major cancers^{60, 61}. Furthermore, since we did not have information on HBV infection, we cannot evaluate the interaction with HBV infection. But our analysis adjusted for a history of chronic hepatitis and liver diseases. Another limitation in our study is the lack of information for both study subjects and family members on some rare monogenic syndromes such as genetic hemochromatosis and α-1 antitrypsin deficiency, which are associated with increased risk of liver cancer. Moreover, since the high number of comparisons, it is difficult to avoid the problem of multiple testing, which could possible increase the false positive errors in the study. With an increasing number of tests, the number of chance associations increases. However, the observed associations between family history of liver cancer and liver cancer risk is biological plausibility and consistency with previous findings. Thus, it may be less likely be caused by false positive errors. With regard to the intriguing findings on liver cancer risk associated with family history of stomach or pancreatic cancer, although it might also be biological possible, they must be explained with caution due to limited study cases and the problem of multiple testing. As the observed “significant” associations between family history of stomach or pancreatic cancer and liver cancer were based upon exploratory analyses, confirmation of these findings in other prospective studies and further considerations of potential mechanisms are still warranted. Of note, the observed associations between family history of stomach or pancreatic cancer and liver cancer may provide some clues for future explorations on carcinogenic mechanisms in environmental or genetic level. Finally, there was a high heterogeneity in the overall interest of the meta-analysis, which may, to some extent, limit the conclusions. However, when we stratified the studies in the meta-analysis by study quality, there was no significant heterogeneity with the subgroup of the high-quality studies and the pooled result from the high-quality studies was consistent with the overall result of the meta-analysis.

In conclusion, our results indicated that individuals having a family history of liver cancer among first-degree relatives are at increased risk of developing liver cancer. This risk is particularly high for those with an affected mother. The “dose-response” of risk with increasing family history score of liver cancer might further facilitate future cancer prevention and screening programs on finding out individuals with the highest potential liver cancer risk.

Supplementary Material

Supp Table S1-S2

Table S1. Adjusted hazard ratios for liver cancer by family history of cancer in the Shanghai Women’s (1997–2000) and Men’s (2002–2006) Health Studies

Table S2. Main characteristics of the studies on liver cancer and family history of liver cancer included in the meta-analysis

Text S1. References 51–61

Novelty & Impact Statements.

Heritable factors likely contribute to a person’s risk of developing liver cancer, but whether a family history of the disease increases that risk remains unclear. Here, in two prospective cohorts derived from the Shanghai Women’s and Men’s Health Studies, an association between family history of liver cancer and increased risk of liver cancer was detected in the general Chinese population. A “dose-response” risk was identified, in which risk increased in tandem with an increased family history score of liver cancer, potentially allowing minimal-, intermediate-, and high-risk individuals to be distinguished in a liver cancer prevention program.

Abbreviations

HBV

hepatitis B virus

HCV

hepatitis C virus

HCC

hepatocellular carcinoma

FH score

family history score

SWHS

the Shanghai Women’s Health Study

SMHS

the Shanghai Men’s Health Study

ICD-9

International Classification of Disease, Ninth Revision

OR

odds ratio

RR

relative risk

HR

hazard ratio

CI

confidence interval

BMI

body mass index

NA

not available