Potential conflict of interest: Nothing to report.
Several studies have indicated that primary biliary cirrhosis (PBC) may be associated with increased risk of some cancers, but the results are controversial. We conducted a systematic review of studies to examine the association of PBC with cancer risk by meta-analysis. We searched the PubMed and EMBASE databases for English-language studies published before November 2011. Studies were included if they reported relative risk estimates with 95% confidence intervals (CIs) or related data for the association between PBC and cancer risk. Approximately 16,300 PBC patients from several countries were included in this analysis. Of the 3510 titles identified, 16 publications involving 17 studies meeting the inclusion criteria were included in the meta-analysis. Compared with the general population, PBC patients had a significantly higher risk of overall cancer (pooled rate ratio [RR], 1.55; 95% CI, 1.28-1.83) and hepatocellular carcinoma (HCC) (pooled RR, 18.80; 95% CI, 10.81-26.79). For stomach and pancreas cancers, the results of one study that only examined male patients with PBC indicated that PBC patients had increased risk of stomach cancer and pancreatic cancer, whereas the results of other studies of mixed-sex patients showed no significant association. Therefore, despite inconsistent results, the meta-analysis could not be conducted for assessing the association. PBC was not significantly associated with increased risk of other cancers. Conclusion: The present systematic review and meta-analysis demonstrate that PBC is closely associated with a greater risk of overall cancer and HCC, but not with other cancers. The data regarding the association between PBC and risks of several cancers need to be further confirmed in future studies. (HEPATOLOGY 2012)
Primary biliary cirrhosis (PBC) is a chronic progressive cholestatic liver disease of unknown etiology characterized by inflammatory destruction of bile ducts, which predominantly affects middle-aged women. The disease presents a variety of disease spectrums from asymptomatic disease state to full-blown cirrhosis. The survival of PBC patients is long because of slow progression and early detection of the disease. Several studies have indicated that PBC may be associated with increased risks of some cancers, such as hepatocellular carcinoma (HCC), breast cancer, pancreatic cancer, and so forth.1-13 If an increased risk for some malignancies can be proved, clinical management, surveillance, and follow-up issues will be carefully addressed. However, the results of previous studies are controversial. The differences noted between the studies may be explained partially by the methodology, sample sizes, and so forth.
To date, no published meta-analyses have successfully established the association of PBC with cancer risk. The aim of the present study was to perform a systematic review and meta-analysis to derive a better estimation of the association.
A literature search of the PubMed and EMBASE databases was conducted for English-language studies published before November 2011 using combinations of the following terms: primary biliary cirrhosis and cancer; malignancy; malignancies; neoplasm; tumor; carcinoma; and lymphoma. All eligible articles were retrieved, and their references were checked for other relevant studies.
Studies were included in the meta-analysis if they fulfilled the following inclusion criteria: (1) cohort or case-control design; (2) PBC as one of the exposure interests; (3) cancer as one of the outcome of interests; (4) rate ratio, hazard ratio, or standardized incidence ratio with 95% confidence intervals (CIs) (or with data to calculate them) available; and (5) independent study. In case of multiple reports on the same population or subpopulation, we included only data from the latest or complete studies with the largest numbers of cases and controls. Studies were excluded if the effect size could not be calculated according to these studies. When studies provided more than one rate ratio (RR) according to the duration of PBC before malignancy was diagnosed, the RRs for individuals diagnosed with PBC more than 1 year prior to the diagnosis of malignancy were extracted and combined.
Quality assessment was performed using the Newcastle-Ottawa Scale (NOS).14 There are a total of eight items in the NOS categorized into three dimensions: selection, comparability, and—depending on the study type—outcome (cohort) or exposure (case-control). A star system of the NOS has been developed for the assessment. A maximum of one star is given for each numbered item within the selection and exposure categories, and a maximum of two stars is given for comparability. Because there is no establishment for standard criteria, we considered a study awarded 0-3 stars, 4-6 stars, or 7-9 stars as a low-, moderate-, or high-quality study, respectively.
All articles were retrieved and assessed independently by two reviewers (Y. L. and Z. Y.) who extracted data that included authors, publication date, country of origin, characteristics of the study population (including sex, age, and mean follow-up years), number of observed and expected cases, and other details of adjustment. Any disagreement was resolved by consensus.
Publications that reported different measures of relative risks such as RR, hazard ratio, standardized incidence ratio (SIR), and proportional incidence ratio (PIR) with corresponding 95% CIs were selected for inclusion in the meta-analysis. The preferred method of data presentation was the calculated RR compared with the general population. For publications without control group, RR was generally estimated as the age- and sex-adjusted SIR. If SIR was not specifically reported in the primary study, it was calculated from the observed and expected incidence rates presented in the study (SIR = number of observed malignancies per number of expected malignancies). Of note, the expected number of cases of a particular cancer by sex and 5-year age bands in a primary study was calculated using data from the International Agency for Research on Cancer CancerBase No. 9.15 The corresponding 95% CIs were estimated using the PAMCOMP program.16 Heterogeneity of effects across studies was assessed using the chi-square statistic and quantified by I2, which represented the percentage of total variation across studies that was attributable to heterogeneity rather than chance (P < 0.10 was considered representative of statistically significant heterogeneity).17 A fixed-effect model was used when there was no heterogeneity of the results of the trials. Otherwise, the random-effect model was used. If statistical heterogeneity was present, the Galbraith plot was used to detect the potential sources of heterogeneity.18 Besides, meta-regression analysis was also applied to perform both general analyses and subgroup analyses to better investigate possible sources of between-study heterogeneity. Subgroup analyses of association of PBC with overall cancers, HCC, and breast cancer were performed by stratifying on region, case ascertainment, the type of effect size, sex, and mean or median age. To assess the stability of results, sensitivity analysis was performed using sequential omission of individual studies or by omitting studies plotted by the Galbraith plot methods as the possible major source of heterogeneity. Funnel plots were performed to estimate the potential publication bias, and an asymmetrical plot suggests a possible publication bias. The asymmetry was assessed using Egger's linear regression test and P < 0.05 was considered to represent statistically significant publication bias.19 All analyses were performed with STATA 11.0 software.
A total of 3510 publications were identified in the initial search, and 3455 records were excluded based on screening of titles or abstracts (Fig. 1). Full text articles were retrieved only for 55 publications and assessed for eligibility. Of these 55 publications, 39 were excluded (26 duplicate publications, one review, four publications containing overlapping data, and eight publications in which SIR and 95% CI could not be calculated based on data provided). Overall, we identified and included 16 publications involving 17 studies that met the inclusion criteria in the systematic review. Notably, there was one publication that involved two cohort studies, one for Spanish patients and the other for Italian patients.20
Characteristics of Included Studies.
The 17 studies were published between 1984 and 2011 (Table 1) and involved a total of 16,368 PBC patients. Study characteristics, demographic information, and adjustment or restriction variables for the included studies are listed in Table 1. Of these 17 studies, four22, 23, 25, 27 were population-based, while the others were hospital-based. Nine studies involving 6766 patients were conducted for relative risk of overall malignancies,11, 12, 21-27 12 involving 13,576 patients for hepatocellular carcinoma,12, 13, 20-25, 28-30 nine involving 5945 patients for breast cancer,1, 11, 12, 21, 23-27 five involving 3221 patients for kidney cancer,1, 12, 21, 25, 26 and five involving 8466 patients for colon cancer.11-13, 21, 26 The numbers for other cancers are listed in Table 3. Of 13 studies using SIR as the measurement of relative risk, three presented SIRs23, 25, 27 and nine did not,11, 20, 24, 26, 28-31 necessitating calculation of the SIRs for the combined population, and one provided PIR rather than SIR.1 Furthermore, this study provided PIR for various site-specific malignancies (e.g., HCC, breast cancer, skin melanoma, colorectal cancer, kidney cancer), but not for overall malignancy. Thus, the PIRs were used as the measurement of relative risk for the combined population for various site-specific malignancies, rather than overall malignancy. In addition, there were, at least partly, overlapping data for HCC risk between this study and another study by Cavazza et al.20 Therefore, the data for HCC risk, but not other site-specific malignancy risks, from the study by Floreani et al.1 were excluded.
Table 1. Characteristics of Studies on PBC and Cancer Risks
No. of PBC Patients
Mean Age at Study Entry, Years
Cancer Type (n)
Mean Follow-up, Years
Estimate of Relative Risk
Adjustment for Covariates
Abbreviations: ICD, International Classification of Diseases; NA, not available.
Table 2. Stratified Analyses of Pooled Relative Risks of Overall Cancer, HCC, and Breast Cancer for PBC
No. of Studies
No. of PBC Patients
Pooled RR or SIR (95% CI)
Meta-regression P Value
Type of effect size
Mean or median age
Type of effect size
Mean or median age
Type of effect size
Mean or median age
All of the studies included were cohort studies. On the basis of the NOS for the cohort studies, the majority of studies included were deemed of high quality (13 studies with score of 7 or more). The quality of three studies was deemed moderate (score of 4-6). Only one study was of low quality (score of 3) (Supporting File 1).
PBC and Risk of Overall Cancer.
As shown in Table 3, the pooled RR with 95% CI was 1.55 (95% CI, 1.28-1.83) in a random-effect model for PBC patients compared with general population. Due to moderate heterogeneity (I2 = 43.6%, P = 0.077), we analyzed the source of the heterogeneity by the Galbraith plot and meta-regression including region, case ascertainment, the type of effect size and age. However, we could not find the possible major source of heterogeneity by the Galbraith plot and by the meta-regression (Fig. 2 and Table 2). Furthermore, we conducted subgroup meta-analyses by region, case ascertainment, type of effect size, sex, and age. The increased risk did not change significantly in various subgroups, with the exception of one subgroup for studies with RR as measurement of risk. There was no significant risk increase in this subgroup, which may have to do with the small number of studies (only three studies) with significant heterogeneity (I2 = 52.4%) (Table 2).
Table 3. Pooled Relative Risks of Various Malignancies
No. of Studies
No. of P atients
Pooled RR or SIR (95% CI)
PBC and HCC Risk.
As shown in Table 3, because of severe heterogeneity (I2 = 79.1%, P < 0.001), a random-effect model was used to evaluate the association of PBC with HCC risk. The pooled RR with 95% CI was 18.80 (95% CI, 10.81-26.79) for PBC patients compared with the general population. In addition, the Galbraith plot was performed to analyze the source of heterogeneity; however, we could not find the possible major source of heterogeneity, because it plotted too many studies as the outliers (Fig. 3). Furthermore, a meta-regression including region, case ascertainment, type of effect size, and age was also performed to analyze the source of heterogeneity. The results indicate that only the type of effect size was significantly associated with the heterogeneity (P < 0.1). When conducting subgroup meta-analyses by these factors, PBC still remained significantly associated with an increased risk of HCC in various subgroup analyses with the exception of two, one for the studies in the USA population (pooled RR 23.88, 95%CI, -9.14-56.89), the other for the population-based studies (pooled RR 8.61, 95%CI, -4.18-21.40), which might result from too small number of studies (only three studies for each subgroup) with significant heterogeneity (Table 2).
PBC and Other Cancer Risk.
As shown in Table 3, there were nine studies reporting RRs or giving information with which SIRs could be calculated for breast cancer risk; five for kidney cancer risk; five for colon cancer risk; four for stomach cancer risk; three for pancreatic cancer, lung cancer, non-Hodgkin lymphoma and colorectal cancer risk; and two for esophageal cancer, rectal cancer, uterine cancer, cervical cancer, prostate cancer, bladder cancer, skin melanoma, skin nonmelanoma, Hodgkin disease, and thyroid cancer risk. Because of no significant heterogeneity (all I2 values were <20% and all P values were >0.1), fixed-effect models were conducted to evaluate the association between PBC and various cancer risks, with the exception of stomach cancer and pancreatic cancer. The results indicate that PBC was not with increased risk of breast cancer, kidney cancer, or colon cancer, as well as other malignancies. For breast cancer, we also conducted subgroup analyses by region, case ascertainment, type of effect size, and age. The results showing no significant association with increased risk of breast cancer did not change in various subgroups (Table 2). For stomach cancer and pancreatic cancer, there were four studies and three studies, respectively. One study by Landgren et al.13 involving a large number of male PBC patients found that PBC patients had increased risk of stomach cancer (RR, 1.66; 95% CI, 1.10-2.51) and pancreatic cancer (RR, 2.06; 95% CI, 1.44-2.96). Other studies showed no significant association between PBC and risk of these two cancers in mixed-sex patient groups.
For analysis of pooling more than three individual studies, sensitivity analysis was performed to examine the stability and reliability of pooled RRs by sequential omission of individual studies. The results indicate that the significance estimate of pooled RRs was not significantly influenced by omitting any single study.
Because it is unlikely that funnel plots will be useful in meta-analyses containing fewer than five studies,32 the publication bias was evaluated by Funnel plot and Egger's test only for meta-analyses of pooling five or more individual studies. Funnel plot shapes showed no obvious evidence of asymmetry, and all the P values of Egger's tests were over 0.05 (Supporting Files 2-6). These results suggest that publication bias was not evident in various meta-analyses.
This is, to our knowledge, the first systematic review and meta-analysis to assess the association between PBC and cancer risk. Using the NOS, we found that the majority of studies included in this meta-analysis were of high quality (13 studies with score of 7 or more), and only one study was of low quality (score of 3). The results of this study indicate that PBC is significantly associated with an increased risk of overall cancer and HCC but not other cancers. In addition, we could not draw a consistent conclusion about the association of PBC with the risks of stomach and pancreatic cancers; this association needs to be examined further in a larger number of studies.
Several studies examining the risk of malignancy in PBC patients have yielded diverse results. Some data have revealed an increased overall cancer risk,11, 12, 23-26 whereas others disagree.21, 22, 27 The present study, with more strong evidence via meta-analysis of published studies, confirmed that there is increased risk of overall malignancy in PBC patients. Compared with non-PBC individuals, PBC patients may have an approximately 55% increased risk of overall malignancies. Furthermore, subgroup meta-analyses showed that PBC still remained significantly associated with increased risk of overall cancers in the majority of subgroups, with the exception of one subgroup for studies with RR as a measurement of risk. The lack of a significant risk increase in this subgroup may be due to the small number of studies (only three) with significant heterogeneity (I2 = 52.4%). Therefore, more high-quality case-control studies on the association of PBC and cancer risk should be performed in the future.
HCC is the most common primary malignant tumor of the liver,34 and its incidence in PBC is not well known. Some studies have reported an increased risk of HCC in PBC patients, whereas others have found a low incidence of HCC in PBC. One reason for the controversy is that PBC is a relatively rare disease. Thus, the sample size was usually small in the majority of studies. For example, HCC was not found in PBC patients in the latest study by Ngu et al,27 though the investigators conceded that because the number of male PBC patients–the highest risk group for HCC—was so small (n = 6), it may have led to bias. The present meta-analysis, with a larger sample size and stronger evidence, demonstrated an increased risk of HCC in PBC patients, which was more than 18.8-fold higher than that of the general population. Another reason for the controversy is that there are some geographical and environmental differences between studies. Therefore, we further conducted subgroup meta-analyses, which confirmed that this increased risk could not be affected by such variables as region (except the United States), age, sex, case ascertainment (except population-based studies), and type of effect size. However, there are still several confounding factors, such as advanced histological stage (stage 4 PBC),1, 5, 8, 9, 21 history of blood transfusion,9, 28 and smoking or drinking habit,33–35 which might be associated with increased probability for HCC development in PBC patients or might be directly associated with PBC development. The interference of these factors cannot be excluded in this meta-analysis, because subgroup meta-analyses were not performed because of the small number of selected studies exploring the association of these factors with HCC risk in PBC patients. This might also be a major reason why there was significant heterogeneity between studies in overall meta-analysis and in the majority of subgroup meta-analyses.
Although the data on the association between PBC and the risks of stomach and pancreatic cancers are inconsistent, meta-analyses could not be conducted for assessing the association. The reason is that one study by Landgren et al,13 who found that PBC patients had increased risk of stomach cancer (RR, 1.66; 95% CI, 1.10-2.51) and pancreatic cancer (RR, 2.06; 95% CI, 1.44-2.96), examined the association only in male patients with PBC. However, other studies showing no significant association with the risks of these two cancers were performed in mixed-sex patient groups. Regardless, the present study suggests that the significant association between PBC and increased risk of stomach and pancreatic cancers cannot be excluded, at least not in male patients. A larger number of studies need to be performed to confirm this association.
Notably, our present meta-analysis with insignificant between-study heterogeneity showed no significant association between PBC and breast cancer risk. These results also did not show any change in various subgroup meta-analyses. Furthermore, increased risk of breast cancer with PBC was found in only three studies,11, 12, 26 all of which were conducted before 1990. These results further quantitatively confirmed the conclusion by Piscaglia and Sagrini,8 who suggested that the incidence of breast cancer was not increased in PBC patients and a higher reported incidence may be attributed to immunosuppressive agents, which were used commonly before 1990.
Moreover, the present meta-analysis also did not show any significant association between PBC and several other site-specific malignancies including colon cancer, rectum cancer, colorectal cancer, lung cancer, kidney cancer, esophagus cancer, uterus cancer, cervical cancer, prostate cancer, bladder cancer, thyroid cancer, skin melanoma, skin nonmelanoma malignancy, Hodgkin disease, and non-Hodgkin lymphoma. However, the conclusion should be addressed with caution, since subgroup meta-analysis could not be performed for these malignancies due to too small number of available studies.
The present study has some limitations that should be addressed. First, the study included a wide variety of articles that looked at risks of malignancy in PBC. Therefore, we had to acknowledge that the specific differences between those articles which could account for different results might be a potential source of bias. Second, we could not include some studies that failed to report data with which relative risk of some cancers could be calculated. These unidentified studies may reduce the power of our analysis, but were unlikely to bias its results. Third, the impact of confounding inherent in the included studies can not be completely excluded, which might bias the results either toward overestimation or underestimation of risk estimates. Although subgroup analyses with some known confounders such as age, sex, region, case ascertainment, and type of effect size were performed for overall cancer, HCC and breast cancer risks, other confounders cannot be excluded as a potential explanation for the observed findings. Furthermore, for other cancer risks, subgroup analyses could not be performed due to the small number of studies. Fourth, as described in the previous study, it is impossible to completely exclude potential publication bias because small studies with null results tend not to be published,36 even though no significant publication bias was found by funnel plot analysis and formal statistical tests. Finally, data regarding PBC and risks of the majority of malignancies were extremely sparse, limiting our ability to draw firm conclusions.
In conclusion, the present systematic review and meta-analysis demonstrated that PBC is significantly associated with increased risk of overall malignancy, especially with HCC. Thus, as recommended by the AASLD Practice Guidelines,37 regular screening for HCC with cross-sectional imaging with or without alpha-fetoprotein at 6- to 12-month intervals should be advised for PBC patients. Our meta-analysis has provided the most comprehensive quantitative evidence for the practice by far. Furthermore, this significant association might also exist between PBC and stomach and pancreatic cancer risks, at least in male patients (which, however, needs to be further confirmed by a larger number of studies). In contrast, there is no significant association between PBC and breast cancer risk, which suggested that PBC patients do not need to be submitted to stricter surveillance programs for breast cancer than the general population. Also, there is no significant association between PBC and other cancer risks; however, this assessment needs to be further confirmed by a larger number of studies.