Surveillance for hepatocellular carcinoma in patients with primary biliary cirrhosis


  • Potential conflict of interest: Nothing to report.


Hepatocellular carcinoma (HCC) occurs with increased frequency in patients with primary biliary cirrhosis (PBC). Effectiveness of surveillance recommendations for HCC is controversial, and data are lacking in patients with PBC. In this study, we attempt to (1) establish the importance of surveillance for HCC in patients with PBC; (2) identify a target population of patients with PBC for HCC surveillance; and (3) propose surveillance recommendations for patients with PBC. We retrospectively identified 36 patients seen at the Mayo Clinic between 1976 and 2007 with a diagnosis of both PBC and HCC. Five patients (14%) were diagnosed incidentally, 17 patients comprised our surveillant population, and 14 patients were diagnosed outside a surveillance program. Patients in the surveillant population were more likely to undergo therapy (88% versus 43%; P = 0.01) and had improved survival (P = 0.002) compared with the nonsurveillant population. All cases of HCC except one were predicted to be at significant risk for HCC based on age, sex, evidence of portal hypertension, and history of blood transfusion using a previous predictive model. Conclusion: We established the importance of surveillance for HCC in patients with PBC. We demonstrated adequate performance of a predictive model and propose it should be refined and used to identify patients with PBC who should be screened for development of HCC. Further studies are needed so that optimal HCC surveillance recommendations in this population can be determined and included in the practice guidelines for PBC. (HEPATOLOGY 2008.)

Hepatocellular carcinoma (HCC) is the most common primary malignancy of the liver,1 and its incidence in the United States is rising.2 Approximately 70% to 90% of all detected HCC cases occur within an established background of chronic liver disease and cirrhosis of all etiologies.2, 3 HCC is one of the main causes of death in patients with liver cirrhosis.4 Primary biliary cirrhosis (PBC) is a chronic cholestatic liver disease that affects predominantly middle-aged women.5 The natural history of PBC usually consists of slow disease progression, which can result in liver cirrhosis, liver failure, and eventually require liver transplantation.6 HCC occurs with increased frequency in patients with PBC, particularly in patients with advanced histological disease.7–20 The exact frequency is unknown but is estimated to be between 0.7% and 16%.19, 21

Even though the effectiveness as well as cost-effectiveness of screening strategies remain controversial, the development of HCC in patients with cirrhosis argues for the opportunity to apply surveillance tests to detect liver cancer at an early and thus potentially curable stage.22 Although HCC surveillance is widely recommended for patients with cirrhosis, the current practice guidelines for PBC by the AASLD do not include specific recommendations for HCC surveillance.23

The AASLD practice guidelines for the management of HCC recognize stage 4 PBC24, 25 as a high-risk patient population and recommend inclusion of those patients in a surveillance program.26 In practice, this strategy has several limitations. For example, the histologic scoring of PBC has many known limitations, including sampling variability and size of biopsy specimen,27, 28 histologic variation after the use of ursodeoxycholic acid (UDCA),29–31 and interobserver and intraobserver variation, even in the hands of experienced liver pathologists. However, the major limitation of this strategy is that serial liver biopsies are not routinely indicated for clinical monitoring; consequently, many patients do not have recent histological evaluation several years into the course of their disease. Accurately determining the time when cirrhosis develops based on clinical information to include a particular patient in a surveillance program for HCC can be challenging. Surveillance for HCC of all patients with PBC might burden medical resources, and identification of a high-risk subgroup for surveillance without relying on histology could optimize the use of economic resources for health care. Older age, male sex, history of blood transfusion, and any signs of portal hypertension or cirrhosis seem to predict HCC in patients with PBC,18, 32 and predictive models might be useful for determining surveillance strategies. Unfortunately, data on HCC surveillance in patients with PBC are lacking.

In this study, we identified a large series of patients with PBC followed at the Mayo Clinic who eventually developed HCC during the course of their disease, and further characterized clinical, laboratory, and radiographic data to provide initial data to: (1) establish the importance of surveillance for HCC in patients with PBC; (2) identify a target population of patients with PBC for HCC surveillance; and (3) propose surveillance recommendations.


AFP, alpha-fetoprotein; CT, computed tomography; HCC, hepatocellular carcinoma; HR, hazard ratio; MRI, magnetic resonance imaging; MRS, Mayo Risk Score; PBC, primary biliary cirrhosis; UDCA; ursodeoxycholic acid.

Patients and Methods

Patient Population and Study Design.

Using the Mayo Medical Index System, 115 possible case subjects with any kinds of biliary cirrhosis complicated by liver malignancy seen at Mayo Clinic Rochester between 1976 and 2007 were identified retrospectively. After Institutional Review Board approval, thorough review of the medical records identified 36 patients who met criteria for both PBC and HCC and were therefore included in this case-series study. Seventeen of those patients, seen between 1976 and 2002, were included in a previously published study.32 Nineteen patients were found to have HCC between 2002 and 2007.

The diagnosis of PBC was defined by presence of two of the following three criteria: (1) chronic cholestatic liver disease for more than 6 months; (2) positive anti-mitochondrial antibody titer ≥1:40 or >0.1 U; and (3) liver histology with features consistent with or diagnostic of PBC, as well as the exclusion of other causes of chronic liver disease. In 25 of 36 cases, the diagnosis of HCC was confirmed via histology derived from needle biopsy examination (n = 11), surgery including resection (n = 5), at liver transplantation (n = 6), or at autopsy (n = 3). In the remaining cases, the diagnosis was confirmed via angiography (n = 3) or magnetic resonance imaging (MRI) (n = 3) or was presumed to have HCC through a combination of clinical (known cirrhosis), laboratory (markedly elevated alpha-fetoprotein level), and/or computerized tomography (CT) (n = 5) findings.

Screening and Detection of HCC.

A program for HCC surveillance in patients with PBC has been in place at the Mayo Clinic since approximately 2000 and is based on performing periodic abdominal ultrasonographic examinations and serologic tests of alpha-fetoprotein (AFP) level every 6 to 12 months in all patients with PBC. Variations due to patient and physician preference, as well as patient and physician compliance, were observed. Alternative imaging techniques were occasionally used for the purpose of surveillance due to patient or physician preference. In general, when the AFP was greater than 10 ng/mL or was more than twice the previous value, triphasic CT was performed to rule out the presence of HCC. Otherwise, when a focal lesion greater than 1 cm was detected via ultrasonography, a fine-needle aspiration or biopsy as well as other imaging tests were performed (CT with contrast media, MRI with contrast media or angiography). Due to patient preference, a few patients were diagnosed clinically using a combination of clinical (known cirrhosis), laboratory (markedly elevated AFP), and/or CT examinations.

Data Collection.

The medical records of the 36 study subjects were retrospectively reviewed. We collected laboratory data, physical findings, medical history, and social history in the year of the HCC diagnosis or the nearest calendar year within 2 years. The variables collected include age at HCC diagnosis, Mayo Risk Score (MRS) and model for end-stage liver disease score, serum AFP levels, findings on radiographic studies (CT, MRI, ultrasonography, and angiogram, when available), histological findings if available, and therapy. The patients were considered to have PBC when this diagnosis was explicitly recorded in the Mayo medical record or outside records, or by patient report if the diagnosis was made prior to the first evaluation at our institution. Duration of PBC was defined as the time between the initial diagnosis of PBC and the initial diagnosis of HCC. Relevant demographic and clinical information was abstracted from the medical records and tabulated in a database.

Statistical Analysis.

Continuous data were expressed as the mean ± standard error of the mean or medians and ranges as appropriate. Categorical data were expressed as the number of subjects (and percentage) with a specified condition or clinical variable. The detection of significant differences in continuous variables between the study groups was performed using nonparametric Wilcoxon tests. Comparison of categorical data between the study groups was performed using Fisher's exact method where appropriate. Overall survival curves were plotted for the groups of interest using the Kaplan-Meier method, and the log-rank test was used in order to compare the survival curves. Multivariate survival analysis was performed using a Cox proportional hazard model, with P < 0.05 in univariate analysis required for entry in the model. The patients whose cancers were diagnosed incidentally at autopsy or transplantation were excluded from these survival analyses. Logistic regression analyses using screening intervals (months) as an independent variable were performed to assess the association of surveillance intervals with clinical outcomes (death and subsequent treatment of HCC). All tests were two-sided, and the chosen level of significance was P < 0.05.


Patient Characteristics.

A total of 36 patients with PBC who were also diagnosed with HCC comprised our study population. The relevant demographic and clinical information for the patients at the time of HCC diagnosis are summarized in Table 1.

Table 1. Characteristics of the Study Subjects at the Time of HCC Diagnosis (n = 36)
VariablesDescriptive Statistics
  • *

    Data are reported as mean ± SEM

  • Based on hepatitis B serologies.

  • Based on hepatitis C serologies or polymerase chain reaction/RNA testing.

Female sex64%
Age (years)*69 ± 2
History of blood transfusion50%
Current or past smoking22%
MRS* (n = 34)6.88 ± 0.30
MELD* (n = 28)12 ± 1
Duration PBC (years)*13.3 ± 1.3
Portal hypertension94%
Radiographic evidence of “cirrhotic configuration of the liver”97%
Clinical decompensation 3 months prior47%
Hepatitis B (n = 27)0%
Hepatitis C (n = 23)0%
UDCA use > 6 months64%

Table 2 shows characteristics of each of the cases described in our series, including diagnosis, treatment, and outcomes. The patients whose cancers were diagnosed based on clinical grounds (cases 6, 14, 18, 21, and 26) had clinical characteristics comparable to those with more specific diagnostic tests and are included in all further analyses.

Table 2. Clinical Characteristics and Outcomes of 36 Patients with PBC and HCC
CaseSexAgeMRSDiagnosisUDCA > 6 MonthsYear of DiagnosisNodulesLargest Diameter (cm)Definite DiagnosisTreatmentStatusSurvival (Months)
  1. Abbreviations: F, female; HAE, hepatic artery chemoembolization; I, incidental; M, male; MRS, Mayo Risk Score; NS, non-surveillant; OLT, orthotopic liver transplantation; PEI, percutaneous ethanol injection; RFA, radiofrequency ablation; S, surveillant; UDCA, ursodeoxycholic acid.

3M545.57NSNo1987210Surgical pathologyNoneDead4
15M696.18SYes1996Multiple8.5Surgical pathologyResectionDead33
16F775.63NSNo199812.4BiopsyRFA x2, PEI x1Dead49
17M735.17SYes200025BiopsyResection, HAE x 9, RFA x2Alive77
18F729.66NSYes200111.5CTHAE x1Dead8
19F784.41SYes200133.7Surgical pathologyResectionDead42
20F85.SNo200213.7BiopsyRFA x1Dead28
21F737.47SYes200221.8AFP/CTRFA x1Dead15
22F885.02SYes200227.5Surgical pathologyResectionDead3
23F646.06SYes200228AngiogramHAE x1, OLTAlive52
24M745.75SYes200225AngiogramHAE x6, PEI x4Dead38
26F797.03NSYes200313CTRFA x1Dead2
27F615.25SYes200331.5ExplantationHAE x1, OLTAlive49
29F747.55SYes200312.5AngiogramHAE x1Dead2
30M847.32SYes200544.5BiopsyTherasphere x2Unknown7
31F836.64NSYes200512.4Surgical pathologyResectionDead14
32F875.73NSNo200515Surgical pathologyResectionDead1
34M545.19SYes200643ExplantationHAE x6, OLTAlive12
35F735.25SYes200721.8MRRFA x1Alive6
36M583.51SYes200712.9ExplantationHAE x2, OLTAlive8

Five patients (14%) were found to have HCC incidentally (two at autopsy and three on liver explantation). The other 31 patients were found to have HCC after further evaluation prompted by an abnormal test, including AFP levels in 5 patients (16%), CT in 10 patients (32%), ultrasonography in 15 patients (48%), and abdominal MRI in 1 patient (3%). Among these 31 patients, 17 (55%) were diagnosed based on tests obtained for the purpose of screening; these patients were referred to as the surveillant population for the purposes of this study. The remaining patients did not fulfill the criteria for surveillance, and HCC was detected using tests obtained for specific evaluation of decompensated liver disease in 8 patients (26%) and for evaluation of other symptoms/other reasons in 6 patients (19%) (such as back pain, abdominal pain, and postoperative evaluation of resected ovarian cancer). These patients will be referred to as the nonsurveillant population for purposes of this study. The reasons why patients did not fulfill criteria for surveillance included patient noncompliance, physician noncompliance, unknown (patient was seen elsewhere), and, most commonly, diagnosis of PBC/HCC before 2000. The demographics and clinical characteristics of the surveillant and nonsurveillant populations were similar (data not shown). Significant exceptions included slightly higher MRS (5.91 ± 0.33 versus 7.63 ± 0.39; P = 0.02), higher likelihood of having a diagnosis of HCC prior to January 1, 2000 (71% versus 12%; P = 0.001), and lower likelihood of having used UDCA for greater than 6 months (94% versus 36%; P = 0.001).

Severity of Liver Disease.

At time of HCC diagnosis, all patients had advanced liver disease based on MRS, evidence of portal hypertension, and radiographic evidence of cirrhotic configuration of the liver (Table 1). None of our patients had serological markers of hepatic fibrosis, transient elastography, or MR elastography performed during the course of their clinical care. Interestingly, 17 patients (47%) had clinical evidence of decompensation over the previous 3 months, which prompted further evaluation that culminated in the diagnosis of HCC and, in one case, both PBC and HCC. Histologic staging was obtained in 14 patients (39%) at the time of the diagnosis of HCC, based on samples from ultrasound-guided biopsy of the liver lesion, surgical specimens from segmental resection, and liver explants at time of orthotopic liver transplantation. One patient (7%) had evidence of stage 2 disease, and two patients (14%) had stage 3 disease on liver biopsy. The remaining 11 patients (79%) had evidence of stage 4 disease. Only six patients (17%) had a liver biopsy available within 5 years prior to their diagnosis of HCC: biopsies were reported as stage 1 in one patient, stage 2 in one patient, and stage 4 in four patients. Twenty patients (56%) had a liver biopsy greater than 5 years from their diagnosis of HCC: four patients with stage 2 PBC, eight patients with stage 3, and eight patients with stage 4. Therefore, only 12 patients (33%) would have been eligible for inclusion in a surveillance program based on a stage 4 biopsy report, according to current practice guidelines.

Number and Size of the Detected Lesions.

Table 2 shows the morphological features of HCC diagnosed incidentally and diagnosed in the surveillant and nonsurveillant population. Of 36 HCCs, 22 (61%) were unifocal and 14 (39%) were multifocal. Only six HCCs in the surveillant population were unifocal (35.3%) compared with 12 (85.7%) in the nonsurveillant population. The diameter of the largest lesion detected was similar in the unifocal and multifocal tumors (4.5 cm ± 0.6 cm versus 4.75 cm ± 0.8 cm, respectively; P = 0.29). However, it did appear that the diameter of the lesions detected in the surveillant population was smaller than those detected in the nonsurveillant population (3.8 cm ± 0.6 cm versus 4.9 cm ± 0.8 cm in unifocal HCC; 4.6 cm ± 0.8 versus 10 cm in multifocal HCC), although this did not reach statistical significance (P > 0.05 for both analyses).

HCC Treatment and Outcomes.

At the time of HCC diagnosis, surveillant patients had lower MRS (5.91 versus 7.27; P = 0.02) and model for end-stage liver disease scores (10 versus 12; P = 0.33) compared with patients in the nonsurveillant population. Patients in the surveillant population were more likely to undergo treatment for HCC compared with the nonsurveillant population (88% versus 43%; P = 0.01). Treatment for HCC in our patients is summarized in Table 2.

In our case series, eight patients underwent liver transplantation; three cases of HCC were diagnosed incidentally at time of transplantation, two patients were listed following the diagnosis of HCC, and three patients had been previously awaiting liver transplantation. One patient underwent retransplantation within 6 months due to hepatic vein thrombosis. Mean time from diagnosis to transplantation in the five nonincidental patients was 3 months (range, 1–10 months). All transplanted patients are currently alive and have been seen at the Mayo Clinic in follow-up within the last 6 months and have no evidence of recurrent malignancy.

Twenty-four deaths (67%) were observed, 15 of which were thought to be a result of HCC development. Four patients died of unknown causes (three of which had been seen within 1 year and were doing well without evidence of recurrent malignancy). Three patients died from decompensated liver disease following surgical resection for HCC (n = 2) and hepatic artery chemoembolization (n = 1). One death was secondary to recurrent metastatic ovarian disease. Ten patients have been last seen within 12 months and are alive. Two patients were lost to follow-up.

Survival Analysis.

Median survival time (after diagnosis of HCC) was 15 months (0–49) and was significantly higher in patients in the surveillant population compared with the nonsurveillant population (28 months [range, 2–82] versus 3 months [range, 0–49]; P = 0.008). Further analysis demonstrated that survival of patients with HCC detected during a surveillance program was significantly longer (log-rank, P = 0.002) than that of nonsurveillant patients (as shown in Figure 1).

Figure 1.

Kaplan-Meier curve of survival in surveillant (dashed line) versus nonsurveillant (solid line) populations (P = 0.002).

In addition to the surveillance status, we identified that MRS and subsequent treatment status were significantly associated with survival (data not shown). Therefore, we assessed the association between the surveillant status and survival after controlling for these possible confounders using the Cox proportional hazard regression method. After controlling for MRS (hazard ratio [HR] = 1.6, P = 0.03) and subsequent treatment status (HR = 1.5, P = 0.15), the effect of the surveillant status on survival was decreased yet borderline significant (from HR = 1.9, P = 0.004 to HR = 1.6, P = 0.06).

Association Between Length of Screening Intervals and Clinical Outcomes.

The median length of the screening interval was 7 months (range, 6–22 months) in the surveillant population. To evaluate the association between length of screening intervals in the surveillant population and clinical outcomes such as death and ability to provide subsequent treatment, individual logistic regression analyses were performed. The analyses conducted in the small surveillant population failed to show any significant association between the length of surveillance interval and clinical outcomes such as death (P = 0.50) or ability to provide subsequent treatment (P = 0.41).

Predicting HCC Using a Proposed Diagnostic Model.

Previously, we developed a model for prediction of the presence of HCC in patients with PBC using four clinical variables: age >70 years, male sex, history of blood transfusions, and evidence of portal hypertension (defined as presence of ascites, peripheral edema, or esophageal varices).32 In this study, we estimated the probability of HCC for the 19 patients not included in the original model who were diagnosed with HCC after 2002, based on the previously described model. The probability of HCC was predicted by using the combinations of these four variables (Table 3). Because they were not found to independently predict HCC in the original study, variables such as bilirubin and platelet count were not evaluated. Overall, 18 patients (94.7%) are identified as high-risk patients. Only one patient who developed HCC was estimated to have less than a 6% risk. This patient was a 58-year-old male with no history of smoking or blood transfusion and no evidence of ascites, peripheral edema, or esophageal varices. However, he did have evidence of splenomegaly and a cirrhotic configuration of his liver on radiographic studies. He underwent liver transplantation as treatment for his HCC, and explantation revealed a cirrhotic (stage 4) liver.

Table 3. Predicted Probability of HCC Among 19 Patients with PBC Based on Combinations of Four Clinical Variables
Clinical Variables*Predicted Probability (%)Number of Patients (n = 19)
Age >70MalePortal HypertensionBlood Transfusion
  • *

    0 = no, 1 = yes

  • Based on previously proposed model (see text for details and reference).



In this study, we describe the largest case series to date of patients with PBC who developed HCC during the course of their disease. Thirty-six cases were identified. Patients who were diagnosed with HCC during a surveillance program were more likely to undergo therapy (88% versus 43%; P = 0.01) and had a significantly better survival (log-rank, P = 0.002) independent of disease severity than those not included in a surveillance program. We determined that the combination of four main variables readily available in the clinical setting and independent of histological findings adequately predicts patients at high risk for HCC and might be useful in identifying a target population for an HCC surveillance program in PBC.

The development of HCC in PBC has been a topic of discussion for several years. Historically, patients with PBC were considered to be at low risk for HCC.33, 34 Even though recent independent studies from geographically diverse areas of the world have unequivocally shown an increased risk for HCC in PBC,10–20 data on surveillance in patients with PBC are lacking.

The risk of developing HCC is a concern due to its very high and rapid mortality rate.17 In our study, the median survival following the development of HCC in our nonsurveillant population of PBC patients was 0.25 years, and 0.90 years in the subgroup of patients who received treatment. These data are similar to the median survival following the development of HCC previously described in patients with PBC10 and that of all patients with HCC reported between 1992 and 1996.35 In contrast with the study conducted by Jones et al.,10 in which no differences in prognosis regardless of treatment modality or in survival between screened and nonscreened cases were found, our data show that patients with PBC diagnosed with HCC during surveillance were more likely to undergo treatment and had significantly improved survival after diagnosis of HCC compared with patients who were not in a surveillance program. Interestingly, Shibuya et al.18 showed that in patients with PBC who received intensive surveillance and aggressive cancer treatment, the development of HCC did not negatively impact the survival of PBC. This positive impact of surveillance on survival of patients with PBC following HCC development further emphasizes the importance of HCC surveillance in PBC.

Periodic screening for HCC of all patients with PBC is not justified. HCC surveillance should be offered when the annual risk is 1.5% or greater26 and therefore not necessary in noncirrhotic stage PBC. In contrast, in patients with advanced, cirrhotic stage PBC, the risk of HCC is comparable to patients with cirrhosis secondary to hepatitis C.13 Patients with histologic evidence of stage 4 PBC who are currently identified as the target population for surveillance26 are likely to represent only a portion of patients with advanced, cirrhotic stage disease who should be included in a surveillance program. Indeed, we found that 21% of the patients had evidence of histologic stages 2 and 3 disease at the time of HCC development. Histological staging obtained for the purpose of confirmation of HCC diagnosis (suspicious nodule biopsy, autopsy) or for HCC treatment (explantation, surgical pathology) excluded, histological staging was not available within 5 years from diagnosis of their malignancy in 83% of the patients. Strikingly, 24 patients (67%) would not have met current criteria for HCC surveillance.

In order to adequately identify all patients with PBC whose annual risk of HCC development justifies inclusion in a surveillance program, interest lies in risk prediction with variables independent of liver histology. Predictive models based on factors readily available in clinical practice and independent of histological findings are likely to be useful in this regard. Based on our recently described model,32 evidence of portal hypertension or a combination of two or more risk factors such as age greater than 70, male sex, and history of blood transfusions adequately identify patients who are at higher risk of HCC and therefore should be included in a surveillance program. Despite having been developed based on cases with advanced HCC at the time of their diagnosis, the model performed quite well in attributing high risk of HCC to the 19 patients who were diagnosed with less-advanced HCC between 2002 and 2007. It was not our intent to validate our previously constructed model, and we did not evaluate its performance in matched controls that did not develop HCC. Therefore, sensitivity, specificity, negative and positive predictive values, and accuracy of the model could not be determined. Our prediction model still requires further perfecting. In our study, the mean age at the time of diagnosis of HCC was 69 years, and one patient with HCC was not identified as a high-risk patient by the predictive model, despite evidence of portal hypertension (i.e., splenomegaly). A modified version of our current model that includes slightly younger patients and assigns higher risk to patients with manifestations of portal hypertension other than varices, ascites, and encephalopathy might determine the target population of patients with PBC for HCC surveillance more accurately. Other means to improve the performance of the model may include the use of serologic markers of fibrosis and radiographic modalities such as transient elastography (Fibroscan)36 or MR elastography37 for the detection of cirrhosis in patients with PBC. Multiple serologic markers of fibrogenesis have been studied and are available for use in commercial “panels,”38 but their clinical utility has yet to be firmly established. Recent studies have shown promising performance of serum hyaluronic acid, tissue inhibitor of matrix metalloproteinase 1, and amino terminal propeptide of type III collagen compared with liver histology in assessment of progression of fibrosis in patients with PBC.39 Unfortunately, to the best of our knowledge, no published studies address this important issue.

The AASLD HCC management guidelines recommend that surveillance should be performed every 6 to 12 months.26 The surveillance interval is related to HCC factors rather than specific liver diseases/cirrhosis etiology. The median 7-month interval of surveillance in our population seemed to confer a significant survival benefit compared with the nonsurveillant population. The observed survival benefit was not correlated to the length of the interval between surveillance tests. This suggests that the benefit is mostly from the inclusion of patients in a surveillance program, even if the patients (or physicians) are not perfectly compliant with a specific surveillance schedule.

Our study has some limitations. Due to the study design, only HCC cases were evaluated; therefore, we were unable to examine performance characteristics of our proposed model in patients who did not develop HCC, and furthermore we could not evaluate the impact of treatment with UDCA on the incidence of HCC. Because of the retrospective nature of our study and small sample size, we did not have enough power to control for the year of diagnosis of HCC, which can be a significant confounder in survival analysis. In addition, our power to evaluate the effect of surveillance interval on survival was limited. However, a prospective study with adequate power to evaluate these issues would require a large number of patients with several years of follow-up, particularly if UDCA has any chemoprotective effect, as recently suggested,40–44 and therefore is unlikely to be feasible. Another limitation includes the lack of randomization into surveillant and nonsurveillant groups, but this limitation is not unique to our study. The effectiveness of surveillance programs for HCC in other chronic liver diseases has been suggested in nonrandomized trials and one randomized controlled trial conducted in patients with hepatitis B.45 Our nonsurveillant population included patients who were evaluated for decompensation (but did not meet criteria for surveillance otherwise) and are not the ideal comparision group. However, due to the widespread use of surveillance in clinical practice, assembly of an ideal control group is not practically feasible. Nonetheless, in the absence of randomized controlled trials, observational studies such as our study can contribute useful information to answering questions related to surveillance for HCC in patients with PBC.

In conclusion, we identified 36 patients with PBC who developed HCC during the course of their disease. Patients who were diagnosed with HCC during a surveillance program were more likely to undergo therapy and had a significantly better survival than patients who were diagnosed outside of a surveillance program. Our study provides data to suggest that patients with advanced stages of PBC determined by documented histologic stage or alternatively by combinations of clinical variables such as advanced age, male sex, previous blood transfusions, and evidence of portal hypertension are at higher risk for development of HCC and constitute a target population suitable for HCC surveillance. We propose that models should be used to determine the population of patients with PBC for HCC surveillance, but the ideal model has yet to be determined. Further studies on HCC development in patients with PBC are needed so that optimal HCC surveillance recommendations in this population can be determined and included in the practice guidelines for PBC.