Community-Based Mass Ultrasonographic Screening of Hepatocellular Carcinoma among Thrombocytopenic Adults

In 2000, 414,900 men (rank 3 of male cancer death; age-adjusted mortality 13.6/10⁵) and 191,000 women (rank 6; 6.4/10⁵) died of liver cancer, including hepatocellular carcinoma (HCC; ref. 1). Taiwan is one of prevalent areas of liver cancer. It was the first cause of cancer death of men (40.5/10⁵) and the second of women (15.0/10⁵) in 2006 (2). Cancer screening is useful in that early diagnosis and prompt treatment improve the survival rates for some cancers. HCC (3) have encouraged us to attempt even earlier detection of such tumors. Benefits of HCC screening have been documented by cohort studies (4) and randomized control trials (5). Regular follow-up or surveillance of high-risk subjects for early detection of HCC has become commonplace in hospitals, especially in areas with a prevalence of chronic hepatitis B virus (HBV; ref. 6) or hepatitis C virus (HCV; ref. 7) infections. Hospital-based surveillance of patients resulted in increased early detection of HCC and improved patient survival rates (8). Although some community-based HCC screening projects have been reported previously (5, 9-13), it has not been routine. In earlier community-based screenings, subjects with elevated α-fetoprotein (AFP) were referred to hospitals for further examination after an entire community had been screened for AFP (9-11). It is relatively easy to collect blood samples in the traditional or filter paper methods and send them back to the AFP laboratory (10, 11). However, AFP alone is neither sensitive nor specific enough to detect HCC (12, 14). Therefore, AFP alone should not be recommended for screening unless ultrasonography (US) is not available (15). A combination of AFP and US is the most acceptable screening tool for HCC documented thus far (14, 16). Regrettably, it would be impossible to perform US for the entire population of any community, so a two-stage community-based HCC screening program, in which US is limited only to high-risk groups, seems more feasible. The first screening identifies the high-risk group among the larger population, and the second-stage US screening is used only for the high-risk group. A community-based HCC screening program has been conducted in Taiwan since 1991. Serum hepatitis B surface antigen (HBsAg), anti-HCV antibody, aspartate aminotransferase, alanine aminotransferase, AFP, and a family history of HCC were selected as markers for identifying members of the high-risk group (4, 12, 17, 18). Although ≥90% of HCC cases in each community were covered by inclusion criteria of first-stage screening, ∼30% of the participants were identified as belonging to the high-risk group for second-stage US screening (4, 12), and performing US on up to 30% of a community's adult population would be a heavy load. Because most HBV- and/or HCV-related HCC occur together with cirrhotic liver parenchyma (19), limited HCC screening on cirrhotic patients might therefore be cost-effective. Our pilot studies suggested that thrombocytopenia (platelet counts <150 × 10⁹/L) is a valid surrogate for liver cirrhosis and can be used to identify high-risk candidates for US HCC screening in dual HBV- and HCV-endemic areas (20). Based on these findings, this study's HCC mass screening was conducted in a HCC-endemic county that has not been screened before.

As shown in Fig. 1, the study area was Tainan County in southern Taiwan. Of the 31 townships, 15 have high HCC mortality rates (male >50/10⁵; ref. 2). The population of Tainan County is ∼1,105,000, of which 475,957 residents (43.1 %) are ≥40 years old. Focusing on the population ages ≥40 years, the countywide prevalence of anti-HCV antibody and HBsAg were 10.2% and 10.9%, respectively (21).

During April and November 2004, the Tainan County Health Bureau conducted a countywide community health screening in 216 different locations. All residents ages ≥40 years were invited by mail, telephone, and media to undergo a comprehensive health examination. Blood samples were obtained for measurements of HBsAg, anti-HCV, and AFP levels (ELISA, General Biological), biochemical tests (Hitachi 747 auto-analyzer, Hitachi), and complete blood counts (Sysmex SE9000 analyzer, Sysmex UK). A total of 17,551 males and 39,151 females eligible for the health examination were enrolled. The mean age of these 56,702 subjects was 63.0 ± 11.5 years for males and 59.9 ± 11.7 years for females (20). Subjects with thrombocytopenia (platelet count <150 × 10⁹/L) or elevated AFP (>20 ng/mL) were defined as candidates for the high-risk group and were enrolled for second-stage US screening. US was done in their communities 1 week after blood tests by nine experienced hepatologists of a medical center using a hand-carried US apparatus with a 5 to 2 MHz convex probe (Titan, SonoSite). Once suspected hepatic focal lesions were found, the subjects were referred to medical centers for further confirmation and treatment. Medical records were abstracted from referral medical centers. Diagnostic criteria of HCC were based on pathology/cytology, a combination of AFP >400 ng/mL and a positive arterial image, or arterial images of at least two modalities (22). Study protocol was approved by the Institutional Review Board of Chang Gung Memorial Hospital, Taiwan. Each participant provided informed written consent.

For estimating the sensitivity of the first screening, our previous hospital-based study of 4,042 HCC cases was cited as a reference. In the reference, prevalence of thrombocytopenia was 42% in HBsAg(+)/anti-HCV(-) HCC cases, 63% in HBsAg(-)/anti-HCV(+) cases, 50% in HBsAg(+)/anti-HCV(+) cases, and 30% in HBsAg(-)/anti-HCV(-) cases (20).

For evaluating the validity of this screening, we linked our file with the National Cancer Registration Database to find HCC cases, who were reported before the end of 2005, among our whole study population. There is National Cancer Registration System in Taiwan. All hospitals that have ≥50 beds should report the newly diagnostic cancer cases to this system. These hospitals covered most but not all incident cases of malignancy. Not only pathologically proven cases but also clinical diagnosed cases should be reported. International Classification of Diseases Codes 9th edition (ICD9) was used in this registration system. All the primary liver cancer (ICD9=155) would be included in our analysis. In Taiwan, >90% of primary liver cancers are HCC (23). Because the screening ended at the end of 2004, the analysis was divided into three periods: (I) during screening (from April to December 2004), (II) within 6 months after screening (from January to June 2005), and (III) between 6 and 12 after screening (from July to December 2005).

The continuous variables were described by mean ± SD. Categorical variables were presented by percentage. The difference between groups was determined by χ² test, and the trend was examined by χ² test for a linear trend. The α level was set at 0.05. Diagnostic validity was expressed by sensitivity, specificity, accuracy, positive predictive value (PPV), and negative predictive value (NPV).

A total 3,242 (5.7%) residents (male/female, 1,415/1,827; ages 66 ± 10 years) meeting the criteria for “high risk,” including 2,928 (90.3%) with thrombocytopenia, 87 (2.7%) with AFP elevation, and 273 (7.0%) with both, were enrolled for US screening. Of these, 465 (14.3%) were positive for HBsAg, 1,291 (39.8%) were positive for anti-HCV, and 123 (3.8%) were positive for both and the remaining 1,363 (42.1%) were neither (Table 1).

Figure 2 is the flowchart of this screening. A total of 2,983 (92.0%) underwent US examination. In addition to 42 known HCC cases, 137 suspected cases of HCC were referred to medical centers for confirmation; 124 (90.5%) cases complied with referrals and 72 patients (58.1%; PPV of second-stage US screening) were confirmed as cases of HCC. Overall PPV of first-stage screening was 2.41% (72 of 2,983). Table 2 shows the number and percentage of confirmed HCC cases in groups by screening markers and viral etiology. Subjects with AFP elevation alone had the highest PPV (25.4%, 17 of 67, 1 in 4 cases), subjects with thrombocytopenia alone had the lowest (0.78%, 21 of 2,697, 1 in 128 cases), and 15.5% (34 of 219, 1 in 6 cases) of the subjects had both AFP elevation and thrombocytopenia. The PPV increased from cases without HBV or HCV infection (0.47%, 6 of 1,286, 1 in 214 cases), with HBsAg alone (3.18%, 14 of 440, 1 in 31 cases), and with anti-HCV alone (4.09%, 47 of 1,151, 1 in 24 cases) to dual infection (4.72%, 5 of 106, 1 in 21cases; P < 0.001, χ² test for linear trends). The PPV of cases with HBV and/or HCV infection (3.89%, 66 of 1,697, 1 in 26) was significantly higher than that of without (0.47%, 6 of 1,286, 1 in 214 cases; P < 0.001). Three of 6 detected non-B, non-C HCC cases had normal AFP levels. Using thrombocytopenia as the only marker of first-stage screening, 23.6% (17 of 72) of the HCC cases would have failed to have been detected, whereas using AFP alone would have failed to detect 29.2% (21 of 72). As mentioned in Materials and Methods, our previous hospital-based study was cited as a reference (20). As shown in Table 3, we estimated that there should be 101 HCC cases among the entire population of participants. Using AFP > 20 ng/mL as the only marker for the first screening, 0.64% (364 of 56,702) of the population would be identified as belonging to the high-risk group, of which 50.5% (51 of 101, sensitivity) might be determined to be cases of HCC. Using thrombocytopenia, 5.33% (3,205 of 56,702) would be placed in the high-risk group, with a sensitivity of HCC detection should be 55.5% (55 of 101). Using a combination AFP > 20 ng/mL and thrombocytopenia as first-stage screening markers, 5.7% (3,242 of 56,702) of the population should be identified as members of high-risk group, and the estimated sensitivity should be 71.3% (72 of 101).

As shown in Table 4, there should be 119 HCC cases that have been diagnosed as HCC in whole screening population during this period. Besides 72 confirmed HCC cases, 17 subjects among low-risk group and another 30 from high-risk group have been registered as cases of HCC before the end of 2005. The incidence of primary liver cancer in the high-risk group (102 of 3,242 = 3,146/10⁵) is 100 times higher than that in the low-risk group (17 of 53,460 = 32/10⁵; P < 0.001). Among 259 nonresponders to second-stage US, 7 were registered as cases of HCC during screening period and another 3 were reported within 1 year. Eight of the 10 were with AFP levels >20 ng/mL. Among 2,804 subjects negative for second-stage US screening, 15 were registered to be cases of HCC within 1 year after screening. None was registered during screening period and 11 were diagnosed between 6 and 12 months after screening. All of them were with low platelet counts and 6 of them were also with high AFP level. Another 5 US-positive cases were reported as cases of primary liver cancer by the National Cancer Registration Database, and 3 of them were registered during screening period. These 5 cases should be defined as true positive HCC cases in the second-stage screening. The total number of true positive should be 77 (Table 5). As summarized in Table 6, the sensitivity, specificity, accuracy, PPV, and NPV of first screening were 85.7%, 94.5%, 94.4%, 3.15%, and 99.97%, respectively. Those of the whole two-stage screening were 64.7%, 99.9%, 99.8%, 56.2%, and 99.9%, respectively. Limited to the high-risk group selected by first-stage screen, those of second-stage US screening were 75.5%, 98.1%, 97.4%, 56.2%, and 99.2%. Ten of the 25 false-negative cases were US nonresponders.

The 72 confirmed cases, 41 men and 31 women, were ages 68.1 ± 8.8 years. Their largest tumor diameters were ≤3 cm in 27 (37.5%) cases, between 3 and 5 cm in 23 (31.9%) cases, and ≥5 cm in 22 (29.2%) cases. Only 5 (6.9%) cases had progressed beyond the intermediate stage of Barcelona Clinic Liver Cancer staging system and were recommended to receive conservative treatment only (17). Of 53 cases treated by modern medicine, initial treatment modalities were surgical resection in 13 cases, local ablation in 8 cases, and transcatheter arterial embolization in the remaining 32 cases.

Although there were several reasons of thrombocytopenia, the major reason should be chronic hepatitis C infection especially in HCV-endemic area. In a published article using the same population of this study, the village- and township-specific rates of thrombocytopenia are significantly correlated to their rates of anti-HCV. It concluded that excess rates of thrombocytopenia than general population should be due to high prevalence of chronic HCV infection (21). A pilot study showed that thrombocytopenia is an effective surrogate of liver cirrhosis in HCV-endemic area but not in low HCV prevalent areas (20). In this study, the over all prevalence of anti-HCV was as high as 10.2%. Thrombocytopenia can be used as a screening marker. The aim of first-stage screening in a two-stage screening design is identification of high-risk groups. Our pilot studies reported only ∼50% of HCC cases can be detected by such means (20). Although the recommendation 4 in American Association for the Study of Liver Diseases HCC practice guideline wrote “AFP alone should not be used for screening unless ultrasound is not available (level II) (15),” Trevisani et al. reported that the best discriminating AFP value was 16 ng/mL. A value of 20 ng/mL (above which investigations for HCC are recommended) had equivalent sensitivity (60.0 versus 62.4%) and specificity (90.6 versus 89.4%; ref. 24). To increase the sensitivity of mass screening, serum AFP was added as a another screening marker.

To understand the validity of cancer screening, the number of false negative should be estimated or identified. In the article, we designed two approaches to elucidate this issue. Firstly, we cited distributions by viral etiology of HCC cases in a nearby medical center. Under the assumption of different rates of thrombocytopenia in groups of different viral etiology, the numbers of false-negative cases in each group were estimated. Based on the estimation, there should be 101 HCC cases and 72 were detected in this screening with a sensitivity of 71.3%. The criticism of this design is that the distributions of viral etiology between hospital and community might different. Because the viral etiology of HCC is with geography difference (25), large sample size in the nearby hospital might minimize the bias. Secondly, we linked our file with National Cancer Registration Database to find registered cases of liver cancer (ICD9=155) within 1 year after screening as the false-negative cases. The results of data linkage showed that there should be 119 cases of liver cancer and 77 were detected in the screening with a sensitivity of 64.7%. Although there were still some limitations of this design, such as incomplete registration of National Cancer Registration Database, inclusion cases of primary liver cancer other than HCC, and misclassification of some newly developed true-negative cases as false negative, it is the most reliable and available information. The estimations of these two quite different designs were compatible. This newly designed mass HCC screening showed that only 5.7% of participants could be defined as members of the HCC high-risk group and had a high coverage or sensitivity rate of 64.7% to 71.3% of all HCC cases. This screening design is therefore feasible and sensitive.

The sensitivity (64.7%) of the two-stage screening, estimated according to National Cancer Registration Database, seems unsatisfied. There were 25 HCC cases in the high-risk group not detected by second-stage US screening (false-negative). Ten of the 25 false-negative cases were nonresponders of the second-stage US screening. Eight of the 10 were with AFP levels >20 ng/mL and 6 of the 8 cases were reported during screening period. It means that some subjects skipped US screening, and visited hospital for further study, once AFP elevation were informed. In this study, the response rate of subjects with high AFP was only 78.6% (286 of 364). Their high absent rate resulted in lower sensitivity of this screening. US technique is operator dependent. Among 2,804 subjects with negative results of second-stage US screening, 15 were reported to been cases of HCC within 1 year after screening. None was reported during screening period, and 11 were diagnosed between 6 and 12 months after screening. All of them were with low platelet counts and 6 of them were also with high AFP level. Some of the 15 HCC cases should be newly developed after screening rather than lost by the previous US screening. Inclusion of these cases might overestimate the case number of false-negative and decrease the sensitivity of this screening. Based on the above two points of view, the sensitivity (64.7%) of this screening design should be underestimated.

In this large-scale study, a total of 56,702 adults ages ≥40 years were screened and 72 detected cases were confirmed as HCC. The prevalence of HCC was 127/10⁵ or 3 to 5 times higher than the annual incidence or mortality in Taiwan as a whole (2). This high prevalence rate may result from more asymptomatic HCC cases being detected earlier. Only 5 (6.9%) of 72 confirmed cases were beyond the intermediate Barcelona Clinic Liver Cancer stage (13) and were recommended to receive conservative treatment only. More than half (n = 37; 51.4%) were detected in the very early or early stages and detected early enough to have a chance for effective treatment.

The mean age of detected cases was 68.1 years, and most of the confirmed cases (52 of 72; 72%) in the current study were positive to anti-HCV. The mean ages of HCV-related HCC patients in Taiwan is 65.1 years (25). A study from the United States showed that the median age of HCC patients at diagnosis is 74 years, and most patients (91%) were >65 years (26). Of Japanese HCC patients, ∼80% were anti-HCV positive and were >60 years old (16). An Italian study concluded that elderly patients (≥70 years old) with HCC have a worse prognosis compared with nonelderly patients. Such a difference seems to be a consequence of undertreatment (27). On the other hand, a Japanese study reported that an advanced stage of HCC, not advanced age, influenced the survival rates of these elderly patients (≥80 years old; ref. 28). Therefore, older high-risk cases should be screened and older HCC cases should be treated in the same manner as nonelderly patients. As mentioned earlier, the other factor was a platelet count of <100 × 10⁹/L. Thrombocytopenia, indicating advanced liver cancer, has been reported to be a poor prognostic factor of curative treatments, such as resection (29), radiofrequency ablation (30), and percutaneous ethanol injection therapy (31). Besides the HCC itself, hepatic failure is another major cause of death among HCC patients. HCC screening can detect early HCC but not early liver cancer. This indicates that medical care for the complications of liver cancer might improve patients' survival rates. Antiviral treatment for cases with chronic HBV (32, 33) and HCV (34) infections decreased their incidence of HCC and hepatic failure. Marked AFP elevation (>400 ng/mL) correlated with poor differentiation and extended invasion (35). It has also been one of the poor prognostic factors in the Cancer of the Liver Italian Program score (36) and several analyses of the survival rates for resection (37), radiofrequency ablation (38), and transcatheter arterial embolization (39). In the present study, we used platelet count (<150 × 10⁹/L) and AFP level (>20 ng/mL) as screening markers. The screen levels of screening were not so severe than those from the previously mentioned reports of poor prognoses, such as platelet count <100 × 10⁹/L or AFP level >400 ng/mL.

Abdominal ultrasound as a surveillance tool has shown a sensitivity of 71% to 73%, a specificity of 93%, and a PPV of 14% to 73% in some studies (14). It is a well-documented tool for HCC screening (15, 16). The rates are similar to those of this study. Due to cost-benefit considerations, it is impossible to conduct a US screening on the entire population of any community. As a result, the two-stage HCC screening design has tried to limit US screening to members of the high-risk group only. However, there still remains the controversy of “Who should be defined as being in the high-risk group for HCC and should be screened by US?” Table 7 shows four community-based two-stage HCC screenings with different designs. Cost of screening tools, sensitivity of the screening tests, and PPV of the high-risk group are all important issues. In the earliest study (model 1), the subjects with elevated AFP were referred to a hospital for further examination after the entire community's screening for AFP (11). Although it is easy to collect blood samples in community, AFP alone should not be recommended for screening (15). The other community-based HCC screening program (model 2) has been conducted in seven townships in Taiwan since 1991. Six markers, including HBsAg, anti-HCV, aspartate aminotransferase, alanine aminotransferase, AFP, and a family history of HCC, were selected as the first-stage markers for high-risk group identification (4, 12, 17, 18). Although the sensitivity of the first screening markers should be ≥90%, ∼30% of all participants should be enrolled in second-stage US screening. Unfortunately, the high cost of screening tools in the first stage and the heavy load of second-stage US are two limitations of this design. Thrombocytopenia was the only marker for first-stage screening, and AFP and US were employed as tools for the second-stage screening (20), creating what an economical version of our previous design (model 3). However, the sensitivity is only ∼50%. The current design used thrombocytopenia and AFP as markers for first-stage screening (model 4). Shifting AFP from the second stage (model 3) to the first stage (model 4) increased sensitivity to an estimated 71.3% but also increased the cost of the new design. Only 5.7% of the participants were enrolled in the second-stage US. The six-marker design (model 2) had higher sensitivity but also a higher cost, so for budget and feasibility considerations, the present design should be a reasonable alternative. Another issue is postscreening surveillance of the high-risk group. Surveillance is recommended for patients with chronic HBV and HCV infections (15). Model 2 provides the information on HBsAg, anti-HCV, and US diagnosis of liver parenchyma after the first cross-sectional screening, which is essential for long-term surveillance. In the present study (model 4), >40% of the subjects in the high-risk group were negative for HBsAg and anti-HCV. In the second-stage US screening, thrombocytopenic subjects without chronic HBV and HCV infection were proven to have lower risk for HCC. We therefore recommend that HBsAg and anti-HCV should be checked among the high risk identified by the current design and that high-risk subjects who are either HBsAg or anti-HCV positive should be enrolled into the surveillance program.

Response rates influence results, but this is inevitable in any community-based study. This comprehensive screening, which included not only liver tests but also lipids, sugars, chest X-rays, and bone marrow density, should have helped minimize selection bias. Large numbers (n = 56,702) of participants from every part of the study area increased the representativeness of the study population. A total of 2,983 (92.0%) identified high-risk cases responded to US, and 124 (90.5%) suspected cases complied with referrals for further confirmation. Although the PPV may have been underestimated, such high response rates may have limited the negative influence.

For obvious high-risk groups, well-documented screening tools, and good treatment results of early-stage cases, HCC should be screened for in either a hospital-based or a community-based design in high endemic areas. Multiple factors, such as budget, manpower, validity, and feasibility, should be considered in the design of the screening methods. In this study, we present a new design of mass HCC screening, discussing its effectiveness and limitations. This economical design enrolled only 5.7% of participants for US and could detect 64.7% to 71.3% of all HCC cases early enough for them to be successfully treated. Screening for HCC using the current design should prove to be economical and effective and should be applied to those communities with a high prevalence of HCC, especially those areas with limited budgets, and seldom or never been screened before.

No potential conflicts of interest were disclosed.

We thank all 31 primary health-care centers in the study area for cooperation.