Pyogenic Liver Abscess as a Warning Sign for Primary Liver Cancer: a Nationwide Population-based Study

Primary liver cancer (PLC), including hepatocellular carcinoma and intrahepatic cholangiocarcinoma, were the fifth most common cancers worldwide (Jemal et al., 2011). Taiwan is one of the most endemic countries with an ageadjusted annual incidence of 36.82 per 100,000 persons, and a mortality rate of 27.12 per 100,000 persons in 2008 (Taiwan Cancer Registry, 2012). The fatality of PLC is extremely high despite some progress having been made in combined modality treatment (Montomoli et al., 2011; El-Serag HB et al., 2011). Hence, identifying patients at higher risk for PLC is critical to improve the treatment outcome. Pyogenic liver abscess (PLA), as the result of bacterial infection with subsequent inflammatory reaction and pus formation of the liver parenchyma, is endemic in Taiwan. The annual incidence of PLA has increased steadily in Taiwan from 11.15 per 100,000 individuals in 1996 to 17.59 per 100,000 individuals in 2004 (Tsai et al., 2008). PLA has been regarded as the initial manifestation of hepatobiliary neoplasms in several reports with a range of 3 to 10 cases (Okuda et al., 1991; Yeh et al., 1998;


Introduction
Primary liver cancer (PLC), including hepatocellular carcinoma and intrahepatic cholangiocarcinoma, were the fifth most common cancers worldwide (Jemal et al., 2011). Taiwan is one of the most endemic countries with an ageadjusted annual incidence of 36.82 per 100,000 persons, and a mortality rate of 27.12 per 100,000 persons in 2008 (Taiwan Cancer Registry, 2012). The fatality of PLC is extremely high despite some progress having been made in combined modality treatment (Montomoli et al., 2011;El-Serag HB et al., 2011). Hence, identifying patients at higher risk for PLC is critical to improve the treatment outcome.
Pyogenic liver abscess (PLA), as the result of bacterial infection with subsequent inflammatory reaction and pus formation of the liver parenchyma, is endemic in Taiwan. The annual incidence of PLA has increased steadily in Taiwan from 11.15 per 100,000 individuals in 1996 to 17.59 per 100,000 individuals in 2004(Tsai et al., 2008. PLA has been regarded as the initial manifestation of hepatobiliary neoplasms in several reports with a range of 3 to 10 cases (Okuda et al., 1991;Yeh et al., 1998; Wen-Kuan Huang 1 , Yung-Chang Lin 1 , Meng-Jiun Chiou 2 , Tsai-Sheng Yang 1 , John Wen-Cheng Chang 1 , Kuang-Hui Yu 3 , Chang-Fu Kuo 3 , Lai-Chu See 2 * Chong et al., 2009;Huang et al., 2009;Li et al., 2012). Lin et al. reported 2.15% of patients with underlying hepatocellular carcinoma (HCC) presented with PLA as the initial manifestation (Lin et al., 2011). They arbitrarily defined patients with PLA as the initial manifestation of underlying HCC when the diagnosis date of HCC was within 60 days of diagnosis of PLA, which might lead to misclassification bias regarding HCC status (Copeland et al., 1977). In addition, lack of a matched non-PLA control group in Lin's study is a drawback.
The high incidences of both PLA and PLC in Taiwan provide a unique opportunity to examine the association between these two diseases. Hence, in this study, we used a nationwide population-based cohort in Taiwan to estimate the incidence of PLC after the diagnosis of PLA and to explore the risks of PLC by including a non-PLA control group. Data is further stratified by time after the diagnosis of PLA to explore the role of PLA on PLC.

Taiwan National Health Insurance Research Dataset
The primary data source of this study was retrieved

Study design
A hybrid study of cohort with comparative group was used. Both the 2000 and 2005 longitudinal cohorts were used for this study to obtain patients with PLA (PLA group) and without PLA (control group). The PLA group were adult (age≥18) patients who were hospitalized due to a new diagnosis of PLA (International Classification of Diseases, 9th Revision, Clinical Modification [ICD-9-CM], code 572.0) between 2000 and 2008. The diagnosis of PLA was based on evidence of imaging studies and antibiotics treatment during admission. The index date was defined as the date of the first hospitalization for PLA. The control group was randomly selected from the remaining patients in the database, which matched the PLA group in terms of sex, year of birth, and the index date when patients sought treatment for diseases other than PLA. The ratio of the PLA group and the control group was 1:1. Subjects were followed until December, 2008. We excluded patients diagnosed with PLC before the index date of PLA for both the PLA group (n=93) and the control group (n=752).

Statistical analysis
Descriptive statistics, such as mean, standard deviation (SD), frequency and percentage were used to summarize characteristics of the sample. Chi-square test and independent t-test were used to compare the data between the two study groups, where appropriate. The incidence rates of PLC were computed using the Kaplan-Meier method. The univariate Log-rank test and the multivariate Cox hazard model were used to examine the effect of PLA on PLC. Stratification of the data by different time intervals after diagnosis of PLA was made to examine the role of PLA on PLC. The significance level is 0.05. All statistical analyses were performed using SAS (version 9.1, SAS Institute, Cary, NC).

Characteristics of the patients
From 2000 to 2008, 1,987 PLA cases and 1,987 matched controls were obtained. The mean age was 60.2 years (SD, 14.8 years) and 63.8% were male. The PLA group had a significantly higher rate of liver cirrhosis, hepatitis B infection, hepatitis C infection, chronic liver disease, nonalcoholic fatty liver disease, cholelithiasis, cholangitis and diabetes mellitus than the control group ( Table 1).

Incidence of primary liver cancer
PLC occurred in 56 patients (1.4%), including 48 (2.4%) in the PLA group and 8 (0.4%) in the control group. 50% of PLC occurred within one month after the diagnosis of PLA, and 73% of PLC occurred within one year after the diagnosis of PLA. The incidence rate of PLC was higher in the PLA group than in the control group (601.5 vs. 93.3 per 100,000 person-years) ( Table 2).

Risk factors of primary liver cancer
Univariate analysis reveals that liver cirrhosis, hepatitis B infection, hepatitis C infection, chronic liver disease, non-alcoholic fatty liver disease, PLA, and sex was significantly associated with PLC. The unadjusted hazard ratio (HR) of PLC was 6.2 times greater (95% CI, 2.9-13.0; p < 0.0001) for patients with PLA than for those without PLA. Cox proportional hazard regression shows that liver cirrhosis, hepatitis B infection, hepatitis C infection, PLA, and sex were independently associated with a higher risk of PLC. The adjusted hazard ratio (HR) of PLC was 3.4 times greater (95% CI, 1.6-7.3; p =0.0015) for patients with PLA than for those without PLA (Table  3). The hazard ratio of PLC stratified by follow-up time after initial PLA diagnosis is shown in Figure 1. The PLC risk for the PLA group was significantly higher within the first year of follow-up after PLA diagnosis (HR: 35.4, 95% CI: 4.9-258.5) as compared with the control group and became insignificant (HR: 2.0, 95% CI: 0.8-4.9) after the first year after PLA diagnosis.

Discussion
To our knowledge, this is the first large populationbased cohort study to estimate the incidence of PLC in patients with PLA. Our results show that the incidence rate of PLC (601.5 per 100,000 person-years) for the PLA group was much higher than that of the control group (93.3 per 100,000 person-years). The PLC risk for the PLA group was significantly higher within the first year after PLA diagnosis (HR: 35.4, 95% CI: 4.9-258.5) as compared with the control group and became insignificant (HR: 2.0, 95% CI: 0.8-4.9) after first year after PLA diagnosis.
The overall increased risk of PLC in patients with PLA can be partly explained by the high proportions of risk factors of PLC in the PLA group (Table 1). The short temporal duration between PLA and PLC suggests that PLC masquerades as PLA before its definite diagnosis. The mechanisms that probably explain how PLC manifests as PLA are spontaneous tumor necrosis or biliary obstruction caused by tumor thrombi superimposed with bacterial infection (Yeh et al., 1998).
On the other hand, the link between infection/chronic inflammation and cancer has been demonstrated by several studies (Lochhead et al., 2007;Matsuzaki et al., 2007;Grivennikov et al., 2010). Chronic infection with hepatitis B virus and hepatitis C virus were considered the dominant risk factors in hepatocellular carcinoma development (Beasley et al., 1981;Tsukuma et al., 1993). Chronic cholangitis and intrahepatic stone disease were strongly associated with intrahepatic cholangiocarcinoma (Su et al., 1997). PLA, as an inflammatory process, damaged hepatic parenchyma tissue and is likely to participate in the development of liver tumorigenesis. However, it takes considerable time in inflammation-mediated carcinogenesis (Coussens et al., 2002). Our findings revealed that PLC risk was significantly higher within the first year of follow-up after PLA diagnosis but not afterward, suggesting that PLA is a warning sign rather than a trigger factor of PLC. The reasons to differentially diagnose between PLA and PLC are as follows. First, PLC mimicking PLA remains a diagnostic challenge. It is difficult to distinguish between PLA and PLC by specific clinical symptoms, laboratory tests and imaging studies (Shimizu et al., 2011;Li et al., 2012). Yeh et al. suggested male gender, hepatitis B and/or C infection and cirrhosis might provide meaningful clues to underlying liver cancers (Yeh et al., 1998), which was supported by our findings. Furthermore, histopathologic examination should be taken into consideration when physicians raised concern about the possibility of PLC. Second, in the study conducted by Lin et al, patients diagnosed with HCC who initially manifested as PLA had a worse prognosis than those without PLA (Lin et al., 2011). They supposed that management of PLA may lead to delayed diagnosis and that drainage procedures may cause tumor to spread if physicians are not on heightened alert for PLC. In the study by Yeh et al., the prognosis of patients with HCC presented as PLA was poor, with a mean survival of 3.5 months (Yeh et al., 1998). Taken together, it is worth noting that patients with PLA should be intensely monitored to detect PLC as soon as possible.
The strength of our study is the use of a nationwide database as a population-based sample that was highly representative of the Taiwanese population. A large sample size, a long follow-up period, and rich co-morbid disorders allow us to examine the risk of PLA in PLC longitudinally and multivariately. The study has some limitations inherent to administrative data. First, although major risk factors were adjusted, other possible risk variables, such as smoking, alcohol use, and obesity, were not examined (McGlynn et al., 2011). Second, the lack of microbiologic data was an obstacle to investigation of the association between causative pathogens and PLCs. Finally, information regarding the severity and duration of PLA were not reliably available.
In conclusion, patients with PLA were associated with a significantly higher risk of PLC relative to patients without PLA, especially within the first year after the diagnosis of PLA, suggesting that PLA is a warning sign for PLC. Hence, patients with PLA should be intensely monitored to quickly detect PLC. Further studies are needed to clarify the mechanism underlying the relationship between PLA and PLC.