Pneumonia is an independent risk factor for pyogenic liver abscess: A population-based, nested, case-control study

Background Bacteremic pneumonia is considered a potential cause of distal organ abscess formation. Therefore, we hypothesize that pneumonia is a risk factor for pyogenic liver abscess (PLA).The aim of this study is to explore the association between pneumonia and PLA. Methodology/Principal findings A nationwide, population-based, nested, case–control study was conducted using data from the Taiwan National Health Insurance Research Database. In total, 494 patients with PLA and 1,976 propensity score matched controls were enrolled. Conditional logistic regression was used to estimate adjusted odds ratios (aORs) in patients with exposure to pneumonia before PLA. After matched and adjusted for confounding factors including age, sex, urbanization, income, chronic liver disease, alcohol-related disease, biliary stone, chronic kidney disease, diabetes mellitus, chronic liver disease, and cancer, hospitalization for pneumonia remained an independent risk factor for PLA with an aORs of 2.104 [95% confidence interval (CI) = 1.309–3.379, p = 0.0021]. Moreover, the aORs were significantly higher among patients hospitalized for pneumonia within 30 days (aORs = 10.73, 95% CI = 3.381–34.054), 30–90 days (aORs = 4.698, 95% CI = 1.541–14.327) and 90–180 (aORs = 4.000, 95% CI = 1.158–13.817) days before PLA diagnosis. Conclusion Pneumonia is an independent risk factor for subsequent PLA. Moreover, hospitalization for pneumonia within 180 days before PLA diagnosis was associated with an increased risk of PLA.


Introduction
Pyogenic liver abscess (PLA) is a potentially fatal bacterial infection of the hepatic parenchyma, with a 5.6%-23% mortality rate [1,2]. Moreover, complications, such as metastatic infection to the lungs, central nervous system, and eyes, increase patient morbidity and mortality [3,4]. Most PLA cases develop because of systemic bacteremia or intraabdominal infections [5]. An immunocompromised status, diabetes mellitus, liver cirrhosis, and advanced age are wellknown predisposing risk factors for PLA [6,7].
However, the etiologies of and risk factors for PLA have continued to evolve; recently reported risk factors include zolpidem and proton pump inhibitor use and splenectomy [8][9][10]. In Taiwan, the most common pathogen of PLA has also changed from Escherichia coli to Klebsiella pneumoniae since the 1980s [11]. Notably, in addition to causing PLA, K. pneumoniae is also a dominant pathogen of bacteremic community-acquired pneumonia in Taiwan [12]. In South Africa, pneumonia accounted for up to 62% of K. pneumoniae bacteremia [13]. Furthermore, many pneumonia survivors may acquire bacterial infections after their original pneumonia infection. Brain abscess, psoas muscle abscess, para-aortic arch abscess, thyroid abscess, splenic abscess, and infraorbital abscess have been reported after pneumonia [14][15][16][17][18][19]. Bacteremic pneumonia is considered a potential cause of distal organ abscess formation. Therefore, we hypothesize that pneumonia is a risk factor for PLA.
Understanding the risk factors for PLA is clinically crucial because rapid diagnosis followed by appropriate management of PLA can improve patient outcomes and prevent complications. Here, we conducted a case-control study by using a national database from Taiwan and explored the association between pneumonia and PLA.

Database and settings
This nested case-control study was conducted using registration and claims data of 2009-2013 obtained from the Longitudinal Health Insurance Database 2010 (LHID2010), a subset of the National Health Insurance (NHI) Research Database (NHIRD), managed by the Taiwanese National Health Research Institutes. This dataset contains all information regarding sociodemographic status such as sex and date of birth; outpatient, inpatient, and emergency care; and prescription drugs of 1 million of the NHIRD beneficiaries, randomly sampled from the 2010 registry of 23 million of the NHIRD beneficiaries. The disease diagnosis is based on the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes. The diagnosis coding in this dataset is highly reliable because all insurance claims have been monitored by medical reimbursement specialists and through peer review. Our study protocol was approved by the Institutional Review Board of Chung Shan Medical University Hospital (CSMU No.: 15061). The requirement for written informed consent from the participants was waived because the LHID2010 contains deidentified data.

Study population
Adult patients aged >20 years diagnosed as having new onset PLA (ICD-9-CM: 572.0) between January 2010 and December 2013 were categorized as the case group. The date of first hospitalization for PLA was defined as the index date. Patients with previous PLA diagnoses in the data set before January 2010 were excluded (n = 141).
For each PLA case, the first round matching was conducted to select controls at a 1:100 full matching by age, sex, urbanization and income. The second round matching was applied to reduce the confounding of co-morbidities such as diabetes mellitus (DM; ICD-9-CM: 250),

Definition of pneumonia
We defined pneumonia (ICD-9-CM: 481, 482, 483, and 485-486) as the primary diagnosis, which required hospital admission and intravenous antibiotic treatment. A study demonstrated that the identification of inpatient pneumonia using this method has a positive predictive value of 88% [20].

Statistical analysis
Categorical variables were delineated as numbers and percentages and compared using the chi-squared or Fisher exact test, where appropriate. Continuous data were delineated as means ± standard deviations and compared using the independent t test. Conditional logistic regression was used to estimate crude and adjusted odds ratios (aORs) with a 95% confidence interval (CI) for case group compared with control group. In multivariate analysis, we adjusted for comorbidities that may have predisposed a patient to PLA. In the subsequent analyses, we stratified pneumonia events into the number of patients hospitalized for pneumonia and duration of the hospitalization. Statistical analysis was performed using SPSS (version 18.0; SPSS Inc., Chicago, IL, USA) and SAS (version 9.4; SAS Institute, Cary, NC, USA). A p value <0.05 indicated statistical significance.

Results
After propensity score matching, a total of 494 cases of new onset PLA and 1,976 control patients without PLA were enrolled in this study for final analysis. Table 1 lists the demographic characteristics of the patients and their distribution in the case and control groups. The mean age was 60.11 ± 14.43 and 60.09 ± 14.46 years in the case and control group, respectively; no significant difference was noted in age, sex, and socioeconomic status between the two groups after matching. However, compared with the control group, the case group had a significantly higher proportion of the biliary stone (4.05% vs 2.02%) and previous hospitalization for pneumonia (5.67% vs 2.83%). During the study period, the trend of incidence of PLA was correlated to the trend of incidence of hospitalization for pneumonia (Fig 2). Table 2 showed the result by unconditional logistic regression analysis, hospitalization for pneumonia was an independent risk factor for PLA (aORs = 2.136, 95% CI = 1.289-3.537, p = 0.0032) after the adjustment for confounding factors, such as age, sex, urbanization, income, co-morbidities(including, COPD, alcohol-related disease, biliary stone, CKD, DM, chronic liver disease, and cancer). Another risk factors for PLA was biliary stone (aORs = 1.959, 95% CI = 1.112-3.452).
Our main findings, the results of conditional logistic regression that adjusted for co-morbidities, are shown in Table 3. The aORs of PLA for exposure to pneumonia within 2 year prior to index date was 2.104 (95% CI = 1.309-3.379) in model 1. To further explore the association between hospitalization for pneumonia and PLA onset, the time interval between date of last hospitalization for pneumonia and the index date was calculated for each patient. The aORs for the time interval between hospitalization for pneumonia and PLA onset were listed, including hospitalized for pneumonia 30 days (aORs = 10.73, 95% CI = 3.381-34.054), 30-90 days (aORs = 4.698, 95% CI = 1.541-14.327) and 90-180 (aORs = 4.000, 95% CI = 1.158-13.817) days before the index date, were significantly higher than those of patients not hospitalized for pneumonia in model 2. Table 4 indicates the five specific models we used for subgroup analysis to explore the risk of PLA and number of pneumonia admissions within 1, 3, 6, 12, and 24 months before the index date. These five models demonstrated that patients having an increased number of hospitalizations for pneumonia before the index date had a significantly higher aORs for the development of PLA.  Table 5 shows the underlying comorbidities of patients who were hospitalized for pneumonia in PLA and no PLA group. Compare with no PLA group, the ORs of pneumonia among all comorbidities were not higher in PLA group. Table 6 lists the etiology and pathogens of pneumonia in patients hospitalized for pneumonia. Only the infection of K. pneumoniae demonstrated a significant difference between the case and control groups (p = 0.0383).

Discussion
According to our review of relevant literature, this is the first study to demonstrate that previous hospitalization for pneumonia is an independent risk factor for subsequent PLA (aORs = 2.104, p = 0.0021) by using a nationwide administrative database. Our findings also demonstrate that time interval between hospitalization for pneumonia and PLA onset as well as the number of hospitalizations for pneumonia is associated with the risk of PLA. A potential explanation for our results is the bacterial seeding that occurs in the hepatic parenchyma from the bloodstream during pneumonia. In Taiwan, Mycoplasma pneumoniae, Chlamydia pneumoniae, Streptococcus pneumoniae, and K. pneumoniae are common pathogens of community-acquired pneumonia [21]. Among these micro bacteria, K. pneumoniae is the pathogen most commonly isolated from patient with pneumonia of higher severity [12,21]; it is also the major cause of bacteremic pneumonia and PLA in Taiwan. The situation in Taiwan is different from Western countries: the prevalence of liver abscess is lower in most Western populations [22]. In the United States, the most common bacterial pathogens of pneumonia are Strep. pneumoniae, M. pneumoniae, Staphylococcus aureus, and Legionella pneumophila [23]. Notably, K. pneumoniae infection is relative uncommon in Western populations [13]. The high rate of infection of the K1 and K2 capsular serotype strains of K. pneumoniae in East Asian countries may explain the aforementioned association between pneumonia and PLA in Taiwan [24][25][26]. Our study also demonstrated a significant difference in K. pneumoniae pneumonia between the case and control groups, but these groups contained only 5 and 2 patients with K. pneumoniae infection, respectively (Table 6). This low prevalence of K. Another explanation for the development of PLA was the dysfunction of gut barriers during pneumonia. A previous study demonstrated that enterocyte injury is common in patients critically ill with pneumonia [27]. Therefore, after gut barrier dysfunction, bacteria may translocate to the liver through the portal vein system, eventually causing abscess formation in liver. DM patients are more likely to develop PLA because of their higher gut permeability [28]. Moreover, antibiotic therapy during pneumonia, which alters the composition and functions of gut microbiota and the intestinal immune system, eventually increases the risk of bacterial invasion [29]. Another potential mechanism contributing to PLA is the direct influence of pneumonia on the immune system. An immunocompromised status is a well-known potential risk factor for PLA. Studies have demonstrated that apoptosis of neutrophils and necroptosis of macrophages are common during bacterial pneumonia [30,31]. A critical finding reported in this study is that hospitalization for pneumonia within 180 days before the index date was associated with an increased PLA risk. Our results also indicated that increased hospitalization for pneumonia was correlated with higher PLA risk. These findings are of clinical significance to physicians because a timely diagnosis of PLA is difficult [32]. The most common signs and symptoms of PLA, such as fever and abdominal pain, are nonspecific; even laboratory findings are nonspecific and non-diagnostic. PLA is usually diagnosed using imaging techniques such as sonography and computed tomography. This study established an association of pneumonia and development of PLA. A follow up imaging in selected patients treated for pneumonia may lead to early diagnosis of PLA especially if they have history of gall stones and symptoms suggestive of PLA.
The strength of this case-control study is that we used data from LHID2010, the subset of Taiwan National Health Insurance Research Database, a nationwide dataset comprising 1 million beneficiaries randomly selected from the 2010 NHI registry. The Taiwan NHI system, established in 1995, covers the medical expenses of approximately 98% of the population of Taiwan; therefore, the included data accurately represents conditions in Taiwan. Moreover, the LHID2010 has a longitudinal design to minimize selection biases. Furthermore, the possibility of recall bias was reduced by using the administrative database for analysis. In fact, several high quality articles publications were generated by this database set [33][34][35].
Our study also has several limitations. First, the LHID2010 does not contain information regarding patients' clinical presentation, laboratory data, microbiological culture data, and pneumonia severity index score; however, the risk of bacteremia is correlated with the severity of pneumonia. Second, although the cases and controls were matched with comorbidities, our sample size was insufficient to match for all comorbidities (Table 1); therefore, we performed multivariate analysis ( Table 2, Table 3) to adjust for these potential confounding factors. Third, liver abscess is a disease endemic to Taiwan and East Asian countries; thus, our results may not be generalizable to Western populations. Fourth, small size of patient population and weakness of propensity matching methodology nature which does not account for unmeasured confounders in this study should be taken into consideration.
In conclusion, pneumonia is an independent risk factor for subsequent PLA. Moreover, hospitalization for pneumonia within 180 days before PLA diagnosis was associated with an increased PLA risk. Therefore, clinical physicians should consider PLA as a differential diagnosis for patients with infections after hospitalization for pneumonia.