Aspirin use is associated with reduced risk for recurrence of pyogenic liver abscess: a propensity score analysis

Aspirin possesses anti-bacterial activity that may prevent recurrence of Klebsiella pneumoniae pyogenic liver abscess (KP-PLA). In ex-vivo study, aspirin was administered before bactericidal assay against serotype K1 K. pneumoniae. We identified 5,912 patients with PLA who had no known pre-existing hepatobiliary diseases or malignancy in Taiwan from 1999 to 2013 from nationwide cohort study. Multivariate Cox proportional hazards regression models was used to estimate the hazard ratios [HR] for the association between aspirin use and recurrent PLA. The PLA recurrence rate in patients taking aspirin daily for 30 or more days, from 90 days before to 90 days after the first PLA episode (aspirin users), and aspirin non-users was 42.5 and 74.6 per 1,000 person-years of follow-up, respectively. The population-based study showed a HR for PLA recurrence in aspirin users of 0.50 (95% confidence interval, 0.35–0.69), relative to that in non-users, after adjustments for confounders. An ex-vivo study indicated that aspirin was able to significantly enhance bacterial killing by leukocytes, whether collected from diabetic patients with KP-PLA recurrence or from healthy volunteers. Our results suggest that aspirin is associated with reduced risk for PLA recurrence among Taiwanese with PLA who had no preexisting hepatobiliary diseases or malignancy.

Effects of aspirin in patients with PLA recurrence. Follow-up summation was performed for 23,780 person-years of follow-up [PYFU] during the study period. The mean follow-up time was 3.4 years in aspirin users and 3.9 years in non-users. A total of 1,449 (23.7%) patients had PLA recurrence after a first episode. The PLA recurrence rate was 42.5 per 1,000 PYFU in the aspirin-user group and 74.6 per 1,000 PYFU in the non-user group. Compared with patients in the non-user group, those in the aspirin-user group had a significantly lower risk of PLA recurrence after adjustments for age, gender, diabetic, hypertension, cancer, amoxicillin or ampicillin, metformin, other OADs, insulin, statins, aspirin used (adjusted HR: 0.62; 95% CI: 0.44-0.87, p < 0.01); however, the differences in all-cause mortality (adjusted HR: 0.92; 95% CI: 0.75-1.13, p = 0.45) and risk of cancer (adjusted HR: 0.76; 95% CI: 0.54-1.08, p = 0.12; Table 2) were not statistically significant between the two groups.
For development of the propensity score, a logistic regression model was applied (Table S1). In the propensity-score matched and competing adjusted-risk model, aspirin users still had a lower risk of PLA recurrence, with an adjusted HR of 0.65 (95% CI: 0.45-0.95; p = 0.02; Table 3). Meanwhile, comorbidity with diabetes or hypertension and use of metformin or insulin were also risk factors for PLA recurrence ( Table 3). The  Kaplan-Meier analysis for cumulative incidence of PLA recurrence showed a significantly lower risk of recurrent PLA in the aspirin-user group compared with that in the non-user group both before propensity-score matching (p < 0.001; Fig. 2A) and after propensity-score matching (p = 0.014; Fig. 2B).
The aspirin users in subgroup analysis had lower competing adjusted risk for PLA recurrence after a first episode of PLA; this was especially significant for females, those with diabetes or a non-cancer comorbidity, amoxicillin and ampicillin non-users, other anti-diabetic drug non-users, and insulin non-users (Fig. S1). Cause-specific hazard analysis also yielded similar results (data not shown). Aspirin use in the 6-12 months and 3-6 months prior to a first episode of PLA did not show a lower risk for PLA recurrence. The HR for PLA recurrence among patients with aspirin use at the time of the first episode of PLA was 0.62 (95% CI: 0.41-0.93; p = 0.02), and the HR for PLA recurrence among patients with aspirin use after the first episode was 0.98 (95% CI: 0.97-0.98, p < 0.01), with both groups having a lower risk of PLA recurrence ( Table 4).
The restricted cubic spline Cox model indicated that aspirin use was associated with a lower recurrent risk for PLA, especially in diabetic patients who did not receive an OAD or those who received a high-dose OAD for glucose control (Fig. 3). Aspirin use did not show a significant difference in risk for PLA recurrence in diabetic patients with and without insulin therapy. effects of aspirin on ex vivo leukocyte bactericidal activity. Of 5 diabetic patients admitted to KCGMH between 2013 and 2016 with a diagnosis of recurrent KP-PLA, 3 were males and 2 females, with a mean age of 59.8 years (range: 54-68 years). These 5 patients did not take aspirin before or after a first episode of  Table 2. Incidence of recurrent pyogenic liver abscess events and relative incidence in patients with pyogenic liver abscess. Incidence rate, 1000 person-years. Note: HR, hazard ratios. Adjusted HR: adjusted age, gender, diabetic, hypertension, cancer, amoxicillin or ampicillin, metformin, other oral anti-diabetic drugs, insulin, statins, aspirin used.
KP-PLA. The average time between first occurrence and KP-PLA recurrence was 1.7 years (range: 0.8-2.4 years).
Poor glycemic control with hemoglobin A1c above 9.0% (average: 10.7%; range: 9.4-12.5%) was observed in these 5 patients. The demographic information and data gathered from these patients are shown in Table S2. The difference between the rate of KP killing by leukocytes from these 5 diabetic patients and 5 healthy volunteers before oral administration of 100 mg of aspirin was significant (p < 0.01; Fig. 4). KP was significantly more susceptible to killing after incubation with leukocytes from healthy volunteers. The ex vivo bactericidal activity of leukocytes collected 1 h following oral administration of 100 mg of aspirin from 5 diabetic patients with KP-PLA recurrence and 5 healthy male volunteers was significantly greater than that of leukocytes collected from the same patients or from volunteers before aspirin treatment (p < 0.01; Fig. 4). The results indicated that aspirin was able to significantly enhance bacterial killing by leukocytes, whether collected from diabetic patients with KP-PLA recurrence or from healthy volunteers.

Discussion
We provided evidence that exposure to aspirin for more than 30 days, from 90 days before to 90 days after the first episode, was associated with decreased risk for recurrence of PLA in a nationwide population-based study in Taiwan. The duration relationship for aspirin use (both at the time of the first episode and at the time after the first episode of PLA; Table 4) was evident in the current study. Furthermore, the effect of aspirin use on a lower risk for PLA recurrence was found especially in diabetic patients who did not receive an OAD or in those on a higher OAD dose for glucose control (Fig. 4). The risk of recurrence may vary with glycemic status in diabetic patients; however, no laboratory tests that could serve as a proxy for glycemic status (HbA1c) were available in this dataset. Our findings imply that, with aspirin use, PLA may be preventable, especially in patients with a new diagnosis of diabetes and in diabetic patients with poor glucose control. We further demonstrated that aspirin enhanced leukocyte bactericidal activity against KP in diabetic patients with KP-PLA recurrence and in healthy volunteers in an ex-vivo study. To our knowledge, this is the first longitudinal study that specifically examined the role of aspirin in the prevention of PLA recurrence, using both population-based and ex-vivo findings.
Recurrent KP-PLA has seldom been reported 14,15 . Recently, a tertiary teaching hospital in China reported that 4 (4.1%) of 98 cases of KP-PLA had a prior history of PLA; therefore, these 4 cases were thought to represent recurrences, with onset between 7 and 96 months after the first episode 16 . In another study conducted in Taiwan that included 601 patients with PLA with a mean follow-up of 6 years, the cumulative recurrence rate of PLA was lower in both the patients with cryptogenic PLA (2.0%) and diabetic patients with PLA (4.4%) than those pre-existing hepatobiliary diseases with PLA (23.8%) 1 ; moreover, 80% of diabetic patients initially infected with  www.nature.com/scientificreports www.nature.com/scientificreports/ KP in this study were again infected with KP when the PLA recurred, which suggests that the recurrence rate of PLA in diabetic patients was low and most PLA recurrence was due to KP 1 . However, these single-center retrospective studies were limited by selection and recall bias 1,16 , and our large-scale population-based study might overcome these limitations. Furthermore, our cohort excluded patients with cholangitis, hepatobiliary cancer, or splenomegaly to reduce the complexity of analysis. Meanwhile, we also excluded patients with prior myocardial infarction or stroke to ensure comparability between aspirin users and non-users. The results showed that aspirin therapy decreased the risk of PLA recurrence. This finding was also confirmed in multivariate analysis after adjusting for a number of potential confounders, which was further reinforced by the duration relationship for aspirin use.
Adding to the long list of benefits, researchers have found that aspirin can reduce virulent bacteria associated with serious infections. One study found that SAL, a byproduct of aspirin, disrupted bacterial ability to adhere to host tissue, reducing propagation and spread of infection. Animals treated with aspirin have fewer bacteria at sites of infection, and develop smaller abscesses 17 . Additionally, our previous study has demonstrated that SAL could affect CPS production in KP, therefore attenuating the pathogenicity of KP strains. Aspirin also promotes human leukocyte phagocytosis and bactericidal activity against KP 13 . The majority of KP-PLA cases are preceded by colonization of the gastrointestinal tract, as suggested in epidemiological studies 9 . We found that patients who had used aspirin during the month prior to the diagnosis of community-acquired KP bacteremia appeared to have a lower risk for invasive syndromes including PLA 13 . As a result, aspirin may not be curative, but it could reduce the ability of bacteria to cause infection. We speculate that, with time, the immune system can adapt to compensate for or overcome the deficits, thereby decreasing the risk of PLA associated with use of aspirin. More studies are needed to confirm this hypothesis. The ex-vivo study still suggests that aspirin renders the host susceptible to KP. However, aspirin may simply arrest invasion by KP through the loss of resistance to colonization by KP. The consequences of an altered intestinal microbiota due to aspirin therapy are not fully understood. We did not explore how aspirin administration disturbs host-microbiota homeostasis that leads to KP-PLA. This also presents an opportunity for intervention, if established intestinal colonization can be eliminated. Use of aspirin to enhance innate immune defenses and inhibit pathogen colonization of the intestinal lumen and translocation across mucosal barriers may provide new strategies for management of this emerging infectious disease.

Figure 2. (A)
Cumulative proportion of patients developing recurrent pyogenic liver abscess events during follow-up according to aspirin uses and non-uses, before propensity-score matching. Kaplan-Meier analysis for cumulative incidence of pyogenic liver abscess recurrence showed significantly lower risk in the aspirin users group than in the non-users group before propensity-score matching (p < 0.001). (B) Cumulative proportion of patients developing recurrent pyogenic liver abscess events during follow-up according to aspirin users and non-users, after propensity-score matching. Kaplan-Meier analysis for cumulative incidence of pyogenic liver abscess recurrence showed significantly lower risk in the aspirin users group than in the non-users group after propensity-score matching (p = 0.014).  www.nature.com/scientificreports www.nature.com/scientificreports/ This study had some limitations. First, this was an observational study and could not prove a causal relationship. However, the ex-vivo leukocyte bactericidal experiment showed that aspirin was able to significantly enhance bacterial killing by leukocytes, subsequently resulting in lower risk for PLA recurrence. Second, smoking, drinking, exercise habits, diet, and other behaviors were not included in our model, but these may influence the relationship between aspirin and PLA recurrence. To minimize bias, we used propensity-score matching and subgroup analyses, which showed a strong relationship with health-seeking behaviors, but still indicated that aspirin was associated with reduced risk for PLA recurrence. Third, the disease and drug records collected from the databases may have been incorrect. We considered PLA as a critical event. Although, and used ICD-9-CM code 041.3 to identify sources of infection other than the lung. This code may have been omitted in the dataset. Our PLA cohort excluded patients with a history of hepatobiliary malignancy, cholangitis, common bile duct stones, and portal hypertension with splenomegaly to limit PLA to cryptogenic cases 18 . Figure 3. Dose-response association between oral anti-diabetic drug use (daily defined dose) and hazard ratio for recurrent pyogenic liver abscess in diabetic patients. Aspirin use was associated with lower risk for recurrent pyogenic liver abscess, especially in diabetic patients who did not receive an oral anti-diabetic drug (OAD) or those who received a high OAD dose for sugar control. Aspirin use did not show a significant difference in risk for recurrent pyogenic liver abscess in a comparison between diabetic patients with and without insulin therapy. Note: ddd, daily defined dose. codes 140-208) were defined as diagnoses made at 2 outpatient visits or during 1 hospitalization within 3 years before the first episode of PLA. Prescriptions for amoxicillin or ampicillin, metformin, other oral anti-diabetes drugs (OADs) (sulfonylureas, alpha-glucosidase inhibitors, thiazolidinediones, dipeptidyl peptidase-4 inhibitors), insulin, and statins were also defined as dosing for more than 30 days, from 90 days before to 90 days after the first episode; these were treated as covariates in this study.
Those who did not receive aspirin during the study period were considered controls and were assigned an index date corresponding to the same year of the case. Patients younger than 20 years and those with prior stroke or acute myocardial infarction were excluded. The etiology of cryptogenic PLA was nearly always KP in Taiwan 18 . Therefore, we further excluded patients with a history of hepatobiliary malignancy, cholangitis, common bile duct stones, or portal hypertension with splenomegaly in order to limit our cohort to PLA caused by KP. Based on intention-to-treat principle, analyses were conducted according to initial assignment of aspirin modality, regardless of any subsequent changes. outcome variables: pLA recurrence, complications, and mortality. The observation period started 30 days after discharge for the first episode of PLA and ended when the patient died or developed recurrence, or December 31, 2013, which ever occurred earlier. The incidence of cancer or myocardial infarction was also followed up until death or December 31, 2013. Mortality records were retrieved from patients who died in the hospital or withdrawn from the NHIRD. Study design and participants for ex-vivo human leukocyte bactericidal activity assay. Diabetic patients admitted to KCGMH between January 1, 2013 and December 31, 2016 with community-acquired mono-microbial bacteremia caused by KP were included to determine the patients with KP-PLA recurrence 21 . PLA recurrence was defined as having a typical clinical presentation, with new imaging findings of abscess recurrence after the first episode of abscess had fully resolved. All isolated KP strains were stored at −80 °C before use. Peripheral blood samples were collected from 5 healthy male volunteers aged 25-40 years and from participants with recurrent KP-PLA for an ex-vivo bactericidal assay as described below. All participants provided written informed consent.
Determination of K. pneumoniae serotype and rmpA genes. Genotyping of 7 clinically significant capsular types (K 1 , K 2 , K 5 , K 20 , K 54 , K 57 , and K N1 ) for all KP isolates was performed with a polymerase chain reaction (PCR) assay using primers designed for the cps variable region 22 . PCR was also used to determine the presence of rmpA. Previously published primers for rmpA were used, and the expected PCR products of rmpA were 535 bp in length 23 .
Ex-vivo human leukocyte bactericidal activity assay. Bactericidal activity was measured using a standard assay method 24 . The KP isolate from each enrolled patient during the first episode of KP-PLA (study for each patient, respectively) and serotypes K1 KP, KP-M1 (study for volunteers) were cultured on Tryptic Soy Agar medium overnight at 37 °C, and then opsonized with addition of 10% pooled serum collected from 5 healthy male volunteers who did not take any aspirin prior to donation. The suspension was mixed at a 10:1 ratio of infected to whole blood leukocytes, which had been collected from diabetic patients with recurrent KP PLA and 5 volunteers, either before or 1 h after a 100-mg oral dose of aspirin. Samples were collected 1 h later and diluted in H 2 O (pH 11.0, adjusted with NaOH) to lyse the leukocytes and disperse the bacteria for the power-plate colony assay 25 . All tests were performed in triplicate to ensure reproducibility. www.nature.com/scientificreports www.nature.com/scientificreports/ Statistical analysis. Baseline characteristics were compared using either a 2-sided t-test or χ 2 test as suitable. In multivariate Cox proportional hazards regression models, the effects of aspirin were adjusted for age, sex, comorbidities, and prescriptions used. Results were expressed as hazard ratios (HRs) and compared with those aspirin non-users. All time-independent covariates were compared using estimated log-log survival curves to ensure a constant HR over time. Adjusted HRs and 95% confidence intervals (CIs) for PLA recurrence with aspirin use were based on participant characteristics (see below). After adjusting for covariates, the cumulative hazards of PLA recurrence over time were compared between aspirin users and non-users with Cox proportional hazards regression models and a cause-specific, PLA recurrence and mortality, competing adjusted-risk model 26 . Propensity-score matched logistic regression analysis for the likelihood of aspirin use was performed using the baseline covariates. Adequacy of balance for the covariates in the matched sample was assessed using the standardized mean difference between the 2 groups, with differences less than 10% reflecting good balance.
In subgroup analysis for subjects with diabetes, restricted cubic spline curves were used to examine non-linear associations of aspirin users and non-users with recurrent PLA, with adjustments for confounding variables, to further characterize the nature of the relationships between OAD doses with the risk of PLA recurrence. Four knots were chosen to produce a curve that appeared adequately smooth 27 . To control for consistent medication effect across different durations, the aspirin used after the first episode of PLA was assessed as a duration relationship covariate in the Cox model. The counts of surviving bacteria in the bactericidal activity assay were analyzed using the log rank test. The 2-tailed test was used in statistical significance testing and a p value < 0.05 was considered significant. All statistical analyses were carried out using STATA 15.1 (STATA Corp., College Station, TX, USA) and SAS 9.4 (SAS Institute Inc., Cary, NC, USA).