PDF
Citation
Print
INTRODUCTION
Infective endocarditis (IE) is a rare but highly lethal disease.1) Previous studies conducted worldwide have reported incidence rates (IRs) of approximately 3–7 cases per 100,000 person-years, with a 30% mortality rate within 1 year of disease onset.2)3)4)
Some studies have indicated an increased short-term risk of IE associated with traumatic cutaneous wounds,5) while others have found that bacteria cultured from the blood of patients with IE are predominantly derived from the skin surface, such as Staphylococcus aureus.6) These studies suggest the existence of a potential pathogenic pathway where bacterial exposure through skin wounds increases the likelihood of future IE development.
Not all wounds are caused by traumatic events; non-traumatic cutaneous wounds can occur in daily life. If non-traumatic cutaneous wounds are not promptly disinfected or if treatment is delayed, there is still a risk of developing infection symptoms. However, the association of these non-traumatic wounds with IE has not been discussed in previous studies.5) Moreover, whether there is a difference in the temporal sequence of IE occurrence between traumatic and non-traumatic wounds has not been investigated in any previous literature.
Therefore, this study aimed to investigate the association between cutaneous wounds and the occurrence of IE using data from the National Health Insurance Research Database (NHIRD) in Taiwan.
METHODS
Ethical statement
This study utilized a secondary database for data analysis, precluding direct access to primary data or patients. Ethical approval was obtained from the Institutional Review Board of the Hualien Tzu Chi Hospital (approval number: IRB109-245-B).
Data source
In Taiwan, all citizens are required to enroll in the National Health Insurance (NHI) system, and healthcare provider services are reimbursed through the NHI Administration. All claims data from this single-payer system, including outpatient, emergency department (ED), inpatient, dental, and pharmacy services, are compiled in the NHIRD. The NHIRD is managed by the Health and Welfare Data Science Center of the Ministry of Health and Welfare. Researchers can access and utilize the NHIRD through research project applications, ensuring information security and patient privacy. The disease diagnosis coding systems used are the International Classification of Diseases, Ninth and Tenth Revisions, Clinical Modification (ICD-9-CM until 2015; and ICD-10-CM after 2016). Additionally, dates, causes, and locations of death for deceased citizens are linked with the National Death Registry.7)
Study design
All NHI-insured persons from 2013 to 2022 were included in this study. This study employed a self-controlled case series (SCCS) design, which included patients who experienced both the exposure (cutaneous wounds) and outcome (IE) of interest within the study observation period.8) We monitored these patients for the occurrence of wounds and IE events until death or until the end of the available data (December 31, 2022). We applied a 3-year wash-out period (2010–2012) for our primary outcome, IE. The strength of the SCCS design lies in estimating the temporal association between a transient exposure (cutaneous wound) and an outcome event (IE). This design enables control for time-invariant factors such as genetics and the immune system. Furthermore, the SCCS design allowed us to estimate relative risks by comparing the IRs of IE events during exposure (cutaneous wound) and non-exposure periods.9)
Inclusion and exclusion criteria
We only included patients who experienced both cutaneous wounds (exposure) and IE events (outcome) during the observation period according to the principle of SCCS design.9) We only enrolled patients with cutaneous wounds treated in outpatient clinics. The IE events were only extracted from the inpatient admission records. Patients were excluded from the study population if they met any of the following conditions during the observation period: having IE history, having age at enrollment <20 years, having valve replacement history, having more than one event of IE occurrence, having cutaneous wounds treated in the ED, and having cutaneous wounds with symptoms of infection (including fever, urinary tract infection, pneumonia, bacteremia, sepsis, and cellulitis; the diagnosis codes were listed in Supplementary Table 1) when presenting to the clinics.
Exposure periods
The primary exposure variable was cutaneous wounds, defined as patients receiving treatment for such wounds in outpatient departments; relevant reimbursement codes are listed in Supplementary Table 2. If the interval between wound treatments was <14 days, these treatments were considered part of the same wound episode. The index date was defined as the first date of a wound episode, and we followed the risk over the subsequent 4 weeks after the initiation of wound treatment. Previous studies have suggested that this period is associated with a high risk of developing IE.5)
Figure 1 provides a visual representation of the different observation intervals and their corresponding calculated effects.
Figure 1
Data structure in SCCS design of current study. Figure 1 illustrates how the observation periods and exposure status were defined in this study. All samples were followed starting from 2013, with the date of the first insured as the initial date. Samples with any record of IE diagnosis before the start of follow-up were excluded. In the data structure of this study, each patient has one or multiple observation time-intervals. Firstly, for each patient, the whole observation period was divided by observation year (e.g., first, second, third years), which fulfills the requirement of counting age effects in the SCCS design. Then, for each incident of wound treatment, the patients were observed for the followed 4 weeks, generating a new observation record for each week. Except for the period of wound treatment and the subsequent 4 weeks, all other time intervals are considered as the baseline period. The first date of every observation interval was the index date for that interval. Through this data structure, we can calculate the occurrence of events, duration of observation interval, and incidence rates across different observation interval.
IE = infective endocarditis; SCCS = self-controlled case series.
Considering the variations in severity, progression, and potential confounders, we used disease diagnosis codes to distinguish between traumatic wounds and non-traumatic wounds. In our study, "traumatic wounds" include injuries to the head, neck, chest, shoulders, pelvis, and limbs, while "non-traumatic wounds" refer to all wounds other than traumatic wounds. We assumed that the source of infection was exposed at the time of the traumatic wound occurrence. In contrast, other non-traumatic wounds may have been exposed to contaminants for some time before seeking medical care. The interval from bacterial exposure to medical encounter may differ between traumatic wounds and other non-traumatic wounds (Supplementary Table 2).
Primary outcome
The primary outcome was the IE event, defined as an inpatient admission claim with any of the following diagnostic codes, which were validated in a previous study with a positive predictive value of 0.83: ICD-9-CM, 421, 036.42, 098.84, 112.81, 115.04, 115.14, 115.94; ICD-10-CM, I33, and I39.10)11) The first IE event during the observation period was selected based on the SCCS design assumption.9)
Covariates and confounders
Potential confounders were collected for each observation period’s index date. Age, sex, and financial status were extracted from the NHI registry. Financial status was categorized into 4 levels based on monthly income-based insurance premiums: financially dependent, 15,840–29,999, 30,000–44,999, and ≥45,000 New Taiwan Dollar. Comorbidities were defined as having at least 1 inpatient diagnosis or 2 outpatient diagnoses in the year prior to the index date, including diabetes mellitus (DM), end-stage renal disease (ESRD), hypertension, cancer, drug abuse, congenital heart disease (CHD), and heart valve disease (HVD). Disease diagnosis codes for these comorbidities are provided in Supplementary Table 1.
Previous clinical guidelines have indicated that patients with implanted pacemakers have a higher risk of developing IE.1)12) Therefore, we tracked whether there was a history of cardiac implantation before each observation period. Some exposures, including invasive dental treatment (IDT),10)13) head- and neck-related injury or surgery,14)15) are thought to increase the risk of IE. We defined these risk factors as present if any claim codes were found in the previous year. Additionally, chronic conditions such as home care (often associated with bedsores or chronic wounds) and port-A catheter implantation (involving needle wounds), increase the risk of infection through skin wounds. These factors were considered present if any outpatient or inpatient records were found in the previous year. Disease treatment codes for these procedures are provided in Supplementary Table 2.
We included the number of outpatients, ED, and inpatient visits in the year prior to the index date as covariates to adjust for variations in patient health status and healthcare utilization.
We also collected data on concurrent medications, defined as drugs prescribed for ≥30 days within the year before the index date, including immunosuppressants,16) histamine type-2 receptor antagonists, and proton pump inhibitors (which suppress gastric acid secretion, induce bacterial overgrowth in the gastrointestinal tract, and subsequently increase the risk of infection).17)18)19)20) The World Health Organization (WHO) Anatomical Therapeutic Chemical codes for these medications are provided in Supplementary Table 3.
Statistical analysis
Baseline characteristics at the enrollment and IE onset dates were described using the mean ± standard deviation for numerical variables and number (percent) for categorical variables. Conditional Poisson regression was used to calculate the adjusted IR ratios (aIRRs) of IE, using the baseline period as a reference. The duration of the observation period was included as an offset variable in the Poisson regression model. The patients' unique identification numbers were used to determine which observation records should be assigned to the same cluster (patient). All baseline characteristics shown in Table 1, including demographics, comorbidities, associated factors for IE, and concurrent medication, healthcare visits, were adjusted in the multivariable Poisson regression model. Data management and statistical analyses were performed using R software, version 4.4.0 (R Foundation for Statistical Computing, Vienna, Austria).
Table 1
Patients characteristics at the date of enrollment, IE onset date, and outpatient clinic visit for wound (n=3,241)
Values are presented as mean ± standard deviation or number (%).
CHD = congenital heart disease; DM = diabetes mellitus; ESRD = end-stage renal disease; H2 blockers = histamine type-2 receptor antagonists; HTN = hypertension; HVD = Heart valve disease; IDT = invasive dental treatment; IE = infective endocarditis; NTD = New Taiwan Dollar; PPI = proton pump inhibitor.
Subgroup and sensitivity analyses
We conducted stratified analyses based on the presence of antibiotic prescriptions for wound treatment at the initial encounter. Subgroup analyses were also conducted in patients with DM and ESRD, who are known to have an increased risk of developing infectious diseases.21)22) To ensure the validity of the outcome events, we excluded inpatients with IE who did not receive antibiotic treatment.
Multiple sensitivity analyses were performed to verify the consistency and reliability of the relationship between cutaneous wounds and IE. First, we excluded subsets known to have a high risk of IE and performed subgroup analyses, such as those with a pacemaker, CHD, and HVD.1)12)23)24)25) Second, we excluded patients with long-standing cutaneous wounds due to receiving home care or having a port-A catheter. Third, to account for potential IE risks associated with factors other than cutaneous wounds, we excluded patients who had recently undergone IDT, head- and neck-related injuries, or surgeries.
Additional sensitivity analyses for self-controlled case series design
Some assumptions of the SCCS design should be noted. We addressed these assumptions by performing sensitivity analyses and their data are shown in the Supplementary Method 1.
E-value approach for unmeasured confounders
In every observational study, unmeasured confounding factors may exist, such as wound size, depth, or duration.26) To assess the potential impact of these factors on the association between IE events and cutaneous wounds, we calculated E-values. These values represent the minimum strength of association that an unmeasured confounder would need to have with both the exposure and outcome to fully explain the observed association.27)
RESULTS
Characteristics of eligible study population
Supplementary Figure 1 displayed a study population selection flow diagram of this study. In total, 3,241 eligible patients were included. At the enrollment of the study, the average age was 60.6 years, with males comprising 59.9% of the cohort. Additionally, 28.5% of patients had a history of DM. At the time of IE occurrence, 0.9% had received a cardiac pacemaker implantation, and 10.2% were ESRD with hemodialysis (Table 1). The average elapsed time from the enrollment date to the IE onset date was 5.24 years (SD, 2.85 years). Data on all characteristics of patients at the time point of outpatient clinic visits for the wounds are also reported in Table 1. Supplementary Table 4 displayed the treatment outcomes within 180 days of first IE admission events.
Based on the whole population of the nationwide database analyzed in this study, the IR of IE among outpatients with treated cutaneous wounds was 17.1 per 100,000 person-years. The age-standardized IR, based on the WHO 2000 population composition, was 11.6 per 100,000 person-years.
Association between wound and infective endocarditis events
An increased risk of IE occurrence was found during the second week following a traumatic wound and the first 2 weeks following a non-traumatic wound, compared to the baseline period using a multivariable Poisson regression model after adjusting for all covariates that are listed in Table 1. The aIRR (95% confidence interval [CI]; p value) were 2.16 (1.07–4.35; 0.032) for the second week after a traumatic wound, and 1.56 (1.17–2.09; 0.002), 1.58 (1.19–2.10; 0.002) for the first 2 weeks following a non-traumatic wound, respectively (Table 2). However, there were no significant increase or decrease in aIRR was found during the first week of traumatic wounds (aIRR, 0.82; 95% CI, 0.26–2.56; p value=0.731). Overall, 98.0% of IE admissions were managed with antibiotic treatment. The results, after excluding patients with IE events without antibiotic treatment (Supplementary Table 5), remain similar.
Table 2
Risks of infective endocarditis occurrence in different observation periods of the study population
Sensitivity analyses
Figure 2 shows the results after excluding patients who had received cardiac pacemaker implantation, had a history of CHD, or had a history of HVD. There was an increased risk of IE occurrence during the second week following a traumatic wound and the first 2 weeks following a non-traumatic wound, compared to the baseline period.
Figure 2
Sensitivity analyses after excluding patients with pacemaker implantation, history of CHD, or HVD.
aIRR = adjusted incidence rate ratio; CI = confidence intervals; CHD = congenital heart disease; HVD = heart valve disease.
Supplementary Figure 2 shows the results after excluding patients who had home care or a port-A catheter, indicating those who had long-term wounds. Supplementary Figure 3 shows the results after excluding patients who had recently undergone IDT, experienced head- and neck-related injury, or head- and neck-related surgery, indicating those who had short-term risk factors associated with IE. There were increased risks of IE occurrence during the second week following a traumatic wound and the first 2 weeks following a non-traumatic wound, compared to the baseline period, in these sensitivity analyses.
Supplementary Figure 4 shows the results using the subgroup of patients with DM and ESRD. There was an increased risk of IE occurrence during the second week following a traumatic wound and a non-traumatic wound in patients with DM, compared to the baseline period.
Supplementary Table 6 shows the results after excluding patients who died within 90 days of IE admission, as these patients have an increased probability of death that may violate the assumptions of the SCCS design. Supplementary Table 7 shows the results after adding a pre-treatment variable to control for the confounding by indication issue, thus testing the robustness of the SCCS design. These results were consistent with the main analytical findings.
Figure 3 shows the stratified analysis results according whether antibiotics prescription for cutaneous wound. In the current study, among 10,938 wound episodes, 31% were treated with antibiotics, as identified by prescriptions of antibacterial medications using ATC codes.
Figure 3
Subgroup analysis stratified by antibiotics prescription for wound treatment at initial encounter.
aIRR = adjusted incidence rate ratio; CI = confidence intervals; N/A = not applicable due to zero event occurrence that were not able to calculate the aIRR.
There were increased risks of IE occurrence during the third week following a traumatic wound without antibiotics (aIRR, 2.46; 95% CI, 1.01–5.99; p value=0.047), the first week following a non-traumatic wound with antibiotics (aIRR, 2.22; 95% CI, 1.40–3.51; p value <0.001), the second week following a non-traumatic wound without antibiotics (aIRR, 1.91; 95% CI, 1.41–2.58; p value <0.001), compared to the baseline period.
E-value estimation
E-values were calculated using the main significant findings presented in Table 2. The E-values for the second week following a traumatic wound were 3.74, and the first 2 weeks following a non-traumatic wound were 2.49 and 2.54, respectively. This indicates that there would need to be at least one unmeasured confounder with an increased risk of >3.74 fold in both exposure (cutaneous wound) and outcome (IE) to potentially change our study findings.
DISCUSSION
This study had 3 main findings. First, the relative risk of IE occurrence increased in the second week after patients sought treatment for traumatic wounds, and the risk of IE occurrence was elevated throughout the first 2 weeks after patients sought treatment for non-traumatic wounds. For clinical implications, it is important to monitor traumatic cutaneous wounds for signs of progression into deep tissue or organ infections within the following 2 weeks. On the other hand, non-traumatic cutaneous wounds are often associated with delayed visits and infection onset. Therefore, it is crucial to promptly assess for infection symptoms or any abnormalities in cardiac examinations. Second, through multiple sensitivity analyses, we confirmed that even after excluding known risk factors associated with IE (such as dental treatments and recent surgeries) and adjusting for potential drawbacks in the study design (such as the pre-treatment period), the trends remained consistent. These findings provide robust evidence for the association between cutaneous wounds and the occurrence of IE in the current SCCS study. Third, patients had an increased risk of IE occurrence in the second week following non-traumatic wounds without antibiotic prescription. This implies delayed physician-seeking behavior or an underestimation of potential bacteremia coexisting with non-traumatic wounds.
The study results supported our hypothesis that bacteria could invade through a cutaneous wound and cause IE within 2 weeks after traumatic wounds (Table 2). These findings align with the natural history of wound occurrence and bacterial invasion leading to IE, further enhancing the reliability of the study results in inferring the temporal association in a clinical context.
In a previous study using 2 Japanese databases, it was found that the one-month period following a traumatic wound is a high-risk period for the occurrence of IE. However, the 2 databases contained less than 5% of the population and may not be representative of the Japanese population.5) Our study, conducted with a nationwide database, is representative of the Taiwanese population. We showed that cutaneous wounds could increase the risk of IE in Taiwanese patients.
We observed that the IR of non-traumatic wounds in the study population was approximately 8 times higher than that of traumatic wounds. The term 'non-traumatic wound' might encompass different definitions, such as chronic wounds, delayed-treated wounds, or not-yet-healed wounds. It is conceivable that the risk of infection increases with the time delay of wound healing. Although the causes and duration of delayed medical care for such non-traumatic wounds are unknown in the current study, on average, non-traumatic wounds carry an increased risk of progression to IE compared to baseline periods.
In this study, the results in Figure 3 indicate that, in the subgroup with traumatic wounds, there was no significant difference in the aIRR between antibiotic users and non-antibiotic users during the first 4 weeks following treatments. Also, in the subgroup with traumatic wounds, an increased aIRR was found only in the antibiotic users in the post 1st week, while it was found only in the non-antibiotic users in the post 2nd week. Thus, there is no solid evidence suggesting the use of antibiotics as a predictor for IE. In contrast, the results in Table 2 indicate that patients with traumatic wounds had an increased aIRR only at the post 2nd week, whereas patients with non-traumatic wounds had an increased aIRR at both the post 1st week and post 2nd week. As such, we speculated that the nature of wounds could be a more important predictor of IE. Patients with non-traumatic wounds treated with antibiotics may have more severe wounds. This situation may possibly lead to more screening for IE alongside the use of antibiotics, resulting in more diagnoses of IE.
This study has at least 4 major strengths. First, we conducted a comprehensive 10-year observation using a nationwide database, providing a large sample size for detailed analysis of the risk of IE in shorter time intervals. This approach yielded more informative evidence about the temporal sequence than the current literature.5) Second, the study utilized the SCCS design, which effectively detected time-sensitive events, allowing us to highlight the second week after a traumatic wound as a critical time-point for IE occurrence. A patient with discomfort and a recent cutaneous wound should have a heart evaluation to detect IE at an earlier stage. Third, the SCCS design controls for factors that do not vary over time but differ among individuals, including genetics, immune system function, and wound healing capacity, enabling the adjusted calculation of the average relative risks in study participants.9) Lastly, few studies have explored the relationship between non-traumatic wounds and the risk of IE in the general population. Our findings on the risk of developing IE during the entire observation period for both traumatic and non-traumatic wounds relative to the non-cutaneous wound period provide real-world evidence addressing this issue.
This study has several limitations. First, as this was an observational study, there may have been several unobserved confounding factors. Through the E-value calculation, we found that at least one unmeasured confounder with an increased risk of >3.74 in both exposure (cutaneous wound) and outcome (IE) would be required to change our study conclusions. However, after we adjusted for many currently known risk factors, the possibility of the existence of an unmeasured confounder, which can change our study conclusions, might be low. Second, the administrative claim database used in this study did not include medical information such as pathogens, wound sizes, and the exact timing of trauma occurrence. Thus, the current study cannot confirm whether the primary causative pathogens of IE originate from common skin bacteria such as S. aureus.6) We also cannot explore the association between the risk of IE and wound sizes or timing of trauma occurrence. These may require further investigation through clinical data research in the future. Third, patients with relatively severe wounds are typically treated in the ED, and the exclusion of these patients from the analysis was one limitation of the current study. Fourth, some small wounds that did not visit a clinic could still lead to endocarditis, and these cases were classified into the baseline period in the current study. As such, the possibility of such classification errors may exist. Fifth, when patients visited clinics for a wound, it is likely that antibiotics were used in many cases. This would differ from the risk of endocarditis caused by a wound in cases where antibiotics were not used. Sixth, this study did not find a consistent pattern of the time-dependent results. The possibility of multiple comparisons should be considered, and the results should be interpreted with caution. Last, this study was conducted using a nationwide database from Taiwan, an East Asian country. Our findings may vary owing to differences in age composition, genetics, healthcare systems, or public health policies in different countries. Further research involving diverse populations from various countries is required to validate our findings.
In conclusion, in patients who experienced traumatic or non-traumatic cutaneous wounds, clinicians should be vigilant for symptoms or signs associated with IE in the first 2 weeks after wound treatment to avoid delays in diagnosis and treatment. This approach helps identify potential cases of IE.
XML Download