Real-World Data in Pharmacovigilance Database Provides a New Perspective for Understanding the Risk of Clostridium difficile Infection Associated with Antibacterial Drug Exposure

Antibacterial drug exposure (ADE) is a well-known potential risk factor for Clostridium difficile infection (CDI), but it remains controversial which certain antibacterial drugs are associated with the highest risk of CDI occurrence. To summarize CDI risk associated with ADE, we reviewed the CDI reports related to ADE in the FDA Adverse Event Reporting System database and conducted disproportionality analysis to detect adverse reaction (ADR) signals of CDI for antibacterial drugs. A total of 8063 CDI reports associated with ADE were identified, which involved 73 antibacterial drugs. Metronidazole was the drug with the greatest number of reports, followed by vancomycin, ciprofloxacin, clindamycin and amoxicillin. In disproportionality analysis, metronidazole had the highest positive ADR signal strength, followed by vancomycin, cefpodoxime, ertapenem and clindamycin. Among the 73 antibacterial drugs, 58 showed at least one positive ADR signal, and ceftriaxone was the drug with the highest total number of positive signals. Our study provided a real-world overview of CDI risk for AED from a pharmacovigilance perspective and showed risk characteristics for different antibacterial drugs by integrating its positive–negative signal distribution. Meanwhile, our study showed that the CDI risk of metronidazole and vancomycin may be underestimated, and it deserves further attention and investigation.


Introduction
Clostridium difficile is an anaerobic, spore-forming Gram-positive bacillus that usually colonizes in the human gut [1]. It is an opportunistic pathogen that is able to abnormally proliferate, produce toxins and result in diarrhea, especially in patients with changes in the indigenous colonic microbiota following antibiotic use [2], and it is reported that the attributable mortality of C. difficile infection (CDI) should be at least 5.99% [3]. In recent decades, the increasing incidence, severity and mortality of CDI have made it a challenging clinical problem for medical personnel [4]. In response to this challenge, diagnosis and treatment guidelines have been developed in recent years to optimize the management of CDI [5][6][7][8][9]. In primary prevention for CDI, the careful selection of antibacterial drugs and, whenever possible, the avoidance of high-risk antibacterial drug exposure (ADE) is the mainstay because most cases of CDI are both iatrogenic and nosocomial [4]. Meanwhile, some studies have shown that strict antimicrobial stewardship is beneficial in reducing CDI

Report Basic Information and Patient Characteristics
A total of 16,010,899 reports were recorded in the FAERS database from 1 January 2004 to 31 December 2022. Using the Preferred Terms (PTs) in Table 1 to retrieve target reports, a total of 30,937 reports considered CDI-related were returned and downloaded. As the culprit drug of CDI may be indecisive and can be attributed to multiple drugs, there were a total of 222,971 drugs contained in those CDI-related reports. After excluding drugs missing generic names, duplicated drugs and drugs that were not under J01 of the Anatomical Therapeutic Chemical (ATC) classification system, a total of 99 drug names were classified into "antibacterials for systemic use (J01)". The 99 drug names were used to match reports that CDI occurrence was related to antibacterial drug use, and finally, a total of 8063 (26.1%) reports were identified for further analysis. As some of the 99 drug names were synonymous (e.g., ampicillin and ampicillin sodium), we integrated drugs with the same ingredient manually, and finally, there were 73 drugs included in the final antibacterial drug list to detect ADR signals. The detailed processing flow is shown in Figure 1.
Information in the 8063 antibacterial drug use-related CDI reports was extracted and collected. The annual number of reports from 2004 to 2022 was presented in Figure 2A, among which 2019 was the year that FAERS received the greatest number of CDI reports associated with ADE. With regard to report sources, health professionals (73.8%) were the main submitters ( Figure 2B), and the USA was the leading reporting country ( Figure 2C). The demographic characteristics of patients were summarized, and the result showed that there were fewer male patients than female patients ( Figure 2D) and the age of those patients was mainly located in the 71-80 age group ( Figure 2E). In terms of patient outcome, Information in the 8063 antibacterial drug use-related CDI reports was extracted and collected. The annual number of reports from 2004 to 2022 was presented in Figure 2A, among which 2019 was the year that FAERS received the greatest number of CDI reports there were fewer male patients than female patients ( Figure 2D) and the age of those patients was mainly located in the 71-80 age group ( Figure 2E). In terms of patient outcome, CDI usually resulted in hospitalization (67.3%), and even the death of 1282 (15.9%) patients were associated with CDI ( Figure 2F).

ADR Signal Detection Results
After integrating synonymous drugs, 73 antibacterial drugs were used to detect ADR signals at Standardized MedDRA Queries (SMQ) level and PT level. The signal detection results at the SMQ level are shown in Table 2, and it showed that metronidazole (a = 2004) was the most reported antibacterial drug followed by vancomycin (a = 1793), ciprofloxacin (a = 1176), clindamycin (a = 823) and amoxicillin (a = 566), while metronidazole (ROR = 22.10, 95% CI 21.10-23.14) had the highest positive signal strength followed by vancomycin (ROR = 21. 30 18-17.47). In addition, the signal detection results for 10 different PT levels are shown in Tables S1-S10.

ADR Signal Detection Results
After integrating synonymous drugs, 73 antibacterial drugs were used to detect ADR signals at Standardized MedDRA Queries (SMQ) level and PT level. The signal detection results at the SMQ level are shown in Table 2, and it showed that metronidazole (a = 2004) was the most reported antibacterial drug followed by vancomycin (a = 1793), ciprofloxacin (a = 1176), clindamycin (a = 823) and amoxicillin (a = 566), while metronidazole (ROR = 22.10, 95% CI 21.10-23.14) had the highest positive signal strength followed by vancomycin ( 18-17.47). In addition, the signal detection results for 10 different PT levels are shown in Tables S1-S10.   Note: The classification of antibacterial agents is based on the Anatomical Therapeutic Chemical classification system, and the bold represents the drug category and its code. Abbreviations: AE, adverse event; CI, confidence interval; ROR, reporting odd ratio.
As metronidazole and vancomycin were usually used as therapeutic agents for CDI, we further reviewed the indications for metronidazole and vancomycin recorded in the "patient.drug.drugindication" field. In order to eliminate the influence of this factor on ADR signal detection results as much as possible, if the indication of metronidazole and vancomycin was related to the treatment of CDI, the report was excluded. The adjusted signal detection results for metronidazole and vancomycin at the SMQ level and PT level are shown in Tables 3 and 4, respectively.

Distribution of ADR Signals
There were 11 ADR signal detection results for each of the 73 antibacterial drugs, including one for the SMQ level and 10 for the PT level. In addition, signal detection results can be divided into three states, namely positive signals, negative signals and not reported for target drug-AE combinations. The distribution of signal detection results for 73 antibacterial drugs is presented in Figure 3. It showed that 58 antibacterial drugs had at least one positive ADR signal detection, while another 15 antibacterial drugs did not show any positive signals at the SMQ level or PT level, although there were CDI cases reported. Of these antibacterial drugs with positive signals, only ceftriaxone had 11 positive signals.

Discussion
Antibacterial drugs, one of the greatest achievements of human beings in the field of medicine, have played an extremely important role in improving human health level and ensuring life safety. However, with the extensive use of antibacterial drugs in clinical practices, various ADRs associated with antibacterial drugs have emerged, among which CDI is one of the most noteworthy potentially life-threatening ADRs [20]. Therefore, it is necessary to determine the risk of CDI induced by different antibacterial drugs. In this study, we reviewed CDI reports associated with ADE in the FAERS database between 2004 to 2022 and found that 73 antibacterial drugs were recorded as potential culprit drugs. At the same time, based on the aforementioned antibacterial drug list, we conducted a disproportionality analysis to evaluate the risk correlation between the occurrence of CDI and

Discussion
Antibacterial drugs, one of the greatest achievements of human beings in the field of medicine, have played an extremely important role in improving human health level and ensuring life safety. However, with the extensive use of antibacterial drugs in clinical practices, various ADRs associated with antibacterial drugs have emerged, among which CDI is one of the most noteworthy potentially life-threatening ADRs [20]. Therefore, it is necessary to determine the risk of CDI induced by different antibacterial drugs. In this study, we reviewed CDI reports associated with ADE in the FAERS database between 2004 to 2022 and found that 73 antibacterial drugs were recorded as potential culprit drugs. At the same time, based on the aforementioned antibacterial drug list, we conducted a disproportionality analysis to evaluate the risk correlation between the occurrence of CDI and ADE. As far as we know, this is the first study using a pharmacovigilance database to evaluate the risk of CDI occurrence for ADE.
Although it is widely recognized that any antimicrobial therapy increases the risk of CDI and there is a difference among different antibiotics, it remains controversial which certain antibiotics or classes of antibiotics are related to the highest risk of CDI. A previous study showed that fluoroquinolones were the antibacterial agent most strongly associated with CDI, while all the third-generation of cephalosporins, macrolides, clindamycin and intravenous beta-lactam/beta-lactamase inhibitors were intermediate-risk antibacterial agent [13]. Another study showed that the risk of hospital-acquired CDI was greatest for cephalosporins and clindamycin, while the importance of fluoroquinolones should not be overemphasized [21]. A recent study suggested that the highest-risk antibacterial agents related to CDI occurrence included second-generation and later cephalosporins, carbapenems, fluoroquinolones and clindamycin, while doxycycline and daptomycin were related to a lower CDI risk [22]. However, due to the difference in the region, patient inclusion and exclusion criteria, study design, drugs involved in the evaluation and the definition of risk classification, it is difficult to unify the CDI risk of antibacterial agents. In this regard, by using a unified standard to detect the ADR signals for each antibacterial agent at the PT and SMQ level, our study added new evidence for understanding the risk of CDI induced by ADE from a pharmacovigilance perspective. In comparison to the studies mentioned above, the advantage of this study is that it makes full use of real-world data to get a complete antibiotics list leading to CDI occurrence, involving 73 antibiotics commonly used in clinical settings. Therefore, our study can provide a more comprehensive overview of the risk of antibacterial drugs, facilitating a comparison of risks between them and providing a reference for antimicrobial stewardship.
Consistent with previous studies [13,21,22], our ADR signal detection results showed a high risk of CDI in most fluoroquinolones, cephalosporins, carbapenems, macrolides, clindamycin and beta-lactam/beta-lactamase inhibitors, which proved the credibility of our results to some extent. However, it is noteworthy that, in ADR signal detection, metronidazole and vancomycin have a surprising number of reports and high signal strengths at the SMQ level and PT level, although they have been reported as possible causes of CDI in previous studies [23,24]. There are several possible explanations for this noteworthy result. First, metronidazole and vancomycin were used as therapeutic drugs for CDI [5][6][7][8][9], which may lead us to ignore their risk of inducing CDI, while our research detected this neglected risk relationship. Second, 26.2% of reporters in this study were non-health professionals, so they may confuse therapeutic and etiological drugs and misjudge culprit drugs, which may result in biased results. Third, due to FAERS being a database with a voluntary reporting nature, underreporting of other antibacterial agents may exist [25], which may highlight the CDI risks of metronidazole and vancomycin. In order to reduce the influence of misreporting due to the overlap of indications and AEs for metronidazole and vancomycin, we excluded the reports that the indication of metronidazole and vancomycin was related to the treatment of CDI, but the adjusted signal detection results for metronidazole and vancomycin still showed conspicuous high potential risk (Tables 3 and 4). In this regard, our results showed a warning that we should pay more attention to the CDI risk of metronidazole and vancomycin, which may have been previously neglected. Although the true relationship between CDI occurrence and vancomycin and metronidazole still needs a well-designed study to verify, we think there are two main potential values for these data. First, it provides evidence of the potential high risk of CDI induced by vancomycin and metronidazole, so it may help us identify previously neglected CDI cases induced by vancomycin and metronidazole. In this way, we can timely take measures, such as stopping taking medicine, changing medicine and etiological treatment, to protect patients from unnecessary sustained injury. Second, due to the potential high CDI risk signals of vancomycin and metronidazole, our results provided an opportunity to investigate further the CDI risks of vancomycin and metronidazole, which may affect future clinical practice in primary prevention of CDI and antimicrobial stewardship.
In addition to detecting ADR signals of 73 antibacterial drugs at the SMQ level and PT level and adjusting signal detection results for metronidazole and vancomycin, we also integrated the positive-negative distribution of their ADR signals, and the total number of positive signals was between 0 to 11 for each antimicrobial drug. If an antimicrobial drug has a relatively large total number of positive signals, it may mean that its risk of CDI is relatively high [26]. For example, in this study, ceftriaxone, one of the antimicrobial drugs belonging to the third generation of cephalosporins and one of the well-known high-CDI-risk antimicrobial agents, was the only drug showing 11 positive signals. In this regard, this indicator concisely summarized the CDI risk characteristics of antimicrobial drugs, facilitating to get a quick understanding of the risk of different antibacterial agents.
Although this study comprehensively summarized the CDI risk of antibacterial drugs by using a pharmacovigilance database, there were also some inevitable limitations in this study. First, due to the intrinsic limitations of the pharmacovigilance database, the fact that un-peer-reviewed data, underreporting, Weber effect and notoriety bias may lead to biased results [25,27,28]. Second, due to the total number of patients exposed to each antibacterial drug is unclear, the incidence of CDI for an antibacterial drug cannot be determined. Third, patient gender, age, concomitant therapeutic drugs, dose and duration of antibiotic use, and comorbidities may influence the occurrence of CDI, but it is almost impossible to shield the potential interference of those factors to our results due to the intrinsic limitations of the pharmacovigilance database. Fourth, the ADR signal result only represents the strength of the statistical association between the drug of interest and AE of interest, so a well-designed study is still needed to verify whether there is a true causality.

Data Source
The data in this study were obtained from the FAERS database, a large international pharmacovigilance database with voluntary reporting nature, which recorded ADRs information related to post-market, FDA-approved medications as well as natural substances, vaccines and medical devices [29]. It currently publicly opens more than 16 million drug post-marketing AEs records reported by manufacturers, consumers and healthcare professionals and updates quarterly. The recorded information in the database includes but is not limited to patient demographic information, report sources, medication information, AEs involved and patient outcomes [30]. Those data are highly structured and can be retrieved, collected and downloaded from the openFDA platform by constructing an appropriate retrieval statement through an application programming interface (API) [31]. In this study, we summarized and analyzed CDI reports related to ADE between 1 January 2004 and 31 December 2022 in FAERS.

Identification of CDI Reports Associated with Antibacterial Drug Use in FAERS
The FAERS reporting system uses the PTs in Medical Dictionary for Regulatory Activities (MedDRA) to standardize AEs occurring in patients [30]. SMQs are a series of PT sets that potentially indicate the same medical condition, which was developed to optimize data retrieval and signal detection in pharmacovigilance activity [32]. Within an SMQ, PTs can be further divided into narrow-scope PTs and broad-scope PTs according to the degree of association with the condition or area of interest [33]. Among them, the PTs with a narrow scope are closely related to the condition or area of interest, while such association is relatively weak for PTs with a broad scope.
Pseudomembranous colitis is an inflammatory condition of the colon characterized by the presence of yellow-white exudative plaques that coalesce to form pseudomembranes on the mucosa, and it is usually a marker of severe CDI [34]. Meanwhile, it is also one of the SMQs in MedDRA that includes many PTs potentially indicating CDI, so it can be used to identify CDI-related reports in FAERS. In order to improve the accuracy of case identification and signal detection, in this study, only narrow-scope PTs of pseudomembranous colitis (SMQ) in MedDRA 23.0 were selected to retrieve CDI-related reports in FAERS (Table 1). According to the ATC classification system, if one of the generic drug names recorded in the "patient.drug.openfda.generic_name" field can be classified into "antibacterials for systemic use (J01)" in a report, this report is considered CDI reports associated with ADE and included in the final analysis.

ADR Signal Detection Method
Disproportionality analysis is a kind of technology used to detect ADR signals at present. Based on the classical two-by-two contingency table (Table 5), researchers can compare the differences between the occurrence frequency and background frequency for target drugs and target AEs. The reporting odd ratio (ROR) is one of the well-established disproportionality analysis methods, which calculates the ratio of the odds of a selected drug versus all other drugs for a certain AEs compared to the odds of the same drugs for all other AEs recorded in FAERS to detect potential ADR signals [35]. In this study, we used the ROR and its corresponding 95% confidence intervals (CIs) to identify ADR signals, and the ROR and its 95% CI can be calculated by the following formula: (2) Table 5. Two-by-two contingency tables for disproportionality analysis.

Drug of Interest Other Drugs Total
When the lower-bound 95% CI of ROR was above 1.0 with at least three cases (a ≥ 3 in Table 5), it was considered a positive signal, suggesting a potential risk of the target AE caused by the target drug; instead, if the lower-bound 95% CI of ROR and the number of cases cannot meet the above-mentioned threshold, it was regarded as a negative signal, suggesting the statistical connection between target AE occurrence and target drug use is weak [26,36]. To some extent, the ROR value represents a statistical correlation between the drug of interest and AE of interest, and the ROR value is larger, the stronger the statistical correlation. Using this indicator, we can highlight the AE that may be induced by a certain drug and conduct a further investigation so as to inform the possible risk; on the other hand, we can also use it to compare the risk of different drugs causing the same AE, so as to guide the selection of therapeutic drugs or discontinuation of a culprit drug [17].

Data Collection and Analysis
With reference to the API build guideline issued by the openFDA (https://open.fda. gov/apis/drug/event/how-to-use-the-endpoint/, accessed on 1 January 2023), we can retrieve and download the target reports for further analysis. The specific data collection and analysis steps of this study are as follows.
Firstly, by using the R package "httr" to call API, PTs in Table 1 were used to retrieve target reports in FAERS, and the returned dataset was downloaded in "json" format. Secondly, the R package "jsonlite" was used to read the downloaded dataset and extract the reports information, including Safety Report ID number, patient demographics, report time, report sources, medication use and outcomes. Thirdly, generic drug names recorded in the "patient.drug.openfda.generic_name" field were used to further identify reports associated with antibacterial drug use, and report characteristics were summarized. Fourthly, the ADR signals at the SMQ level and PT level were detected by calculating the ROR value and its 95% CI by using disproportionality analysis, and 11 signals were generated for each antibacterial drug. Finally, the positive-negative distribution of signals was summarized.
In this study, R version 4.1.0 (R Foundation for Statistical Computing, Vienna, Austria) was used for data processing and analysis.

Conclusions
The vastness, authenticity and accessibility of FAERS data have made it an important resource for evaluating drug safety cost-effectively. In this study, CDI reports associated with ADE in FAERS were summarized, and the CDI risk of different antibacterial agents was explored. As the first study to evaluate CDI risk related to antibacterial drug exposure using a pharmacovigilance database, our study provided a preliminary picture of CDI induced by antibacterial drugs in the real world that can help to better primary prevention for CDI and antimicrobial stewardship. Meanwhile, the potentially high CDI risk of metronidazole and vancomycin that may have been previously overlooked was detected, and it deserved further attention from regulators, health professionals and others involved in antimicrobial stewardship. Of particular note, however, our study as a pharmacovigilance study using the FAERS database only provided a statistical association between CDI occurrence and antibacterial drugs, so further well-designed study is still necessary to validate the causality.
Supplementary Materials: The following supporting information can be downloaded at: https:// www.mdpi.com/article/10.3390/antibiotics12071109/s1, Table S1: Pharmacovigilance signal detection results for antibiotic associated colitis; Table S2: Pharmacovigilance signal detection results for clostridium bacteremia; Table S3: Pharmacovigilance signal detection results for clostridium colitis; Table S4: Pharmacovigilance signal detection results for clostridium difficile colitis; Table S5: Pharmacovigilance signal detection results for clostridium difficile infection; Table S6: Pharmacovigilance signal detection results for clostridial infection;