Electrolyte Disorders Associated with Piperacillin/Tazobactam: A Pharmacovigilance Study Using the FAERS Database

Electrolyte disorders (EDs) can disrupt normal bodily functions and lead to life-threatening complications. We evaluated whether piperacillin–tazobactam (TZP), a widely used antibiotic for moderate-to-severe infections, is associated with electrolyte imbalances via a disproportionality analysis of a self-reporting pharmacovigilance database. We searched The US Food and Drug Administration (FDA) Adverse Event Reporting System (FAERS) from 2004 to 2018 for EDs related to TZP and calculated three data-mining indices—the proportional reporting ratio (PRR), the reporting odds ratio (ROR), and the information component (IC)—compared to all other drugs. Signals were defined when one of the three criteria of the indices was met. For the signals detected in the initial analysis, further disproportionality analyses in relation to other penicillins were conducted using the same method. A total of 9829 reports related to TZP with 36,207 TZP–adverse event pairs were retrieved. Among 10 EDs, hypokalemia was detected as the only significant signal (PRR 2.61; ROR 2.61, 95% CI: 2.17–3.14; IC 95% lower CI: 1.11) compared to all other drugs. Compared with other penicillins, hypokalemia remained a significant signal for TZP using IC (95% lower CI: 0.26). In conclusion, TZP was significantly associated with hypokalemia.


Introduction
Electrolytes are essential for basic life functioning [1]. They play a crucial role in maintaining electrical neutrality in cells, generating and conducting action potentials in nerves and muscles including cardiac and respiratory muscles, and regulating osmolality and volume of body fluids and blood pressure [1,2]. Thus, electrolyte imbalances can cause a variety of medical problems, from mild abnormal body functions to life-threatening complications, depending on the involved electrolytes and severity.
Considering the broad use of TZP, it is important to evaluate TZP-associated key electrolyte disorders. We conducted a disproportionality analysis using a self-reporting pharmacovigilance database to detect signals of TZP-associated EDs in relation to all other drugs. For the EDs detected in the initial analysis, further analyses were performed to identify whether these EDs remained significantly associated with TZP compared to other penicillins (PCNs). Additionally, factors affecting EDs showing significant association in the initial analysis were explored. The clinical characteristics of the TZP-associated reports are presented in Table 1. Patients were mostly adults (age ≥ 19 years). Males accounted for a larger proportion than females. Most of the reports were associated with serious outcomes (95.6%), of which hospitalization (51.1%) was the most frequent. Health professionals and France were the main reporting source and country, respectively.  (Table 4). Hypokalemia was identified as a signal in more PCNs when compared to all other drugs than when compared among PCNs.

Other Outcomes
Of 9829 reports related to TZP, only 8382 contained information regarding both sex and age. These reports were included in a logistic regression analysis to explore the risk factors for TZP-associated hypokalemia. Univariate logistic regression analysis identified female sex as a risk factor for hypokalemia and indicated that age did not affect hypokalemia development in TZP reports. Female sex remained a predictor of hypokalemia after adjusting for age ( Table 5). As per the results of the Hosmer-Lemeshow test, the degree of fit was good (p = 0.847).

Discussion
Electrolytes play a vital role in maintaining basic homeostasis in the body [1]. EDs can interrupt normal bodily functions and may lead to life-threatening complications depending on their seriousness [1,2]. TZP is one of the most commonly used antibiotics for treating severely infected patients who are prone to electrolyte imbalances [3][4][5]. Therefore, it is necessary to determine whether TZP is associated with EDs. We conducted a disproportionality analysis using the FAERS database and found that hypokalemia was the only signal of 10 TZP-related EDs and remained a signal when compared to other PCNs. Hypokalemia is indicated as an adverse effect on the TZP label. Unexpected signals of EDs for TZP were not detected despite the presence of other ED-related reports. To our knowledge, this is the first study to evaluate TZP-related EDs using a pharmacovigilance database.
Hypokalemia is a common ED in clinical practice. Mild hypokalemia is generally asymptomatic; however, severe hypokalemia can result in life-threatening complications [19][20][21]. Cases of serious hypokalemia that caused arrhythmia or were refractory to potassium treatment and required discontinuation of TZP therapy have been widely reported [10][11][12][13][14][15][16]. Although TZP-associated hypokalemia is considered a rare adverse event, recently, a randomized clinical trial [7] and two observational studies [8,9] reported a remarkably high incidence of hypokalemia during TZP therapy (12.6%, 13.9%, and 24.8%, respectively). This study adds support to previous results by detecting the hypokalemia signal for TZP via a disproportionality analysis of a large real-world dataset.
Hypokalemia was identified as a signal in most PCNs compared to all the other drugs in our study. PCN-associated hypokalemia has been reported across penicillin classes [7][8][9][10][11][12][13][14][15][16][22][23][24][25][26][27][28][29][30][31]. Penicillins act as non-reabsorbable anions that generate a transmembrane potential gradient in the cortical collecting tubules and increase potassium secretion, leading to hypokalemia [21,23]. Studies [8,9,23,24,29] that assessed the incidence of hypokalemia associated with PCNs showed various related incidences. Studies that investigated TZP-associated hypokalemia reported incidences of 13.9% and 24.8%, respectively [8,9]. Other studies that assessed the incidence of hypokalemia with flucloxacillin reported a 23.7-42% incidence [23,24]. Another study that evaluated the safety of nafcillin and oxacillin, including the risk of hypokalemia, reported that the nafcillin group showed a higher incidence of hypokalemia (51%) than the oxacillin group (17%) [29]. The results of these studies indicated nafcillin is associated with a higher incidence of hypokalemia, followed by flucloxacillin, TZP, and/or oxacillin. Similarly, in this study, the signal strength of hypokalemia differed between each PCN and was almost consistent with the results of previous observational studies. Nafcillin showed the strongest signal, followed by flucloxacillin, TZP, and/or oxacillin, in comparison to all other drugs or other PCNs. This finding will provide useful information in selecting specific PCN among those with similar antibacterial spectra. For example, the antistaphylococcal PCNs nafcillin and oxacillin are recommended as the primary choice for severe and invasive methicillin-susceptible Staphylococcus aureus infection [32][33][34]. Considering the similar costs and the likely equal efficacy of these agents, clinicians may consider prescribing oxacillin over nafcillin for invasive methicillin-susceptible Staphylococcus aureus infection, especially in patients at a risk of developing hypokalemia [35].
Previous studies have reported several risk factors for developing hypokalemia. Female sex was a frequently mentioned risk factor for hypokalemia [8,[36][37][38]. Our study confirmed these results. Most of the potassium in the body is found in muscle cells. In general, females have lower muscle mass than males. Thus, females might have low total exchangeable potassium levels, resulting in a higher risk for hypokalemia [36]. Older age has been reported as another risk factor owing to the accompanying low body mass, polypharmacy, and malnutrition [8,9,36,39]; however, this remains controversial, with one study reporting that younger age was a significant predictor of hypokalemia [40]. In this study, there was no significant difference among the age groups in the number of hypokalemic and non-hypokalemic reports. These results, however, should be interpreted with caution, because analysis with different variables was not performed, since the data provided were limited in scope.
This study had some limitations. First, as part of the pharmacovigilance data, underreporting and reporting bias may exist. The reporting rate may vary depending on the AEs [41], but it is generally~6% on average [42]. Specifically, non-severe or non-serious AEs may have been under-reported [43]. Most TZP-related reports included in this study were of serious cases. Second, there are duplicate reports in the FAERS database. Although we manually deleted all the duplicates that we detected, it is almost impossible to exclude all the duplicates because the same report may be submitted by different reporters with different identification numbers and at different times. Third, it has been suggested that AE-reporting increases over the first two years after regulatory approval of a drug, and then declines continuously (i.e., Weber effect) [44,45]. All the PCNs included in our study were launched long before 2004, and the use of some of them decreased due to the emergence of alternative agents. This may have affected the results of the disproportionality analysis for each PCN. Therefore, generalizing the findings should be done with caution, and well-structured comparative studies are warranted. Despite these limitations, this study is valuable, and its findings are noteworthy. Our study is the first to assess TZP-associated EDs using a pharmacovigilance database and includes the first analysis that evaluates the relative signals of hypokalemia among PCNs. These results will provide valuable information for selecting PCNs and improving their safe use. Moreover, given that this study used the FAERS database, which includes worldwide reports, and a long study period, from 2004 to 2018, we can guarantee the representativeness of the international population.

Data Sources
The FAERS database, which contains information on AEs and medication error reports submitted to the FDA voluntarily by healthcare professionals and consumers, as well as mandatory reporting by manufacturers and distributors [46], was mined. We used reporting data files published by the FDA on a quarterly basis on its website from 2004 to 2018 [47].
This database system transitioned from the Legacy Adverse Event Reporting System (LAERS) to the current system on September 10, 2012. Some changes were introduced to the FAERS database structure and existing field contents. Both LAERS and FAERS data consist of seven data tables: patient demographics and administrative information (DEMO), drug information (DRUG), adverse events (REAC), patient outcomes (OUTC), report sources (RPSR), drug therapy start and end dates (THER), and indications for use/diagnosis (INDI) [48]. The main differences between the LAERS and FAERS data are the renaming of the key fields in the DEMO table: "isr" and "case" in LAERS to "primaryid" and "caseid" in FAERS, respectively. The "isr" and "primaryid" are the primary link fields (primary key) between the tables [48].
Drugs in the DRUG table are classified as primary suspect, secondary suspect, concomitant, or interacting [48]. AEs and indications in the REAC and INDI tables are coded according to the standardized terminology of a hierarchical system in the Medical Dictionary for Regulatory Activities (MedDRA) at the level of the preferred term (PT) [48,49]. Each report can include more than one drug and more than one AE, resulting in more than one drug-AE pair per report.
The PCNs used as comparators are listed in Table 6. We selected PCNs that had been used in the United States during the study period and flucloxacillin, which was not available in the US but is commonly used in European Union countries.

Data Extraction
The quarterly LAERS and FAERS ASCII data files from 2004 to 2018 were downloaded from the FDA website. A case may have multiple versions, i.e., an initial case version and/or one or more follow-up case versions. The latest version of a case contains the most up-to-date information. Thus, the final version of each case was selected for the disproportionality analysis. In addition, specific reports regarded as erroneous on the FDA website were omitted [48]. Data reported in different units, such as age, were standardized. Cases that contained missing or inappropriate data in the identification key row, drug name, or reaction row were removed. Drugs reported as concomitant in the 'role cod' field of the DRUG table were excluded.

Statistical Analyses
Cases were classified by a specific AE versus all other AEs and a specific drug exposure versus all other drug or other PCNs exposure. The specific AE was hypokalemia or one of the other EDs, and the specific drug was TZP or one of the other PCNs. Two-by-two contingency tables were generated for the disproportionality analysis. Disproportionality analysis is a statistical method used to detect AE signals. In this study, for each drug-AE pair, frequentist and Bayesian methods were used to calculate disproportionality using PRR, ROR, and the Bayesian confidence propagation neural network (BCPNN) of IC [50][51][52][53]. The PRR is the reporting rate of one specific event among all events for a given drug compared to those for all other drugs in the database [50]. For the PRR, a signal is detected if the frequency is ≥3 and PRR is ≥2 with an associated χ2 value ≥4 [51]. The ROR is the odds ratio of one specific AE versus all other events for a given drug compared to those for all other drugs in the database [52]. For the ROR, a signal is detected if the frequency is ≥3, ROR is ≥2 with an associated χ 2 value ≥4, and the lower limit of the 95% two-sided confidence interval (CI) exceeds 1 [51,52]. The signal metric in BCPNN is IC, which can be additional information obtained on the probability of the event by specifying a drug, and a signal is considered when the lower 95% CI exceeds 0 [53]. In this study, AEs were defined as signals when at least one of three indices met the criteria indicated above. The characteristics of TZP-associated reports were expressed as the frequency and proportion of variables. To explore the risk factors for each ED detected as a signal, univariate and multivariate analyses were performed for age and sex using the enter method based on the reports that included information on both sex and age, and p < 0.05 was considered statistically significant. The Hosmer-Lemeshow test was used to examine the goodnessof-fit for logistic regression models, and p > 0.05 was considered statistically significant. Statistical analyses were conducted using PASW Statistics 18.0 (IBM, Armonk, NY, USA). This study was approved by the institutional review board of Kangbuk Samsung Medical Center (approval no. 2020-08-041).

Conclusions
Hypokalemia was the only significant signal of TZP-associated EDs compared to all other drugs and was still significant when compared to other PCNs.