Acute kidney injury (AKI) is a rapid decline in the kidneys’ ability to filter metabolic waste products from the blood, which manifests clinically as an abrupt and sustained increase in the serum levels of urea and creatinine with an associated disruption of salt and water homeostasis [1]. It is a common disorder of renal function that affects 7 to 18% of hospital inpatients and 20–200 per million population in the community worldwide [2], significantly increasing the risk of death, economic burden, and the probability of progression to end-stage renal disease and chronic kidney disease [3, 4, 5].

There are many known etiologies of AKI, mainly including sepsis, critical illness, circulatory shock, burns, trauma, cardiac surgery, major non-cardiac surgery, poisonous plants and animals, radiocontrast agents, and nephrotoxic drugs [6, 7]. Among them, drugs are the third to the fifth leading cause of AKI [8], accounting for about 18%–27% of all AKI cases [9], and even as high as 66% in elderly patients [10]. Due to the prominent position of drugs in the etiology of AKI, the 2012 AKI guideline of the Kidney Disease Improving Global Outcomes (KDIGO) group recommended that the management of potentially nephrotoxic drugs should be the basic intervention measures throughout the four stages of AKI (high risk of AKI, AKI stage 1, 2, and 3) [11]. In this context, it is of great significance for medical professionals to be familiar with the potential nephrotoxicity of drugs to ensure the management quality of AKI.

In clinical practice, medical professionals usually rely on primary, secondary, and (or) tertiary information sources to obtain information on nephrotoxic drugs [8]. However, this process may not run smoothly because of the substantial fragments and discrepancies in nephrotoxicity information provided by various sources [12, 13]. In response to this situation, some researchers have carried out literature review studies to integrate the nephrotoxicity information of drugs, highlight the high-risk drugs causing AKI, and recommend management measures [6, 9, 14, 15, 16, 17, 18, 19]. However, although these studies have made an excellent critical review of the literature and simplified the process for medical professionals to achieve and apply information about drug nephrotoxicity, they cannot provide comprehensive quantitative and comparative risk data of these drugs, which partly limits the practicability of the research results in high-risk drug monitoring, primary suspected drug identification, and drug withdrawal selection. Thus, it is necessary to explore a method to quantify and compare the AKI risk of drugs comprehensively.

At present, the massive real-world adverse event (AE) data in the spontaneous reporting system bring opportunities for understanding the risk characteristics of drugs [20, 21]. In terms of kidney safety, the pharmacovigilance database has been widely used to understand the AKI risk of specific drugs [22, 23], investigate the most common AKI pathogenic drugs [24], compare the risk differences of AKI among drugs in the same category [25, 26, 27], and explore risk characteristics of drug-induced AKI in the special population [28]. To some extent, these studies confirmed the feasibility of using pharmacovigilance databases to explore the risk characteristics of drug-related AKI in the real world. However, although the aforementioned studies provide quantitative AKI risk information of some drugs to varying degrees, it is almost impossible to comprehensively integrate the quantitative risk of drug-induced AKI based on these results due to the differences in drugs investigated, research methods, time span, and population selection. In addition, the potential AKI risks of many drugs have also not been quantitatively evaluated in pharmacovigilance research, which also hinders the comprehensive understanding and integration of drug-related AKI risk characteristics.

In this context, our objective is to assess the AKI risk of all the reported nephrotoxic drugs in the FDA Adverse Event Reporting System (FAERS) in a unified standard, trying to present the whole picture of drug AKI risks from the perspective of pharmacovigilance, so as to provide more comprehensive renal safety information for clinical AKI management.

Drug-induced AKI is an important clinical problem widely concerned by health workers. In the clinical management of drug-induced AKI, timely identification and withdrawal of potential nephrotoxic drugs play a cornerstone role [11]. However, due to the lack of unified quantitative indicators to judge the risk of AKI caused by different drugs, the determination and withdrawal of pathogenic drugs are still highly dependent on the personal experience of medical staff. Aiming at this problem, we reviewed 327,561 AE reports related to AKI in the FAERS database over the past 20 years to investigate the association between the occurrence of AKI and drug use. To our knowledge, such a large sample is unprecedented, which provides a solid foundation for our comprehensive investigation of the nephrotoxicity risk of drugs.

In this study, we present the etiological drug characteristics of AKI from multiple dimensions. First, we summarized a list of 1456 reported nephrotoxic drugs, involving 81 drug classes (Fig. 2). Among the 1456 reported nephrotoxic drugs, our results showed that the largest number of drugs belong to antineoplastic agents (Fig. 2b). Although it is well-known that the kidney is one of the main target organs affected by the toxic effects of antineoplastic agents, either traditional chemotherapeutics or novel molecularly targeted agents [35], antineoplastic agents did not show an obvious dominant position among reported nephrotoxic drugs in previous pharmacoepidemiologic studies [9, 13, 14, 15, 24, 36]. There are two possible explanations for the high proportion of antineoplastic agents in the drug list. One possible explanation is that there is a large base of antineoplastic agents used in the real world, which may increase the proportion of antineoplastic agents in the integrated drug list. Another possible explanation is that antineoplastic agents do play an important role in the occurrence of AKI in the clinical setting, whereas our study opportunistically caught a glimpse of it. However, for whatever reason, with the development and application of new antineoplastic drugs and therapies, the problem of antineoplastic agent-induced AKI has become increasingly prominent and has attracted wide attention [37, 38, 39]. Therefore, it is necessary for medical staff to have a deeper understanding of the AKI risk related to antineoplastic agents and provide timely prevention and treatment.

To further understand the causative drugs of AKI, based on the integrated drug list obtained from the FAERS database, we made statistics on the most frequently reported nephrotoxic single drugs and drug classes at the SMQ level and PT level, respectively (Figs. 3,4). As a whole, “antineoplastic agents” is the drug class with the third largest number of AKI reports at the SMQ level (Fig. 4), which further demonstrates that the use of antineoplastic agents is an important trigger of AKI. In addition, it is noteworthy that although only 17 (1.17%) of the 1456 reported nephrotoxic drugs were classified as “drugs for acid related disorders” (Fig. 2b), it is still the drug class with the largest number of AKI reports among the 81 kinds of drug classes. Meanwhile, in the statistics of single drugs at the SMQ level, four drugs (omeprazole, pantoprazole, esomeprazole, and lansoprazole) are among the top 10 most frequently reported nephrotoxic drugs (Fig. 3). The existing literature also suggested that “drugs for acid related disorders” and the drugs they contain are closely related to AKI [40, 41, 42]. Therefore, the above evidence suggests that we should pay more attention to the use of “drugs for acid related disorders” in future clinical practice, to identify and intervene in possible renal damage in the process of drug use in time. Similarly, the AKI risks of some other listed high-reporting drug classes (e.g., diuretics [43], antibacterials [44], and agents acting on the renin-angiotensin system [45]) and drugs (e.g., furosemide [46], vancomycin [47], and enalapril [48]) have also been confirmed in previous literature. Therefore, to some extent, the results mentioned above reflect the most commonly reported nephrotoxic drugs in the real world and can provide some reference information for managing nephrotoxic drugs. However, we also noticed that some drugs that are considered to have low nephrotoxicity (e.g., metformin, amlodipine, and metoprolol) also have high reporting frequency. There are two possible explanations for this abnormal report accumulation. One possible explanation is that it reflects the unknown renal injury model. For example, a recent animal study showed that metformin, even at low doses, exacerbated experimentally induced AKI and increased mortality in mice [49], but this effect has not been confirmed in the population. Another possible explanation is that diabetes and cardiovascular disease are two of the four main types of chronic diseases [50], but as mainstream therapeutic drugs, metformin, amlodipine, and metoprolol inevitably have a huge consumption base and corresponding high AKI report accumulation, even though they have a low risk of kidney injury.

Therefore, it should be kept in mind that although the high reporting frequency of drugs in the FAERS database partly reflects the possible nephrotoxicity of some drugs, the high reporting frequency is not absolutely equivalent to the high AKI risk, because the difference in drug use base in the real world will greatly affect this figure. However, the inherent limitations of the spontaneous reporting database make it impossible for us to calculate the actual incidence of AKI of various drugs with existing data to evaluate and compare the AKI risks of various drugs. In this context, we introduced the disproportionality analysis to evaluate the risk of AKI caused by different drugs and used the risk signal detection result as an alternative quantitative index of incidence to represent the AKI risk of drugs [20]. Although previous studies have also tried to quantify the AKI risk of drugs by applying the disproportionality analysis, the drugs they evaluated were usually limited to a specific class, such as immune checkpoint inhibitors [39], antibiotics [51], and sodium-glucose-cotransporter-2 inhibitors [52]. Meanwhile, due to the differences in research methods and research time span, the above research results cannot be integrated to perform an AKI risk comparison across drug classes. By contrast, our study expanded the evaluated drugs to 1456 drugs reported in the FAERS database in the recent 20 years (Supplementary Table 2), which provides resources for comparing AKI risks among drugs in a wide range under a unified standard, thus providing better support for timely identification and withdrawal of drugs causing AKI. For example, if a patient suspected drug-induced AKI is using cefazolin and diclofenac at the same time, cefazolin (ROR, 5.36; 95% CI, 5.00–5.74) is more likely to be the culprit drug than diclofenac (ROR, 2.03; 95% CI, 1.97–2.09) according to the signal detection results (Supplementary Table 2). In this regard, we can stop cefazolin or adjust cefazolin to other alternative medications accordingly in clinical practice. Similarly, if we find that there are drugs with high AKI risk signal strength in the prescription, we can especially strengthen the monitoring of patients’ renal function, so as to identify emerging kidney injury in time.

In addition, based on the risk signal detection results at the SMQ level and PT levels, we further summarized the AKI risk characteristics of 1156 drugs. On the one hand, we made statistics on the sum of positive risk signal counts of each drug and found that 1021 drugs have at least one positive risk signal at the SMQ level or PT levels (Table 4), revealing the statistical relationship between these drugs and AKI. To some extent, the sum of positive risk signal counts of drugs can be used as a simplified indicator reflecting the correlation strength between drugs and AKI [31]. For example, vancomycin with 15 positive risk signals theoretically has a higher AKI risk than cefixime with three positive risk signals (Supplementary Table 2), suggesting that the nephrotoxicity of vancomycin should be focused on in clinical use. Accordingly, we can locate some key drugs, and paying attention to these drugs is beneficial to the prevention, diagnosis, and treatment of drug-induced AKI. However, for some drugs with a large number of positive risk signals, the interpretation of AKI risk still needs to be cautious. For example, although there are case reports suggesting that nifedipine can induce acute interstitial nephritis [53], larger studies have shown the beneficial effect of nifedipine (calcium channel blocker) on renal outcomes and can be used as a selectable first-line therapy in non-proteinuric chronic kidney disease [54, 55]. Therefore, for drugs that are potentially beneficial to renal outcomes or treat kidney-related diseases, positive risk signals should perhaps be understood as a statistical relationship between drugs and diseases rather than a causal relationship. On the other hand, we integrated the risk signal distribution characteristics of the main drug classes at the SMQ level and PT levels (Fig. 5). Our results showed that “antineoplastic agents” was the drug class with the largest number of positive signal drugs at the SMQ level and most PT levels, which once again highlights its triggering effect on AKI and reminds us to pay enough attention to it. Besides, our result also reveals the differences in the proportion of risk signal distribution between some drug classes. Theoretically, a drug class with a higher proportion of positive risk signals has a higher overall AKI risk than a drug class with a lower proportion [31]. In this regard, our results provide a new perspective to understand the AKI risk difference among drug classes.

Besides summarizing the reporting frequency and risk signals of drugs, we also explored the histological injury type distribution of the main drugs and drug classes (Table 5). In previous studies, although it is known that the same drug may cause multiple histological types of renal injury [56], the distribution of drugs for histological injury type is not clear. Our study showed that tubulointerstitial injury is the most commonly reported histological injury type for most drugs and drug classes. In drug-induced tubulointerstitial injury, the drugs for acid related disorders (omeprazole, pantoprazole, esomeprazole, and lansoprazole) were obviously dominant. In a previous study, Gérard AO and colleagues [57] used the World Health Organization Safety Database (VigiBase) to identify drugs responsible for tubulointerstitial nephritis and also highlighted the central role of proton pump inhibitors (PPIs), which also supported our results. Therefore, for patients with a rapid decline in renal function during drug (particularly PPIs) use, tubulointerstitial injury may be the primary suspicion after excluding other non-drug causes. However, it should also be noted that some other drugs may have different histological injury characteristics. For example, it is reported that both tubulointerstitial injury and tubular injury can be induced by non-steroidal anti-inflammatory drugs [56], but our results showed that tubular injury induced by acetylsalicylic acid is more frequently reported than tubulointerstitial injury. In this regard, it is necessary to understand the histological characteristics of renal injury caused by different drugs further because they are of great significance in treatment selection.

Additionally, we further evaluated the time relationship between the occurrence of AKI and the start of medication. In previous studies, the average time of vancomycin-induced AKI was 24.9 days from the start of therapy [58]. Our results showed that the median time of vancomycin-induced AKI is 5 days, which is significantly shorter than the above-mentioned time interval. However, it has also been reported that the higher trough values of vancomycin are associated with shorter time intervals and increased histopathological damage [47, 59]. For the patients with initial vancomycin trough values exceeding 20 mg/L, the median time to AKI of vancomycin is about 6 days [60], which is basically consistent with our research results. Therefore, one possible explanation for the discrepancy is that the reporting preference (more likely to do spontaneous reports about severe reactions) makes the cases with high initial vancomycin trough values concentrated [61], and leads to a significantly shorter time interval. In our study, lansoprazole has the longest median time interval of AKI (1724 days), and we noticed that other PPIs also have similar long-time intervals. In previous studies, although the association between PPIs and AKI occurrence has been widely confirmed, the time relationship has not been elaborated in detail. In fact, due to unrecognized tubulointerstitial injury, long-term PPI use was usually associated with an increased risk of chronic kidney disease and kidney failure [56], which may partly explain the long-term interval of PPIs. However, it is noteworthy that although the median time interval of different drugs varies greatly, they are all classified as early failure types in Weibull’s shape parameter $\beta$, indicating that the AKI risk decreases with time. Undoubtedly, different drugs have different onset time characteristics in inducing AKI, so it is necessary for us to further understand the onset time characteristics of different drugs to help identify the possible culprit drugs.

Although this study presents the etiological drug characteristics of AKI from multiple dimensions, it also has to emphasize its limitations. First, although multiple PTs were used in this study to enhance the identification of drug-induced AKI cases, a non-standard diagnostic approach may still bring bias, and it is also difficult to ensure that the inherent AKI in a specific clinical scenario (e.g., the COVID-19 pandemic) will not be wrongly attributed to drugs [62]. Second, as an AE reporting system with a voluntary nature, the problems of underreporting, duplicate reporting, Weber effect, and notoriety bias inevitably exist in the FAERS database, which may damage the accuracy of our AKI risk signal detection results [33]. Third, many factors (e.g., patient age, sex, and accompanying disease) can affect the occurrence of AKI [63], but we can’t shield the influence of these factors on the AKI risk signal detection results. Fourth, our study mainly focused on the AKI risk of a single drug, but an AKI report usually records multiple drugs. The drug-drug interaction may indeed influence the quantification of AKI risk of drugs, but it is not deeply explored in this study. Fifth, the Weibull distribution analysis was used to explore the changes of AKI risk caused by different drugs with time in this study, but this method is usually used to identify lifetime distribution in reliability engineering and may bring unexpected methodological bias, so it is necessary to supplement the verification of external validity in future research. Finally, due to the lack of a model linking drug report accumulation with drug prescription/drug use frequency data, we cannot precisely compare AKI risks of different drugs by calculating the incidence. Although the results of risk signal detection are used as a substitute index of incidence to quantitatively evaluate the AKI risk of different drugs, its essence is the statistical association between drugs and AKI occurrence rather than causality. In some cases, such an association is indirect or even misleading, so it is necessary to apply additional evaluation tools (e.g., Bradford-Hill criteria) to verify the causal relationship between drug use and AKI [64].

The FAERS database has a large population, wide geographic coverage, and publicly available accessibility, which have qualified it as an important resource for investigating the potential adverse reaction risk of drugs. In this study, we provided a comprehensive review of the culprit drugs of AKI in the FAERS database, summarized a list of 1456 reported nephrotoxic drugs in the past 20 years, and made statistics on the top 10 frequently reported nephrotoxic drugs and drug classes. In addition, we applied disproportionality analysis to evaluate the AKI analysis of 1456 drugs uniformly, integrated the distribution characteristics of AKI risk signals of drugs on this basis, and found that 1021 drugs were associated with the occurrence of AKI to different degrees. Meanwhile, based on co-reported PT and available time data, we further explored the histological injury characteristics and onset time characteristics of reported culprit drugs. Although our study has unavoidable limitations due to the inherent nature of spontaneous reporting databases, it also provides a real-world perspective for understanding the AKI risk of drugs and can provide reference information for the identification and withdrawal of causative drugs in the clinic.

The original contributions presented in the study are included in the article/Supplementary Material; further inquiries can be directed to the corresponding authors.

Conceptualization, DL and QD; methodology, JZ and RZ; software, JZ and DL; data curation, JZ, RZ, YW, FW, TZ and DL; formal analysis, JZ, RZ, YW, FW, TZ and DL; writing—original draft preparation, JZ and RZ; writing—review and editing, RZ, JZ, YW, FW, TZ, DL, QD; visualization, DL; funding acquisition, DL and QD. All authors read and approved the final manuscript. All authors have participated sufficiently in the work and agreed to be accountable for all aspects of the work.

Not Applicable.

This research was funded by the Research Incubation Project of the Third Affiliated Hospital of Chongqing Medical University (KY24038), Senior Medical Talents Program of Chongqing for Young and Middle-aged, and Key Specialty Construction Project of Clinical Pharmacy in Chongqing.

The authors declare no conflict of interest.

Supplementary material associated with this article can be found, in the online version, at https://doi.org/10.31083/IJP44094.