Nephrotoxic drug burden among 1001 critically ill patients: impact on acute kidney injury

Acute kidney injury (AKI) frequently occurs in critically ill patients and is associated with high morbidity and mortality in the short and long term [1‐4]. AKI is of multifactorial origin in most critically ill patients [5, 6]. Beyond fluid loading and general perfusion pressure maintenance, research failed until now to identify effective specific interventions to prevent AKI [7‐9]. Pathophysiology of AKI is complex, involving multiple pathways including inflammation, hypoxia, and oxidative stress, leading to renal cellular injury [4‐6]. Drug toxicity acting through several of those pathways is a potentially modifiable risk factor for AKI [10‐14]. Indeed, given the high risk of short-term death, physicians may consider the overall benefit/risk ratio in favor of giving drugs despite their potential for damaging the kidneys. Nevertheless, even modest kidney dysfunction is associated with adverse patients’ outcome [15, 16]. The multifactorial nature of AKI and the numerous nephrotoxic drugs potentially delivered to critically ill patients make it very challenging to delineate attributable risk of AKI to specific drugs. Epidemiologic studies identified drug toxicity as a contributing factor in 15–25% of AKI cases [17‐20]. Conversely, studies investigating specific contribution of individual drugs to AKI either did not identify any attributable risk, showed conflicting results, or confirmed clinical relevance of toxicity [21‐26]. Large AKI epidemiology studies did not specifically evaluated nephrotoxic causes of AKI which represents a important limit. Conversely, studies specifically addressing nephrotoxicity were focused on one drug class only and did not evaluate the global nephrotoxic burden patients experience. To the best of our knowledge, the literature is lacking a large multicenter descriptive study of nephrotoxic prescription pattern in ICU patients evaluating the link between nephrotoxicity and AKI. Such knowledge is not only descriptive or academic in the sense that identifying nephrotoxic prescription burden and its impact on AKI may lay the foundations for interventional trials minimizing this potential aggression. Tackling drug toxicity represents a unique and simple mean to actively impact AKI in the ICU. The objective of this study was to give a large multicenter description of nephrotoxic drug prescription in ICUs, to semi-quantitatively evaluate the nephrotoxic burden experienced by critically ill patients, and to investigate its relationship with AKI.

This multicenter prospective descriptive cohort study was conducted in ten ICUs in France. Reporting follows the STROBE guidelines for observational studies (http://​www.​equator-network.​org). The study lasted 8 weeks, divided into 2 inclusion periods of 4 weeks in each participating center: one in the winter and one in the subsequent summer in the years 2014–2015. All adult patients were included except patients who underwent renal replacement therapy within the past week, on a chronic basis or the day of ICU admission and patients with expected very short stay (< 48 h).

Data were prospectively collected in an electronic clinical record file. None of the participating centers had a clinical pharmacist involved at the bedside. Aside of demographic data, admission diagnosis, co-morbidities, and severity of illness (simplified acute physiology score 2 (SAPS2) [27]) exposition to nephrotoxic medication and occurrence of AKI were recorded.

Nephrotoxic drug delivery was recorded daily during the first 7 days in the ICU, using a predefined list of drugs based on a literature review performed prior to study initiation (Table 1) [10‐12, 28]. Beta-lactam antibiotics were recorded as potentially nephrotoxic only when given at high dose for neurologic infections or endocarditis.

AKI incidence was recorded over the same period, based on urine output and serum creatinine concentration according to the Kidney Disease Improving Global Outcome classification (KDIGO), see Additional file 1 for details [29].

Patients’ outcome was further evaluated recording renal replacement therapy requirement, ICU, and hospital lengths of stay and mortalities.

Characteristics of the 1001 patients included are presented in Table 2. No missing data were observed on the main variables of interest. Six hundred and seventeen patients (62%) received at least one nephrotoxic drug during their first 7 days in the ICU. Among those, 184 (30%) received two and 119 (19%) received three or more different nephrotoxic drugs. Most frequent nephrotoxic drugs received in the ICU are presented in Table 3.

AKI occurrence is depicted in the study flow chart (Fig. 1): 396 (40%) patients experienced AKI the day of ICU admission. Among the 286 patients with KDIGO stage 1 or 2 at ICU admission, 138 (48%) experienced AKI worsening with development of stage 2 and/or 3 of the KDIGO classification within 7 days. Thus, a total of 351 (35%) patients experienced either occurrence or worsening of AKI in the ICU (cases), subsequent to potential nephrotoxic drug prescription in the unit. The median day of occurrence or worsening was day 2 (interquartile range days 2–4). Forty-nine patients (5%) underwent renal replacement therapy. AKI occurrence as well as its worsening during the first 7 days in the ICU were independent risk factors of death in the hospital (death hazard ratio (HR) of, respectively, 1.45, IC₉₅ [1.06–1.98], p = 0.021 and 2.00, IC₉₅ [1.38–2.88], p < 0.001) for AKI after adjusting for admission SAPS2.

Of note before ICU admission, a large proportion of patients received nephrotoxic medication: 644 (64%) received at least one nephrotoxic medication within the 48 h of ICU admission, either as chronic maintenance therapy (n = 530, 53%), or as a newly initiated prescription (n = 297, 30%).

Among the 351 cases, 327 could be matched to 327 controls (representing 195 distinct patients due to matching with replacement). Main characteristics of cases, controls, and patients not included in the matched cohort are presented in the Additional file 1: Table S1. The proportion of cases of AKI worsening as compared to control patients increased with increasing nephrotoxic burden prior to the index day, exhibiting a dose–response relationship (Fig. 2). Conditional logistic regression among matched cases and controls showed a significant increased risk of AKI development/worsening with increasing nephrotoxic burden prior to the index day: OR 1.20 IC₉₅ [1.04–1.38] (p = 0.015) in the whole matched population. Nephrotoxic burden prior to the index day was 0.86 ± 1.30 drug.days among control patients and 1.20 ± 1.76 drug.days among cases (median and interquartile ranges of, respectively, 0 [0; 1] and 1 [0; 2] (Fig. 3). Among cases and controls, AKI worsening was independently associated with hospital mortality (HR = 3.17 [1.81; 5.54] p < 0.001, adjusted on admission SAPS2). Among the most frequently prescribed nephrotoxic drugs, namely iodinated contrast media, diuretics, and antibiotics, the relative contribution of each drug to the difference in nephrotoxic burden observed between cases and controls is presented in the supplemental digital content (Additional file 1: Table S2).

Sensitivity analysis according to severity of disease showed that the link between semi-quantitative evaluation of nephrotoxic burden and AKI was apparent in the patients with lower severity of disease (admission SAPS2 below the median value of the population of 48): OR of 1.26 IC₉₅ [1.03–1.56], p = 0.026. Among the sickest patients (admission SAPS2 higher than 48), the association did not reach statistical significance: OR 1.13 IC₉₅ [0.92–1.37], p = 0.245 (Fig. 3). Nephrotoxic burden among cases of AKI occurrence/worsening and of controls was plotted according to severity of disease in Fig. 4. Interestingly, whereas nephrotoxic burden steadily increased with severity of disease among control patients, the opposite was observed for cases of acute kidney injury occurrence/worsening (Fig. 4). Sensitivity analysis performing matching without replacement of controls showed results similar to the primary analysis: OR 1.19 IC₉₅ [1.12–1.27] (p = 0.004), as well as using wider matching margins (data not shown). Sensitivity analysis among the subgroup of patients admitted without AKI showed results similar to the whole population in favor of a significant impact of nephrotoxic burden on AKI occurrence: 192 cases matched with 192 controls, OR of nephrotoxic burden for AKI occurrence 1.37 IC₉₅ [1.12–1.69] (p = 0.003). Adjusting the conditional logistic regression on sepsis diagnosis, the day of ICU admission showed similar results to the main analysis: OR = 1.18 IC₉₅ [1.02; 1.37], p = 0.023. Specifically evaluating diuretics, they were more frequently prescribed among cases of AKI occurrence or worsening (23%) than among control patients (16%), p < 0.001. No interaction (p = 0.77) was observed with the timing of diuretic administration in the course of the ICU stay (first 3 days versus only later on).

Both prescription of nephrotoxic drugs and AKI appeared very common among critically ill patients. Adjusting for confounders, the former, as semi-quantitatively evaluated through the drug.days nephrotoxic burden, was independently associated with the latter, overall, and among the patients with lower severity of disease. As AKI was confirmed to be an independent mortality risk factor, one may speculate on a negative link between nephrotoxic drug administration and patient outcome, a hypothesis to be tested in an interventional trial. Austin Bradford Hill proposed consensual criteria to assess a potential causal relationship between exposure and disease [32]. Whereas plausibility, analogy, and coherence may be assumed based on preclinical data, the present work adds evidence in favor of a causal link among several other Hill criteria. (1) Temporality was taken into account as nephrotoxic burden was considered only before the index day among cases and controls. (2) Consistency is reinforced by the large multicentric nature and the sensitivity analyses of the present work. However, given that nephrotoxic burden, as quantified in the present study, is a relatively novel approach, it should be further validated among independent patient cohorts. (3) Biological gradient (i.e., dose–response relationship) was observed with an increasing risk of AKI for increasing nephrotoxic burden (Fig. 2). (4) Specificity of the link between nephrotoxic burden and AKI may be assumed as case–control matching accounts for known and potentially for unknown confounders. (5) Strengths of the association may be considered important with an odd ratio of 1.20 for each 1 drug.day increase in nephrotoxic burden, i.e., one additional nephrotoxic drug delivered or one additional day of therapy with such a drug (e.g., 3.60 OR for a 3-day course of aminoglycosides). Last but not least, Hill criteria, i.e., clinical experimental evidence, cannot be assumed given the observational design of the present work. Nevertheless, the present observations give important information to design an interventional trial on the topic. A trial evaluating a reduced nephrotoxic burden bundle (restricting prescription indications, dose, and duration) to prevent AKI and associated poor outcome would need to specifically address iodinated contrast media, loop diuretics, and aminoglycosides, the main contributors to excess nephrotoxic burden among AKI cases (Additional file 1: Table S2, Additional digital content), and target patients with lower severity of disease, early in the ICU stay. Furthermore, evaluation of the impact of nephrotoxic burden on AKI recovery may deserve evaluation. In the meanwhile, monitoring nephrotoxic burden by clinical pharmacists and/or the use of electronic alerts may be interesting means to make clinicians aware of the nephrotoxic burden imposed to patients and its potentially deleterious consequences. The simplicity of drug.days nephrotoxic burden calculation may enable to envision relatively easy clinical implementation.

Our results are consistent with interventional studies showing an increased incidence of AKI when aminoglycosides are associated with beta-lactam antibiotics to treat sepsis compared to monotherapy [26]. Among pediatric critically ill patients, nephrotoxic drug prescription was associated with AKI, with the same drugs involved and observation of a similar dose–response relationship as observed in the present work [33].

Conversely, our results may seem in contradiction with evidence showing a lack of association between some potential nephrotoxic drugs and AKI, e.g., iodinated contrast media [23]. The present study assumes an additive nephrotoxic effect, thus revealing a total clinically relevant toxicity which may not be apparent when evaluating drugs separately. This additive toxicity hypothesis may be challenged given the heterogeneity of toxicity mechanisms. However, concerning iodinated contrast media toxicity for example, most experimental evidence was observed in two hit models, e.g., hypovolemia (such as may be induced by diuretics) prior to contrast administration, supporting additive phenomenon [34]. In a large case–control study, observing an association between nephrotoxic drug delivery and AKI similar to the present work, more than half of the patients experiencing AKI received several nephrotoxic medications supporting additive toxic phenomenon [35].

The present study has important limitations. First, despite matching, one cannot exclude the presence of some residual confounders. Only an interventional trial may definitively prove a causal relationship between nephrotoxic drug prescription and AKI occurrence. Second, one may discuss the list of nephrotoxic medications recorded. In particular for immune-allergic mechanisms, the potential drug list may be endless. Likewise, growing evidence in favor of nephrotoxicity of the combination of piperacillin–tazobactam with vancomycin emerged since study design [36]. However, only 77 patients (8% of the cohort) received vancomycin, and thus, the combination of piperacillin–tazobactam with vancomycin at best concerned a minority of patient unlikely to alter overall results. Recording potential nephrotoxicity of beta-lactam antibiotics only when given at high dose also represents a limitation. Indeed, aside of dose-related high urinary drug concentration representing a well-established risk factor, low pH and low urine output may also favor crystalluria, sometimes observed in patients receiving normal antibiotic doses [37‐39]. Conversely, the nephrotoxic potential of some drugs recorded may be debated such as acetylsalicylic acid or gadolinium recorded in a minority of patients [40]. However, one cannot exclude that drugs with a low potential nephrotoxicity may participate to the additive nephrotoxic burden. Given the lack of consensus concerning the drugs with the highest nephrotoxic potential in the ICU setting, the choice was made of a relatively large drug panel to be recorded. Third, various mechanisms of nephrotoxicity may come into play for a given drug, e.g., diuretics may induce AKI through hypovolemia and/or interstitial nephritis. Indeed, in the setting of fluid overload and congestive heart failure, diuretics may even prove beneficial on overall cardiovascular dynamics and thus on renal function. However, precise effective volemia is difficult to measure and could not be recorded in the present large-scale study; trials evaluating diuretics as a mean to prevent AKI occurrence or worsening in the critical care setting were negative [9]. Therefore, like others [33], we made the pragmatic choice of considering the potential deleterious effect of diuretics, mediated through induced hypovolemia, as predominant in the population under study. Finally, when calculating nephrotoxic burden, the same weight was attributed to all drugs. Weighting the contribution of each drug according to its nephrotoxic potential may theoretically be an appealing method to overcome this limit; however, set weighting coefficients would be highly debatable given the literature scarcity. In the same line, the administered dose and potential overdosing were not taken into account. This pragmatic choice represents a limitation of the study, but the simplicity of the nephrotoxic burden concept evaluated here may facilitate its adoption by clinicians. Indeed, physicians may easily calculate or be alerted of a nephrotoxic drug burden increase above a set threshold. Withholding one nephrotoxic drug prescription or 1 day of therapy with such a drug may represent a meaningful action for physicians to keep nephrotoxic burden low. Such clinical strategies will need to be investigated, defining thresholds and prescription scenarios.

The frequently observed prescription of nephrotoxic drugs to critically ill patients may be evaluated semi-quantitatively through computing drug.day nephrotoxic burden, an index significantly associated with subsequent AKI occurrence and worsening among patients with lower severity of disease. Those results call for the design of interventional trials prospectively testing strategies to reduce the nephrotoxic drug burden in the ICU.