Acute kidney injury in patients treated with vancomycin and piperacillin-tazobactam: A retrospective cohort analysis

Data Source

Patient data were collected from the University of Kentucky Center for Clinical and Translational Science Enterprise Data Trust (EDT). The EDT contains clinical data from the inpatient population of UKMC from 2006 to present. Data stored and updated nightly by the EDT includes: demographics, financial classification (Medicare, Medicaid, private insurance), provider-level detail (service line), medical diagnosis (ICD-9 codes), medical procedures (Current Procedural Terminology [CPT] codes), lab tests and results, medication administration details, visit details (age, length of stay, etc), and vital signs. This study was approved by the UKMC Institutional Review Board.

Data collected for each patient included: demographic data, visit details (length of stay, admitting and primary diagnosis codes, etc.), severity of underlying illness as defined by the Charlson Comorbidity Index (CCI), all serum creatinine levels drawn per visit, medication administration information (dose, date, and time administered), all VAN trough levels, receipt of other nephrotoxic agents, blood pressures, and receipt of vasopressors.

Outcome Ascertainment

The definition of AKI was based on the RIFLE (Risk, Injury, Failure, Loss, End-stage) criteria,⁹ with risk defined as a 25% to 50% decrease in estimated glomerular filtration rate (GFR), injury as a 50% to 75% decrease in estimated GFR, and failure defined as a greater than 75% decrease in estimated GFR. Loss and end-stage classifications were not assessed because of this study’s follow-up period. The adjusted Cockcroft and Gault equation¹⁰ was used to estimate GFR due to the inconsistency of weight availability in the dataset and concordance with the institution’s practice. Baseline creatinine clearance was calculated with the first serum creatinine obtained, and the minimum creatinine clearance was calculated using the maximum serum creatinine during each patient’s visit. The percent decrease in creatinine clearance was calculated from these 2 values. AKI status was defined as meeting any of the RIFLE criteria. Mortality was assessed for all patients and defined as the composite of inhospital mortality and discharge or transfer to hospice care.

Exposure Ascertainment

Hypotension exposure was defined as experiencing 1 of the following: mean arterial blood pressure less than 60 mm Hg, a diagnosis of hypotension by a physician, or receipt of vasopressors or inotropic agents. Days of therapy for each drug were obtained and combination days of therapy were calculated by including only those days in which the patient received both medications. Total days of therapy were calculated by the sum of all days receiving at least 1 study agent. Exposure to other nephrotoxic agents (eg, acyclovir, angiotensin converting enzyme [ACE] inhibitors, angiotensin II receptor antagonists, aminoglycosides, amphotericin B, cyclosporine, foscarnet, loop diuretics, nonsteroidal anti-inflammatory drugs, sulfonamides, tacrolimus, and tenofovir) were defined as receipt of at least 1 dose of the agent during hospitalization.

Statistical Analysis

Characteristics between groups were described with basic descriptive statistics. Continuous variables were compared with 1-way analysis of variance (ANOVA) or the Kruskal-Wallis test. Categorical variables were compared with chi-square or Fisher exact test. Yearly AKI trends were assessed with Pearson correlation coefficient. To control for differences in underlying severity of illness between groups, a subanalysis was performed in which the cohort was split into 4 groups (0, 1, 2 to 4, and ≥5 points) based on CCI. Univariate models for all covariates were created with probability of AKI as the outcome. Covariates significant after univariate were incorporated into the multivariate model, which was subsequently adjusted to achieve the highest predictive accuracy by minimizing the Akaike information criterion (AIC). Nephrotoxic agent exposures were included in the final multivariate model regardless of statistical significance in univariate analysis. Model fit was assessed with a standardized Hosmer-Lemeshow goodness-of-fit test.¹¹ All statistical analyses were completed with RStudio v 0.98 running R v 3.1.2 (R Foundation for Statistical Computing, Vienna, Austria).¹² All tests were 2-tailed and significance was defined at an alpha of 0.05.

Of 17,879 patients initially screened, 11,650 patients were evaluated, of which 5,497 received VAN and PTZ (VAN/PTZ), 3,055 received VAN alone, and 3,098 received PTZ alone. Table 1 contains basic demographic information. The mean age of patients was 52.5 years ± 16.8 years with 6,242 (53.6%) males. Patients receiving VAN/PTZ had higher CCIs than either monotherapy group and had significantly increased length of hospitalization. While patients in the combination therapy group were more likely to experience hypotension, concomitant nephrotoxic agent exposure was more common in the VAN monotherapy group.

RIFLE-defined AKI occurred in 1,647 (14.1%) across the entire cohort. AKI occurred in 21% of VAN/PTZ patients, 8.3% of VAN patients, and 7.8% of PTZ patients (P < 0.0001). RIFLE-defined risk, injury, and failure occurred more frequently in the VAN/PTZ cohort compared to the VAN and PTZ monotherapy groups (Figure). There were no differences in AKI rates between years studied (r² = 0.4732, P = 0.2). Patients in the VAN/PTZ group experienced AKI on average of 8.0 days after treatment initiation, compared to 8.7 days and 5.2 days for VAN and PTZ monotherapy groups, respectively. The composite of inhospital mortality and transfer-to-hospice care was more common in VAN/PTZ patients (9.6%) compared to monotherapy groups (VAN, 3.9%; PTZ, 3.4%), most likely due to the increased severity of illness.

In the subgroup analysis of patients with similar CCI, AKI incidence increased with severity of illness. When CCI was 0, 7.5% of patients experienced AKI compared to 11.2%, 16.4%, and 18.9% of patients when CCI was 1, 2 to 4, and ≥5, respectively (P < 0.0001). VAN/PTZ (range = 12.1% to 26.5%) was associated with greater AKI incidence than either VAN (range = 4.8% to 11.5%) or PTZ (range = 3.8% to 10.4%) alone in each subgroup (P < 0.0001 for all subgroups).

Factors associated with AKI in univariate analyses included treatment with VAN/PTZ, days of therapy, baseline creatinine clearance, transfer from outside hospitals, CCI, admission type, length of hospitalization, dehydration exposure, and hypotension exposure. Exposure to aminoglycosides, amphotericin B, ACE inhibitors, nonsteroidal anti-inflammatory drugs, tacrolimus, foscarnet, loop diuretics, sulfonamides, and tenofovir were all associated with increased odds of AKI in simple univariate logistic regression. Gender, age, year of treatment, angiotensin II receptor antagonist exposure, and cyclosporine exposure were not significantly associated with AKI incidence.

After multivariate logistic regression, monotherapy with VAN or PTZ was associated with decreased odds of AKI compared to VAN/PTZ therapy (aOR_VAN,0.48; 95% CI_VAN,0.41-0.57; aOR_PTZ, 0.43; 95% CI_PTZ, 0.37-0.50). No difference in AKI incidence was observed between VAN and PTZ groups (aOR_PTZ:VAN, 0.88; 95% CI, 0.73-1.08). Table 2 describes the relationship between AKI and other covariates included in the model. Increased odds of AKI were seen with concomitant administration of ACE inhibitors, amphotericin B, tacrolimus, loop diuretics, and tenofovir. Radio-contrast dye administration was associated with lower odds of AKI. Patients admitted urgently and emergently were at higher risk of AKI, while those admitted via the trauma center were less likely to experience AKI compared to patients who were electively admitted. Increased length of stay and duration of therapy were both associated with increased likelihood of AKI, independent of treatment group; however, durations of therapy beyond 12 days was not associated with increased AKI. Hypotension, as defined, and diagnosed dehydration both independently increased AKI odds. Aside from those older than 80 years of age, increasing age was not associated with increased AKI risk. Male gender was associated with a slight decrease in AKI rate. No evidence of overfitting was observed with the standardized Hosmer-Lemeshow P-value of 0.683, and the model provides good predictive accuracy with a C-statistic of 0.788.