Contrast-Induced Nephropathy in Intensive Care Unit Patients: a Propensity Score-Adjusted

Contrast-induced nephropathy in intensive care unit patients: A propensity score-adjusted study

Supplemental data

Supplemental Methods

Supplemental Tables

Supplemental Table 1: Absolute standardized differences of model covariates before and after propensity score matching

Supplemental Table 2: Changes in SCr following CT scan

Supplemental Table 3: eGFR 45 ml/min/1.73m2subset outcomes after propensity score analysis and adjustment for IV fluids given day of and day after scan.

Supplemental Table 4: eGFR >45 ml/min/1.73m2subset outcomes after propensity score analysis and adjustment for IV fluids given day of and day after scan.

Supplemental Table 5: Demographics of complete cohort before and after propensity score adjustment

Supplemental Table 6: Complete cohort unadjusted and adjusted outcomes

Supplemental Table 7: Demographics of eGFR 60 ml/min/1.73m2subset before and after propensity score adjustment

Supplemental Table 8: eGFR 60 ml/min/1.73m2subset unadjusted and adjusted outcomes

Supplemental Table 9: Demographics of eGFR 30-59 ml/min/1.73m2subset before and after propensity score adjustment

Supplemental Table 10: eGFR 30-59 ml/min/1.73m2subset unadjusted and adjusted outcomes

Supplemental Table 11: Demographics of eGFR <30 ml/min/1.73m2subset before and after propensity score adjustment

Supplemental Table 12: eGFR <30 ml/min/1.73m2subset unadjusted and adjusted outcomes

Supplemental Table 13: eGFR 45 ml/min/1.73m2 with stable preCT SCr cohort unadjusted and adjusted outcomes

Supplemental Table 14: eGFR >45 ml/min/1.73m2 with stable preCT SCr cohort unadjusted and adjusted outcomes

Supplemental Table 15: eGFR 45 ml/min/1.73m2 with no preCT-AKI cohort unadjusted and adjusted outcomes

Supplemental Table 16: eGFR >45 ml/min/1.73m2 with no preCT-AKI cohort unadjusted and adjusted outcomes

Supplemental Table 17: eGFR 45 ml/min/1.73m2 ED admission cohort unadjusted and adjusted outcomes

Supplemental Table 18: eGFR >45 ml/min/1.73m2 ED admission cohort unadjusted and adjusted outcomes

Supplemental Table 19: eGFR 45 ml/min/1.73m2 inpatient cohort unadjusted and adjusted outcomes

Supplemental Table 20: eGFR >45 ml/min/1.73m2 inpatient cohort unadjusted and adjusted outcomes

SUPPLEMENTAL METHODS

Study Design and Clinical Data Retrieval

Study design and implementation for this retrospective study were overseen by our institutional review board and conformed to Health Insurance Portability and Accountability Act guidelines on patient data integrity and the Declaration of Helsinki. The need for informed consent for patients who provided research authorization was waived. All clinical data were extracted from our electronic medical record (EMR) using a combination of relational database software (DDQB, IBM, Armonk, New York) and manual chart review. Data extraction of most clinical variables has been described previously (1-3). Patients with any history of dialysis, but who were not on dialysis at the time of ICU admission and CT scan, such as patients with a prior AKI event necessitating transient dialysis, were identified by a combination of CPT codes and manual chart review as previously described (3). Patients who received mechanical ventilation (Current Procedural Terminology (CPT) 31500 or International Classification of Diseases-9(ICD) code 96.7x) or who received blood transfusion (CPT 36430) in the 24 hours prior to CT scan were identified. Additional ICU clinical variables were retrieved for this study using the Multidisciplinary Epidemiology and Translational Research in Intensive Care (METRIC) database (4) as follows. Acute Physiology and Chronic Health Evaluation (APACHE) III scores and Sequential Organ Failure Assessment (SOFA) scores on day one of the ICU admission were retrieved for each patient. The average mean arterial pressure (MAP) in the 24 hours prior to CT scan was calculated for each patient.

Study Population

Many patients in the current study were included in previous publicationsthat examined the incidence of AKI, emergent dialysis, and mortality in patients who received a contrast-enhanced or unenhanced CT scan (2, 3, 5, 6). However, none of these studies specifically examined ICU patients or included acute clinical covariates such as APACHE and SOFA scores, mechanical ventilation, and transfusions in the propensity score models.

Adult patients (18 years or older) who provided research authorization were included in the current study if they 1) were admitted to an ICU from January 2006 to December 2014 at our institution; 2) received an unenhanced (noncontrast group) or IV contrast-enhanced (contrast group) abdominal, pelvic, and thoracic CT scan on the day of or any time during their ICU admission; and 3) had at least two pre-scan (within 7 days prior) and at least one post-scan (24-72 hours post) serum creatinine (SCr) result. Patients were excluded if they 1) had insufficient pre- and post-CT SCr results; 2) received intravenous or intra-arterial contrast material from another imaging study or procedure within seven days before or three days after the CT; or 3) had pre-existing renal dialysis requirements. For patients with multiple ICU admissions and/or who underwent multiple CT examinations over the study timeframe, only the first CT scan in the most recent ICU admission were included to avoid confounding bias.

At our institution there are no mandatory protocols for avoiding CIN; any prophylactic measures are used at the provider’s discretion. Volume repletion is commonly performed. Specific CIN prophylactic measures such as NAC and sodium bicarbonate are not recommended, as there is no strong evidence of their effectiveness.

PreCT Renal Function

All SCr data associated with each CT scan record were extracted from the EMR and temporally sorted with respect to the date of the scan. PreCT eGFR was calculated for each patient from the mean SCr result(s) 24 hours prior to CT scan using the MDRD (Modification of Diet in Renal Disease) equation based on the National Kidney Foundation Kidney Disease Outcomes Quality Initiative (KDOQI) recommendations as previously described (5). Patients were stratified by PreCTeGFR into >45 and 45 ml/min/1.73m2subsets for two reasons. First, when using traditional KDIGO CKD eGFR cutoffs of 60, 30-59, and < 30 ml/min/1.73m2subsets we had very few patients with eGFR < 30 ml/min/1.73m2who received contrast. These KDIGO-stratified subsets were assessed as described in the Sensitivity Analyses section below. Second, there is precedent for stratifying patients into eGFR > and 45 ml/min/1.73m2subsets because of the increased and under-recognized risks associated with Stage 3B CKD(7).

Patients were classified as having stable pre-CT SCr if the maximum difference between pre-scan SCr results was less than 0.5 mg/dL. Patients with differences greater or equal to 0.5 mg/dL were further classified as unstable, increasing or unstable, decreasing depending on the trend in SCr leading up to the scan. Multiple SCr results within 3 days of scan were first examined. If a patient did not have two SCr results in that time period the results within 7 days of scan were then examined. Patients with AKI in the 7 days prior to CT scan (PreCT-AKI) were identified through a combination of ICD-9 diagnostic codes and manual review of clinical notes. PreCTand maximum serum creatinine (SCr) results of patients identified as having PreCT-AKI were used to stratify the AKI into Stages 1-3 using KDIGO criteria.

Outcome Variables

The outcomes examined in this study were acute kidney injury (AKI), emergent dialysis, and death following CT scan. AKI was classified based on SCr and urine output (UOP)-defined KDIGO criteria(8). The term “Post CT-AKI” was applied to both contrast material–dependent renal injury after contrast-enhanced CT (PC-AKI), and contrast material–independent renal injury after unenhanced CT to provide a uniform definition of AKI [11]. The use of the term PC-AKI is used to specifically describe a significant excess risk of AKI following intravenous contrast exposure. PreCT SCr was defined as the mean SCr result 24 hours prior to CT scan. The maximum SCr in the 7 days following CT was used to determine whether patients met AKI criteria and to classify patients into Stage 1-3 AKI. Stage 3 AKI was defined using only KDIGO SCr criteria; patients who received dialysis without concurrent Stage 3 increases in SCr were not classified as having Stage 3 AKI. Urine output was calculated from hourly data provided by METRIC. Patients with UOP <5 ml/kg/hr for at least 6 hours in the 72 hours post-CT were identified as meeting KDIGO UOP criteria. Overall AKI incidences were defined using either SCr criteria or both SCr and UOP criteria. Cases of emergent dialysis within 7 days and death within 30 days of CT scan were identified as previously described (3). Patients who underwent dialysis were automatically classified with Stage 3 AKI, regardless of changes in SCr. The decision to start emergent dialysis was provider-specific and not based on strict institutional criteria. Total ICU length of stay and total hospitalization length of stay were also examined.

Comprehensive changes in SCr following CT were examined in the entire cohort and in the eGFR >45 and 45 ml/min/1.73m2subsets both before and after propensity score matching. Mean SCr results were calculated for each patient 24hrs pre-CT and 24hrs, 48hrs, and 72hrs post-CT. Absolute and relative changes in SCr were then calculated at 24, 48, and 72 hrs post-CT in comparison to the 24 hr pre-CT result.

Propensity Score Analysis

Propensity score generation, stratification by deciles, and matching for patients in the contrast and noncontrast groups were performed using the R package MatchIt as previously described (2, 9). Propensity score estimates, representing the probability of intravenous contrast administration for each patient in both the contrast and noncontrast groups, were generated for each eGFRsubsetusing a logistic regression model derived from the clinical variables noted with (*) in Table 1. The amount of IV fluids administered to patients in the 24 hours prior to their CT scan was included as a covariate in the propensity score model.

The amount of IV fluids administered on the day of and 24 hours following CT scan were not included as covariates since they took place after the decision to administer contrast material and could potentially confound the result. These two posthoc covariates were instead added as adjustment covariates to a conditional logistic regression model following matching with pre-scan IV fluids and other Table 1 covariates. Covariates that remained marginally imbalanced following propensity score matching (p 0.15) were also included in this conditional logistic regression model.

Stratification into deciles based on propensity score was performed using the Subclassification routine of MatchIt. Matching was performed on the logit of the propensity score using nearest neighbor matching (Greedy-type matching) and a caliper width of 0.15 of the standard deviation of the logit of the propensity score. One-to-one matching was performed for all eGFR subsets except for the eGFR <30 ml/min/1.73m2subset and eGFR45 inpatient subset, where 1:3 and 1:2 (contrast to noncontrast) matching was performed to account for the very small contrast group sample size.

Sensitivity Analyses

Three sensitivity analyses of additional patient subgroups were performed to help validate study findings. First, propensity score analysis of the complete ICU patient cohort, with patients not stratified into specific eGFR subsets, was performed. Second, a sensitivity analysis was performed where patients were stratified into 60, 30-59, and <30 ml/min/1.73m2subsets to mirror the KDOQI classification of chronic kidney disease (10). Third, eGFR 45 ml/min/1.73m2and eGFR > 45 ml/min/1.73m2patients with stable renal function prior to their CT scan and patients without PreCT-AKI were separately examined. Finally, since patients who had been admitted from the emergency department in the 24 hours prior to ICU admittance and their CT scan (“ED”) were likely more acutely ill than patients who were already hospitalized when transferred to the ICU (“inpatient”), ED and inpatient subgroups, further stratified into eGFR 45 ml/min/1.73m2 and eGFR > 45 ml/min/1.73m2subsets, were separately examined. Propensity score stratification and matching were independently performed on all sensitivity subgroups and differences in outcomes were measured using conditional logistic regression as described below.

Statistical Analysis

Statistical analyses were performed using R (version 3.0.3, R Foundation for Statistical Computing, Vienna, Austria) (11). Dichotomous variables were displayed as counts with percentages, categorical data were shown as relative frequencies (%), and continuous data were presented as medians with interquartile ranges (IQR). Differences in clinical characteristics and rates of AKI, emergent dialysis, and mortality between the contrast and noncontrast groups prior to matching were assessed using the Wilcoxon rank-sum test for continuous clinical characteristics and Fisher’s Exact test or Pearson’s chi-square test for categorical clinical characteristics and outcomes. Differences in clinical features and rates of AKI, emergent dialysis, and mortality following 1:1 matching (eGFR 45, 45, 60, and 30-59 ml/min/1.73m2subets) or 1:3 matching (eGFR <30 ml/min/1.73m2subsets) were measured using conditional logistic regression, conditioned on the unique ID assigned to each match. Changes in absolute standardized differences of each covariate before and after propensity score matching were calculated using MatchIt. Differences in outcomes following decile stratification were measured using Cochran-Mantel-Haenszel estimates of risk via conditional logistic regression, conditioned to the strata identity.

REFERENCES

1.McDonald JS, McDonald RJ, Lieske JC, et al. Risk of Acute Kidney Injury, Dialysis, and Mortality in Patients With Chronic Kidney Disease After Intravenous Contrast Material Exposure. Mayo Clin Proc. 2015;90(8):1046-53.

2.McDonald RJ, McDonald JS, Bida JP, et al. Intravenous contrast material-induced nephropathy: causal or coincident phenomenon? Radiology. 2013;267(1):106-18.

3.McDonald RJ, McDonald JS, Carter RE, et al. Intravenous contrast material exposure is not an independent risk factor for dialysis or mortality. Radiology. 2014;273(3):714-25.

4.Herasevich V, Pickering BW, Dong Y, Peters SG, Gajic O. Informatics infrastructure for syndrome surveillance, decision support, reporting, and modeling of critical illness. Mayo Clin Proc. 2010;85(3):247-54.

5.McDonald JS, McDonald RJ, Carter RE, Katzberg RW, Kallmes DF, Williamson EE. Risk of intravenous contrast material-mediated acute kidney injury: a propensity score-matched study stratified by baseline-estimated glomerular filtration rate. Radiology. 2014;271(1):65-73.

6.McDonald JS, McDonald RJ, Lieske JC, et al. Risk of Acute Kidney Injury, Dialysis, and Mortality in Patients With Chronic Kidney Disease After Intravenous Contrast Material Exposure. Mayo Clin Proc. 2015;90(8):1046-53.

7.Wu MJ, Shu KH, Liu PH, et al. High risk of renal failure in stage 3B chronic kidney disease is under-recognized in standard medical screening. J Chin Med Assoc. 2010;73(10):515-22.

8.Khwaja A. KDIGO clinical practice guidelines for acute kidney injury. Nephron Clin Pract. 2012;120(4):c179-84.

9.Ho DE, Imai K, King G, Stuart EA. MatchIt: Nonparametric Preprocessing for Parametric Causal Inference. 2011.

10.K/DOQI clinical practice guidelines for chronic kidney disease: evaluation, classification, and stratification. American journal of kidney diseases : the official journal of the National Kidney Foundation. 2002;39(2 Suppl 1):S1-266.

11.R Development Core Team. R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing, 2012.

12.Bryer J. PSAboot: Bootstrapping for Propensity Score Analysis2016. Available at: Accessed 8/11/2016.

1

Supplemental Table 1: Absolute standardized differences of model covariates before and after propensity score matching

* = covariate included in the propensity score model.

Supplemental Table 2: Changes in SCr following CT scan

Supplemental Table 3: eGFR 45 ml/min/1.73m2subset outcomes after propensity score analysis and adjustment for IV fluids given day of and day after scan.