Excessive diagnostic testing in acute kidney injury

Background The patterns, performance characteristics, and yield of diagnostic tests ordered for the evaluation of acute kidney injury (AKI) have not been rigorously evaluated. Methods We characterized the frequency of AKI diagnostic testing for urine, blood, radiology, and pathology tests in all adult inpatients who were admitted with or developed AKI (N = 4903 patients with 5731 AKI episodes) during a single calendar year. We assessed the frequency of abnormal test results overall and by AKI stage. We manually reviewed electronic medical records to evaluate the diagnostic yield of selected urine, blood, and radiology tests. Diagnostic yield of urine and blood tests was determined based on whether an abnormal test affected AKI diagnosis or management, whereas diagnostic yield of radiology tests was based on whether an abnormal test resulted in a procedural intervention. In sensitivity analyses we also evaluated appropriateness of testing using prespecified criteria. Results Frequency of testing increased with higher AKI stage for nearly all diagnostic tests, whereas frequency of detecting an abnormal result increased for some, but not all, tests. Frequency of detecting an abnormal result was highly variable across tests, ranging from 0 % for anti-glomerular basement membrane testing to 71 % for urine protein testing. Many of the tests evaluated had low diagnostic yield. In particular, selected urine and blood tests were unlikely to impact AKI diagnosis or management, whereas radiology tests had greater clinical utility. Conclusions In patients with AKI, many of the diagnostic tests performed, even when positive or abnormal, may have limited clinical utility.

The traditional approach to the differential diagnosis of AKI emphasizes three broad categories: pre-renal AKI, which results from inadequate perfusion of the kidneys; post-renal AKI, which results from obstruction to the flow of urine; and intrinsic causes of AKI, which can be due to injury or dysfunction of the glomeruli, other blood vessels, tubules, or the interstitium. A large number of diagnostic tests are available for the investigation of AKI. Determining which test to order in which patient with AKI should be based on estimates of pre-test probability from the clinical context and an appreciation of the diagnostic performance characteristics of the various tests available.
Previous studies have examined the utility of some AKI diagnostic tests in isolation, such renal ultrasonography [10] and urine microscopy [11], but none have comprehensively assessed the patterns and clinical utility of a wide range of diagnostic tests in patients with AKI. We performed this study to identify the frequency with which various urine, blood, radiology, and pathology tests are ordered in patients with AKI, and to assess their diagnostic yield.

Study design
We conducted a retrospective cohort study among inpatients at Brigham and Women's Hospital (BWH). All protocols were approved by the BWH Institutional Review Board.

Study cohort
BWH is a 777-bed tertiary care academic medical center that provides care to an ethnically and socioeconomically diverse population. We analyzed all hospitalizations of adult patients (age ≥ 18) who were admitted in 2010 and developed AKI. While participants could only be considered to have a single episode of AKI per admission, AKI during subsequent admissions was also included and considered a separate AKI event. Thus, data were analyzed at the AKI-event level. We excluded patients with endstage renal disease (ESRD) on maintenance hemo-or peritoneal dialysis and those who had undergone kidney transplantation.

Data collection
We obtained data through the Research Patient Data Registry, a central clinical data warehouse maintained by Partners Healthcare System, of which BWH is a member. We obtained information on patient demographics (age, gender, race, comorbidities) and on patterns of diagnostic testing for AKI, including urine, blood, and radiology testing, as well as renal biopsy.

Definition and staging of AKI
We defined and staged AKI according to the serum creatinine (SCr)-based criteria established by the Kidney Disease Improving Global Outcomes (KDIGO) Work Group [12]. Urine output data were not available. AKI was defined as an increase in SCr ≥0.3 mg/dl over any 48 h time period during hospitalization or ≥50 % increase in SCr over baseline known or presumed to have occurred within 7 days. For patients meeting criteria for AKI, we identified the AKI stage based on the ratio of maximum SCr during hospitalization divided by the baseline SCr, according to consensus criteria [12]. Baseline SCr was defined as the mean of all available outpatient SCr measurements obtained 7 to 365 days prior to admission, a validated approach described previously [13]. If no outpatient SCr values were available, the lowest SCr measurement during hospitalization was used as the baseline.

AKI etiology
We manually reviewed the electronic medical record (EMR) of 100 randomly-selected patients and recorded the AKI etiology where it was known or suspected. In cases where AKI etiology was suspected to be multi-factorial, we recorded the two etiologies judged to be of primary importance.
Urine and blood test results were determined to be normal or abnormal based on reference ranges provided by the BWH central laboratory. However, some tests such as urine sodium have normal physiologic variation and thus reference ranges are not applicable. These tests were therefore not categorized as normal or abnormal.

Radiographic analysis
We manually reviewed all radiology reports. We considered CTAP tests as an AKI diagnostic test only if the indication for the test specifically mentioned AKI, acute renal failure, elevated SCr, decreased urine output, or concern for hydronephrosis. We categorized renal ultrasound and CTAP tests as abnormal if any abnormality was documented except simple cysts and non-obstructing stones and masses noted on prior imaging and found to be stable in size.

Diagnostic yield of abnormal tests
We evaluated the diagnostic yield of one urine test (eosinophils), four blood tests (SPEP, free light chains, ANCA, and cryoglobulins), and two radiology tests (renal ultrasound and CTAP). To evaluate diagnostic yield of urine and blood tests, we manually reviewed the EMR of patients with abnormal test results to determine whether the test affected the patient's AKI diagnosis or management during that hospitalization or within 30 days of discharge. For example, for patients with an abnormal SPEP or serum free light chain test result we determined from progress notes and discharge summaries whether they were subsequently diagnosed with a plasma cell dyscrasia likely to be contributing to their AKI (e.g., cast nephropathy) or whether the test affected their subsequent management (e.g., intravenous fluids, chemotherapy, or additional testing such as renal biopsy or bone marrow biopsy). Additional examples are provided in Table 1. Since radiology tests often reveal abnormal findings that appear chronic or otherwise have limited immediate clinical relevance, we assessed the diagnostic yield of radiology tests based on whether an abnormal test resulted in a subsequent procedural intervention. For renal ultrasonography, abnormalities other than hydronephrosis (e.g., atrophy, cortical thinning, and other chronic changes) were not included in this analysis, since interventions are not typically performed for these findings.

Sensitivity analysis #1
Some diagnostic tests commonly ordered for the workup of glomerulonephritis (ANCA, anti-GBM, and complement levels) and paraproteinemias (SPEP, UPEP, and serum free light chains) may also be ordered for purposes unrelated to AKI. We therefore manually reviewed the EMR of 50 randomly-selected patients who were tested for one of the above glomerulonephritis labs and 50 randomly-selected patients who were tested for one of the above paraproteinemia labs, and we determined the approximate frequency with which these tests were obtained for purposes related or unrelated to AKI. We classified each test into one of three categories: 1) obtained for AKI evaluation; 2) not obtained for AKI evaluation; 3) unable to determine whether ordered for AKI evaluation.

Sensitivity analysis #2
To evaluate the appropriateness of testing for glomerulonephritis and paraproteinemias in AKI, we manually reviewed the EMR of the same 100 patients above. We developed criteria to evaluate appropriateness of testing ( Table 2) and applied those criteria to each patient. Further, we evaluated whether appropriateness of testing differed based on patient location (floor vs. ICU), ordering team (primary team vs. renal consult), and AKI stage.

Sensitivity analysis #3
Evaluation of diagnostic yield (described above) focused on the utility of abnormal tests only. We therefore manually reviewed the EMR of the same 100 patients above to determine the diagnostic yield of normal test results. Specifically, we determined the number of normal or negative tests that affected the patient's AKI diagnosis or management during that hospitalization or within 30 days of discharge. Examples of how a normal test result could have influenced a patient's management include: continuing with supportive care (e.g. intravenous fluids for pre-renal azotemia); performing or deferring a renal biopsy; and starting or discontinuing immunosuppressive medications or plasmapheresis.

Statistical methods
Most of the data shown are descriptive. Continuous data are shown as median (interquartile range [IQR]). Fisher's exact test was used to compare the frequency of appropriate testing based on patient location, ordering team, and AKI stage. All tests were two-tailed, and P < 0.05 was considered significant.
The most commonly ordered tests in AKI were urinalysis and automated urine sediment examination ( Fig. 1). Blood tests for vasculitis, glomerulonephritis, and paraproteinemias were ordered in fewer than 10 % overall, and more frequently for higher stages of AKI. Radiology testing with renal ultrasonography or CTAP was obtained in 10 % and 1 %, respectively. Kidney biopsies were performed in 28 patients (0.5 % of all AKI episodes).

Diagnostic yield of selected tests
Not all diagnostic tests during the course of AKI are obtained specifically for clarification of the cause or treatment of AKI (e.g., automated urinalysis, urine sodium, urine total protein). We therefore focused on those tests  Albumin, g/dL 3.5 (3.0-3.9) 3.6 (3.1-4.0) 3.4 (2.9-3.9) 3.2 (2.8-3.7) Stages 1, 2, and 3 refer to AKI severity according to the criteria established by the Kidney Disease Improving Global Outcomes (KDIGO) Work Group [12] as follows: stage 1, increase in serum creatinine 1.5 to 1.9 times baseline or absolute increase ≥0.3 mg/dl; stage 2, increase in serum creatinine 2.0 to 2.9 times baseline; stage 3, increase in serum creatinine ≥3.0 times baseline, or increase ≥0.5 mg/dl to an absolute level ≥4.0 mg/dl, or initiation of renal replacement therapy used most commonly to diagnose specific causes of AKI such as interstitial nephritis, paraprotein-related kidney disease, glomerulonephritides, and obstructive uropathy. Data on the number of tests performed, the number that were abnormal, and the number that affected AKI diagnosis or management are shown in Fig. 2. Additional data on frequency of abnormal results by AKI stage are shown in Table 4.

Urine eosinophils
Urine eosinophil testing was obtained in 7 % of AKI episodes (3 %, 9 %, and 24 % in patients with AKI stages 1, 2, and 3, respectively). Abnormal results (urine eosinophils ≥ 1 %) were found in 7 % of tests. Among the abnormal results, 24 % were in stage 1, 24 % were in stage 2, and 52 % were in stage 3. The median urine eosinophil % among those with abnormal results was 2 (IQR, 1-3). Among those with abnormal results, the presumptive diagnosis was AIN in 28 %. Kidney biopsy was performed in two patients with eosinophiluria and showed AIN in one, and vasculitis in the other.

SPEP
SPEP was ordered in 6 % of all AKI episodes and was abnormal in 19 %. Among AKI episodes with an abnormal SPEP, 75 % were IgG, 12 % were IgA, 8 % were IgM, and in 5 % only faint free light chains were identified; overall, 63 % consisted of monoclonal kappa and 37 % consisted of monoclonal lambda; the median M-spike concentration was 0.9 (IQR, 0.5-2.4 g/dl). In only 9 cases (2.6 % of the tests) did a positive SPEP affect AKI diagnosis or management (Fig. 2).

ANCA and anti-GBM
ANCA testing was ordered in 3 % of AKI episodes. Abnormal results were found in 11 % of tests, with median titers of 1:20 (IQR, 1:20 to 1:180). Among the abnormal results, 80 % were positive for p-ANCA and 20 % for c-ANCA. In only 5 cases (3 % of the tests) did a positive ANCA affect diagnosis or management (Fig. 2). In 3 of those cases, pauci-immune crescentic glomerulonephritis was confirmed by renal biopsy; in the remaining 2 cases, a presumptive diagnosis of ANCA vasculitis was made clinically. Anti-GBM antibodies were tested in 1 % of AKI episodes and were negative in all.

Radiology testing
Renal ultrasound was ordered in 567 (10 %) of AKI episodes. Abnormal results were detected in 196 (35 %) of the tests. Severe abnormalities such as severe or bilateral hydronephrosis were less frequently found (2 % and 10 %, respectively; Table 5). Hydronephrosis was detected in 109 (19 %) of the tests, among which 35 (32 %) resulted in a procedural intervention (Fig. 2). Among the 35 cases of that resulted in an intervention, a clinical history of metastatic or genitourinary cancer (N = 26), hemorrhagic cystitis (N = 2), and other comorbidities raising suspicion for obstruction were present in each case. CTAP testing was ordered in 0.9 % of AKI episodes. Abnormal results were detected in 18 out of 54 (33 %) tests. Among the abnormal tests, 61 % resulted in an intervention (Fig. 2). CTAP, but not renal ultrasonography, was more likely to be abnormal among patients with higher AKI stages (Table 4). In general, the frequency of detecting abnormalities was greater with radiology tests than with urine or blood tests.

Sensitivity analyses
We conducted the following sensitivity analyses among 50 randomly selected patients who were tested for one of the "glomerulonephritis" labs (ANCA, anti-GBM, or complements), and an additional 50 randomly selected patients who were tested for one of the "paraproteinemia" labs (SPEP, UPEP, and serum free light chains): Sensitivity analysis #1frequency of testing related or unrelated to AKI Among the patients who were tested for one of the glomerulonephritis labs, 46 % were ordered specifically for AKI, 12 % were ordered for reasons unrelated to AKI, and 42 % were ordered for reasons which were not clearly documented. Among the 50 patients who were tested for one of the paraproteinemia labs, 44 % were ordered specifically for AKI, 6 % were ordered for reasons unrelated to AKI, and 50 % were ordered for reasons which were not clearly documented.

Sensitivity analysis #2appropriateness of testing
Using prespecified criteria (Table 2), we determined appropriateness of glomerulonephritis and paraproteinemia testing in AKI. Overall, glomerulonephritis and paraproteinemia testing was appropriate in 50 % and 54 % of tests, respectively (Fig. 3). When testing was ordered by the renal consult team compared to the primary team, appropriateness of testing was higher for glomerulonephritis testing (P=0.047), and showed a similar trend that did not reach statistical significance for paraproteinemia testing (P=0.09, Fig. 3). Similarly, when testing was ordered in patients with higher AKI stages, we found that appropriateness of testing was higher for paraproteinemia testing (P=0.04) but not for glomerulonephritis testing (P=0.13, Fig. 3).

Sensitivity analysis #3diagnostic yield of normal tests
Among the patients who underwent one of the glomerulonephritis tests, testing was normal in 45 out of 50 (90 %) tests, and affected AKI diagnosis or management in only 1 out of the 45 (2.2 %) normal tests (and in 0 out of the 5 abnormal tests). Specifically, in one case a negative ANCA test resulted in discontinuation of immunosuppressive medications. Among the patients who underwent one of the paraproteinemia tests, testing was normal in 39 out of 50 (78 %) tests, and affected AKI diagnosis or management in 0 out of the 39 normal tests (and in 4 out of the 11 abnormal tests).

Discussion
In this retrospective study we characterized the patterns and clinical utility of a wide variety of AKI diagnostic tests, including urine, blood, radiology, and pathology tests. Our key findings are: 1) frequency of testing increases with higher AKI stage for nearly all diagnostic tests; 2) frequency of abnormal results increases with higher AKI stage for some, but not all, diagnostic tests; and 3) many of the tests evaluated have low clinical utility. In particular, selected urine and blood tests are unlikely to impact AKI diagnosis or management. Cumulatively, these data suggest a paucity of actionable diagnostic tests in AKI.
Previous studies have examined the utility of a limited number of AKI diagnostic tests in isolation, including urine microscopy [11], urine eosinophils [14], the fractional excretion of sodium [15] and urea [16], and renal ultrasonography [10]. Most of these studies have focused on a single diagnostic test, have been limited by modest sample sizes, and have focused on test performance without inclusion of data on frequency of testing. None, to our knowledge, have simultaneously evaluated the patterns and test performance of a wide variety of AKI diagnostic tests in a large AKI cohort.
To assess diagnostic yield, we focused mostly on abnormal test results. Although normal test results can also be clinically useful by ruling out diagnoses, our sensitivity analysis of normal test results for glomerulonephritis and paraproteinemia labs revealed very low diagnostic yield. We therefore focused on abnormal test results, and evaluated diagnostic yield based on whether the result influenced AKI diagnosis or management. We found selected tests (e.g. urine eosinophils) are unlikely to be clinically useful in AKI. In contrast, renal ultrasonography and CTAP were much more likely to be useful, especially when the clinical history was suggestive. This was best exemplified by renal ultrasonography: among the cases of hydronephrosis that were detected and intervened upon, 100 % had a clinical history raising suspicion for obstruction. Thus, diagnostic testing should be based on both pre-test probability from the clinical context as well as an appreciation of the diagnostic value and actionability of the data provided by the test.
Among the urine, blood, and radiology tests examined in Fig. 2, the ratio of the number of tests ordered to the number of tests that had clinical utilitya concept similar to number needed to screen (NNS)ranged substantially: from 5 in the case of CTAP to 60 in the case of urine eosinophils. In general the ratio was much higher for urine and blood tests, suggesting lower diagnostic yield, compared to radiology tests. In the case of renal ultrasonography, for example, the 567 tests ordered led to 35 interventions, resulting in a NNS of 16. This is similar to the NNS observed among high risk patients in a prior study that evaluated the utility of renal ultrasonography in AKI [10]. For urine and blood tests, on the other hand, we would speculate that clinicians may be aware of their poor predictive value, and therefore base most of their diagnostic and management decisions on other data such as clinical history and physical examination. For example, a recent study evaluated the utility of urine eosinophils in the diagnosis of AIN, using renal biopsies as the gold standard, and found that the presence of urine eosinophils performed very poorly in distinguishing AIN from other etiologies of AKI [14]. Abnormal radiology tests, on the other hand, may provide more actionable data, such as amelioration of obstruction.
We acknowledge several limitations. Some test results may have been considered abnormal based on reference ranges but may have been clinically insignificant. Other tests may have been ordered for reasons unrelated to AKI, such as routine urinalysis on admission, which may have led to overestimation of the frequency of AKI diagnostic testing. We attempted to minimize this bias by focusing on tests which are most commonly ordered for the diagnostic evaluation of AKI and not for other purposes. In the case of CTAP, testing was only included if ordered specifically for AKI. We also performed sensitivity analyses on tests for glomerulonephritis and paraproteinemias to determine the fraction of tests that were ordered for reasons unrelated to AKI. We found that only a minority of tests were ordered for reasons unrelated to AKI, though a large portion were also ordered for reasons which were not clearly documented.   Due to institutional restrictions we were unable to show data on the cost of diagnostic testing. An additional limitation is that because this was a single-center study the findings may not be generalizable to other settings; however, this limitation may be at least partially offset by the large sample size and heterogeneous nature of our patient population. Finally, we were unable to evaluate the diagnostic yield of AKI testing in selected subgroups, such as patient location (general medical floor versus ICU), ordering team (primary team versus renal consult), or AKI severity. However, we conducted sensitivity analyses to evaluate whether appropriateness of diagnostic testing varied based on these factors. Although we found that appropriateness of testing may increase for some tests when ordered by the renal consult rather than the primary team, patients who require renal consultation may have greater severity of AKI; thus, we cannot exclude the possibility of confounding, and therefore this finding should be interpreted with caution.

Conclusions
In summary, a large number of diagnostic tests are available for the evaluation of AKI, and determining which test to order should be based on estimates of pre-test probability and performance characteristics of the test. Our data suggest that many of the currently available tests have limited clinical utility, even when they are abnormal or "positive". Thus, developing better diagnostic tests that provide reliable and actionable data on AKI diagnosis or management should be a priority in AKI research. The dearth of such tests may be a key reason why therapeutic advances to improve outcomes in AKI have been largely unsuccessful.