False-positive AKI and the perfectly imperfect biomarker

1344 6

E:Renal Fellow Networkfalse AKI.png
A small absolute change in serum creatinine level, 0.3 mg/dl, is used by Acute Kidney Injury Network (AKIN) and Kidney Disease Improving Global Outcomes (KDIGO) guidelines to define the presence of Acute Kidney Injury (AKI). The base of this definition was formed by several studies findings of strong association between adverse outcomes and minor changes in serum creatinine level. Subsequently, evidence emerged suggesting that this may not be true to the same extent in people with pre-existing CKD, because variations in serum creatinine concentration are common in these individuals.

As with all other laboratory tests, serum creatinine measurements are affected by within- and between-sample coefficients of variation, intra-individual variation and biologic variation. Biological variation may result from variations in diet, muscle mass and breakdown, tubular secretion, variability in volume homeostasis and from medications uses. The variation in measured serum creatinine level could be as high as 9%. Because only a small increase in serum creatinine is needed to meet AKI criteria, random variation in creatinine level may be a significant contributor to AKI diagnosis in the absence of a true reduction in GFR. This is called a false-positive AKI. It has been shown that high variation in serum creatinine in the period, of days, preceding the development of AKI was not associated with the anticipated inpatient mortality or dialysis. This observation supports the existence of false-positive AKI.

Lin et al demonstrated, using the KDIGO definition, an 8% overall false-positive rate for AKI diagnosis. This rate was much higher, 31%, for the subgroup of CKD patients with serum creatinine ≥1.5 mg/dl. Therefore, an absolute change in serum creatinine of 0.3 mg/dl may represent a relative inconsequential change in GFR in CKD patients rather than a superimposed acute injury.

In my opinion, false-positive AKI could largely explain why most randomized trials for early intervention in AKI have been unsuccessful in improving outcomes. AKI is misclassified under frameworks that do not reflect true GFR reduction. Consequently, patients with false-positive AKI are included in AKI studies and dilute observed effect sizes. This potentially leads to false-conclusions that certain interventions are ineffective and do not improve outcomes. The underlying severe disease is quite likely the actual mediator of adverse outcomes seen in AKI. Therefore, small changes in serum creatinine may be nothing more than a reflection of the severity of the underlying disease process. This point remains a topic of hot debate. Moreover, AKI definition using small increments in serum creatinine level has not been validated among patients with CKD.
It is obvious now that serum creatinine is an imperfect AKI biomarker; especially that it is being used on the basis of a relative change in value of a continuous variable instead of the crossing of a particular threshold. The ideal biomarker would accurately detect true reduction in GFR, be detectable early in the course of renal dysfunction to allow for timely intervention, and predict outcomes. It is likely that current AKI criteria will eventually be modified at least in part by sensitive and specific biomarkers of kidney injury. The use of such biomarkers will help in the development of a new paradigm for classifying AKI that is not only dependent upon serum creatinine. Meanwhile, the awareness about false-positive AKI should be highlighted and the limitations of serum creatinine, as an AKI biomarker, should be re-emphasized.

Authored by Mohammed A. Kaballo, Nephrology Fellow, Ireland


  1. Thank you for your kind words Mohammed. I certainly agree tat there won't be one tool/biomarker that can be used solely to predict, risk-stratify & diagnose AKI. I think ADQI were thinking that way when they suggested a matrix in which a damage biomarker or creatinine or urine output could be used to diagnose.

    I would very much like to see a prospective study to test this method – particularly the prediction of a fall in creatinine, but not only. Unfortunately, I've not heard of anyone doing this (and I'm not in a position to try anything locally). It is not difficult, just needs a protocol and willing players (maybe multiple sites doing a few each at minimum expense). I'm available to help with protocol of course. There are some limitations as you suggest. I wouldn't recommend it post cardiac arrest, where I believe that creatinine production has been suspended. The estimating equation for creatinine production is a limitation of course – I tried several equations with no noticeable difference in performance. It'd be good if something like the creatine to creatinine ratio (or some other combination) was always correlated with production, but I know of nothing and I think it would be another research project entirely to try and find if there was such a correlation.

    Finally, if someone were to set up a study, I'd like to see a direct comparison with nephrocheck to predict a doubling of creatinine in 12h.

  2. Hi Swapnil and John. Thank you very much for enriching this discussion.
    John, I have read, with great interest, a lot of your publications about creatinine kinetics in AKI. This is a great opportunity for me to discuss this topic with an expert.

    I think using the creatinine excretion rate to production rate ratio to diagnose AKI, and detect recovery, is really helpful in the clinical setting. It is a cheap test, albeit powerful. Its power resides in its ability to determine GFR loss and recovery without the need for prior knowledge of the patient’s baseline creatinine level. As you know, because of unknown baseline creatinine, most nephrologists frequently face the diagnostic dilemma of whether the elevated creatinine is due to an undiagnosed CKD or due to an AKI. Waiting for renal imaging and parathyroid hormone level to provide clues usually result in significant delays in diagnosing AKI in this group of patients.

    Another strength point of the creatinine excretion rate to production rate ratio is its capability of reflecting immediately whether GFR has recovered or not. This will definitely help to overcome the interpretation problem arising from the lag of creatinine changes behind changes in GFR.

    Unfortunately, like any other test in medicine, it has limitations. Its main limitation is its dependence on estimated creatinine production rates. As you know these rates vary considerably between individuals and during acute/chronic illnesses. Another limitation is urine collection. In day-to-day practice, it is easy to obtain an accurate urine collection over several hours for creatinine clearance in the critical care setting because most patients are catheterized. But in non-catheterized patients outside the critical care unit, the urine collection process is often challenging, and inaccurate, and hence the utility of the creatinine excretion rate to production rate ratio might be limited.

    I am not sure if many people agree with me, but I think it is unrealistic to expect that we can develop a new PERFECT AKI biomarker that can SOLELY predict, risk-stratify & diagnose AKI. Most likely we are going to end up, hopefully in the near future, with a new paradigm comprising several biomarkers, of both kidney function & kidney damage. Certainly we need to develop new methods to determine creatinine production, but John’s method for predicting the direction of creatinine changes using creatinine excretion rate to production rate ratio, is novel and promising. I would like to see it tested in a large-scale study, as I believe it is going to be a major contributor to a big leap forward in the development of a successful AKI diagnostic and prognostic model.

    John, are you aware of any studies that are testing/planning to test your method on a large number of patients in various clinical settings?

    Thank you

  3. Dear Mohammed
    Thanks for the response. I agree entirely about the problems of dichotomizing continuous variables. That is why when we ran the EARLYARF trial I used as an outcome measure the "relative average value of creatinine" which was a continuous measure taking into account the change of creatinine over time from baseline (see Evaluation of trial outcomes in acute kidney injury by creatinine modeling. Clinical Journal of the American Society of Nephrology,2009 4(11), 1705–1715. http://doi.org/10.2215/CJN.00820209).

    The analogy to troponins is interesting, but it has some problems. Troponins measure an injury to the myocardium and AMI is defined on the basis of a rise or fall with at least one above the 99th percentile of a normal population. We have several kidney injury markers which do the same. The difference is that we haven't quantified well what their levels are in a "normal" population, nor have we quantified well their relationship with hard outcomes (mortality in particular), and we have no interventions that directly address the pathological processes associated with the kidney damage markers (and make a difference to outcomes). I increasingly think this should be the direction of research in the area and we should forget about trying to find markers that predict a change in creatinine. As you so rightly point out, it is imperfect.

    A couple of other thoughts. Predicting the use of the RRT is interesting because it depends so much on local practice. I think it a valid area to pursue if there are reasonable treatments to try which could prevent the need for RRT. "Progression" can mean two things, a further increase in creatinine or further decrease in GFR. I think it should only be the latter, but a lot of literature uses a further increase in creatinine to mean progression. However, I think often all this is is that the kidney some time earlier lost sufficient filtration capacity that the creatinine has yet to reach a new steady state when it is measured at a later time point and so it continues to rise reflecting the extent of earlier GFR loss, not true progression.

    Thank you for your comments about creatinine clearance. I think it has some value. More recently I published a little paper using the ratio of creatinine excretion rate to estimated creatinine generation rate which can be measured at the same time as creatinine clearance. I think the worth of these two in combination can tell the attending physician if the elevated creatinine they have observed is likely to fall or not – ie if the GFR has already resolved (but not yet seen in creatinine) or not. This may prevent unnecessary fluids etc. I'll be interested in what you think? (http://www.hindawi.com/journals/bmri/2014/542069/).

    I am enjoying this discussion – thanks.

  4. Hi JP. Thanks for your valuable comment. You are absolutely right that by "early intervention" I meant as soon as an increase in creatinine is observed.

    Regarding your question about “crossing a particular threshold”:
    Yes, creatinine is a continuous variable, and as you know, generally it is a bad idea to dichotomize a continuous variable. Unfortunately, we are dichotomizing creatinine to define a binary outcome i.e. AKI vs No AKI, and we are using thresholds that are completely arbitrary. When small changes in creatinine are used to define AKI, its sensitivity will go higher at the expense of specificity and vice-versa when larger changes are used instead.
    The biggest problem we are facing currently with all the novel AKI biomarkers is that serum creatinine is being used as the gold standard against which these biomarkers are tested. The validation process is largely hindered by the use of an imperfect gold standard, as this can result in misinterpretations of the diagnostic performance of the investigated biomarker. So, what we actually need is a Renal “Troponin”; something similar to the cardiac troponin, a highly-sensitive marker with a level cut-off point beyond which the progression of kidney injury and the need for dialysis could be confidently predicted. Of course to find such a marker, first we need to find a perfect gold standard for the validation process.

    Recently, I have read an article about the utility of repeated 4-hour creatinine clearance measurements in the detection of AKI. It was interesting to see how useful this test was in differentiating resolving from ongoing AKI, in addition to its ability to detect AKI earlier when compared to serum creatinine level alone. I believe that this study is underrated and such short duration creatinine clearance tests should be studied further, in-depth, using larger cohorts and different clinical settings. This will help proof and validate its utility, perhaps as an important component of an AKI risk prediction model.

    Finally, I would like to thank you again for generating this interesting discussion.

  5. Hello Mohammed. Some Nephrologists on twitter pointed out your post to me. You've made some interesting points. I agree with all you say in your first three paragraphs. There is certainly an issue with false positives with AKI. Lin's analysis is certainly worth reading. You posit that it is the false positive rate that is the reason most AKI RCTs have not improved outcomes. I agree that could be a factor, but I think there are several, perhaps more serious issues with these trials. First, by "early intervention" do you mean simply as soon as an increase in creatinine is observed (eg >0.3 mg/dL), or do you mean before an increase is observed? There has only been one published trial (the EARLYARF study of EPO in the ICU) which had a measure of kidney damage and intervention before a change in creatinine (there are some trial aimed at prevention which intervene prophylactically on all at risk of AKI). So, I will assume that you mean simply when creatinine is first observed to be elevated. In those cases, according to creatinine kinetics, patients will have had an earlier loss of GFR (usually a day or two earlier) – hence they are a long time after the insult to the kidney. I think this delay plays a major roll in why studies have failed. Also, an elevated creatinine does not mean that the patient still has a reduced GFR. It may be that the creatinine is yet to return to baseline. Again, a confounding factor for trials. Finally, I think many compounds are aimed at preventing further injury rather than reversing GFR – so measuring creatinine may not be the right surrogate outcome.

    In your final paragraph I'm not sure what you mean by "crossing a particular threshold". Are you saying that we should take care that the thresholds used for AKI surrogates are arbitrary because creatinine (and urine output) are continuous variables? I'd agree with that. You mention that an accurate biomarker would accurately detect true reduction in GFR. But this is merely a measurement of change in GFR – this requires having a "GFRometer" which we don't have (yet). I've published on short duration creatinine clearance as an alternative. Most of the "novel" biomarkers being investigated respond to injury rather than change in kidney function per se (the exception being plasma cystatin C). Thus, when they are compared to a later rise in creatinine (which they inevitably are), not only are there the problems with creatinine being an imperfect biomarker as you say, but also we are comparing apples with oranges. This is why studies and meta-analyses of injury biomarkers with hard outcomes are important – ie so they can stand on their own as markers of kidney damage.

    Thanks for the post!

Leave a Reply