Tag: MDRD

Posted on

1A? No way. KDOQI has lost its way

It started with a tweet. It always does.

Daniel Coyne is one of my heroes in nephrology. Starting in fellowship and for the first decade of being a nephrologist, I was suckered into the normalization of hemoglobin delusion. The idea that we could heal our patients’ hearts by treating the ubiquitous anemia of CKD was seductive. The theory made sense. The retrospective data looked amazing. Many really smart nephrologists promoted the idea. Living through the medical reversal of CHOIR, CREATE, and finally TREAT was both academically traumatic and a profoundly formative moment in my career. One of the major components of the CKD-anemia delusion was the role that KDOQI played in promoting the high hemoglobin targets before the clinical data showed the folly of this strategy. Dr. Coyne is one of the first nephrologists to ring the warning bells. We highlighted his story in NephMadness 2016. So when Dr. Coyne criticized the nutrition guideline my ears were perked.

The thread highlighted above is convincing and I amplified it.

Subsequently,` I received a polite email from one of the guideline authors suggesting I may have gone too far in criticizing the guideline. Time to put up or shut up. Here are my thoughts, at longer than tweet-length, regarding the protein restriction guideline. Much of this comes from Dr Coyne’s thread, so read that first, so you know where the smart stuff comes from.

Guidelines should only express certainty when there is certainty. When guidelines flip-flop with each new edition doctors lose credibility while taking care of patients. Consistency is important. Because of this I want guidelines to be late adopters rather than on the cutting edge, and for guidelines to be humble about the frailties of human knowledge. 

This puts me in opposition to many guidelines but this 1A guideline on low protein diet seems particularly egregious. The guideline I’m focused on is:

The Detailed Justification of this guideline provides this guidance

Research reports a beneficial effect of protein restriction (0.55-0.6 g/kg per day) on ESKD/death in adults with CKD. In adults with CKD, 5 RCTs reported findings on the effect of protein restriction on survival/deaths. Three studies clearly indicated a beneficial effect of moderate restriction in dietary protein on the development of ESKD/death.153,164,168


So there are 5 RCT that looked at protein restriction for the development of ESRD/Death. Of these 5 studies (of which I could only find 4), 3 were positive, but one of them, Locatelli, is only positive if you consider a P value of 0.06 as significant.

It is so strange to see this reported as a positive trial, because when I would answer questions about low protein diets for my patients with relatively early kidney disease, I would say, this has been looked at in a number of studies but there were two big studies, one in Italy (this one) and one in the US (MDRD) and neither were able to show improvement in outcomes. But here, I guess, this is good enough to qualify for “clearly indicated a beneficial effect.”

The second reference is a small 82 person study by Henrik Hansen. I find this troubling because even though the KDOQI guideline specifically excludes people with diabetes, Hansen restricted enrollment to patients with diabetic nephropathy.

The last of the triplets is Rosman. I could not find Reference 168. This sounds like a study that was presented at a conference, Proceedings of the European Dialysis and Transplant Associations, European Renal Association 1985. Here is the pubmed listing. The abstract is pretty thin. I am particularly concerned because a later publication that seems to be long-term follow-up this cohort (no I don’t know how they picked up an additional 50 patients) and it spins a different tale. The title says it all:

I also highlighted a concerning line from the abstract, “The diet appeared to have a selective effect on the progression rate of renal failure: only patients with primary glomerular disease responded to the diet.” If this is the case this is something that should have given pause to the authors trying to generalize this recommendation for all people without diabetes and CKD 3-5 not on dialysis.

The three “positive” trials were published in 1991, 2002, and 1989. This means the cohorts were pre-ACEi/ARB. In fact, Locatelli’s study actively discouraged the use of ACEi.

Lastly, of the two negative trials (of the five that looked at ESRD/death) the authors brush off Cianciaruso for being, “a relatively small sample size.” Let’s take a look at the trials.

Cianciaruso had an N of 423, twice the size of Rosman’s study and five times the size of Hansen’s. The only study in the group of 5 larger was Locatelli with 456. Relatively small sample size? Come on.
We are currently in the midst of a revolution in evidence based nephrology with the data emerging about the effects of SGLT2i. One of the compelling findings with each subsequent blockbuster study is how consistent the data is regardless of how the outcome is assessed:

  • Change in GFR? Yup, the SGLT2i is better
  • Change in GFR slope? Check, SGLT2i again
  • Fraction of patients that reach death or dialysis? SGLT2i still got you covered.

This is not the case for this guideline. The authors trumpet guideline 3.0.1 claiming a reduction in need for dialysis, but this benefit appears to be fragile because if instead we look at low protein diets (LPD) through the lens of the change in GFR, the data no longer points to a benefit. From the guideline:

This mismatch between the 1A conclusions about development of ESRD or death and the lack of biologic plausibility that one can prevent ESRD without preventing loss of GFR should be addressed in the guideline mentioned and probably should have forced the guideline committee to rethink the 1A grade they awarded to protein restriction. 

Coyne’s first tweet was in response to this

Dr Fouque expresses outrage that someone would question the safety of low protein diets. Well safety is a concern. As I wrote on this blog (see this and the follow up here). These posts are regarding long term follow-up of the MDRD cohort. MDRD is famous for not showing any difference in ESRD or death with either a low or very low protein diet, the long term follow up revealed that among people that did progress to dialysis, the patients randomized to the very low protein had nearly a two times higher risk of death (HR 1.92). So Dr. Fouque, with all due respect, I do think it is quite reasonable to question the risks of a low protein diet.

Dr. Fouque comment also highlights what is so dangerous about the 1A grade for protein restriction, it says this is settled science, we no longer need to investigate this. A 1A guideline insulates protein restriction from questions that need to be asked. This is not settled science. Not even close.

When the KDIGO Diabetic Kidney Disease Guidelines were published a lot of people went nuts over the fact that ACEi did not warrant a 1A recommendation.

If you find this curious, I advise you to listen to Freely Filtered #27 where Ian de Boer describes the logic for that 1B grade. Whether or not you find that admirable or absurd it is clear that the authors of the diabetic kidney disease guideline respect a score of 1A. They hold it in reverence and reserve it for only the most rigorous evidence. This argument came from the mind of Nayan Arora:

I can’t find the same respect for grading the evidence in the KDOQI nutrition guidelines.

Posted on

What is missing from the race and eGFR discussion

The ASN and NKF have a joint task force working toward a response to the race and eGFR problem and they are now inviting people to submit oral and written testimony. I signed up. Hopefully I get an opportunity, but suspect there will be too many people for them to hear even a fraction of the applicants.

I have been thinking about race and eGFR and this is where I am at…

  • Race is a social not a biological construct
  • People identified as black (or self-identified as black) have higher measured GFR for the same serum creatinine as non-blacks
  • These higher eGFR results in black people (self-identified or not), a marginalized group and a population already at increased risk of adverse kidney outcomes, being denied transplant listing and CKD referral.

But what is not ever seem to be questioned in this discussion is the perverse use of estimated GFRs to make critical binary decisions in individual patients. The eGFR equations are amazing how well they predict GFR for groups of patients with minimal bias, but their reliability in an individual is stunningly imprecise. Accuracy in individual patients is measured by P30, the likelihood that the true value will be within 30% of the measured value. The P30 is 84% with CKD-Epi, a bit better compared to the 80% in MDRD. This means that for the critical decision of whether to list a patient for a kidney transplant, a patient with an eGFR of 21 will have an actual measured GFR somewhere between 15 and 27 in 84% of cases. This 30% spread is greater than the 16% adjustment for black race.

Though using race for eGFR should be stopped, and we can do that today by making cystatin-c the coin of the realm, this doesn’t change the problem of over indexing on eGFR for individual patient decisions. Cystatin C is no better than creatinine in providing a precise estimate in eGFR (P30 86%). Decisions like transplant listing and CKD referral should not rely on a measurement with so much uncertainty. We report eGFR on lab reports but give physicians no sense of the imprecision hidden in that number.

I think if eGFR were reported as a range (±30%), we would stop using sharp cut-off limits for critical decisions like transplant and referral.

The use of sharp cutoff for decisions like transplant and CKD referral harms all patients with CKD, not just black people. We should immediately to remove race from eGFR calculations by standardizing cystatin-C as the way to assess eGFR but at the same time we should start the process of unwinding guidelines and individual patient decisions from being wedded to inaccurate estimates of GFR.

Posted on

A couple of new Tweetorials

The first was in response to Robert Centor’s excellent description of how he uses reciprocal creatinine. Honestly I had not thought about reciprocal creatinine in a long time. It was fun diving into some of the literature around it. Here is the tweetorial:

Today I did a second tweetorial on hyperosmotic hyponatremia

Here I had some technical problems. I wrote the entire tweetorial using chained tweets in Safari on MacOS. When I went to upload all tweets, Twitter hung and failed to upload more than the first 8. I had to then go through and re-post the remainder of the tweetorial. I was frustrated and failed to attach two of the animated gifs I made. I added them as additional tweets but they break the flow. Tweetorials are like writing email newsletters, once you publish the tweets (or hit send on the newsletter) there is no opportunity for editing.

Posted on

Is estimated GFR racist?

Update from January 2021: This is an old post and I have evolved my thoughts on this issue. I leave this here mainly as bread crumb on the trail of my evolving thoughts about this topic.

 

 

Zachery Berger published this epic tweet storm last week about estimated GFR. It starts here:

The conclusion is that using race in the MDRD formula (and by extension the  CKD-epi formula) is inherently racist.

I do not think this is the case. Trying to estimate GFR from a serum creatinine and a few demographic variables is impossible, the best we can hope for is a reasonable guess. To see how bad we are take a look at the wide variability at high GFRs with the current CKD-Epi formula:

So that GFR of 60 has a 95% CI of being between 35 ml/min and 92 ml/min. Not so reassuring.

One of the primary reasons for this imprecision is that creatinine production varies from body to body. When one person produces more creatinine than another, for a set rate of creatinine excretion his serum creatinine concentration (what we measure on a blood test) will be higher. Who produces more creatinine? People with more muscle mass.

  • Larger people produce more creatinine than smaller people
  • More muscular people produce more creatinine than less muscular people
  • People with four limbs produce more creatinine than people with 3 limbs
  • Men produce more creatinine than women, on average
  • Young people produce more creatinine than older people, on average
  • Vegetarian Indians produce less creatinine than westerners
  • Black people produce more creatinine than non-black people, on average

The data is shown in figure 1 of Levey’s 1999 study.

Even though Dr. Berger did not draw the conclusion that estimated GFR is inherently sexist, let’s look at gender first. I have recolored the two graphs and superimposed them on one another. Men are in red and women are in blue:

It is clear that for any given GFR the men tend to have a higher creatinine than the women. This is not perfect and it is not hard to pick out individuals where this generalization fails, but in general this is a fair generalization. Levey comments and quantifies this gender difference:

At any given GFR, the serum creatinine concentration is significantly higher in men than in women (P 0.001).

The figure, without any recoloring, provides the curves for black (solid line) compared to non-black (dotted line) patients. Again it is clear that the average GFR is higher for black patients at any set creatinine. Levey comments and quantifies the racial difference:

At any given GFR, the serum creatinine concentration is significantly higher in men than in women and in black persons than in white persons (P=0.001).

Dr. Berger misses this fact:

How do we know *that* to be true? BECAUSE THEY MEASURED IT!

The refernces are just there to show that this is not a new and novel finding. This was an expected finding. The study does not rest on these references. The investigators in the MDRD study measured the serum creatinine, GFR, and asked patents if they were white, black or hispanic. The data shows that black people had, on average, 18% higher GFR for any measured creatinine. The fact that the prior work on this subject was deplorable does not alter the findings.

Berger is so upset that the estimated GFR differentiates black and white people that he misses the real problem with the MDRD study, the embarrassing lack of black people in the original data set. Only 12% of that cohort was African American, less than 200 people. A group that has the greatest incidence of end-stage kidney disease should be over-represented in a study about reducing the progression of CKD, not under-represented. Remember, Levey was using the data already collected for the Modification of Diet on Renal Disease study. This was not de novo data collected for the purpose of generating this equation. This weakness was corrected in the CKD-Epi equation where there were nearly 3,000 African Americans representing 30% of the cohort. The adjustment for race went from an 18% bump in GFR for a given creatinine down to 15.9%. Not much difference.

We use race, gender, and age not because we are racists, sexists, and agists, but rather because there are physiologic differnces between the races, the genders, and the aged. We exploit those differences to improve the accuracy of our estimate. All of these adjustment are just attempts to use demographic variables to squeeze a better correlation of GFR from a serum creatinine.

Posted on

eGFR: the problem of false positives

Last week I saw a 58 year old African American woman who was referred to me for an eGFR (i.e. MDRD equation) of 58. Her insurance company notified her primary care doctor about this decreased GFR. I saw the letter that Blue Cross sent and it did not give the physician any guidance on what to do with this information. The insurance company just wanted to make sure the physician was aware that the patient was flagged as having CKD. The primary care doctor sent her to me for further evaluation.

The patient, however, was completely freaked out. She went on the internet and started to learn about kidney disease and to her horror found (correctly) that her lisinopril and simvastatin could cause kidney disease. Since both of these medications had been started in the last few years she suspected (wrongly) that they were the cause of her kidney disease and stoppd both of them.

Her GFRs for three years before referral had been: 65, 63, 65 and 58. When I repeated her GFR it was 62.

This is a classic case of what Dr. Harold Feldman was writing about in the Feburary CJASN (PDF). Here is a patient who stopped the two most important drugs for her future health (statin, ACEi) because of a false positive eGFR.

This article uses a Markov chain Monte Carlo method to simulate use of serum Cr or serum Cr plus eGFR for CKD screening. The model they used is illustrated below:

In the model patients gets screened once a year (a cycle) from age 60 to 78. There are 6 states patients must be assigned to:

  1. No kidney disease (CKD stage 0)
  2. No kidney disease but false positive screening test (my patient)
  3. CKD, diagnosed
  4. CKD undiagnosed (false negative screening test)
  5. ESRD
  6. Death

In each cycle every patient must be assigned to a state. Dead patients must remain dead, ESRD patients can remain ESRD or die. Patients must develop CKD (state 3 or 4) at least one cycle prior to progressing to ESRD. Patients in any living state can die. Patients with CKD (state 3 or 4) can not tranition to no CKD (state 1 or 2). And according to the text but not the figure, patients without kidney disease but false positive screening (state 2) would go back to state 1 for the next cycle.

Some assumptions in the calculation:

  • Incidence of CKD 0.7% until age 65 then 2.3%
  • Mortality without CKD 0.97% from age 60 to 70, then 2.4% after age 70
  • Mortality with CKD 0.050% (why this would be half the rate of non-CKD makes no sense)
  • Mortality with ESRD age 62-67: 15%; 67-75: 19%; and 75+: 26%
  • Annual rate of progressing from CKD to ESRD 0.076%
  • Treatment of CKD had no effect on mortality
  • Treatment of CKD reduced the annual rate of progression to ESRD by 21% (from 0.076% to 0.055%)
The Baysean test characteristics for eGFR and sCr:
  • Sensitivity of eGFR: 0.924
  • Specificity of eGFR: 0.835
  • Sensitivity of serum Cr: 0.559
  • Specificity of serum Cr: 0.950

The model used the following evaluation of CKD

  • Two clinic visits with a nephrologist
  • Limited renal ultrasound
  • Renal function panel
  • U/A, urine protein, urine creatinine

The costs for the different states are outlined in the table below:


In the initial analysis eGFR was more accurate and more cost-effective than the serum Cr. Use of eGFR kept patients off dialysis (29 patients) and reduced deaths (13 patients) at the expence of an ocean of false positives:


But when you assigned a false positive CKD a slightly lower quality of life than a true negative, 0.98 versus 1.0, the serum creatinine came out more cost effective per quality adjusted life year (QALY).

Summary: the better test (as measured by the area under the curve of a receiver operator characteristics curve) loses to the worse test because of the decreased quality of life that results from a false positive reading. The false positives were so much more prevelant that they overwelmed the benefit from the decrease in death and ESRD found with the more accurate eGFR test.

This study has received tremendous publicity, likely because noone had looked at eGFR in this way before and it had a contrarian view. While the whole nephrology community has been pushing for routine eGFR reporting along with creatinine, Feldman comes and publishes a scathing indictment. Additionally, it makes good copy to say that nephrologists developed a new way to measure renal function that dramatically increases the demand for nephrology services.

My primary concearn with this study is two fold:

  1. The eGFR equation is used as a screening test in this study and in real- life. Screening tests need to be as sensitive as possible even at the expense of specificity. The thought is that the increased false positives will be picked up with secondary testing. But a screening test never wants to give patients a clean bill of health when in actuality they have smoldering unrecognized disease. In Feldman’s analysis the eGFR works perfectly as a screening test by picking up nearly all of the patients with CKD but the decreased specificity results in numerous false positives. It is interesting that it is not the cost of evaluating the false positives that results in the cost ineffectiveness of eGFR but it is the decreased quality of life that results from the anxiety associated with the initially positive diagnosis of CKD. In my mind this means we need to do a better job educating patients and providers to the nature of the eGFR test.
  2. The other problem with the study is it threatens to throw out the baby with the bath water. Even if the study does show that eGFR is not cost effective, part of the problem is that the only utility given to the eGFR is in the early diagnosis of CKD to prevent ESRD. However, I more often use the eGFR to dose adjust medications, to estimate the risk of contrast nephropathy, guide the use of loop versus thiazide diuretics. All of these uses of the eGFR cannot be replaced by a serum creatinine because the serum creatinine does not account for age, gender and race.

False positive diagnosis of CKD by the eGFR are real problems and Dr. Feldman has done the nephrology community a favor by bringing this issue to light. It would be interesting for Feldman to re-run his Monte Carlo simulation with various definitions of CKD, does an eGFR of 50 ml/min reduce the false positives enough to reduce the cost below the benefits? What about 45 ml/min (sometimes called CKD 3b)? It is important for a dialog to be initiated among primary care doctors, nephrologists and payers to come up with better definitions of CKD that don’t freak our patients unnecessarily while providing the best care we can.

Posted on

Journal Club: low protein diet

Effect of a very low protein diet on outcome: long-term follow-up.

This is the long-term follow-up of the B group from the original MDRD study.
Enrollment criteria:
  • Age: 18-70
  • Abnormal Cr 1.2-7 women 1.4-7 in men.
  • MAP of 125 or less (160/100)
  • Proteinuria less than 10g per day
  • No diabetics
GFR 13-24 mL/min for the B study (low protein versus very low protein diet). Higher GFR were enrolled in the A study (normal protein versus low protein diet).
Protein was restricted for 3 years.
9 months after the study every nutritional parameter was the same between the two groups.

The primary end-point was a composite of death or dialysis and just about every patient in both groups (95.7%) reached this end-point preventing a separation between the groups (p=0.5). Likewise there was no separation with regards to time to dialysis (p=0.4).

The surprising finding occurs when they looked at death after the initiation of dialysis. There were 34 deaths in the very low-protein group and 19 deaths in the low-protein group (p=0.01).

The separation begins around 15 months and grows over time. This difference was statistically significant and grew to a 2-fold increased risk of death after 6 years.

My take is this fits well with what I tell my patients when they ask me about protein restriction. I have always counseled patients against protein restriction. The two largest RCT were both negative trials (The Modification of Diet in Renal Disease and the Northern Italian Cooperative Study Group). Additionally my patients do not have the benefit of dedicated and repeated nutritional couseling that the patients in these trials receive. My fear is that with little therapeutic upside there is signifigent risk of malnutrition from overzealous protein restriction.

This study probably does not apply to my worry as I doubt patients would adhere to a very low-protein diet.

My other concearn regarding low-protein diets is patients need to get calories from somewhere. Calories can only come from protein, carbohydrates or fat. Considering that the vast majority of CKD patients are destined to die before dialysis I worry that my advice for protein restriction will result in increased carbohydrates (bad for diabetes and possibly CV disease, see Richard Johnson’s fructose hypertension research) and/or increased fats (bad for CV disease) and enhance the risk of death from the more likely outcome.

Posted on

Over collection or just a big guy


A patient came to my office with a creatinine of 2.2 indicating a GFR of 33mL/min by the MDRD formula. 

His primary care doctor ordered a 24 hour urine for creatinine and protein as part of her work-up for CKD:
  • 24-hour urine creatinine was 3,232 mg 
  • 24-hour urine protein was below the level of detection (<183>
To calculate the CrCl multiply the urine cr (total mass, not the concentration) by 100 then divide the product by 1440 (the number of minutes in 24-hours) and then by the serum creatinine (in mg/dl).
  • His CrCl is 102 mL/min
This is a huge discrepancy: 
  • Advanced Stage 3b CKD by MDRD
  • Normal kidney function by 24-hour urine collection
The first thing you should do is determine if the 24-hour urine was an adequate sample. Usually I worry about under-collections of urine due to a missed void or spillage. In this case I worried that an over-collection was masking renal failure.  (i.e. Did he collect his urine for more than 24-hours? Did his wife join in and contribute to the collection?) The average man produces 23 mg/kg of creatinine. The average woman produces 18 mg/kg. I am unaware of the proper figures for children.
His body weight is 123 kg and the 24-hour creatinine collection was 3,232 mg. This yields 23 mg/kg, right on the money for an average adult male.
This is just a big guy and this is where the MDRD can fail us.
Supporting the diagnosis of CKD stage zero was a normal renal ultrasound, a lock of proteinuria and a normal U/A and microscopic exam.